Intensive case management for severe mental illness

Summary of findings for the main comparison. Intensive Case Management versus standard care for severe mental illness

Intensive Case Management versus standard care for severe mental illness
Patient or population: people with severe mental illness Settings: community Intervention: Intensive Case Management versus standard care
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control	Intensive Case Managementversus standard care
Service use: 1. Average number of days in hospital per month ‐ by about 24 months	‐	The mean service use: 1. average number of days in hospital per month ‐ by about 24 months in the intervention groups was 0.86 lower (1.37 lower to 0.34 lower)	‐	3595 (24 studies)	⊕⊕⊝⊝ low^1,2
Adverse event: 1b. Death ‐ suicide ‐ by long term	20 per 1000	13 per 1000 (6 to 30)	RR 0.68 (0.31 to 1.51)	1456 (9 studies)	⊕⊕⊝⊝ low^1,4
Global state: 1. Relapse ‐ by long term	‐	‐	‐	‐	‐	No data available
Global state: 1. Leaving the study early ‐ by long term	331 per 1000	225 per 1000 (192 to 262)	RR 0.68 (0.58 to 0.79)	1798 (13 studies)	⊕⊕⊝⊝ low^1,3
Social functioning: 2. Employment status (various measurements) ‐ by long term ‐ not employed at the end of the trial	766 per 1000	536 per 1000 (375 to 766)	RR 0.7 (0.49 to 1)	1129 (4 studies)	⊕⊝⊝⊝ very low^1,5
Mental state: not improved to an important extent ‐ by long term	‐	‐	‐	‐	‐	No data available
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Downgraded one step for risk of bias: randomisation not well described; problematic to blind. ²Downgraded one step for inconsistency: substantial heterogeneity (I² = 74%). ³Downgraded one step for selective reporting bias: only 13 studies reported fully on the flow of participants through the study. ⁴Downgraded one step for imprecision: the 95% CI includes both appreciable benefit and appreciable harm. ⁵Downgraded two steps for inconsistency: considerable heterogeneity (I² = 94%).

Summary of findings 2. Intensive Case Management versus non‐Intensive Case Management for severe mental illness

Intensive Case Management versus non‐Intensive Case Management for severe mental illness
Patient or population: people with severe mental illness Settings: community Intervention: Intensive Case Management versus non‐Intensive Case Management
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control	Intensive Case Management versus non‐Intensive Case Management
Service use: 1. Average number of days in hospital per month ‐ by about 24 months	‐	The mean service use: 1. average number of days in hospital per month ‐ by about 24 months in the intervention groups was 0.08 lower (0.37 lower to 0.21 higher)	‐	2220 (21 studies)	⊕⊕⊕⊝ moderate¹
Service use: 3b. Average number of admissions (skewed data ‐ sample size ≧ 200) ‐ by long term	‐	The mean service use: 3b. average number of admissions (skewed data ‐ sample size ≧ 200) ‐ by long term in the intervention groups was 0.18 lower (0.41 lower to 0.05 higher)	‐	678 (1 studies)	⊕⊕⊕⊝ moderate¹
Adverse event: 1b. Death ‐ suicide ‐ by long term	12 per 1000	11 per 1000 (3 to 35)	RR 0.88 (0.27 to 2.84)	1152 (3 studies)	⊕⊕⊝⊝ low^1,3
Global state: 1. Relapse ‐ by long term	‐	‐	‐	‐	‐	No data available
Global state: 1. Leaving the study early ‐ by long term	159 per 1000	111 per 1000 (83 to 151)	RR 0.7 (0.52 to 0.95)	1970 (7 studies)	⊕⊕⊝⊝ low^1,2
Social functioning 2. Employment status ‐ by medium term ‐ spent > 1 day employed	111 per 1000	162 per 1000 (50 to 527)	RR 1.46 (0.45 to 4.74)	73 (1 study)	⊕⊝⊝⊝ very low^1,4
Mental state: not improved to an important extent ‐ by long term	‐	‐	‐	‐	‐	No data available
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Downgraded one step for risk of bias: randomisation not well described; problematic to blind. ²Downgraded one step for selective reporting bias: only 7 studies reported fully on the flow of participants through the study. ³Downgraded one step for imprecision: the 95% CI includes both appreciable benefit and appreciable harm. ⁴Downgraded two steps for imprecision: the 95% CI includes both appreciable benefit and appreciable harm, and only 73 participants were included.

Background

Description of the condition

Worldwide, more than 25% of people develop one or more mental or behavioural disorders during their lifetime (WHO 2001). Schizophrenia is one illness that heavily contributes to the numbers of people considered severely mentally ill. The lifetime prevalence of schizophrenia is 0.58% in the adult population (Warner 1995). It is currently 26th on the list of diseases ranked according to contribution to overall burden in term of disability‐adjusted life years (DALYs). Its ranking is projected to rise to 20th by the year 2020, with more than 17 million DALYs lost (accounting for 1.25% of overall burden) (Murray 1996). However, other psychiatric/psychological conditions can also profoundly affect a person's functioning. Many people with other types of non‐organic psychotic illness, or even non‐psychotic disorders such as personality disorder, can be considered to be severely mentally ill.

There has been lack of consensus over the definition of 'severe mental illness', but the most common dimensions used to identify this group are i.) diagnosis, ii.) disability, iii.) duration, and iv.) abnormal behaviour. However, there is little consistency between dimensions and thresholds used in different settings (Slade 1997). The definition of severe mental illness with the widest consensus is that of the National Institute of Mental Health (NIMH) (Schinnar 1990). Their definition is based on three criteria: i.) diagnosis of non‐organic psychosis or personality disorder; ii.) duration characterised as involving 'prolonged illness' and 'long term treatment' and operationalised as a two‐year or longer history of mental illness or treatment; and iii.) disability, which includes dangerous or disturbing social behaviour, moderate impairment in work and non‐work activities, and mild impairment in basic needs (National Institute of Mental Health 1987).

A survey conducted in Europe to calculate prevalence rates of severe mental illness according to the NIMH definition put the total population‐based annual prevalence at approximately 2 per 1000 (Ruggeri 2000).

Description of the intervention

Since the 1960s, there has been an almost worldwide trend towards the closure of institutions for the mentally ill. Coupled with these closures, many government policies have focused on reducing the number of hospital beds for people with severe mental illness in favour of providing care in a variety of non‐hospital settings ‐ outpatient clinics, day centres, or community mental health centres. These changes were consistent with the increasing shift from hospital‐based care in favour of a more community‐focused approach (Malone 2007).

Assertive Community Treatment and Case Management (Table 1) are community‐based packages of care developed in the early 1970s. They were initially conceived to co‐ordinate the care of severely mentally ill people discharged from closing mental hospitals. However, they were soon more widely applied as a means of caring for severely mentally ill people who did not require immediate admission (Thompson 1990).

Table 1. Case Management and Assertive Community Treatment

1. Case Management (CM)

The key principle of case management is that a single person ‐ the 'case manager' ‐ takes primary responsibility for a defined group of patients in the community. The case manager is responsible for (Holloway 1991):

assessing the patient's needs;
developing a care plan;
arranging suitable care from community services;
keeping contact with the patient.

Initially, in its simplest form (referred to as 'brokerage'), case managers were not mental health professionals, did not provide any direct care, and worked independently.

2. Assertive Community Treatment (ACT)

Assertive Community Treatment should be practiced according to a defined and validated model (Stein 1980), based on the consensus of an international panel of ACT experts (McGrew 1994; McGrew 1995). A key aspect of ACT is that it is a team‐based approach, characteristically a multidisciplinary team including social workers, nurses, and psychiatrists, caring exclusively for a defined group of patients (McGrew 1995; Olfson 1990). Team members share responsibility for their clients, so it is common for several team members to work together in treating the same patient. Other characteristics of ACT are (Stein 1980):

provide all necessary care themselves, rather than arranging for it to be provided by other services;
provide care at home or in workplaces;
carry low caseloads (usually 10 to 15 patients per member);
practice 'assertive outreach', meaning that they persist in attempts to engage unco‐operative clients;
place particular emphasis on medication compliance;
provide 24‐hour emergency cover.

Core features of Assertive Community Treatment (ACT) were clearly stated since the first paradigm‐shifting study of Stein and Test (Stein 1980), and successively critical ingredients of ACT have operationally defined by developing fidelity scale (McGrew 1994; McGrew 1995).

Case Management was not likewise defined; brokerage case management was rapidly abandoned in favour of Clinical Case Management (Holloway 1995), and more sophisticated, but poorly defined models were developed. In these models case managers have clinical training, provide at least some clinical services, and operate with low caseloads (Rubin 1992; Solomon 1992).

Assertive Community Treatment and Case Management do share common goals such as maintaining contact, reducing hospitalisation (and hence cost), and improving outcome. However, there are, at least in theory and with respect to the original models, important structural distinctions between them. Nonetheless, across time through clinical practice the two interventions have evolved and tended to converge into a package of care known as Intensive Case Management, which contains elements from the two original models (Burns 2008; Scott 1995). In both clinical trials and clinical practice, what is currently called 'Case Management' is thus likely to contain some elements of ACT practice. These models can be called 'Clinical Case Management', 'Intensive Case Management', and 'Strengths Case Management' (Solomon 1992). However, 'Intensive Case Management' is a broader term often used interchangeably with Assertive Community Treatment but distinguished from it on the grounds that it often lacks one or more ACT programme elements (Burns 2001). Intensive Case Management (ICM) emphasises the importance of small caseload (usually considerably fewer than 20) and high‐intensity input. Intensive case managers are usually clinicians who act as therapist in addition to their case management duties (Marshall 2008).

Until a few years ago, the approaches to care within community mental health teams differed. These approaches (evolved over the last 30 years) fell into two main categories: i.) services with well‐delimited aims, such as crisis resolution and home treatments teams, vocational rehabilitation, and early intervention service; and ii.) services aimed at meeting a wide range of patient needs, such as ACT and Case Management (CM) (merging in the Intensive Case Management model) (Ruggeri 2008).

In the last decade such a distinction has no longer been so relevant. Intensive Case Management partly lost its purity and closeness to the original models (ACT and CM), where many services are offering a less intensive but more flexible and responsive form of assertive outreach (Drukker 2008), investing on the 'critical ingredients' of ICM research helped to identify (Burns 2007) (Killaspy 2012). Many emerging practices are developed within ICM framework, where their aim is to address specific target populations and outcome domains (Bond 2015). Specifically, many specialised models of intervention within community mental health teams are based on adaptation of key principles of ICM, where they are addressing specific population subgroups (difficult to engage in traditional settings, high‐risk and revolving door, with comorbidity) (Brewer 2015). Among these, there are packages of care for homeless populations with severe mental illness (Coldwell 2007); populations with severe mental illness and substance abuse (Pettersen 2014), substance abuse, Kirk 2013 or alcohol dependence only (Gilburt 2012), and early intervention in first‐episode psychosis (Brewer 2015). The recent proliferation of models inspired by ICM that focus on a special issue was permitted by the structure and flexibility of the original ACT model, but exploring this emerging area goes beyond the objectives of this review.

How the intervention might work

The theory behind care in the community is that it enables people to live as independently as possible within their own homes or 'homely settings' out of hospital, because unnecessary hospital care is wasteful, untherapeutic, and stigmatising. It was hoped that living in the community would increase opportunities for people with severe mental illness to achieve their full potential as autonomous members of society (Department of Health 1990). Community care policies are also aimed at promoting choice and independence for people experiencing mental health difficulties.

Intensive Case Management is an intervention at the level of local service organisation. It is a way of organising teams, rather than a specific treatment model (Johnson 2008). Intensive Case Management should provide a mental health service that is a reliable, systematic, flexible, and co‐ordinated care method, addressed to answer the unique combination of health and social care needs of people with severe mental illness. It represents a long‐term intensive approach to the patient in the community (Killaspy 2008), providing a comprehensive range of treatment, rehabilitation, and support services (Scott 1995); in the last decade ICM has absorbed the recovery principle of promoting emancipation, through policies encouraging graduation (Finnerty 2015). Intensive Case Management aims to help people with severe mental illness acquire material resources (such as food, shelter, clothing, and medical care) and to improve their psychosocial functioning; to provide sufficient support to keep the patient involved in community life and to encourage growth towards greater autonomy; to develop coping skills to meet the demands of community life; and to ensure continuity of care among treatment agencies (Stein 1980). Key purposes of ICM are to improve outcome, reduce hospitalisation, and prevent loss of contact with services.

A cornerstone in the research field was a study by Burns and colleagues exploring the mechanism for ICM to be effective (Burns 2007). It suggested that the success of ICM depends on its fidelity to the ACT model (i.e. if a team approach is properly implemented) and on the setting (i.e. it would work better where there is a high baseline level of bed use).

Why it is important to do this review

With the evolution of the original intervention models, there was a need to update and merge two previous relevant Cochrane reviews (Marshall 2000a; Marshall 2000b), and to take into account the findings of work by the same authoring team (Burns 2007). During the last 15 years, not only have intervention models been modified, merged, and become more difficult to distinguish in practice, but also research has been more widespread, with new studies evaluating these approaches outside of the USA.

Since early 2000, ICM has been a very implemented and widespread intervention in the community care setting, with many nations in Europe, North America, and Australia, investing great efforts and resources in its promotion and dissemination (in England Care Programme Approach promoting ACT team (Department of Health 1999)).

Since then, research providing long‐term follow‐up outcomes and data on the impact of ACT teams on inpatient service use in specific national settings has been published with emerging data casting doubt on the opportunity of such an initial enthusiastic approach, especially in England, one of the nations where there had been stronger investments in it (Glover 2006). This topic is therefore still under an international debate (Burns 2009; Burns 2010; Burns 2012; Killaspy 2012; Rosen 2013). Almost in the same years (since the mid‐2000s), ICM landed in Asia, where the idea of comprehensive community programmes is gradually catching on, and wide implementation of both programs has inspired programmes highly faithful to ICM (Low 2013; Nishio 2014).

The effects of the currently implemented packages of care in different settings should be fully understood across a range of outcomes.

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We considered all relevant randomised controlled trials, and economic evaluations conducted alongside included randomised controlled trials. We excluded quasi‐randomised studies, such as those allocating by alternate days of the week. Where trials were described in some way as to suggest or imply that the study was randomised and where the demographic details of each group's participants were similar, we included these trials and undertook a Sensitivity analysis.

Types of participants

We required the majority of participants to be:

within the age range of 18 to 65 years;
suffering from severe mental illness, preferably as defined by National Institute of Mental Health 1987, or in the absence of this, from illness such as schizophrenia, schizophrenia‐like disorders, bipolar disorder, depression with psychotic features or/and personality disorder; and
not acutely ill.

We did not consider substance abuse to be a severe mental disorder in its own right, however studies were eligible if they dealt with people with both diagnoses, that is those with severe mental illness plus substance abuse. Dementia and mental retardation are not considered to be severe mental disorders, hence we excluded studies focusing on these populations. We considered only participants treated in the community care setting.

Types of interventions

We considered only interventions and management packages not focused primarily on alternatives to acute hospital admission.

1. Intensive Case Management

We defined Intensive Case Management as where the majority of people received:

a. a package of care shaped on the:

Assertive Community Treatment model, being based on the Treatment in Community Living, Assertive Community Treatment (Stein 1980);
Assertive Outreach model (Witheridge 1982; Witheridge 1991) (i.e. multidisciplinary team‐based approach, practising 'assertive outreach', offering 24‐hour emergency cover, providing care themselves) (McGrew 1995); or
Case Management model (Intagliata 1982), however described as such in the trial report.

b. with a caseload up to and including 20 people.

2. Non‐Intensive Case Management

We defined non‐Intensive Case Management as where the majority of people received:

a. a package of care shaped on the:

Assertive Community Treatment model, being based on the Treatment in Community Living, Assertive Community Treatment (Stein 1980);
Assertive Outreach model (Witheridge 1982; Witheridge 1991) (i.e. multidisciplinary team‐based approach, practising 'assertive outreach', offering 24‐hour emergency cover, providing care themselves) (McGrew 1995); or
Case Management model (Intagliata 1982), however described as such in the trial report.

b. with a caseload over 20 people.

3. Standard care

We defined standard care as where the majority of people received a community or outpatient model of care not specifically shaped on either the model of Assertive Community Treatment and Case Management, and not working within a designated named package or approach to care. If data were available on the standard care caseload, we undertook a final sensitivity analysis testing how prone the primary outcomes were to change when trials comparing Intensive Case Management with standard community care only (caseload greater than 20) were included.

Types of outcome measures

We grouped outcomes by time into short term (up to and including 6 months), medium term (7 months to up to and including 12 months), and long term (over 12 months). Where available, 24 months was the preferred follow‐up point for calculating mean days per months in hospital. If more than one follow‐up point within the same period was available, we reported the latest one. During this period, participants remained allocated in their trial arm.

We grouped outcomes assessed after active intervention was stopped or after participants could choose to which arm they were transferred, by time into short‐term follow‐up (up to and including one year), medium‐term follow‐up (from one to five years), and long‐term follow‐up (over five years). We calculated follow‐up length as time since intervention stopped.

To simplify distinguishing between outcomes assessed during and after the active intervention, we entered ones explicitly reporting follow‐up (FUP) length.

Primary outcomes

1. Service use

1.1 Hospitalisation: mean number of days per month in hospital
1.2 Not remaining in contact with psychiatric services

Secondary outcomes

1. Service use

1.1 Admitted to hospital
1.2 Hospital admission rate
1.3 Use of services outside of mental health provision (i.e. emergency services)

2. Adverse effects

2.1 Death ‐ all causes and suicide

3. Global state

3.1 Leaving the study early (lost to follow‐up)
3.2 Relapse (as defined in trial)
3.3 Not improved to a clinically meaningful extent (as defined in trial)
3.4 Not improved
3.5 Average endpoint score
3.6 Average change score
3.7 Compliance with medication
3.8 Average endpoint score
3.9 Average change score

4. Social functioning

4.1 Contact with legal system (i.e. police contacts, arrests, imprisonments)
4.2 Employment status (number unemployed at end of study)
4.3 Accommodation status (number homeless or not living independently during or at the end of the study, mean days homeless and mean days in stable accommodation per month in study)
4.4 Alcohol use
4.5 Illicit drug use
4.6 Average endpoint score
4.7 Average change score

5. Mental state

5.1 General symptoms
5.1.1 Not improved to a clinically meaningful extent (as defined in trial)
5.1.2 Not improved
5.1.3 Average endpoint score
5.1.4 Average change score

5.2 Specific symptoms
5.2.1 Positive symptoms (delusions, hallucinations, disordered thinking)
5.2.1.1 Not improved to a clinically meaningful extent (as defined in trial)
5.2.1.2 Not improved
5.2.1.3 Average endpoint score
5.2.1.4 Average change score

5.2.2 Negative symptoms (poor volition, poor self care, blunted affect)
5.2.2.1 Not improved to a clinically meaningful extent (as defined in trial)
5.2.2.2 Not improved
5.2.2.3 Average endpoint score
5.2.2.4 Average change score

5.2.3 Mood depression
5.2.3.1 Not improved to a clinically meaningful extent (as defined in trial)
5.2.3.2 Not improved
5.2.3.3 Average endpoint score
5.2.3.4 Average change score

6. Behaviour

6.1 General behaviour
6.2 Not improved to a clinically meaningful extent (as defined in trial)
6.3 Not improved
6.4 Average endpoint score
6.5 Average change score
6.6 Specific behaviours (i.e. self harm; injury to others or property)

7. Quality of life

7.1 Not improved to a clinically meaningful extent (as defined in trial)
7.2 Not improved
7.3 Average endpoint score
7.4 Average change score

8. Satisfaction

8.1 Participant satisfaction
8.1.1 Not improved to a clinically meaningful extent (as defined in trial)
8.1.2 Not improved
8.1.3 Average endpoint score
8.1.4 Average change score

8.2 Carer satisfaction
8.2.1 Not improved to a clinically meaningful extent (as defined in trial)
8.2.2 Not improved
8.2.3 Average endpoint score
8.2.4 Average change score

9. Costs

9.1 Direct costs of psychiatric hospital care
9.2 Direct healthcare costs (including all medical and psychiatric care and the costs of case management, but excluding accommodation other than hospital care)
9.3 Direct costs of all care (including costs of accommodation and subtracting benefits, such as earnings, where these are known)

Summary of findings

We used the GRADE approach to interpret findings, Schünemann 2008, and GRADEpro, GRADEpro, to import data from Review Manager 5, Review Manager, to create 'Summary of findings' tables. These tables provide outcome‐specific information concerning the overall quality of evidence from each included study in the comparison, the magnitude of effect of the interventions examined, and the sum of available data on all outcomes we rated as important to patient care and decision making. We selected the following main outcomes for inclusion in the 'Summary of findings' table.

Service use: average number of days in hospital per month by about 24 months
Service use: average number of admissions (skewed data ‐ sample size ≥ 200) by long term (> 12 months)
Adverse event: death ‐ suicide by long term (> 12 months)
Global state: relapse
Global state: leaving the study early by long term (> 12 months)
Social functioning: employment status – spent less than 1 day employed ‐ by medium term (6 to 12 months)
Mental state: not improved to an important extent

Search methods for identification of studies

Electronic searches

1. Cochrane Schizophrenia Group’s Trials Register

On 10 April 2015, the Information Specialist searched the Register using the following search strategy:

(*ca?e manag* OR *cpa* OR *community treat* OR *community team* OR *community cent* OR *community care* OR *madison model* OR *outreach* OR *hostel* OR *aftercare* OR *residential* OR *housing* OR *transitional* OR *post?hospital* OR *partial hospitali?ation* OR *foster* OR *guardianship* OR *daily living program* OR *crisis interven* OR *early interven* OR *ambulatory* OR *community liv* OR *social support* OR *patient care team* OR *community mental health* OR *patient participation* OR *drop?in* OR *day hospital* OR *day care* OR *day treat* OR *day cent* OR *day unit* OR *intensive care* OR *intensive interven* OR *intensive treat* OR *intensive therap* OR *intensive manag* OR *intensive model* OR *intensive program* OR *intensive team* OR *intensive service* OR *mobile care* OR *mobile interven* OR *mobile treat* OR *mobile therap* OR *mobile manag* OR *mobile model* OR *mobile program* OR *mobile team* OR *mobile service* OR *community interven* OR *community therap* OR *community manag* OR *community model* OR *community program* OR *community service* OR *community base* OR *home care* OR *home interven* OR *home treat* OR *home therap* OR *home manag* OR *home model* OR *home program* OR *home team* OR *home service* OR *home base* OR *broker* OR *care program*) in Title, Abstract, and Index Terms of REFERENCE OR (*ca?e manag* OR *community* OR *outreach* OR *hostel* OR *aftercare* OR *residential* OR *hous* OR *transitional* OR *foster* OR *crisis interven* OR *early interven* OR *ambul* OR *social support* OR *drop‐in* OR *day * OR *(intensive)* OR *(home)* OR *care program*) in Intervention of STUDY

The Cochrane Schizophrenia Group’s Register of Trials is compiled by systematic searches of major resources (including MEDLINE, Embase, AMED, BIOSIS, CINAHL, PsycINFO, PubMed, and registries of clinical trials) and their monthly updates, handsearches, grey literature, and conference proceedings (see Group Module). There is no language, date, document type, or publication status limitations for inclusion of records into the register.

For search methods of previous versions of this review, please see Appendix 1.

Searching other resources

References

Should an included or excluded study suggest that another study was relevant, we identified the reference and acquired the full text.

Personal contact

We contacted authors of trials for additional data where required. We did not systematically contact all authors for additional papers.

Data collection and analysis

Methods used for this version are presented below; previous methods are presented in Appendix 2.

Selection of studies

Two review authors (HB, MK) inspected results of the update search and identified potentially relevant reports. Disagreements were resolved by discussion, or where there was still doubt, we aquired the full article for further inspection. We obtained the full articles of relevant reports for reassessment and inspected them carefully to decide on inclusion or exclusion (see Criteria for considering studies for this review). Review authors were not blinded to the names of the authors, institutions, or journal of publication. Where difficulties or disputes arose, we discussed; if we were unable decide, we added these studies to those awaiting assessment and contacted the authors of the papers for clarification.

Data extraction and management

1. Extraction

1.1 Data extraction for criteria and outcomes

Three review authors (MD, HB, MK) independently extracted data from the included studies and compared results of the data extraction. We would have discussed any disagreements, documented our decisions, and contacted the authors of studies for clarification where necessary. Whenever possible, we would have extracted data presented only in graphs and figures and included the data if two review authors independently reached the same result. In order to obtain any missing information or for clarification, we attempted to contact authors through an open‐ended request. Where possible, we would have extracted data relevant to each component centre of multicentre studies separately.

1.2 Additional data extraction

1.2.1 Fidelity

We rated fidelity of the ICM intervention to ACT on the 'team membership' and 'team structure and organisation' subscales of the Index of Fidelity to Assertive Community Treatment (IFACT) (McGrew 1994). This index was derived from a survey of 20 clinical experts in ACT and validated in a survey of 18 programmes.

a. The 'team membership' subscale comprises four items:

ratio of patients to staff;
total size of team;
extent of psychiatric input;
extent of nursing input to the team.

b. The 'structure and organisation' subscale comprises seven items, whether the team is:

the primary source of care for its patients;
situated away from the hospital;
meets daily;
shares responsibility for caseloads;
is available 24 hours a day;
has a team leader who is also a case manager;
offers unlimited time for its services.

We chose IFACT because the subscales are brief and can be completed from published or unpublished text. For each item on the index, a score of 1 indicates high fidelity to the model. Score ranges from 0 to 11, where the maximum score available on the 'team membership' subscale is 4 and on the 'structure and organisation' subscale is 7, with higher scores indicating higher fidelity to the model.

We obtained fidelity data from published and unpublished trial reports, direct contact with trialists, and data previously obtained directly from trialists reported by previous reviews (Burns 2001; Burns 2007; Catty 2002). Two raters (MD and CBI) independently combined these data into a single fidelity score. Multicentre trials of ICM often struggle to implement a uniform approach, with centres operating at differing degrees of fidelity. Where possible, we rated each component centre separately.

1.2.2 Baseline hospital use

We extracted data relating to the average number of days per month in hospital for all participants in the two years before the study began. We obtained this data from published and unpublished trial reports and from direct contact with trialists.

1.2.3 Service use: hospitalisation

We obtained the primary outcome mean number of days per month in hospital for the included studies from published and unpublished trial reports, direct contact with trialists, and data previously obtained directly from trialists reported by a previous review (Burns 2007).

2. Management

2.1 Forms

Two review authors (HB, MK) extracted data onto simple, standard forms.

2.2 Data from multicentre trials

For the original version, where possible review authors MD and CBI verified independently calculated centre data against original trial reports.

2.3 Scale‐derived data

We included continuous data from rating scales only if:
a. the psychometric properties of the measuring instrument had been described in a peer‐reviewed journal (Marshall 2000);
b. the measuring instrument was not written or modified by one of the trialists for that particular trial; and
c. the measuring instrument was either i.) a self report or ii.) completed by an independent rater or relative.

2.3 Endpoint versus change data

Both endpoint and change data have advantages. Change data can remove a component of between‐person variability from the analysis. On the other hand, calculation of change needs two assessments (baseline and endpoint), which can be difficult in unstable and difficult‐to‐measure conditions such as schizophrenia. We decided to primarily use endpoint data, and only use change data if the former were not available. When relevant, we combined endpoint and change data in the analysis, as we aimed to use mean differences rather than standardised mean differences throughout (Higgins 2011).

2.4 Skewed data

Continuous data on clinical and social outcomes are often not normally distributed. To avoid the pitfall of applying parametric tests to non‐parametric data, we applied the following standards to relevant continuous data before inclusion.

We entered all relevant data from studies of more than 200 participants in the analysis irrespective of the following rules, because skewed data pose less of a problem in large studies. We also entered all relevant change data, as when continuous data are presented on a scale that includes a possibility of negative values (such as change data), it is difficult to tell whether data are skewed or not.

For endpoint data from studies of fewer than 200 participants, we used the following methods.

a. If a scale started from the finite number zero, we subtracted the lowest possible value from the mean, and divided this by the standard deviation (SD). If this value is lower than 1, it strongly suggests a skew, and we excluded these data. If this ratio is higher than 1 but below 2, there is suggestion of skew. We entered these data to test whether their inclusion or exclusion changed the results substantially. Finally, if the ratio was larger than 2, we included these data, because skew is less likely (Altman 1996; Higgins 2011).

b. If a scale starts from a positive value (such as the Positive and Negative Syndrome Scale (PANSS), which can have values from 30 to 210) (Kay 1986), we modified the calculation described above to take the scale starting point into account. In these cases skew is present if 2 SD > (S ‐ S min), where S is the mean score and 'S min' is the minimum score.

Exception to above rules ‐ mean number of days in hospital

We implemented one exception to the above rules in order to present more data, recognising that this is a post hoc decision, but also that the rules with regards to management of skewed data and how robust skewed data are within meta‐analysis are unclear (Higgins 2011). Where mean number of days in hospital data were skewed, and they were provided by studies of fewer than 200 participants, we entered those data into a subgroup of the overall analysis. We also presented the overall effect from all data pooled.

2.5 Common measure

To facilitate comparison between trials, we converted variables that can be reported in different metrics, such as days in hospital (mean days per year, per week, or per month) to a common metric (e.g. mean days per month).

2.6. Conversion of continuous to binary

Where possible, we attempted to convert outcome measures to dichotomous data. This can be done by identifying cutoff points on rating scales and dividing participants accordingly into 'clinically improved' or 'not clinically improved'. It was generally assumed that if there had been a 50% reduction in a scale‐derived score such as the Brief Psychiatric Rating Scale (BPRS), in Overall 1962, or the PANSS (Kay 1986), this could be considered to be a clinically significant response (Leucht 2005a; Leucht 2005b). If data based on these thresholds were not available, we used the primary cutoff presented by the original authors.

2.7. Direction of graphs

Where possible, we entered data in such a way that the area to the left of the line of no effect indicated a favourable outcome for ICM.

Assessment of risk of bias in included studies

For this version of the review, two review authors (HB, MK) assessed risk of bias of all new included studies using the tool described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

This set of criteria is based on evidence of associations between overestimate of effect and high risk of bias of the article such as sequence generation, allocation concealment, blinding, incomplete outcome data, and selective reporting.

If the raters had disagreed, we planned to make the final rating by consensus. Where inadequate details of randomisation and other characteristics of trials were provided, we contacted the authors of the studies to obtain further information. We would have reported non‐concurrence in quality assessment, and if disputes had arisen as to which category a trial was to be allocated, again, we would have resolved this by discussion.

We noted the level of risk of bias in Risk of bias in included studies, summary of findings Table for the main comparison, summary of findings Table 2, and Figure 1.

Figure 1

Methodological quality summary: review authors' judgements about each methodological quality item for each included study.

Measures of treatment effect

1. Binary data

For binary outcomes, we calculated a standard estimation of the random‐effects risk ratio (RR) and its 95% confidence interval. It has been shown that RR is more intuitive than odds ratios (OR), and that clinicians tend to interpret ORs as RR (Boissel 1999; Deeks 2000). Within the 'Summary of findings' table, we aimed to calculate the lowest control risk applied to all data. We assumed the same for the highest‐risk groups. We used the 'Summary of findings' table to calculate absolute risk reduction for primary outcomes.

2. Continuous data

For continuous outcomes, we estimated the mean difference between groups. We preferred not to calculate effect size measures (standardised mean difference). However, if in future versions of this review scales of very considerable similarity are used, we will presume there is a small difference in measurement and will calculate effect size and transform the effect back to the units of one or more of the specific instruments.

Unit of analysis issues

1. Cluster trials

Studies increasingly employ 'cluster randomisation' (such as randomisation by clinician or practice), but analysis and pooling of clustered data pose problems. Firstly, authors often fail to account for intraclass correlation in clustered studies, leading to a 'unit of analysis' error (Divine 1992), whereby P values are spuriously low, confidence intervals unduly narrow, and statistical significance overestimated. This causes type I errors (Bland 1997; Gulliford 1999).

Where clustering is not accounted for in primary studies, we would present data in a table, with a (*) symbol to indicate the presence of a probable unit of analysis error. If we find such studies in subsequent versions of this review, we will attempt to contact first authors of studies to obtain intraclass correlation coefficients for their clustered data and to adjust for this by using accepted methods (Gulliford 1999). Where clustering has been incorporated into the analysis of primary studies, we would present these data as if from a non‐cluster randomised study, but adjust for the clustering effect.

We have sought statistical advice and have been advised that the binary data as presented in a report should be divided by a 'design effect'. This is calculated using the mean number of participants per cluster (m) and the intraclass correlation coefficient (ICC) [Design effect = 1 + (m ‐ 1)*ICC] (Donner 2002). If the ICC is not reported, we would assume it to be 0.1 (Ukoumunne 1999).

If cluster studies have been appropriately analysed taking into account intraclass correlation coefficients, and relevant data documented in the report, synthesis with other studies would be possible using the generic inverse variance technique.

2. Cross‐over trials

A major concern of cross‐over trials is the carry‐over effect, which occurs if an effect (e.g. pharmacological, physiological, or psychological) of the treatment in the first phase is carried over to the second phase. As a consequence, on entry to the second phase the participants can differ systematically from their initial state despite a wash‐out phase. For the same reason, cross‐over trials are not appropriate if the condition of interest is unstable (Elbourne 2002). As both effects are very likely in severe mental illness, we only used data from the first phase of cross‐over studies.

3. Studies with multiple treatment groups

Where a study involved more than two treatment arms, we presented the additional treatment arms in comparisons if relevant. Where the additional treatment arms were not relevant, we did not reproduce these data.

Dealing with missing data

1. Overall loss of credibility

At some degree of loss of follow‐up data loses credibility (Xia 2009). For any particular outcome, should more than 50% of data be unaccounted for, we did not reproduce these data or use them within analyses. If, however, more than 50% of participants in one arm of a study were lost, but the total loss was less than 50%, we marked such data with (*) to indicate that such a result may well be prone to bias.

2. Binary

Where attrition for a binary outcome was between 0 and 50%, and where these data were not clearly described, we presented data on a 'once‐randomised‐always‐analyse' basis (an intention‐to‐treat analysis). We assumed all those leaving the study early to have the same rates of negative outcome as those who completed, with the exception of the outcome of death. We undertook a Sensitivity analysis testing how prone the primary outcomes were to change when we compared data from only those who completed the study with intention‐to‐treat data using the assumption outlined above.

Where number of deaths was more than 10% of the sample overall, we applied the above statement but did not impute attrition due to death.

3. Continuous

3.1 Attrition

Where attrition for a continuous outcome was between 0 and 50%, and data from only those who completed the study were reported, we reproduced these.

3.2 Standard deviations

3.2.1 General

Where there were missing measures of variance for continuous data, but exact standard errors or confidence intervals for group means, or either ‘P’ or 't' values for differences in means, we calculated standard deviation value according to the method described in Section 7.7.3 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). If standard deviations were not reported and could not be calculated from the available data, we asked authors to supply the data. In the absence of data from authors, we used the mean standard deviation from other studies.

3.2.2 Standard deviation mean number of days per month in hospital

For the primary outcome mean number of days per month in hospital, if standard deviations were not reported and could not be calculated from available data, we asked the authors for additional information. In the absence of data from authors, we imputed the missing standard deviations using a regression analysis of SD against mean from those trials that provided both. We documented in what studies we imputed SDs according to the above technique in Table 2.

Table 2. Average number of days in hospital per month ‐ at about 24 months ‐ entering meta‐regression

Intensive Case Management versus standard care	ICM	ICM	ICM	SC	SC	SC	Note
Study ID	Mean	SD	Total	Mean	SD	Total	Note
Audini‐UK 1994	0.95	2.84*	33	0.93	2.03*	33	*SD imputed
Bjorkman‐Sweden 2002	0.83	3.13	33	2.15	4.13	44
Bond‐Chicago1 1990	3.22	4.55	42	5.3	5.42	40
Bond‐Indiana1 (A)	1.28	3.17*	29	7.72	8.99*	32	*SD imputed
Bond‐Indiana1 (B)	2.72	4.54*	34	3.62	5.24*	30	*SD imputed
Bond‐Indiana1 (C)	0.05	1.89*	21	3.38	4.98*	21	*SD imputed
Chandler‐California1 (A)	0.47	2.34*	102	0.78	1.84*	101	*SD imputed
Chandler‐California1 (B)	0.67	2.55*	115	0.96	2.07*	114	*SD imputed
Curtis‐New York 1992	1.77	1.79	146	1.02	1.18	143
Ford‐UK 1995	3.07	6.9	39	1.76	3.67	38
Hampton‐Illinois (A)	1.75	3.63*	48	4.83	6.49*	47	*SD imputed
Hampton‐Illinois (B)	3.25	5.01*	34	3.42	5.02*	36	*SD imputed
Holloway‐UK 1996	2.4	5.1	34	1.2	3	26
Jerrell‐SCarolina1 1991	0.53	2.40*	40	0.8	1.86*	40	*SD imputed
Lehman‐Maryland1 1994	3.04	5.15	77	5.41	7	75
Marshall‐UK 1995	1.04	2.18	40	1.56	4.45	40
Muijen‐UK2 1994	2.53	5.55	41	2.45	5.83	41
Muller‐Clemm‐Canada 1996	1.68	3.56*	61	1.63	2.93*	57	*SD imputed
OPUS‐Denmark 1999	5.11	7.7	263	6.57	8.73	244
Quinlivan‐California 1995	1.09	2.65	30	5.53	8.65	30
Rosenheck‐USA‐GMS (A)	3.63	3.89	44	3.71	2.76	35
Rosenheck‐USA‐GMS (B)	6.99	4.85	47	4.23	5.18	47
Rosenheck‐USA‐NP (C)	18.52	11.16	50	19.16	12.19	43
Rosenheck‐USA‐GMS (D)	2.8	3.31	49	3.26	3.98	53
Rosenheck‐USA‐NP (E)	4.13	5.24	34	3.05	4.61	33
Rosenheck‐USA‐GMS (F)	2.39	3.16	43	2.58	2.45	35
Rosenheck‐USA‐NP (G)	7.68	7.72	40	12.2	10.65	31
Rosenheck‐USA‐NP (H)	4.63	8.58	59	11.21	13.38	55
Rosenheck‐USA‐GMS (I)	5.62	4.67	44	7.8	6.63	44
Sytema‐Netherlands 1999	3.4	5.4	58	4.3	7.3	57
Test‐Wisconsin 1985	0.42	2.29*	72	2.13	3.54*	41	*SD imputed
Intensive Case Management versus non‐Intensive Case Management	ICM	ICM	ICM	Non‐ICM	Non‐ICM	Non‐ICM	Note
Study ID	Mean	SD	Total	Mean	SD	Total	Note
Bush‐Georgia 1990	1.58	3.46*	14	2.39	3.85*	14	*SD imputed
Drake‐NHamp (A)	0.5	0.94	7	2.17	3.21	9
Drake‐NHamp (B)	0.85	1.43	16	1.41	2.06	14
Drake‐NHamp (C)	2.28	3.2	10	1.67	3.84	12
Drake‐NHamp (D)	1.04	2.44	13	0.63	0.91	11
Drake‐NHamp (E)	1.08	4.15	30	1.39	2.36	27
Drake‐NHamp (F)	1.66	4.49	10	0.84	2.33	13
Drake‐NHamp 1998 G	2.05	3.06	9	0.87	0.92	8
Essock‐Connecticut1 1995	2.87	7.82	130	4.3	9.52	132
Essock‐Connecticut2 2006	0.64	1.9	99	0.72	1.3	99
Harrison‐Read‐UK 2000	2.94	5.74	97	3.76	5.83	96
Johnston‐Australia 1998	4.0	5.75	35	3.08	4.3	33
McDonel‐Indiana (A)	3.15	7.1	61	1.43	2.91	64
McDonel‐Indiana (B)	1.22	3.66	14	0.58	1.29	17
Quinlivan‐California 1995	1.09	2.65	30	2.8	4.74	30
REACT‐UK 2002	9.0	8.9	124	8.0	7.8	119
Salkever‐SCarolina 1999	1.12	3.01*	91	1.3	2.51*	53	*SD imputed
UK700‐UK (A)	3.08	5.77	94	2.64	3.49	95
UK700‐UK (B)	3.2	4.79	77	3.16	4.97	73
UK700‐UK (C)	3.29	5.41	76	2.48	4.71	75
UK700‐UK (D)	2.74	4.69	91	3.79	5.22	98

ICM: Intensive Case Management
SC: standard care
SD: standard deviation
Study ID: Study identification name

3.3 Last observation carried forward

We anticipated that in some studies the method of last observation carried forward (LOCF) would be employed within the study report. As with all methods of imputation to deal with missing data, LOCF introduces uncertainty about the reliability of the results. Therefore, where LOCF data had been used in the trial, if less than 50% of the data had been assumed, we reproduced these data and indicated that they were the product of LOCF assumptions.

3.4 Incomplete data for meta‐regression

We anticipated that in some cases not all IFACT score variables would be available. If we could not calculate IFACT score from the available data, we imputed it by multiple imputation using the Multiple Imputation with Diagnostics (mi) library in R (R 2008). As explained above, we only made these assumptions if we were able to directly rate over 50% of the data. We documented in what studies we calculated IFACT score according to the above technique in Table 3.

Table 3. Covariates entering meta‐regression

Intensive Case Management versus standard care	Baseline hospital use	Baseline hospital use	IFACT	IFACT	IFACT	Note
Study ID	Mean	Total	Total score	Organisation subscale score	Staff subscale score	Note
Audini‐UK 1994	1.08	66	6.7	3.5	3.2
Bjorkman‐Sweden 2002	5.63	77	7	4.5	2.5
Bond‐Chicago1 1990	7.83	88	6	4	2
Bond‐Indiana1 (A)	14.17	61	9.2	7	2.2
Bond‐Indiana1 (B)	4.95	64	2.2	1	1.2
Bond‐Indiana1 (C)	10.86	42	7.4	5	2.4
Chandler‐California1 (A)	0.5	203	8.5	5	3.5
Chandler‐California1 (B)	1.14	229	6.6	5	1.6
Curtis‐New York 1992	0.95*	289	5.8	3.5	2.3	*Mean imputed
Ford‐UK 1995	2.61	77	4.8	2	2.8
Hampton‐Illinois (A)	5.6	95	6	4	2
Hampton‐Illinois (B)	5.2	70	5	3	2
Holloway‐UK 1996	7.37	70	9.3	6	3.3
Jerrell‐SCarolina1 1991	2.85	80	8.8	5.5	3.3
Lehman‐Maryland1 1994	4.94*	152	11	7	4	*Mean imputed
Marshall‐UK 1995	3.31*	80	4.9	4	0.9	*Mean imputed
Muijen‐UK2 1994	8.43*	82	5.4	3	2.4	*Mean imputed
Muller‐Clemm‐Canada 1996	4.07	123	6.2	4	2.2
OPUS‐Denmark 1999	NA	547	8	4	4	*Baseline hospital use: not applicable as first episode
Quinlivan‐California 1995	4.50*	60	6.4	4	2.4	*Mean imputed
Rosenheck‐USA‐GMS (A)	3.96	79	6	2	4
Rosenheck‐USA‐GMS (B)	5.83	94	3.8	2	1.8
Rosenheck‐USA‐NP (C)	19.8	93	7.7	5	2.7
Rosenheck‐USA‐GMS (D)	4.19	102	7	3	4
Rosenheck‐USA‐NP (E)	5.33	67	6.4	3.5	2.9
Rosenheck‐USA‐GMS (F)	3.22	78	6.6	3	3.6
Rosenheck‐USA‐NP (G)	11.42	71	8.4	5	3.4
Rosenheck‐USA‐NP (H)	11.4	114	6.4	4	2.4
Rosenheck‐USA‐GMS (I)	8.28	88	5.8	2	3.8
Sytema‐Netherlands 1999	12.17*	118	7.6*	5.1*	2.5*	*Mean and IFACT score imputed
Test‐Wisconsin 1985	2.33	122	8.5	5.5	3
Intensive Case Management versus non‐Intensive Case Management	Baseline hospital use	Baseline hospital use	IFACT	IFACT	IFACT	Note
Study ID	Mean	Total	Total score	Organisation subscale score	Staff subscale score	Note
Bush‐Georgia 1990	3.99	28	3.1	2	1.1
Drake‐NHamp (A)	2.88	19	8	5	3
Drake‐NHamp (B)	1.72	33	3.8	3	0.8
Drake‐NHamp (C)	3.02	25	8.8	5.5	3.3
Drake‐NHamp (D)	1.78	26	7.8	4.5	3.3
Drake‐NHamp (E)	2.76	66	8.5	4.5	4
Drake‐NHamp (F)	2.34	22	3.5	3	0.5
Drake‐NHamp 1998 (G)	4.1	19	5	2	3
Essock‐Connecticut1 1995	2.81*	262	8.5	4.5	4	*Mean imputed
Essock‐Connecticut2 2006	1.08*	198	10*	7*	3*	*Mean and IFACT score imputed
Harrison‐Read‐UK 2000	4.11	193	7.6	4	3.6
Johnston‐Australia 1998	3.66	71	7.3	3.5	3.8
McDonel‐Indiana (A)	4.2	152	4.2	3	1.2
McDonel‐Indiana (B)	1.16	39	4.4	3	1.4
Quinlivan‐California 1995	2.96*	60	6.4	4	2.4	*Mean imputed
REACT‐UK 2002	7.3	251	10.3	6.5	3.8
Salkever‐SCarolina 1999	3.06	144	7	5	2
UK700‐UK (A)	4.55	196	8.8	5	3.8
UK700‐UK (B)	4.66	153	4.5	3	1.5
UK700‐UK (C)	4.33	158	4.2	2	2.2
UK700‐UK (D)	4.59	200	8.5	5	3.5

Baseline hospital use: average number of days per month in hospital for all participants in the two years before the study began
IFACT: Index of Fidelity to Assertive Community Treatment
NA: not applicable
Study ID: Study identification name

We anticipated that in some cases not all baseline hospital use data would be available. We imputed missing data as for the IFACT scores. As explained above, we only made these assumptions if we were able to directly rate over 50% of the data. We documented for which studies we calculated baseline hospital use data according to the above technique (Table 3).

We undertook a Sensitivity analysis testing how prone the results from meta‐regression were to change when we compared data from only those who completed the studies with the imputed data using the assumption outlined above.

Assessment of heterogeneity

1. Clinical heterogeneity

We considered all included studies initially, without seeing comparison data, to judge clinical heterogeneity. We simply inspected all studies for clearly outlying situations or people that we had not predicted would arise. When such situations or participant groups arose, we discussed these fully.

In addition, we specified two potential sources of heterogeneity a priori (fidelity and baseline level of hospital use) (Data extraction and management). We extracted these data as described above.

2. Methodological heterogeneity

We considered all included studies initially, without seeing comparison data, to judge methodological heterogeneity. We simply inspected all studies for clearly outlying methods that we had not predicted would arise. If such methodological outliers arose, we discussed these fully.

3. Statistical heterogeneity

3.1 Visual inspection

We visually inspected graphs to investigate the possibility of statistical heterogeneity.

3.2 Employing the I²statistic

We investigated heterogeneity between studies by considering the I² statistic alongside the Chi² P value. The I² provides an estimate of the percentage of inconsistency thought to be due to chance (Higgins 2003). The importance of the observed value of I² depends on i.) magnitude and direction of effects, and ii.) strength of evidence for heterogeneity (e.g. P value from Chi² test, or a confidence interval for I²). We interpreted an I² estimate greater than or equal to 50% accompanied by a statistically significant Chi² statistic, as evidence of substantial levels of heterogeneity (Section 9.5.2; Higgins 2011). When we found substantial levels of heterogeneity in the primary outcome, we explored reasons for the heterogeneity (Subgroup analysis and investigation of heterogeneity).

Assessment of reporting biases

Reporting biases arise when the dissemination of research findings is influenced by the nature and direction of results (Egger 1997). These are described in Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We are aware that funnel plots may be useful in investigating reporting biases but are of limited power to detect small‐study effects. We did not use funnel plots for outcomes where there were 10 or fewer studies, or where all studies were of similar size. In other cases, where funnel plots were possible, we sought statistical advice in their interpretation.

Data synthesis

Where possible, we employed a random‐effects model for analyses. We understand that there is no closed argument for preference for use of fixed‐effect or random‐effects models. The random‐effects method incorporates an assumption that different studies are estimating different, yet related, intervention effects. According to our hypothesis of an existing variation across studies, to be explored further in the meta‐regression analysis, despite being cautious that random‐effects methods do put added weight onto the smaller of the studies, we favoured using random‐effects model.

Subgroup analysis and investigation of heterogeneity

1. Subgroup analyses

We anticipated conducting two subgroup analyses. For the first version of the protocol for this review, we did not anticipate any subgroup analyses. On further consideration, we now realise that analysis at separate time periods could be thought of as subgroups. The second subgroup is within the primary outcome and relates to skewed and non‐skewed data. We introduced this late into the protocol, and it could be considered post hoc. However, we are also aware that our original rule for management of these data could be considered overly cautious and result in some important data not being presented (Higgins 2011).

2. Investigation of heterogeneity

2.1 Anticipated heterogeneity ‐ outcome of mean days per month in hospital

Investigation of heterogeneity formed part of the secondary objectives of the review. We hypothesised that the effect of ICM on one of our primary outcomes (mean number of days per month in hospital) differs according to fidelity of intervention to the ACT model and the baseline level of hospital use.

We examined the association of the IFACT score and the baseline number of days in hospital with the treatment effect by performing random‐effects meta‐regression analysis in R (R 2008). The script we used to perform meta‐regression analyses is reported in Appendix 3. We also carried out meta‐regression using both variables within the same model. In addition, we examined the relationship between the treatment effect and the two variables using a thin plate spline. If possible, we aimed to enter data from multicentre studies in the meta‐regression disaggregated into the component centre with outcome and fidelity data for each.

Meta‐regression was performed only if at least 10 studies per comparison were available (Higgins 2011). rWe also tested comparison type as an additional regressor in the model.

2.2 Unanticipated heterogeneity ‐ other outcomes

2.2.1 For outcomes other than the second primary outcome (not remaining in contact with psychiatric services)

We reported if inconsistency was high and undertook no exploration.

2.2.2 For outcome 'not remaining in contact with psychiatric services'

We reported if inconsistency was high. First we investigated whether data had been entered correctly. Second, if data was correct, we visually inspected the graph and successively removed studies outside of the company of the rest to see if homogeneity was restored. Should this occur with no more than 10% of the data being excluded, we presented the data. If not, these data were not pooled.

Should unanticipated clinical or methodological heterogeneity have been obvious, we simply stated hypotheses regarding these for future reviews or versions of this review. We did not anticipate undertaking analyses relating to these.

Sensitivity analysis

1. Implication of randomisation

We aimed to include trials in a sensitivity analysis if they were described in some way as to imply randomisation. For the primary outcomes, we included these studies, and if there was no substantive difference when the implied randomised studies were added to those with a better description of randomisation, then we employed all data from these studies.

2. Standard‐care caseload

If data were available, we undertook a sensitivity analysis testing how prone the primary outcomes were to change when trials comparing ICM to standard community care caseload less than or equal to 20 were compared with trials comparing ICM to standard community care caseload greater than 20. If there was a substantial difference, we reported the results and discussed them but continued to pool the data.

3. Assumptions for lost binary data

Where we needed to make assumptions regarding participants lost to follow‐up (see Dealing with missing data), we compared the findings of the primary outcomes when we used our assumption compared with completer data only. If there was a substantial difference, we reported the results and discussed them but continued to employ our assumption.

4. Assumptions for incomplete data for meta‐regression

Where we needed to make assumptions regarding missing SDs data in studies entering meta‐regression (see Dealing with missing data), we compared the findings of the meta‐regression on our primary outcome when we used our assumption compared with data taken from only those who completed the studies. We tested how prone results from meta‐regression were to change when we compared data from those who completed with imputed data using the assumption outlined above. If there was a substantial difference, we reported results and discussed them but continued to employ our assumption.

Results

Description of studies

Results of the search

We have presented the results of the latest update search below; for previous results, please see Appendix 4.

The April 2015 update search of Cochrane Schizophrenia Group's Register of Trials yielded 299 references. We selected 87 for further inspection. One hundred and twenty‐seven references (corresponding to 96 studies with 31 companion papers) were available from the 'awaiting classification' section of the previous version of the review and were all selected for further inspection. We excluded a total of 85 studies from the review. Only two trials met the inclusion criteria and were included (Chan‐Hong Kong 2000; Cusack‐North Carolina). There were 26 new companion papers to previously included studies such as Morse‐Missouri3 2005, OPUS‐Denmark 1999, REACT‐UK 2002, and UK700‐UK 1999.

We have entered 20 trials in the 'awaiting classification' section and have sought further information. We added five new studies to the ongoing studies.

Problematic trials

There were two problematic trials worth special mention.

Jerrell‐SCarolina1 1991 was a three‐arm trial, with two of the arms qualifying as Intensive Case Management (Programme Assertive Community Treatment and Intensive Broker Case Management) and one a control (standard care). As results were reported separately for each arm, it was not possible to present continuous data from two ICM arms pooled together. One option was to treat each arm as a separate 'site', effectively treating the study as two trials, but with the same control group. A second option was to include only one of the experimental arms. Although aware of excluding potentially useful data on an arbitrary basis, we decided to include only one of the arms compared to standard care, per the second option. The main reason for this was to avoid a unit of analysis error, which would have occurred in the first option. We undertook a sensitivity analysis testing how prone results were to change when this trial was not included in meta‐analysis.

Curtis‐New York 1992 was a trial comparing ICM with standard care presenting two main difficulties. The first was regarding ICM caseload size. The study reported caseload ratio as 1:35 (above the 1:20 ratio defining an ICM intervention). As we derived estimation of caseload size by dividing the number of intervention participants by the number of whole‐time equivalent clinical staff in the team (not just those formally classified as 'case managers'), we found that the actual staff:participant ratio was about 1:17. We therefore found this trial eligible for inclusion in the review. The second issue was regarding the peculiar way this trial provided the ICM intervention. Both experimental and control interventions were community‐oriented and fit fully into the review's inclusion criteria, but the ICM team was located in hospital. While undesirable, the team office being based in hospital is not unusual. In any case, the case management was provided in the community. We therefore confirmed inclusion of the study, not wanting to penalise it because it reported details that were not available for all trials. However, we found a discrepancy in the data the study provided on service use outcomes (average number of days in hospital per month, admitted to hospital). Curtis‐New York 1992 was an outlier, being the only study clearly favouring standard care over ICM. We undertook a sensitivity analysis testing how prone results were to change when this trial was not included in the meta‐analysis. Neither results for the primary outcome 'average number of days per month in hospital' nor for the outcome 'admitted to hospital' changed significantly when this study was dropped, but it did significantly affect the level of heterogeneity. We could just advance the hypothesis that the reason for heterogeneity could be the unusual way the intervention was provided in this trial (Table 4).

See: Summary of findings for the main comparison Intensive Case Management versus standard care for severe mental illness; Summary of findings 2 Intensive Case Management versus non‐Intensive Case Management for severe mental illness

Table 4. Interventions in Curtis‐New York

1. ICM: "Intensive outreach case management" from a multidisciplinary team at Harlem Hospital Center. This team implemented a discharge treatment plan and monitored clinical and social problems. The team did not "assume direct responsibility for care but [...] help[ed] the patient enrol in a day hospital programme, adult mental health clinic, rehabilitation programme, or alcohol treatment programme". Caseload: 1:17. N = 147.

2. Standard care: routine aftercare, within the discharge treatment plan prescribed for each patient from Harlem Hospital Center; "most received at least initial treatment from various divisions of the departments of psychiatry within the Health and Hospitals Corporation". N = 145.

For a summary of the trial selection from the 2015 search, please see Figure 2.

Figure 2

Study flow diagram 2015 update

Included studies

See: Characteristics of included studies.

The previous updates of two reviews included 176 reports describing 38 studies. This review now includes these 38 separate trials with an additional two studies including data on 196 participants (Chan‐Hong Kong 2000; Cusack‐North Carolina). These 40 trials provide data on 7524 randomised people. Twenty‐four of these studies had already been included in two original reviews (as included or awaiting assessment), with 14 more derived from the 2008 search and two more derived from the 2015 search. Twenty‐eight trials provided data for the ICM versus standard care comparison, 11 for the ICM versus non‐ICM comparison, and only one for both comparisons. Both of the two newly included studies provided data for the ICM versus standard care comparison. Please note that it was possible to report data for several separate centres of seven multicentre trials (Bond‐Indiana1 1988 (3); Chandler‐California1 1991 (2); Drake‐NHamp 1998 (7); Hampton‐Illinois 1992 (2); McDonel‐Indiana 1997 (2); Rosenheck‐USA 1993 (9); and UK700‐UK 1999 (4)). Several of these centres are reported separately in the Characteristics of included studies table.

1. Study length

Only one study fell into the short‐term category, with a maximum length of six months (Bond‐Indiana1 1988); nine studies reported medium‐term data, with only one study reporting data by seven months, Okpaku‐Tennessee 1997, and the remaining eight studies reporting data by 12 months (Bond‐Chicago1 1990; Bush‐Georgia 1990; Hampton‐Illinois 1992; Johnston‐Australia 1998; Lehman‐Maryland1 1994; Morse‐Missouri1 1992; Solomon‐Pennsylvania 1994; Sytema‐Netherlands 1999). The remaining 29 studies all fell into the long‐term category, with a maximum length of four years (Test‐Wisconsin 1985), and an average length of 23.5 months.

One study reported data only on short‐term follow‐up (five months of active intervention followed by six months' follow‐up) (Chan‐Hong Kong 2000), not reporting data assessed during the intervention. Newly retrieved companion papers provided data from medium‐ and long‐term follow‐up for two studies already included in the previous version of this review. Those two studies were OPUS‐Denmark 1999, now reporting data at three and eight years after active intervention was discontinued (i.e. 5 and 10 years after randomisation), and REACT‐UK 2002, now reporting data at 18 months and 8.5 years after participants could choose to which arm they were allocated (i.e. 3 and 10 years after randomisation). During the follow‐up period, all participants in OPUS‐Denmark 1999 received the control intervention, where participants in REACT‐UK 2002 could remain in the originally allocated intervention or be transferred to the control one.

2. Design

All included studies were randomised with a parallel longitudinal design. Twelve were multicentre trials, but only seven of these provided data for single centres (see above).

3. Participants

We included a total of 7524 participants from 40 trials. Most trials included from the previous review were conducted in Australia, Canada, and the USA. Specifically, most of the trials included from the previous two reviews were conducted in the USA (16 trials; 3474 participants), and only five were European (345 participants). In the first update of 2010, we added 1964 participants from seven trials conducted in Europe, and 1545 participants from 10 studies conducted in Australia, Canada, and the USA. In the current version of the review, we added 62 participants from one trial conducted in China (Chan‐Hong Kong 2000), and 134 participants from one trial conducted in the USA (Cusack‐North Carolina).

The review now provides data on 27 trials, including 5153 participants, conducted in Australia, Canada, and the USA; 12 trials, including 2309 participants, conducted in Europe; and one trial, including 62 participants, conducted in China.

Twenty studies included participants with severe mental illness. None of these studies provided operationalised definitions addressing dimensions of diagnosis, duration, and disability. However, 14 provided criteria for either the diagnosis, impairment, or level of service use. The remaining six studies provided no criteria for defining serious mental illness. Diagnoses of 'severe mental illness' varied across studies, from schizophrenic disorder alone to wider diagnostic groups including schizophrenic, affective, and personality disorder.

Of the remaining 20 studies, 18 involved participants with various diagnoses, but the great majority had some psychotic disorder, and most trials reported criteria for service use or impairment, or both. Two studies included participants with a high level of impairment or service use due to psychiatric illness, but provided no diagnostic criteria for inclusion (Harrison‐Read‐UK 2000; Okpaku‐Tennessee 1997).

Most trials (23) involved participants that had been diagnosed using operationalised criteria (DSM, ICD, OPCRIT, RDC, SADS, see Characteristics of included studies footnotes), whilst 17 (10 in the group including participants with serious mental illness and six in the group including participants with various diagnoses) did not report using any diagnostic tool, but only stated type of illness or level of impairment. Only OPUS‐Denmark 1999 included participants with a first episode of psychotic illness.

Four studies included a total of 742 dually diagnosed participants (Drake‐NHamp 1998; Essock‐Connecticut2 2006; Morse‐Missouri3 2005; Muller‐Clemm‐Canada 1996), and eight studies included a total of 1337 homeless participants.

Information on mean age was available from 32 trials (6473 participants). The average age was about 38 years old. Only Macias‐Utah 1994 did not report information on participant age.

4. Settings

As stated in the inclusion criteria, all of the included studies took place in a community setting, provided both by private and public mental health services. No study was carried out in a low‐income country, as the included studies were conducted in Australia, Canada, the USA, Europe, and China.

5. Interventions

Twenty‐nine trials were included in the comparison ICM versus standard care, and 12 in the ICM versus non‐ICM comparison. Quinlivan‐California 1995 was a three‐arm trial (ICM, non‐ICM, and standard care) and provided data for both comparisons. We considered REACT‐UK 2002 to be an ICM versus non‐ICM comparison due to our assumption that standard care could be considered to be Care Programme Approach, even if not clearly reported by trial authors. The Care Programme Approach was introduced in England in the mid‐1990s and become standard care thereafter; it is a combination of non‐Intensive Case Management and care from a Community Mental Health Team (Department of Health 1999), hence to be considered as non‐ICM according to the definitions in this review.

5.1 Intensive Case Management

On average, the ICM included in this review was well defined. The majority of experimental interventions were explicitly modelled on the ACT model, being based on the Treatment in Community Living model of Stein 1980. Only a few studies based ICM on the Case Management model. The experimental intervention was provided either by an already existing team or ICM services newly established for the trial. Cusack‐North Carolina was the only trial that included ICM as forensic adaptation of ACT compared with standard care.

5.2 Non‐Intensive Case Management

There were no discernable differences between the practice of non‐ICM and ICM except for the intensity of contact. The degree of training and skill of the staff was similar in the ICM and non‐ICM teams. In some studies, non‐ICM was itself an experimental intervention, but it mostly represented the average standard care, as what in this review we call 'standard care' has increasingly shifted towards non‐ICM across decades. Mental health systems increasingly included elements from ICM, melding them with community mental health service. English mental health policy is one example, where the Care Programme Approach was introduced in the UK in the mid‐1990s and become standard care thereafter. It is a combination of non‐ICM and care from a Community Mental Health Team, hence to be considered as non‐ICM, according to the definitions in this review. Therefore the REACT‐UK 2002 study was considered in the ICM versus non‐ICM comparison due to our assumption that standard care could be considered as Care Programme Approach, even if not clearly reported by trial authors.

5.3 Standard care

On average, the definition of standard care was blurred, as this intervention was modelled on a generalist model. Its core was being provided by a community mental health service, but its features were variable across trials run in different countries at different time periods. Presence of further specialised services, such as rehabilitation or psychotherapist services, were variable within standard care services. In a few studies, both ICM and standard care incorporated services for substance abuse treatment and homelessness care.

6. Outcome measures

6.1 Overall

The outcomes for which we could obtain usable data were: service use, adverse effects, global state, social functioning, mental state ‐ general and specific symptoms, behaviour, quality of life, satisfaction, and costs.

Many trials used different scales in assessing treatment effects in various outcomes (global state, social functioning, mental state ‐ general and specific symptoms, behaviour, quality of life, and satisfaction). As most of the scales were used by only one study in each comparison group, it was not possible to enter these data in a unique analysis. Even where studies used the same scale, they often applied different rating scores. Therefore, again, data could not be entered together in the analysis. Some studies failed to clearly report the rating score they used for a pre‐stated scale. We noted this in the 'Risk of bias' tables (see 'Outcomes' in Characteristics of included studies table). No studies assessed improvement by measuring it on scales. We did not calculate effect size measures (standardised mean difference, see Measures of treatment effect).

6.2 Outcome scales

Only details of the scales that provided usable data are shown below. Reasons for exclusions of data are provided under 'Outcomes' in Characteristics of included studies table.

6.2.1. Global Assessment Scale (GAS) (Endicott 1976): in Audini‐UK 1994, Muijen‐UK2 1994, Rosenheck‐USA 1993

This is an observer‐rated scale for evaluating the overall functioning of a person during a specified time period on a continuum from psychological or psychiatric sickness to health. The score ranges from 0 to 100, where a higher score indicates a better outcome. A modified version of the GAS was included in the Diagnostic and Statistical Manual of Mental Disorders (DSM‐III) as the Global Assessment of Functioning Scale (GAF) (APA 1987): in Bjorkman‐Sweden 2002. Outcomes from the two scales are reported together as GAF, as these two scales are very similar, and they report results on the same score range.

6.2.2 Health of the Nation Outcome Scale (HoNOS) (Stein 1999; Wing 1998): in Harrison‐Read‐UK 2000, REACT‐UK 2002

This scale provides a systematic summary of behaviours and functioning, measuring mental health and social/behavioural functioning. It consists of four areas (behaviour, impairment, symptoms, and social), each assessed through 12 items on a five‐point scale (0 to 4). Ratings are from 0 to 48; high score means severe dysfunction.

6.2.3 Rating of Medication Influence (ROMI) (Weiden 1994): in Harrison‐Read‐UK 2000, REACT‐UK 2002

This 20‐item scale measures the influence of factor on medication adherence. Each item is rated according to the degree of influence on medication‐taking behaviour: none (1), mild (2), and strong (3). It has two subscales: patient‐reported compliance (items 1 to 7) and patient‐reported non‐compliance (items 8 to 20). A high score on the compliance subscale indicates high compliance; a high score on the non‐compliance subscale indicates high non‐compliance. Results from the two studies are presented on a different range score (Harrison‐Read‐UK 2000 range score of 1 to 3; REACT‐UK 2002 range score not clearly reported).

6.2.4 Disability Assessment Schedule (DAS) (WHO 2001a): in Holloway‐UK 1996

The World Health Organization's Psychiatric Disability Assessment Schedule (DAS) is a measure of functioning and disability. It contains 36 items with six domains of functioning, including: understanding and communicating, getting around, self care, getting along with others, household and work activities, and participation in society. Higher scores indicate a worse outcome.

6.2.5 Interview Schedule for Social Interaction ‐ abbreviated version (ISSI) (Henderson 1980; Unden 1989): in Bjorkman‐Sweden 2002

The ISSI scale is a self report scale consisting of 30 items that measures social integration and attachment. The maximum score is 30 points; higher scores indicate better social integration and attachment.

6.2.6 Life Skills Profile (LSP) (Parker 1992; Rosen 1989): in REACT‐UK 2002

The Life Skills Profile is a clinician‐rated questionnaire developed in Australia primarily for use with people with psychotic illnesses. Thirty‐nine items and five subscales assess the general domain of disability over the last three months. The five subscales measure self care, non‐turbulence, social contact, communication, and responsibility. Each of the 39 items on the scale range from 'not at all disabled' to 'extremely disabled'.

6.2.7 REHAB Scale (REHAB) (Baker 1983): in Marshall‐UK 1995

The REHAB Scale is an observer‐rated measure of social functioning, covering social activity, self care, speech disturbance, and community skills. It rates the frequency of items of embarrassing or disruptive behaviour, such as violence, self harm, shouting and swearing, and sexual offensiveness (deviant behaviour ‐ REHAB DB); and lack of general skills (general behaviour ‐ REHAB GB). The scale ranges from 0 to 144, with higher scores indicating poorer functioning.

6.2.8 Role Functioning Scale (RFS) (Green 1987): in Jerrell‐SCarolina2 1994

This is a self report scale whereby the total of four subscales measures global role functioning. Higher scores indicate better functioning.

6.2.9 Social Adjustment Scale (SAS) (Weissman 1971; Weissman 1974): in Audini‐UK 1994, Muijen‐UK2 1994

Measures social functioning in a number of life domains (work, social, extended family, marital, parental, family unit, and economic adequacy). Score ranges from 1 to 7, with higher score indicating poorer outcome.

6.2.10 Social Adjustment Scale‐II (SAS‐II) (Schooler 1979): in Jerrell‐SCarolina1 1991

Revised version of the Social Adjustment Scale (see above), used to assess social adjustment. Self reported scale similar to SAS, but adapted for schizophrenia; it comprises 24 items covering seven areas including social, family, and work functioning. The scoring system of the two versions appears to differ, perhaps because this was an adapted version. Higher score indicates better outcome.

6.2.11 Social Functioning Questionnaire (SFQ) (Tyrer 1990; Tyrer 2005): in Harrison‐Read‐UK 2000

An eight‐item self report scale (score range is 0 to 24). It provides a quick assessment of perceived social function. Higher score indicates poorer social functioning.

6.2.12 Strauss‐Carpenter Outcome Scale (Strauss 1972; Strauss 1974): in Bjorkman‐Sweden 2002

The Strauss‐Carpenter Outcome Scale assesses a 21 items exploring frequency of social contacts, employment duration, symptomatology, and duration of rehospitalisation. The scaling of each item extends from 0 (maximal negative) to 4 (maximal positive). The scoring range of the scale extends from 0 (maximal negative) to 84 (maximal positive).

6.2.13 Alcohol Use Scale (AUS) (Drake 1996; Mueser 1995): in Drake‐NHamp 1998

A five‐point scale based on clinicians' ratings of the severity of the disorder, ranging from 1 (abstinence) to 5 (severe dependence).

6.2.14 Dartmouth Assessment of Lifestyle Interview (DALI) (Rosenberg 1998): in Sytema‐Netherlands 1999

An 18‐item, interviewer‐administered scale addressing the detection of substance use disorder in people with severe mental illness. DALI focuses on alcohol, cannabis, and cocaine use disorders. DALI‐alcohol: scores range from ‐4 to +6, higher scores indicate higher risk of alcohol abuse. DALI‐drugs: scores range from ‐4 to +4, higher scores indicate higher risk of drug abuse. As scale ranges from negative to positive value, skew is difficult to detect. We therefore entered data from this scale in Additional tables rather than into an analysis.

6.2.15 Substance Abuse Treatment Scale (SATS) (Drake 1996; McHugo 1995): in Drake‐NHamp 1998

An eight‐point scale indicating progression toward recovery, ranging from 1 (early stages of engagement) to 8 (relapse prevention). Higher scores indicate greater progression.

6.2.16 Timeline Followback (TLFB) (Sobell 1980): in Drake‐NHamp 1998

Scale administered by an interviewer to assess days of alcohol and drug use over the previous six months. Outcome reported as binary data.

6.2.17 Brief Psychiatric Rating Scale (BPRS) (Overall 1962): in Audini‐UK 1994, Drake‐NHamp 1998, Muijen‐UK2 1994, REACT‐UK 2002, Sytema‐Netherlands 1999 (BPRS 24‐item ‐ Velligan 2005); in Chan‐Hong Kong 2000, Ford‐UK 1995, Rosenheck‐USA 1993 (BPRS 18‐item)

The BPRS measures positive symptoms, general psychopathology, and affective symptoms. The original scale has 16 items, but a revised 18‐item scale is commonly used. Symptoms are reported in several ways (i.e. on a scale of 0 to 6 or 1 to 7), but it is most common for each item to be rated on a seven‐point scale (1 = not present to 7 = extremely severe). The 18‐item scale can range from 18 to 126 or from 0 to 108 (as in Ford‐UK 1995 and Rosenheck‐USA 1993).A further version is a 24‐item scale ranging from 24 to 168. For all of the scales, higher scores indicate more severe symptoms.

6.2.18 Brief Symptom Inventory (BSI) (Derogatis 1983): in Rosenheck‐USA 1993

A brief rating scale used by an independent rater to assess the severity of psychiatric symptoms. Scores range from 0 to 4, with higher scores indicating more symptoms.

6.2.19 Comprehensive Psychopathological Rating Scale (CPRS) (Asberg 1978): in Holloway‐UK 1996, UK700‐UK 1999

This is an interview rating scale covering a wide range of psychiatric symptoms; it can be used in total or as subscales. CPRS consists of 65 items that cover the range of psychopathology over the preceding week (40 symptom items are rated by the participant, and 25 observed items are rated by the rater during the interview). Each item is rated on a 0 to 3 scale ranging from 'not present' to 'extremely severe', with higher scores indicating more severe symptoms.

6.2.20 Colorado Symptom Index (CSI) (Shern 1994): in Lehman‐Maryland1 1994, Shern‐USA1 2000

A brief rating scale used by an independent rater to assess the severity of a range of psychiatric symptoms. A lower score indicates more symptoms.

6.2.21 Krawiecka Scale (KS) alias Manchester Scale (Krawiecka 1977): in Harrison‐Read‐UK 2000

This scale rates severity of psychiatric symptoms. It consists of eight categories of symptoms assessed on a five‐point scale, which are depression, anxiety, hallucinations, delusions, flattened and incongruous effect, psychomotor retardation, incoherence and irrelevance of speech, and poverty of speech. A score of 0 or 1 denotes an absence of pathology, while ratings of 2, 3, or 4 denote the presence of the target symptoms in increasing severity. Rating is from 0 to 36. Higher scores indicate a worse outcome.

6.2.22 Present State Examination (PSE) (Wing 1974): in Audini‐UK 1994, Muijen‐UK2 1994

This is a clinician‐rated scale measuring mental status. The scale rates and combines 140 symptom items to give various syndrome and subsyndrome scores. A short version covering the first 40 'neurotic' symptoms has been used in several population surveys. Score ranges from 1 to 120. Higher scores indicate greater clinical impairment.

6.2.23 Symptom Checklist‐90 (SCL‐90) (Hopkins Symptoms Checklist) (Derogatis 1974): in Bjorkman‐Sweden 2002

This is a self report clinical rating scale of psychiatric symptomatology comprised of 90 symptom‐related questions. Out of 90 items, 83 items represent nine subscales: somatisation (12 items), obsessive‐compulsive (10 items), interpersonal sensitivity (9 items), depression (3 items), anxiety (10 items), anger‐hostility (6 items), phobic anxiety (7 items), paranoid ideation (6 items), and psychoticism (10 items). Seven additional items include disturbances in appetite and sleep. The SCL‐90 also utilises three global distress indices: Global Severity Index (GSI), Positive Symptom Distress Index (PSDI), and Positive Symptom Total (PST). The participant assesses the degree of severity of each symptom. Items are rated on a five‐point Likert scale, ranging from 'not at all distressing' (0) to 'extremely distressing' (4), with higher scores indicating greater symptomatology.

6.2.24 Beck Depression Inventory (BDI) (Beck 1979): in Holloway‐UK 1996

A 21‐item self rating scale for depression. Each item comprises 4 statements (rated from 0 to 4) describing increasing severity of the abnormality concerned. The person completing the scale is required to read each group of statements and identify the one that best describes the way they have felt over the preceding week. Score ranges from 0 to 84, with higher score indicating more severe symptoms.

6.2.25 Hospital Anxiety and Depression Scale (HADS) (Zigmond 1983): Harrison‐Read‐UK 2000

This scale is a questionnaire composed of statements relevant to either generalised anxiety or depression referring to the past week. Seven items in the questionnaire reflect anxiety, and seven reflect depression. The participant answers each item on a four‐point (0 to 3) response category; the possible scores range from 0 to 21 for anxiety and 0 to 21 for depression. Higher score indicates a worse outcome.

6.2.26 Scale for the Assessment of Negative Symptoms (SANS) (Andreasen 1982; Andreasen 1989): in Holloway‐UK 1996, UK700‐UK 1999

This scale assesses five symptom complexes to obtain clinical ratings of negative symptoms in people with schizophrenia over the preceding week. They are: affective blunting, alogia (impoverished speech), avolition/apathy, anhedonia/asociality, and disturbance of attention. The final symptom complex seems to have less obvious relevance to negative symptoms than the other four complexes. Assessments are conducted on a six‐point scale (from 0 indicating 'not at all' to 5 indicating 'severe'). Higher scores indicate a worse outcome.

6.2.27 Scale for the Assessment of Positive Symptoms (SAPS) (Andreasen 1984): in OPUS‐Denmark 1999

The SAPS measures positive symptoms in schizophrenia. It is intended to serve as a complementary instrument to the Scale for the Assessment of Negative Symptoms (SANS). SAPS is split into four domains: hallucinations, delusions, bizarre behavior, and positive formal thought disorder. Within each domain, separate symptoms are rated from 0 (absent) to 5 (severe). Higher scores indicate a worse outcome.

6.2.28 Social Behaviour Schedule (SBS) (Wykes 1986): in Holloway‐UK 1996

The SBS is a 21‐item scale designed to assess a range of areas of functioning in people with long‐term mental illness. The scale covers areas such as social behaviour and communication, self care, and inappropriate behaviour. The respondent's behaviour on each item during the previous month is scored by a person familiar with him or her. Each item is rated on a five‐point Likert scale (from 0 to 4), with higher scores indicating greater deficits.

6.2.29 Lancashire Quality of Life Profile (LQoLP) (Oliver 1996; Oliver 1997): in Bjorkman‐Sweden 2002, Holloway‐UK 1996, UK700‐UK 1999

A structured self report interview with 105 items, combining objective and subjective measures in the following nine life domains (range of values 1 to 7): living situation, social relationships, work and education, legal status and safety, religion, family relations, leisure activities, finances, and health. The LQoLP also measures the following additional areas: positive and negative affect (with the Bradburn Affect‐Balance Scale), self esteem, global well‐being (Cantril's Ladder and Happiness Scale), perceived quality of life, and the quality of life of the patient independent of the patient's own opinion (with the Quality of Life Uniscale). The measures from LQoLP used in Bjorkman‐Sweden 2002 were: overall quality of life (which is the mean of subjective quality of life in nine life domains) and global well‐being. Higher score indicates better subjective quality of life/satisfaction.

6.2.30 Manchester Short Assessment of Quality of Life (MANSA) (Priebe 1999): REACT‐UK 2002, Sytema‐Netherlands 1999

A 16‐item scale composed of 4 objective and 12 subjective questions. The 12 subjective items are rated on a 7‐point scale (from 'couldn't be worse' to 'couldn't be better', scored from 1 to 7, range 12 to 84) assessing satisfaction with life 'in general', and in a range of domains such as vocational, financial, friendships, leisure, personal safety, physical health, and mental health. Four objective items, answered yes or no, assess the existence of a close friend, contacts with friends per week, accusation of a crime, and victimisation of physical violence. Higher score indicates better quality of life. In REACT‐UK 2002 and Sytema‐Netherlands 1999, score is reported as a mean ranging from 1 to 7.

6.2.31 Lehman's Quality of Life Interview (QOLI) (Lehman 1988): in Drake‐NHamp 1998, Lehman‐Maryland1 1994, Shern‐USA1 2000; (Lehman 1993) in Ford‐UK 1995; (Lehman 1983) in Marshall‐UK 1995

The QOLI contains 153 items that measure global life satisfaction as well as objective and subjective quality of life, in eight life domains (living situation, daily activities and functioning, family relations, social relations, finances, work and school, legal and safety issues, and health). The QOLI is rated on a seven‐point scale, with higher scores indicating better quality of life. Subjective assessment of general life satisfaction ranges from 1 to 7 (terrible to delighted). Ford‐UK 1995 reported objective quality of life, and Marshall‐UK 1995 reported subjective quality of life.

6.2.32 Camberwell Assessment of Need interview (CAN) (Phelan 1993): in Bjorkman‐Sweden 2002, Harrison‐Read‐UK 2000, UK700‐UK 1999

The Camberwell Assessment of Need assesses the health and social needs of people with mental health problems. It measures 22 areas to yield numbers of met and unmet needs as rated by the participant. Possible scores range from 0 to 22, with a higher score indicating poorer level of met needs.

6.2.33 Camberwell Assessment of Need Short Appraisal Schedule (CANSAS) (Phelan 1995; Slade 1999): in REACT‐UK 2002, Sytema‐Netherlands 1999

This is an abbreviated form of the above CAN.

6.2.34 Client Satisfaction Questionnaire (CSQ) (Larsen 1979): in Audini‐UK 1994, Muijen‐UK2 1994, OPUS‐Denmark 1999, Sytema‐Netherlands 1999

The CQS is a self report instrument designed to measure patient's global satisfaction with services. Items are concerned with quality of services received, how well services met the client's needs, and general satisfaction. The CSQ is substantially correlated with treatment attrition, number of therapy sessions attended, and change in client‐reported symptoms. It consists of eight items that are scored on four‐point Likert scales (1 to 4). Total score ranges from 8 to 32. Higher scores indicate greater satisfaction.

6.2.35 Client Satisfaction Questionnaire (CSQ) ‐ modified version (Gerber 1999; Larsen 1979): in REACT‐UK 2002

This survey has 35 questions covering the location of services, services clients expect, delays in obtaining services, client's input into treatment, information received about drug treatment, satisfaction with treatment, access to clinical files, satisfaction with the therapist, family involvement in treatment, the treatment process, and overall satisfaction. Possible responses to most items range from 1 (most negative) to 7 (most positive). A rating of zero on certain items enables the respondent to indicate that the question was not relevant to his or her situation. Six items within the questionnaire, also on a seven‐point scale, form a general satisfaction questionnaire. Higher score indicates greater satisfaction.

6.2.36 Patient's satisfaction with health services (Tyrer 1979): in UK700‐UK 1999

A self reporting questionnaire that rates nine components of satisfaction with services, each on a four‐point scale (1 to 4). Scores can range from 9 to 36, with higher values indicating less satisfaction.

6.2.37 Patient Satisfaction Instrument (PSI) (Risser 1975): in Chan‐Hong Kong 2000

The PSI assesses patient and client satisfaction. It is a 26‐item questionnaire designed to measure clients' satisfaction with nursing care in the community setting.

6.2.38 Specific Level of Functioning Scale (SLOF) (Schneider 1983): in Chan‐Hong Kong 2000

The SLOF assesses clients' behavioural functioning and daily skills. It is a 43‐item behavioural rating scale designed for use in clients with chronic mental illness in the community.

6.3 Missing outcomes

No trial reported or rated relapse, mental state: not improved, or carer satisfaction.

Excluded studies

See: Characteristics of excluded studies.

Eighty‐one studies were excluded from the previous versions of this review, while a total of 85 studies were excluded from this version.

The earlier update excluded four trials included in the original reviews as the trials did not match the new inclusion criteria. (De Cangas‐Canada 1994; Franklin‐Texas 1987; Lafave‐Canada 1996; Marx‐Wisconsin 1973). Morse‐Missouri2 1997 was also excluded, as it did not report the number of people randomised to each treatment group.

One further trial (Tyrer‐UK 1995), originally included in the Case Management (CM) review, was now excluded because of a methodological issue. Tyrer‐UK 1995 is a trial comparing ICM to standard care. The first issue was to clarify the ICM caseload. Thanks to further information provided from the author, we clarified that there were 25 key workers in the service looking after 400 patients on the register, therefore each key worker had a caseload of 1 to 16, or 'high‐intensity' case management by this review's definition. The second issue that arose was that the case managers in the treatment group were also workers in the control group. The problem was therefore one of contamination, as requiring someone to carry out close monitoring of one participant in the treatment group could affect their care of a similar participant in the control group in an unpredictable way. We decided to exclude this study on the grounds that we could not be sure that high‐intensity case management was really being compared with standard care.

We also excluded three trials that had been awaiting assessment in the original Case Management and Assertive Case Management reviews (see Other published versions of this review) for varying reasons (Godley‐Illinois 1994; Jerrell‐California 1989; Mulder‐Missouri 1985) (for more details see Characteristics of excluded studies).

Jerrell‐California 1989 was a partially published trial previously classified as 'awaiting further assessment' because we needed information on number of people excluded after randomisation (participants were excluded if they refused to participate after randomisation or if they had not been discharged from hospital within six months of entering the study). As these data have not become available, we have now excluded the trial.

Godley‐Illinois 1994 was an unpublished, two‐centre trial initially classified as 'awaiting further assessment' because we could not determine if the intervention was ACT or CM. During the current update we classified the intervention as ICM versus standard care. We had to exclude this study because it contained no usable data due to incomplete data reporting, and no further information became available (i.e. there was an apparent error in the reporting of numbers admitted to hospital: in one table admission rates were reported as 31/52 experimental group and 33/45 control group, while in another table admission rates were reported as 31/52 experimental and 25/45 control).

Mulder‐Missouri 1985 was a report of data from randomised and non‐randomised participants. We excluded this study because these data were not reported separately and, in addition, the intervention did not fit our inclusion criteria (ICM was compared to acute hospital admission).

Overall, we excluded 42 studies because they were not randomised or because randomisation was compromised (Jerrell‐California 1989). We had to exclude five studies because participants required immediate hospital admission (Fenton‐Canada 1978; Hoult‐Australia 1981; Muijen‐UK1 1992; Mulder‐Missouri 1985; Stein‐Wisconsin), one study because participants were dually diagnosed with intellectual disability and mental illness (Martin‐UK 2005), and one additional study because the majority of participants were simply homeless and not clearly ill (Toro‐New York 1997). Most trials had to be excluded because of the intervention: 54 because the experimental intervention was not ICM. We had to exclude Modcrin‐Kansas 1988 as caseload was not reported in either the experimental or the control group. We excluded 48 trials as the comparison intervention was neither standard care nor non‐ICM. We excluded 10 trials as the intervention administered to the experimental group was not only ICM (Chandler‐California2 2006; COAST‐UK 2004; Cosden‐California 2005; Gold‐SCarolina 2006; Grawe‐Norway 2005; Lehman‐Maryland2 1993; LEO‐UK 1994; McHugo‐Washington DC 2004; Shern‐USA2; Shern‐USA3). We excluded 10 trials as they did not present with relevant comparisons. Finally, we excluded two trials, Godley‐Illinois 1994 and Morse‐Missouri2 1997, because we could extract no usable data from the study report (as previously explained).

Awaiting classification

See: Characteristics of studies awaiting classification.

Five trials described as 'awaiting classification' in the previous review have now been excluded as more data have become available (Agius‐Croatia 2007; Kane‐Virginia 2004; Klotz‐California 2001; Li‐China 2004; NCT00781079); two more have now been excluded based only on the abstract, and so they were not included in the excluded studies section (Huang‐China and Johnson‐UK).

Twenty trials, of which three are in the Chinese language, are awaiting classification, and their authors have been contacted for further clarification.

Ongoing studies

See: Characteristics of ongoing studies.

We are aware of six currently ongoing trials, five more than the ongoing studies in the previous version of this review (Walsh‐Connecticut).

Risk of bias in included studies

For multicentre trials that provided data for individual single centres, we did not assess the risk of bias for each centre. Our judgements regarding the overall risk of bias in individual studies are illustrated in Figure 1.

Allocation

All 40 studies were stated to be randomised, but only 11 provided descriptions of the methods used to generate the sequence. We therefore classified these studies as at low risk of selection bias. The most common method of randomisation was random allocation according to a sequence of random numbers generated by a computer program in one of two sites (Bjorkman‐Sweden 2002; Essock‐Connecticut2 2006; Ford‐UK 1995; Harrison‐Read‐UK 2000; OPUS‐Denmark 1999), while Cusack‐North Carolina employed randomisation using a random number table, assigned in blocks of two. Three trials used permuted block (Marshall‐UK 1995; REACT‐UK 2002; Sytema‐Netherlands 1999), one used a table of random permutation (Pique‐California 1999), and one used coin tossing (Rosenheck‐USA 1993). In one of the two sites of the OPUS‐Denmark 1999 trial (Aahrus site), allocation was performed by drawing lots – from among five red and five white lots from a black box. Overall, however, most studies, including Chan‐Hong Kong 2000, were classified as at unclear risk of selection bias with an overestimate of positive effect, as no description of the methods used to generate the sequence was provided.

Regarding the allocation concealment, we rated only four studies as low risk of bias as they provided descriptions of the methods used to conceal random allocation (OPUS‐Denmark 1999; REACT‐UK 2002; Sytema‐Netherlands 1999; UK700‐UK 1999). All four studies used centralised allocation carried on by telephone, fax, or mail. We classified the remaining 33 studies as at unclear risk of selection bias with an overestimate of positive effect.

Blinding

We classified blinding with respect to only primary outcomes. Due to intervention characteristics, that is being a model of service organisation, we assumed participants and clinicians implicitly not being blind to treatment assignment. We also assumed that primary outcomes were likely to be influenced by participant and clinician lack of blinding, as the knowledge of treatment allocation could determine both performance and attrition bias at a level that is difficult to predict/quantify. Whereas we did not consider the primary outcomes as interviewer mediated, we assumed that lack of interviewer blinding would produce less detection bias. We therefore classified all studies providing primary outcome data as at unclear risk of performance and attrition bias. This creates further potential for overestimate of positive effects and underestimate of negative ones.

We have reported blinding to secondary outcomes in the 'Risk of bias' table, but we did not account for it in the global rating of the study blinding risk of bias. Again, if the secondary outcome was clinician/participant mediated, we rated it as unclear. If it was interviewer rated, we assessed it according to information provided in the study. We rated only Shern‐USA1 2000 as at high risk of bias, as it provided only secondary outcome data and was only interviewer mediated, and it was therefore possible to assess risk of bias for this study with greater confidence. We rated Chan‐Hong Kong 2000 as at unclear risk of performance and detection bias, as it did not clearly report any information on blinding.

Incomplete outcome data

Where information was available, we assessed incomplete outcome data separately for primary and secondary outcomes and presented both assessments in Figure 1. However, we only rated risk of bias with respect to primary outcome. Only three trials provided separate information for incomplete primary and secondary outcome data, and so we could assess the risk of bias separately (Holloway‐UK 1996; Johnston‐Australia 1998; REACT‐UK 2002). We judged nine trials as adequately addressing incomplete outcome data and rated them as at low risk of attrition bias. We so rated four of these studies because there were no missing outcome data (Bush‐Georgia 1990; Holloway‐UK 1996; REACT‐UK 2002; Sytema‐Netherlands 1999); three because they made the number and reason for missing data explicit, and the missing data were balanced across groups (Audini‐UK 1994; Essock‐Connecticut1 1995; Johnston‐Australia 1998); and two because they undertook intention‐to‐treat (ITT) analysis (OPUS‐Denmark 1999; UK700‐UK 1999).

In Cusack‐North Carolina, all analyses were intention to treat and outcomes were observed regardless of active or continued participation. Although Chan‐Hong Kong 2000 reported the analyses as ITT, it was not clearly reported whether any participants left early.

We judged Bond‐Chicago1 1990 and Ford‐UK 1995 as at high risk of attrition bias because, although clearly reporting number and reasons for missing data, the reasons for missing outcome data were likely to be related to true outcome, with imbalance either in numbers or reasons for missing data across intervention group. However, our protocol compensated for this somewhat (see Dealing with missing data), and despite the high rating, information from these studies remains included.

We rated the remaining trials as at unclear risk of attrition bias. Either they did not address this issue or presented insufficient information of attrition/exclusions to permit judgement (i.e. no reasons for missing data provided or number randomised not stated. Jerrell‐SCarolina1 1991 reported only number of randomised participants completing the study period).

Some specific examples may serve to illustrate the difficulty in rating this issue. Essock‐Connecticut2 2006 was not an ITT analysis (seven participants were excluded from the study immediately after randomisation because they were lost to follow‐up), but the authors failed to provide information on to what intervention those participants had been allocated and reason for leaving the study early. As this study provided only continuous outcome data, we reproduced completer‐only data that were reported. No action was undertaken to deal with other missing data.

Macias‐Utah 1994 had three problems. First, the study was not an ITT analysis: seven participants were excluded from the study because they were lost to follow‐up. The authors of this study broke with standard practice by failing to provide any data on participants lost to follow‐up, in particular data on admissions to hospital. Second, one participant (presumably randomised to the treatment group) was excluded after randomisation, having refused to participate (again, it was unclear whether this person had been admitted to hospital). Third, five further participants were added (randomly) to the treatment and control groups part way through the study, some as late as "late 1990" (final assessments took place in February 1991). No further information has become available since first review publication, which potentially could substantially affect findings.

In Morse‐Missouri1 1992, 28 further participants were added (randomly) to the initial randomised sample, to replace participants leaving the study within the first month after entering. As replacement was carried out through randomised assignment, we did not raise any questions on the replacement issue and included the study. We presented data from the final sample, obtained after randomised replacement had occurred.

In Muller‐Clemm‐Canada 1996, the number randomised was not clearly reported; authors declared that "Clients who withdrew from the study within the first 6 months were replaced by other clients".

Finally, Sytema‐Netherlands 1999 randomised 119 participants, but one was excluded because he or she moved to another area directly after randomisation. (We performed ITT analysis on the remaining 118 participants.)

Selective reporting

We rated most of the trials (24) as at high risk of reporting bias, as their data was presented in such a way that we could not consider it to be free of the suggestion of selective outcome reporting (i.e. prespecified outcomes were not reported, or they were reported incompletely so that they could not be entered in the analysis, or outcomes were reported that were not prespecified). We rated 16 studies as at low risk of reporting bias. We assessed two studies as at unclear risk of reporting bias.

Other potential sources of bias

Only Hampton‐Illinois 1992 was rated as at unclear risk of other potential sources of bias, as it was unclear whether the study was interrupted early in one of the two centres.

We rated all of the remaining trials as at low risk of other potential sources of bias, as we found no evidence of other bias. Most of these studies were publicly funded. No declaration of interest was made by authors, and we assume there was none to be made. However, many study authors were active pioneers in the development and implementation of the experimental intervention model across the scientific community and the clinical world. This raises the issue of how researcher beliefs could affect the entire process of evaluating an intervention in a randomised clinical trial. Although conscious of this issue, we decided not to make any attempt in rating it as it is very difficult to judge, and erroneous quantification could drive bias into our conclusions.

Effects of interventions

We categorised studies into two comparisons: Intensive Case Management versus standard care, and Intensive Case Management versus non‐Intensive Case Management. The nine main indices of outcome were:

service use;
adverse effects;
global state;
social functioning;
mental state;
behaviour;
quality of life;
satisfaction; and
direct costs.

We considered each index in turn for each of the two comparisons. We were able to extract numerical data from 40 randomised trials, among which seven multicentre trials provided data for individual single centre.

1. COMPARISON 1: INTENSIVE CASE MANAGEMENT versus STANDARD CARE

(summary of findings Table for the main comparison). There were 45 outcomes in this comparison.

1.1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months

Data were available from five studies presenting skewed data from a sample size greater than or equal to 200 participants and from 19 trials reporting skewed data from sample sizes less than 200. We entered these data in separate subgroups, but we also presented the overall data.

1.1.1 Skewed data (sample size ≧ 200)

In the first subgroup analysis (i.e. skewed data from studies with sample size greater than or equal to 200 participants), we found no significant difference in length of hospitalisation per month (n = 1812, 5 randomised controlled trials (RCTs), mean difference (MD) ‐0.46, confidence interval (CI) ‐0.95 to 0.03), although data suggested a trend favouring ICM (P = 0.06). This subgroup had moderate levels of heterogeneity (Chi² = 6.36; df = 4.0; P = 0.17; I² = 37%; Analysis 1.1).

1.1.2 Skewed data (sample size < 200)

In the second subgroup analysis (i.e. skewed data from study sample size less than 200 participants), there was a significant difference between groups, favouring the ICM group in reducing length of hospitalisation (n = 1783, 19 RCTs, MD ‐1.01, 95% CI ‐1.74 to ‐0.28), but these data were highly heterogeneous (Chi² = 79.27; df = 18.0; P = 0.0; I² = 77%; Analysis 1.1).

1.1.3 Overall data (skewed data: sample size ≧ 200 and sample size < 200)

When synthesising data from the two subgroups, we found that length of hospitalisation was significantly reduced in the ICM group (n = 3595, 24 RCTs, MD ‐0.86, 95% CI ‐1.37 to ‐0.34), but the level of heterogeneity was high (Chi² = 89.43; df = 23.0; P = 0.0; I² = 74%; Figure 3). We investigated the heterogeneity by checking again for correctness of data and removing one outlier study from the analysis (Curtis‐New York 1992), as it was the only study favouring standard care. After excluding Curtis‐New York 1992, the level of heterogeneity was still high (I² = 59%; P < 0.0002). We therefore removed the second‐most outlier study from the analysis (Bond‐Indiana1 (A), one of three centres from a multicentre study), as this was the most extreme result (favouring ICM). By excluding Bond‐Indiana1 (A), data remained significant, favouring ICM (n = 3245, 22 RCTs, MD ‐0.79, 95% CI ‐1.22 to ‐0.36). The heterogeneity was reduced to just within our cutoff point (I² = 49%; P = 0.005). Removing two further outliers (Bond‐Indiana1 (C) and Quinlivan‐California 1995) reduced heterogeneity still further (I² = 36%; P = 0.05) as well as the overall estimate, but ICM still seemed to significantly decrease time in hospital (n = 3143, 20 RCTs, MD ‐0.62, 95% CI ‐1.00 to ‐0.23, Figure 4).

Figure 3

Forest plot of comparison: 1 Intensive Case Management versus standard care, outcome: 1.1 Service use: 1. Average number of days in hospital per month ‐ at about 24 months.

Figure 4

Service use: 1. Average number of days in hospital per month ‐ at about 24 months ‐ restoring homogeneity ‐ 4 studies removed from analysis.

No substantial reporting biases were highlighted when investigated through visual inspection of funnel plot (Figure 5). Two studies ‐ Bond‐Indiana1 (A) and Quinlivan‐California 1995 ‐ seemed most heterogeneous (see above).

Figure 5

Funnel plot of comparison: 1 Intensive Case Management versus standard care, outcome: 1.1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months.

We ran meta‐regression on trials providing data for the primary outcome 'average number of days in hospital per month ‐ at about 24 months' (combining data from all ICM studies within Comparison 1 and 2). Within the meta‐regression we found that i.) the more ICM is adherent to the organisation model, the better it is at decreasing time in hospital ('organisation fidelity' variable coefficient ‐0.36, 95% CI ‐0.66 to ‐0.07, Figure 6); and ii.) the higher the baseline hospital use in the population, the better ICM is at decreasing time in hospital ('baseline hospital use' variable coefficient ‐0.20, 95% CI ‐0.32 to ‐0.10, Figure 7). Combining both these variables within the model, 'organisation fidelity' is no longer significant (regression coefficient ‐0.24, 95% CI ‐0.52 to 0.04, P = 0.089), but 'baseline hospital use' resultstill significantly influences time in hospital, although it seems to lose some of its potency (regression coefficient ‐0.18, 95% CI ‐0.29 to ‐0.07, P = 0.0027) (Figure 8). Figure 8 shows the interaction of the two variables on study outcome graphically through the use of thin plate spline modelling. The plot provides a locally weighted two‐dimensional representation of the collinearity between the variables used in the regression.

Figure 6

Meta‐regression: Scatterplot of IFACT organisation subscore versus mean days per month in hospital.

Figure 7

Meta‐regression: Scatterplot of mean baseline days in hospital versus mean days per month in hospital.

Figure 8

Weighted thin plate spline regression showing combined effect of baseline days in hospital and organisational fidelity score on treatment effect.

1.2 Service use: 1 Number of days in hospital ‐ by follow‐up (skewed data, sample size ≧ 200)

We identified one study relevant to this outcome, providing data by medium‐ and long‐term follow‐up (FUP) and measuring outcome during the previous year.

1.2.1 By medium‐term FUP (3 years) (previous year)

We found one trial to be relevant to this subgroup, with a total of 547 participants. For this subgroup, we did not find evidence of a clear difference between the two treatments (MD 0.1, 95% CI ‐10.26 to 10.46; Analysis 1.2).

1.2.2 By long‐term FUP (8 years) (previous year)

There was a single trial in this subgroup, with a total of 547 participants. There was no clear difference between ICM and standard care within this subgroup (MD 4.3, 95% CI ‐4.63 to 13.23; Analysis 1.2).

1.3 Service use: 2. Not remaining in contact with psychiatric services

We found nine relevant studies (total n = 1633) for this outcome, providing data on different follow‐up length. Overall, when pooling studies from different time subgroups, we found a significant advantage to the ICM group, where people were less likely to be lost to psychiatric services than people in the standard care group (n = 1633, 9 RCTs, risk ratio (RR) 0.43, 95% CI 0.30 to 0.61, Figure 9). Heterogeneity was moderately high for this outcome (Chi² = 15.57; df = 8.0; P = 0.05; I² = 48%).

Figure 9

Forest plot of comparison: 1 Intensive Case Management versus standard care, outcome: 1.2 Service use: 2. Not remaining in contact with psychiatric services by short, medium, long term and overall.

As there were fewer than 10 studies for this outcome, we did not use a funnel plot (see Assessment of reporting biases).

1.3.1 By short term

We included only one short‐term study and found no significant difference between treatment groups (n = 95, 1 RCT, RR 0.54, 95% CI 0.28 to 1.05; Analysis 1.3).

1.3.2 By medium term

Medium‐term data available from three studies showed a significant difference between treatment groups, favouring the ICM group, where participants had a lower risk of not remaining in contact with psychiatric services compared with participants in the standard care group (n = 1063, 3 RCTs, RR 0.51, 95% CI 0.36 to 0.71). Heterogeneity was moderately high for this subgroup (Chi² = 3.02; df = 2.0; P = 0.22; I² = 33%; Analysis 1.3).

1.3.3 By long term

Six long‐term studies data confirmed the trend favouring ICM, showing a significant advantage for the ICM group (n = 653, 6 RCTs, RR 0.35, 95% CI 0.18 to 0.68), but data were heterogeneous (I² = 63%; P = 0.02). Herinckx‐Oregon 1996 seemed to be the sole cause of this, and on further consideration post hoc, we think we were in error in including the outcome from this study because it was defined so differently from the other trials. Herinckx‐Oregon 1996 did not include refusing to re‐interview, moving out, and death ‐ as all the other studies had done ‐ and it was impossible to amend this at this stage. We therefore feel justified in removing this study altogether from this part of the review outcomes. Once Herinckx‐Oregon 1996 was removed, the five remaining trials confirmed the significant advantage for the ICM group (n = 475, 5 RCTs, RR 0.27, 95% CI 0.11 to 0.66), and heterogeneity was restored to a moderate level (Chi² = 7.19; df = 4.0; P = 0.13; I² = 44%; Analysis 1.3).

1.4 Service use: 3a. Admitted to hospital

We identified 16 studies relevant to this outcome and categorised data into five subgroups (in keeping with our protocol).

1.4.1 By short term

Data were available from two short‐term studies and showed no significant differences between treatment groups (n = 244, 2 RCTs, RR 0.61, 95% CI 0.22 to 1.69), but these data were heterogeneous (Chi² = 5.36; df = 1.0; P = 0.02; I² = 81%; Analysis 1.4).

1.4.2 By medium term

Five studies reported medium‐term data, and these favoured the ICM group, which had less admission to hospital across time compared with standard care (n = 1303, 5 RCTs, RR 0.85, 95% CI 0.77 to 0.93; Analysis 1.4).

1.4.3 By long term

Eleven studies provided long‐term data. As with the short‐term data, they showed no significant differences between treatment groups (n = 1516, 11 RCTs, RR 0.96, 95% CI 0.74 to 1.23), but data were heterogeneous (Chi² = 32.88; df = 10.0; P = 0.0; I² = 69%; Analysis 1.4).

1.4.4 By long term‐ during previous 12 months

Only one study reported data by long term, but referring to the number of admissions across the previous year. We therefore could not enter this data in the long‐term data subgroup analysis. This data showed a significant effect favouring the ICM group (n = 547, 1 RCT, RR 0.67, 95% CI 0.52 to 0.86), therefore not consistent with long‐term data shown above. As these findings were based on data from one study only, we consider them less robust than the findings from the 11 long‐term studies Analysis 1.4.

1.4.5 By short term FUP ‐ unplanned admission through emergency department (ED)

We found one trial to be relevant to this subgroup (total n = 62). For this subgroup, we did not find evidence of a clear difference between the two treatments (RR 1.0, 95% CI 0.07 to 15.28; Analysis 1.4).

1.5 Service use: 3b. Average number of admissions per month (skewed data)

Data describing the average number of admissions per month were available from one medium‐term and three long‐term studies. All of these data were skewed and did not enter the analysis. Data from the medium‐term study suggested a trend favouring the ICM group. Data from the long‐term studies showed no trend favouring one group over the other. Audini‐UK 1994 and Muller‐Clemm‐Canada 1996 did not report variance measurements. We assumed consistency between studies and used the fully reported variance for Sytema‐Netherlands 1999, and employed these data for Audini‐UK 1994 and Muller‐Clemm‐Canada 1996 as well.

1.6 Service use: 4a. Admitted to ER ‐ by long term

The only study identified describing 'number admitted to ER' showed a non‐significant difference between the two groups (n = 178, 1 RCT, RR 1.13, 95% CI 0.72 to 1.76).

1.7 Service use: 4b. Average number of admissions to ER (skewed data) ‐ by medium term

The two studies describing the average number of admissions to ER reported skewed data. Skewed data were not consistent, as one study did not show any trend in the direction of effect, and the other study showed a trend favouring the ICM group. As in one study the variance measurement was not reported (Jerrell‐SCarolina1 1991), we carried the standard deviation over from the other available study (Lehman‐Maryland1 1994).

1.8 Service use: 5a. Received day hospital care ‐ by short‐term FUP

We identified only one study relevant to this outcome (total n = 62). There was not a significant difference between ICM and standard care (RR 2.0, 95% CI 0.19 to 20.93; Analysis 1.8).

1.9 Service use: 5b. Outpatient visits ‐ by short‐term FUP (6 months)

We identified only one study relevant to this outcome. There was not a significant difference between ICM and standard care (n = 62, 1 RCT, MD 0.29, 95% CI ‐0.14 to 0.72; Analysis 1.9).

1.10 Service use: 5c. Outpatient visits ‐ by medium term (skewed data)

One more small study provided data for this outcome (n = 134), but as data were skewed and the total sample less than 200, we entered them as 'Other' data. Data showed a trend favouring ICM, where the ICM group received more outpatient visits than the standard care group by medium term (Analysis 1.10).

1.11 Service use: 5d. Received home visits ‐ by short‐term FUP

We found a single study showing a significant difference between ICM and standard care in the mean number of home visits received, favouring the ICM group (n = 62, 1 RCT, MD 4.32, 95% CI 3.42 to 5.22). Note that for this outcome the right graph label favours ICM (experimental group).

1.12 Adverse event: 1a. Death ‐ any cause

We found 14 relevant studies for this outcome, the data from which we divided into five subgroups according to different time to follow‐up. We found similar results in mortality across different subgroups, none of which showed a significant difference between ICM and standard carefor overall mortality.

1.12.1 By short term

We found two trials relevant to this subgroup, with a total of 161 participants. For this subgroup, two deaths occurred in the 81 people treated with ICM compared with two deaths in the 80 people treated with standard care (n = 161, 2 RCTs, RR 1.04, 95% CI 0.16 to 6.91; Analysis 1.12).

1.12.2 By medium term

There were six relevant trials in this subgroup (total n = 901). For this subgroup, five deaths occurred in 453 people treated with ICM compared with six deaths in 448 people treated with standard care (n = 901, 6 RCTs, RR 0.78, 95% CI 0.23 to 2.62; Analysis 1.12).

1.12.3 By long term

We found nine trials relevant to this subgroup, with a total of 1456 participants. For this subgroup, 24 deaths occurred in 741 people treated with ICM compared with 27 deaths in 715 people treated with standard care (n = 1456, 9 RCTs, RR 0.84, 95% CI 0.48 to 1.47; Analysis 1.12).

1.12.4 By medium‐term FUP (3 years)

There was a single trial in this subgroup, in which six deaths occurred in 275 people treated with ICM compared with 10 deaths in 272 people treated with standard care, showing no significant difference between ICM and standard care (n = 547, 1 RCT, RR 0.59, 95% CI 0.22 to 1.61; Analysis 1.12).

1.12.5 By long‐term FUP (8 years)

There was a single trial in this subgroup, in which 14 deaths occurred in 275 people treated with ICM compared with 15 deaths in 272 people treated with standard care, showing no significant difference between ICM and standard care (n = 547, 1 RCT, RR 0.92, 95% CI 0.45 to 1.88; Analysis 1.12).

1.13 Adverse event: 1b. Death ‐ suicide

We found 12 relevant studies for this outcome and categorised data into four subgroups. Our results for mortality due to suicide were similar to those found for mortality due to all causes, that is no significant difference in suicide rate between the two intervention groups.

1.13.1 By short term

Data by short term were available from two studies, where no suicides occurred in 62 people treated with ICM compared with two suicides in 65 people treated with standard care (n = 127, 2 RCTs, RR 0.35, 95% CI 0.04 to 3.27; Analysis 1.13).

1.13.2 By medium term

Data by medium term were available from four studies, where two suicides occurred in 412 people treated with ICM compared with two suicides in 407 people treated with standard care (n = 819, 4 RCTs, RR 0.98, 95% CI 0.17 to 5.60; Analysis 1.13).

1.13.3 By long term

Data by long term were available from nine studies, where 10 suicides occurred in 741 people treated with ICM compared with 14 suicides in 715 people treated with standard care (n = 1456, 9 RCTs, RR 0.68, 95% CI 0.31 to 1.51; Analysis 1.13).

1.13.4 By medium‐term FUP (3 years)

Data by medium‐term follow‐up (3 years) were available from one study, where three suicides occurred in 275 people treated with ICM compared with four suicides in 272 people treated with standard care (n = 547, 1 RCT, RR 0.74, 95% CI 0.17 to 3.28; Analysis 1.13).

1.14 Global state: 1. Leaving the study early

We identified 21 studies relevant to this outcome and categorised data into five subgroups (in keeping with our protocol).

1.14.1 By short term

We included five short‐term studies and found no significant differences between treatment groups for number of participants leaving the study early (n = 598, 5 RCTs, RR 0.79, 95% CI 0.44 to 1.41), but data were heterogeneous (Chi² = 80.24; df = 4.0; P = 0.0; I² = 95%; Analysis 1.14).

1.14.2 By medium term

We included eight medium‐term studies and found the risk of leaving the study early was lower for participants in the ICM group (n = 1699, 8 RCTs, RR 0.60, 95% CI 0.51 to 0.70; Analysis 1.14).

1.14.3 By long term

Data from 13 long‐term studies confirmed data from medium‐term studies, showing a significant advantage for ICM (n = 1798, 13 RCTs, RR 0.68, 95% CI 0.58 to 0.79; Analysis 1.14).

1.14.4 By medium‐term FUP (3 years)

There was a single trial in this subgroup, showing no significant difference between ICM and standard care (n = 547, 1 RCT, RR 1.01, 95% CI 0.84 to 1.21; Analysis 1.14).

1.14.5 By long‐term FUP (8 years)

We found one trial relevant to this subgroup, showing no significant difference between ICM and standard care (n = 547, 1 RCT, RR 0.88, 95% CI 0.7 to 1.09; Analysis 1.14).

1.15 Global state: 2. Average endpoint score (Global Assessment of Functioning Scale, high = good)

We identified five studies relevant to this outcome and categorised data into three subgroups. Note that for this outcome the right graph label favours ICM (experimental group).

1.15.1 By short term

There were four relevant trials in this subgroup (total n = 797). For this subgroup, the Global Assessment of Functioning Scale (GAF) score favoured ICM (MD 2.07, 95% CI 0.28 to 3.86; Analysis 1.15).

1.15.2 By medium term

Medium‐term GAF data from three studies (n = 722, 3 RCTs, MD 0.09, 95% CI ‐3.11 to 3.28) showed no significant difference between groups. This subgroup had important levels of heterogeneity (Chi² = 4.42; df = 2.0; P = 0.11; I² = 54%; Analysis 1.15).

1.15.3 By long term

We found five trials relevant to this subgroup, with a total of 818 participants. For this subgroup, the GAF score favoured the ICM group (MD 3.41, 95% CI 1.66 to 5.16; Analysis 1.15).

1.16 Global state: 3. Not compliant with medication ‐ by long term

We only found data from one long‐term study, which favoured the ICM group (n = 71, 1 RCT, RR 0.35, 95% CI 0.15 to 0.81).

1.17 Social functioning: 1a. Contact with legal system (various measurements)

We found 11 relevant studies for this outcome and categorised data into six subgroups, according to different time to follow‐up and different outcomes described.

1.17.1 By short term ‐ contact with the police

We only found data from one study for short‐term outcomes, describing the outcome 'contact with the police'. This study did not reveal any significant difference in the rate of contact with the police between treatment groups (n = 61, 1 RCT, RR 2.57, 95% CI 0.73 to 9.04; Analysis 1.17).

1.17.2 By medium term ‐ arrested

We found three studies describing the outcome 'number of arrested'. These studies failed to show a significant difference between the two intervention groups (n = 604, 3 RCTs, RR 1.08, 95% CI 0.61 to 1.90; Analysis 1.17).

1.17.3 By medium term ‐ contact with the police

Only one medium‐term study was available providing data on 'contact with the police'. These data favoured the ICM group in reducing the number of contacts with the police (n = 88, 1 RCT, RR 0.23, 95% CI 0.09 to 0.55; Analysis 1.17).

1.17.4 By medium term ‐ imprisoned

We found four medium‐term studies describing 'number of imprisoned'. These studies showed no significant advantage for the ICM group (n = 804, 4 RCTs, RR 0.80, 95% CI 0.39 to 1.64). This subgroup had important levels of heterogeneity (Chi² = 6.21; df = 3.0; P = 0.1; I² = 51%; Analysis 1.17).

1.17.5 By long term ‐ arrested

We found data from one study for long‐term outcomes, describing the outcome 'number of arrested' , and it showed no significant advantage for ICM (n = 178, 1 RCT, RR 0.66, 95% CI 0.32 to 1.37; Analysis 1.17).

1.17.6 By long term ‐ imprisoned

We found also five long‐term studies reporting data on 'number of imprisoned', again not showing any significant advantage for ICM in reducing the number of participants imprisoned by long term (n = 908, 5 RCTs, RR 0.86, 95% CI 0.45 to 1.65; Analysis 1.17).

1.18 Social functioning: 1b. Mean contacts with legal system (skewed data) ‐ by medium term

Data were available from one medium‐term trial, describing three different outcomes: bookings, jail days, and convictions. All these data were skewed and did not enter the analysis, therefore we have presented them in Analysis 1.18. Data on booking and jail days suggested a trend favouring the ICM group, reducing contacts with legal system. Data on convictions did not show a trend favouring one group over the other.

1.19 Social functioning: 2. Employment status (various measurements)

We found six relevant studies for this outcome and categorised data into six subgroups, according to different time to follow‐up and various outcomes described.

1.19.1 By medium term ‐ not competitively employed at the end of the trial

One study reported data on 'not competitively employed at the end of the trial', and these data did not show a significant advantage for ICM in improving the number of competitively employed people (n = 88, 1 RCT, RR 1.00, 95% CI 0.91 to 1.10; Analysis 1.19).

1.19.2 By medium term ‐ not employed at the end of the trial

We found four trials relevant to this subgroup; these also failed to show a significant difference between groups, although data did suggest a trend favouring ICM (n = 1136, 4 RCTs, RR 0.89, 95% CI 0.79 to 1.0). However, heterogeneity was present (Chi² = 11.86; df = 3.0; P = 0.01; I² = 74%; Analysis 1.19).

1.19.3 By long term ‐ not employed at the end of the trial

We found four trials relevant to this subgroup. As in the medium‐term comparison, data failed to show a significant difference, although they suggested a trend favouring the ICM group (n = 1129, 4 RCTs, RR 0.70, 95% CI 0.49 to 1.00). However, again, there was considerable heterogeneity (Chi² = 46.48; df = 3.0; P = 0.0; I² = 93%; Analysis 1.19).

1.19.4 By long term ‐ not working/studying in the previous year

We found one trial relevant to this subgroup, with a total of 547 participants. Data showed a significant difference between groups, favouring ICM in reducing the risk of being 'not working or not studying' compared to standard care (n = 547, 1 RCT, RR 0.86, 95% CI 0.74 to 0.99). Heterogeneity for this outcome was high (Chi² = 0.0; df = 0.0; P = 0.0; I² = 100%; Analysis 1.19).

1.19.5 By medium‐term FUP (3 years) ‐ not working/studying in the previous year

There was a single trial in this subgroup. Data failed to show a signiifcant difference between groups (n = 547, 1 RCT, RR 1.02, 95% CI 0.9 to 1.16; Analysis 1.19).

1.19.6 By long‐term FUP (8 years) ‐ not working/studying in the previous year

We found one trial relevant to this subgroup. Again, we did not find evidence of a clear difference between the two treatments (n = 547, 1 RCT, RR 0.99, 95% CI 0.88 to 1.11; Analysis 1.19).

1.20 Social functioning: 3a. Accommodation status (various measurements)

We found 10 relevant studies for this outcome, the data from which we divided into six subgroups according to different time to follow‐up and various outcomes described.

1.20.1 By short term ‐ homelessness

We found data on the outcome 'homelessness' from one short‐term study. This small study revealed a significant reduction in the rate of homelessness in the ICM group (n = 95, 1 RCT, RR 0.04, 95% CI 0.00 to 0.70; Analysis 1.20).

1.20.2 By medium term ‐ homelessness

Medium‐term data on the outcome 'homelessness' were available from one small study. Data did not reveal any significant difference between groups in the rate of homelessness by the medium term (n = 88, 1 RCT, RR 0.32, 95% CI 0.03 to 2.95; Analysis 1.20).

1.20.3 By medium term ‐ not living independently

We found five trials relevant to this subgroup. These data showed a significant advantage for ICM in reducing the number of participants not living independently (n = 1303, 5 RCTs, RR 0.80, 95% CI 0.66 to 0.97). Heterogeneity for this subgroup was moderately high (Chi² = 5.81; df = 4.0; P = 0.21; I² = 31%; Analysis 1.20).

1.20.4 By long term ‐ homelessness

We found 'homelessness' data in three long‐term studies, which did not reveal any significant difference between intervention groups (n = 418, 3 RCTs, RR 0.78, 95% CI 0.34 to 1.82). This subgroup had moderate levels of heterogeneity (Chi² = 3.27; df = 2.0; P = 0.19; I² = 38%; Analysis 1.20).

1.20.5 By long term ‐ not living independently

There were four relevant trials in this subgroup. Data for this subgroup favoured the ICM group, where the incidence of not living independently was lower compared with the standard care group (n = 1185, 4 RCTs, RR 0.65, 95% CI 0.49 to 0.88). This subgroup had moderate levels of heterogeneity (Chi² = 5.39; df = 3.0; P = 0.15; I² = 44%; Analysis 1.20).

1.20.6 By long term ‐ not living in stable accommodation

The outcome 'not living in stable accommodation' was only available from one study. We found that data favoured the ICM group in reducing the number of participants not living in stable accommodation (n = 168, 1 RCT, RR 0.80, 95% CI 0.65 to 0.98; Analysis 1.20).

1.21 Social functioning: 3b. Accomodation status: mean number of days in supported house (skewed data, sample size ≧ 200)

We identified only one study relevant to this outcome, providing data at different times, but always referring to the previous year of follow‐up. Note that for this outcome the right graph label favours ICM (experimental group).

1.21.1 By long term (previous year)

There was no significant difference between ICM and standard care within this subgroup (n = 547, 1 RCT, MD 0.3, 95% CI ‐13.98 to 14.58; Analysis 1.21).

1.21.2 By medium‐term FUP (3 years) (previous year)

We found evidence of a significant difference between ICM and standard care within this subgroup, with data favouring standard care over ICM (n = 547, 1 RCT, MD ‐22.2, 95% CI ‐38.47 to ‐5.93; Analysis 1.21).

1.21.3 By long‐term FUP (8 years) (previous year)

We did not find evidence of a significant difference between the two treatments for this subgroup (n = 547, 1 RCT, MD ‐6.7, 95% CI ‐19.35 to 5.95; Analysis 1.21).

1.22 Social functioning: 3c. Accommodation status (various measurements, skewed data)

Data on this outcome were available from three studies; as all of these data were skewed and did not enter the analysis, we have presented them in Analysis 1.22.

Two studies provided medium‐term data on 'average days in stable accommodation', which showed a trend favouring the ICM group, consistent with results previously described for 'not living in stable accommodation' by long term.

Two studies provided long‐term data on 'average days per month in sheltered homes'. These data were equivocal, as data from one study favoured ICM, whilst data from the second study favoured the standard care group.

1.23 Social functioning: 4a. Substance abuse

We identified only one study relevant to this outcome, providing various measures at different time of follow‐up.

1.23.1 Alcohol abuse ‐ by long term

We found one trial relevant to this subgroup (total n = 547). There was no significant difference between ICM and standard care within this subgroup (n = 547, 1 RCT, RR 0.55, 95% CI 0.26 to 1.17; Analysis 1.23).

1.23.2 Illicit drug abuse ‐ by long term

As for alcohol abuse, data failed to show a significant difference between groups for this subgroup (n = 547, 1 RCT, RR 0.96, 95% CI 0.63 to 1.47; Analysis 1.23).

1.23.3 Substance abuse ‐ by medium‐term FUP (3 years)

We found one trial relevant to this subgroup (total n = 547). As for the two previous outcomes, data failed to show a significant difference between groups (RR 0.91, 95% CI 0.67 to 1.24; Analysis 1.23).

1.24 Social functioning: 4b. Substance abuse (Dartmouth Assessment of Lifestyle Interview (DALI), skewness not detectable) ‐ by medium term

We were unable to enter two types of data into the analysis: medium‐term data assessing alcohol and drug abuse on DALI scale. As the DALI scale averages values from positive to negative, skew is very difficult to detect, we did not enter these data in the analysis. These data tended to favour the standard care group for both the alcohol and drug abuse outcomes; we have presented them in Analysis 1.25.

1.25 Social functioning: 4c. Substance abuse ‐ days used per month (skewed data)

One study provided medium‐ and long‐term data from the outcome 'days of substance use per month', which were equivocal, but skewed; we have therefore presented them in Analysis 1.25.

1.26 Social functioning: 5a. Average endpoint score (various scales)

Three studies provided data from five different scales (Disability Assessment Scale (DAS), Interview Schedule for Social Interaction (ISSI), Role Functioning Scale (RFS), Social Adjustment Scale (SAS‐adapted version), Strauss‐Carpenter Outcome Scale) assessing social functioning by short, medium, and long term. As no more than one study used the same scale per time period, we did not enter more than one study per subgroup. Data from each available time period failed to show any significant difference between treatment groups, with the exception of two outcomes by long term (1.26.7 on ISSI and 1.26.8 on RFS), the first favouring the standard care group, and the second favouring the ICM group.

1.26.1 By short term (RFS, low = poor)

One trial providing data, no significant difference between groups (n = 80, 1 RCT, MD ‐0.62, 95% CI ‐2.23 to 0.99; Analysis 1.26).

1.26.2 By short term (SAS‐adapted version, low = poor)

One trial providing data, no significant difference between groups (n = 80, 1 RCT, MD ‐3.34, 95% CI ‐7.55 to 0.87; Analysis 1.26).

1.26.3 By medium term ‐ social role performance (DAS, high = poor)

One trial providing data, no significant difference between groups (n = 55, 1 RCT, MD 0.1, 95% CI ‐0.4 to 0.6; Analysis 1.26).

1.26.4 By medium term (RFS, low = poor)

One trial providing data, no significant difference between groups (n = 80, 1 RCT, MD ‐0.86, 95% CI ‐2.72 to 1.0; Analysis 1.26).

1.26.5 By medium term (SAS‐adapted version, low = poor)

One trial providing data, no significant difference between groups (n = 80, 1 RCT, MD ‐3.3, 95% CI ‐7.83 to 1.23; Analysis 1.26).

1.26.6 By long term ‐ social role performance (DAS, high = poor)

One trial providing data, no significant difference between groups (n = 58, 1 RCT, MD ‐0.2, 95% CI ‐0.67 to 0.27; Analysis 1.26).

1.26.7 By long term (ISSI, low = poor)

One trial providing data, showing a significant difference between groups favouring standard careover ICM (n = 62, 1 RCT, MD 3.2, 95% CI 0.11 to 6.29; Analysis 1.26).

1.26.8 By long term (RFS, low = poor)

One trial providing data, showing a significant difference between groups favouring ICM over standard care (n = 80, 1 RCT, MD ‐2.35, 95% CI ‐4.05 to ‐0.65; Analysis 1.26).

1.26.9 By long term (SAS‐adapted version, low = poor)

One trial providing data, no significant difference between groups (n = 80, 1 RCT, MD ‐2.75, 95% CI ‐7.13 to 1.63; Analysis 1.26).

1.26.10 By long term (Strauss‐Carpenter Outcome Scale, low = poor)

One trial providing data, no significant difference between groups (n = 60, 1 RCT, MD 0.1, 95% CI ‐1.17 to 1.37; Analysis 1.26).

1.27 Social functioning: 5b. Average endpoint score (various scales, skewed data)

Skewed data on SAS score were available by short, medium, and long term from two studies. These data were equivocal and not consistent between the two studies. One study provided long‐term data on REHAB Scale score, which tended to favour the ICM group, but were also skewed. We have presented them in Analysis 1.27.

1.28 Mental state: 1a. General symptoms ‐ average endpoint score (various scales)

Two sets of data were available: i.) non‐skewed data or skewed data from a sample size greater than or equal to 200 participants per study: entering analysis together; and ii.) skewed data: not entering analysis.

We identified four studies relevant to this outcome entering analysis, providing data from three different scales (Brief Psychiatric Rating Scale (BPRS‐18 items), Brief Symptom Inventory (BSI), Colorado Symptom Index (CSI)) per different time period.

1.28.1 By short term (BPRS‐18 items, high = poor)

We found two trials relevant to this subgroup. There was no significant difference between ICM and standard care within this subgroup (n = 668, 2 RCTs, MD ‐1.56, 95% CI ‐6.85 to 3.73), but data were heterogeneous (Chi² = 12.58; df = 1.0; P = 0.0; I² = 92%; Analysis 1.28).

1.28.2 By short term (BSI, high = poor)

Short‐term mental state scores assessed with the BSI were available from the same two studies providing BPRS short‐term data. Again, data were not significantly different (n = 668, 2 RCTs, MD ‐0.06, 95% CI ‐0.19 to 0.06; Analysis 1.28); however, different from the BPRS results, these data were homogeneous.

1.28.3 By short term (CSI, low = poor)

One further trial was available providing short‐term data on mental state assessed with the CSI. These data showed a significant difference between groups favouring the ICM group (n = 125, 1 RCT, MD ‐0.56, 95% CI ‐0.84 to ‐0.28; Analysis 1.28).

1.28.4 By medium term (BPRS‐18 items, high = poor)

We found two trials relevant to this subgroup, with a total of 662 people. We did not find evidence of a significant difference between ICM and standard care within this subgroup (MD ‐0.96, 95% CI ‐2.42 to 0.51; Analysis 1.28).

1.28.5 By medium term (BSI, high = poor)

We found two trials relevant to this subgroup (total n = 662). We did not find evidence of a significant difference between ICM and standard care within this subgroup (MD ‐0.02, 95% CI ‐0.15 to 0.1; Analysis 1.28).

1.28.6 By medium term (CSI, low = poor)

There was a single trial in this subgroup. We found evidence favouring the ICM group over the standard care group (n = 125, 1 RCT, MD ‐0.35, 95% CI ‐0.65 to ‐0.05; Analysis 1.28).

1.28.7 By long term (BPRS‐18 items, high = poor)

We found three trials relevant to this subgroup (total n = 777). There was no significant difference between ICM and standard care within this subgroup (n = 777, 3 RCTs, MD ‐1.48, 95% CI ‐3.69 to 0.74), but data were heterogeneous (Chi² = 13.13; df = 2.0; P = 0.0; I² = 84%; Analysis 1.28).

1.28.8 By long term (BSI, high = poor)

We found two trials relevant to this subgroup, with data favouring the ICM group, in which participants reached a better mental state score by long term compared to the standard care group (n = 647, 2 RCTs, MD ‐0.18, 95% CI ‐0.31 to ‐0.06; Analysis 1.28).

1.29 Mental state: 1b. General symptoms ‐ mean change from baseline (CSI, low = poor ) ‐ by long term

Finally, one study assessed the mean change from baseline on the CSI, with data showing no difference between groups (n = 168, 1 RCT, MD ‐0.32, 95% CI ‐0.53 to ‐0.11).

1.30 Mental state: 1c. General symptoms ‐ average endpoint score (various scales, skewed data)

Skewed data were available from six studies assessing mental state by different time periods and with different scales (BPRS‐18 items, BPRS‐24 items, CPRS, PSE, SCL‐90). Data failed to show a significant trend favouring one group over the other, this report being consistent across different studies and different rating scales. We considered these data to not be robust as they were skewed, but they were in accordance with short‐, medium‐, and long‐term data. We have presented them in Analysis 1.30.

1.31 Mental state: 2a. Specific symptoms ‐ depression at follow‐up interview

We found data on specific symptoms from only one study, which provided data on depression incidence per different time period. There was no significant difference between groups by medium term, long term, and medium‐term follow‐up (three years).

1.31.1 By medium term

One trial providing data, no significant difference between groups (n = 547, 1 RCT, RR 0.77, 95% CI 0.56 to 1.04; Analysis 1.31).

1.31.2 By long term

One trial providing data, no significant difference between groups (n = 547, 1 RCT, RR 0.83, 95% CI 0.57 to 1.21; Analysis 1.31).

1.31.3 By medium term FUP (3 years)

One trial providing data, no significant difference between groups (n = 547, 1 RCT, RR 1.25, 95% CI 0.91 to 1.72; Analysis 1.31).

1.32 Mental state: 2b. Specific symptoms ‐ average endpoint score (various scales, skewed data, sample size ≧ 200)

We identified only one study (n = 547) relevant to this outcome, providing data on two different dimensions (positive and negative symptoms) assessed at different times. Although skewed, data entered the analyses, as the sample size was greater than or equal to 200 participants per study. Only one comparison showed a significant advantage for the ICM group, in reducing risk of negative symptoms by long term.

1.32.1 By long term ‐ positive symptoms (Scale for the Assessment of Positive Symptoms (SAPS), high = poor)

One trial providing data, showing no significant difference between groups, although tending to favour the ICM group (n = 547, 1 RCT, MD ‐0.22, 95% CI ‐0.45 to 0.01; Analysis 1.32).

1.32.2 By long term ‐ negative symptoms (Scale for the Assessment of Negative Symptoms (SANS), high = poor)

One trial providing data, showing evidence of a significant difference between groups favouring ICM over standard care (n = 547, 1 RCT, MD ‐0.42, 95% CI ‐0.62 to ‐0.22; Analysis 1.32).

1.32.3 By medium‐term FUP (3 years) ‐ positive symptoms (SAPS, high = poor)

One trial providing data, no significant difference between groups (n = 547, 1 RCT, MD 0.12, 95% CI ‐0.15 to 0.39; Analysis 1.32).

1.32.4 By medium‐term FUP (3 years) ‐ negative symptoms (SANS, high = poor)

One trial providing data, no significant difference between groups (n = 547, 1 RCT, MD ‐0.1, 95% CI ‐0.33 to 0.13; Analysis 1.32).

1.32.5 By long‐term FUP (8 years) ‐ positive symptoms (SAPS, high = poor)

One trial providing data, no significant difference between groups (n = 547, 1 RCT, MD 0.03, 95% CI ‐0.21 to 0.27; Analysis 1.32).

1.32.6 By long‐term FUP (8 years) ‐ negative symptoms (SANS, high = poor)

One trial providing data, no significant difference between groups (n = 547, 1 RCT, MD 0.06, 95% CI ‐0.13 to 0.25; Analysis 1.32).

1.33 Mental state: 2c. Specific symptoms ‐ average endpoint score (various scales, skewed data)

We found a second small study (n = 70) providing skewed data on depression symptoms assessed with the Beck Depression Inventory, and negative symptoms assessed with SANS. Neither data set was entered into the analysis. Skewed depression scores favoured the ICM group at medium and long term, whilst skewed negative symptoms scores by medium and long term were equivocal. We have reported these data in Analysis 1.33.

1.34 Behaviour: 1. Specific behaviour ‐ self harm

We found three relevant studies for this outcome providing data per different time period. We detected no significant difference between ICM and standard care in either reducing the risk for self harm or reducing the risk for attempting suicide.

1.34.1 By medium term

There were two relevant trials in this subgroup, in which 30 events occurred in 312 people treated with ICM compared with 30 events in 308 people treated with standard care. There was no significant difference in number of participants who committed self harm between the groups (n = 620, 2 RCTs, RR 0.99, 95% CI 0.61 to 1.59; Analysis 1.34).

1.34.2 By long term

There is a single relevant trial in this subgroup, in which 2 events occurred in 63 people treated with ICM compared with 2 events in 60 people treated with standard care. There was no significant difference in number of participants who committed self harm between the groups (n = 123, 1 RCT, RR 0.95, 95% CI 0.14 to 6.55; Analysis 1.34).

1.34.3 Attempted suicide ‐ by long term (during last 12 months)

Long term data on self harm during the previous 12 months available from a single study confirmed medium‐ and long‐term data (n = 547, 1 RCT, RR 0.81, 95% CI 0.47 to 1.38; Analysis 1.34), that is failing to show any significant difference between the two groups.

1.34.4 Attempted suicide ‐ by medium‐term FUP (during last 3 years)

The above data were again confirmed, from medium‐term follow‐up data on suicide attempts during the previous three years available from a single study. There was no significant difference between ICM and standard care (n = 547, 1 RCT, RR 0.95, 95% CI 0.56 to 1.62; Analysis 1.34).

1.35 Behaviour: 2. Social behaviour ‐ average endpoint score (Social Behaviour Scale, high = poor)

Skewed data were available from one small study (n = 70) assessing behaviour with the Social Behaviour Scale by medium and long term. These data tended to favour the ICM group.

1.36 Quality of life: 1a. Average endpoint score (various scales)

We found seven studies assessing quality of life with various scales by different time periods, and categorised data into five subgroups. Note that for this outcome the right graph label favours ICM (experimental group).

1.36.1 By short term ‐ general well‐being (Lehman's Quality of Life Interview (QOLI), high = better)

The only significant result we found was by short term: data were available from a single study and showed a significantly higher quality of life in the ICM group as assessed on the QOLI general well‐being subscale (n = 125, 1 RCT, MD 0.53, 95% CI 0.09 to 0.97; Analysis 1.36).

1.36.2 By medium term (Lancashire Quality of Life Profile (LQoLP), high = better)

Medium‐term data assessing quality of life with LQoLP (one study) did not show a significant difference between groups (n = 52, 1 RCT, MD 0.09, 95% CI ‐0.60 to 0.78; Analysis 1.36).

1.36.3 By medium term (Manchester Short Assessment of Quality of Life (MANSA) ‐ range 1 to 7, high = better)

Medium‐term data assessing quality of life with MANSA (one study) did not show a significant difference between groups (n = 81, 1 RCT, MD 0.20, 95% CI ‐0.29 to 0.69; Analysis 1.36).

1.36.4 By long term (LQoLP, high = better)

As with the medium‐term data, long‐term data assessing quality of life with LQoLP (three studies) did not show a significant difference between groups (n = 274, 3 RCTs, MD ‐0.13, 95% CI ‐0.38 to 0.12; Analysis 1.36).

1.36.5 By long term (QOLI, high = better)

Again, as with the medium‐term data, long‐term data assessing quality of life with QOLI (two studies) did not show a significant difference between groups (n = 132, 2 RCTs, MD 0.09, 95% CI ‐0.24 to 0.42; Analysis 1.36).

1.37 Quality of life: 1b. Mean change from baseline (QOLI, high = better, skewed data) ‐ by long term

We found one further study providing data by long term for this comparison, but as data were skewed, measuring mean change from baseline on the QOLI, the study was not entered into the analysis. These data tended to favour the ICM group. We have reported these data in Analysis 1.37.

1.38 Participant satisfaction: 1a. Average endpoint score (Client Satisfaction Questionnaire (CSQ), high = better)

We found three relevant studies for this outcome, providing data per different time period. We found that participant satisfaction assessed with the CSQ was significantly greater in the ICM group compared with the standard care group in all three time period assessments. Note that for this outcome the right graph label favours ICM (experimental group).

1.38.1 By short term

Short‐term data were available from only one small study and showed a significant difference between groups, favouring the ICM intervention (n = 61, 1 RCT, MD 6.2, 95% CI 2.6 to 9.8; Analysis 1.38).

1.38.2 By medium term

Medium‐term data from two studies confirmed the above results (n = 500, 2 RCTs, MD 1.93, 95% CI 0.86 to 3.01; Analysis 1.38).

1.38.3 By long term

Long‐term data also favoured the ICM group (n = 423, 2 RCTs, MD 3.23, 95% CI 2.31 to 4.14; Analysis 1.38).

1.39 Participant satisfaction: 1b. Average endpoint score (CSQ, high = better, skewed data) ‐ by short term

One further small trial provided short‐term data, but the data were skewed: attrition in the standard care arm was higher than 50%. Participant satisfaction was assessed with the CSQ, and it tended to favour the ICM group. We have reported these data in Analysis 1.39.

1.40 Participant need: 1. Average endpoint score (various scales, skewed data)

We found more skewed data from two studies assessing participant need on two other scales (Camberwell Assessment of Need Interview (CAN), Camberwell Assessment of Need Short Appraisal Schedule (CANSAS)). Medium‐term data from one study assessed on CANSAS failed to show any difference between groups. Long‐term data assessed in one study with the CAN tended to favour the ICM group. We have reported these data in Analysis 1.40.

1.41 Costs: 1a. Direct costs of psychiatric hospital care ‐ by medium term (unit cost = USD, fiscal year 1990)

Direct medium‐term costs of psychiatric hospital care were available from two studies reporting skewed data, but with a sample size greater than 200 (Chandler‐California1 (A); Chandler‐California1 (B)). Data favoured ICM (n = 426, 2 RCTs, MD USD ‐143.74, 95% CI ‐272.40 to ‐15.08; Analysis 1.41).

1.42 Costs: 1b. Direct costs of psychiatric hospital care ‐ skewed data

Five additional studies did describe 'direct costs of psychiatric hospital care', but data were markedly skewed. Some of these data showed a trend favouring ICM, while some favoured standard care, therefore we could not highlight any trend confirming the findings from meta‐analysis. We have presented these data in Analysis 1.42.

1.43 Costs: 2a. Direct healthcare costs ‐ by long term (unit cost = USD, fiscal year 1988)

Long‐term data for direct healthcare cost were available from two studies; again studies reported skewed data, but with a sample size greater than 200, and so these data could be entered into a meta‐analysis. These data were inconclusive (n = 873, 2 RCTs, MD USD ‐529.24, 95% CI ‐2143.59 to 1085.1; Analysis 1.43), as they were highly heterogeneous (Chi² = 17.83; df = 1.0; P = 0.0; I² = 94%) with inconsistency in direction of effect.

1.44 Costs: 2b. Direct healthcare costs ‐ skewed data

Other skewed data from two studies with a sample size of less than 200 could not be entered into the meta‐analysis. These studies assessed direct healthcare costs by medium term (one study) and by short‐term follow‐up (the other study). Medium‐term data did not show any significant difference between interventions, whilst short‐term FUP data seemed to favour standard care in reducing direct healthcare costs. We have reported these data in Analysis 1.44.

1.45 Costs: 3. Direct costs ‐ other data ‐ skewed data

Five studies described direct costs for "all care" by short, medium, and long term, and one more study described direct costs for "specific" outcome (outpatient care and prison) by medium term. As these data were skewed and from studies with a sample size of less than 200, they could not be entered into the meta‐analysis. We have presented these data in Analysis 1.45.

Costs for all care by short term seemed to favour the ICM group, where costs were reduced (one study); by medium term one study favoured standard care (where costs were reduced), whilst the other study failed to show any difference between the two groups; and five studies provided data by long term: some of these data showed a trend favouring ICM, and some favoured standard care; these data were therefore inconclusive, and we could not highlight any trend.

Medium‐term data from one study on cost for outpatient care showed costs were higher in the ICM group compared to standard care. Data from the same study on cost for prison showed costs were higher for the standard care group compared to ICM.

2. COMPARISON 2: INTENSIVE CASE MANAGEMENT versus NON‐INTENSIVE CASE MANAGEMENT

summary of findings Table 2. This comparison has 36 outcomes.

2.1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months

We found 21 relevant studies for this outcome and categorised data into two subgroups (skewed data but with a sample size greater than or equal to 200, and skewed data with a sample size of less than 200). Overall, combining the two pools of studies, we found no clear difference between ICM and non‐ICM (n = 2220, 21 RCTs, MD ‐0.08, 95% CI ‐0.37 to 0.21, Figure 10). A funnel plot did not show any significant reporting bias (Figure 11).

Figure 10

Forest plot of comparison: 2 Intensive Case Management versus non‐Intensive Case Management, outcome: 2.1 Service use: 1. Average number of days in hospital per month ‐ at about 24 months.

Figure 11

Funnel plot of comparison: 2 Intensive Case Management versus non‐Intensive Case Management, outcome: 2.1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months.

2.1.1 Skewed data (sample size ≧ 200)

We found three studies reporting skewed data but with a sample size greater than or equal to 200. There was no significant difference between groups for reducing length of hospitalisation (n = 694, 3 RCTs, MD ‐0.58, 95% CI ‐1.93 to 0.76; Analysis 2.1). These findings were in accordance with the second subgroup analysis, skewed data from studies with a sample size of less than 200 participants.

2.1.2 Skewed data (sample size < 200)

There were 18 relevant trials in this subgroup, with a total of 1526 people. Again, these data did not show a significant difference between groups (n = 1526, 18 RCTs, MD ‐0.03, 95% CI ‐0.33 to 0.28; Analysis 2.1).

2.2 Service use: 1a. Average number of days in hospital per month ‐ by medium/long‐term follow‐up (skewed data, sample size ≧ 200)

One study provided data by medium‐ and long‐term follow‐up, and as the sample size was greater than 200, skewed data entered the analysis.

2.2.1 By medium‐term FUP (18 months)

One study provided data on medium‐term follow‐up (18 months), confirming the results described above by 24 months. There was no clear difference between ICM and non‐ICM within this subgroup (n = 237, 1 RCTs, MD 0.6, 95% CI ‐1.25 to 2.45; Analysis 2.2).

2.2.2 By long‐term FUP (8.5 years)

The same study provided data on long‐term follow‐up (8.5 years), again showing no significant differences between interventions in reducing number of days in hospital (n = 203, 1 RCTs, MD 0.8, 95% CI ‐1.47 to 3.07; Analysis 2.2).

2.3 Service use: 2. Not remaining in contact with psychiatric services

We found four relevant studies for this outcome and categorised data into three subgroups, by different time period. Short‐term data were not available. When we pooled studies from medium and long term, data did not show significant differences between interventions, but heterogeneity was substantial (n = 1255, 4 RCTs, RR 0.63, 95% CI 0.27 to 1.49, I² = 81%, P = 0.001). We addressed this finding by checking again for correctness of data and explored heterogeneity by dropping each study out of the analysis. Only by removing both Drake‐NHamp 1998 and UK700‐UK 1999 was homogeneity restored. As we could not ascertain any clear reason for the heterogeneity, we have therefore chosen not to pool these data, as it could be misleading to quote an average value for the intervention effect ‐ particularly in this case, when there is inconsistency in direction of effect. We did not use a funnel plot for this outcome because there were fewer than 10 studies (see Assessment of reporting biases).

2.3.1 By medium term

One small study provided medium‐term data showing a significant difference favouring the ICM group (n = 73, 1 RCT, RR 0.27, 95% CI 0.08 to 0.87; Analysis 2.3).

2.3.2 By long term

Long‐term data were available from three studies. Pooled data were not statistically significant (n = 1182, 3 RCTs, RR 0.82, 95% CI 0.34 to 1.98), but data were heterogeneous (Chi² = 10.86; df = 2.0; P = 0.0; I² = 81%) with inconsistency in direction of effect (Analysis 2.3).

2.3.3 By medium‐term FUP (18 months)

One single trial provided data for this outcome, and we found no evidence of a clear difference between the two treatments (n = 251, 1 RCT, RR 0.42, 95% CI 0.17 to 1.05; Analysis 2.3).

2.4 Service use: 3a. Admitted to hospital ‐ by long term

Binary data describing this outcome were available from three long‐term studies, reporting 'number of admitted to hospital'. These data showed a non‐significant difference in number of participants admitted to hospital between groups (n = 1132, 3 RCTs, RR 0.91, 95% CI 0.75 to 1.12; Analysis 2.4). Heterogeneity was high for this outcome (Chi² = 5.26; df = 2.0; P = 0.07; I² = 62%).

2.5 Service use: 3b. Average number of admissions (skewed data ‐ sample size ≧ 200)

We identified two studies relevant to this outcome, one providing data by long term, the other by medium‐ and long‐term follow‐up. Data were skewed, but as the trial sample size were greater than or equal to 200, we entered these data in the analysis. Data from different time periods failed to show any significant differences between ICM and non‐ICM in average number of admissions.

2.5.1 By long term (24 months)

Data from one long‐term study failed to show any significant differences between groups (n = 678, 1 RCT, MD ‐0.18, 95% CI ‐0.41 to 0.05; Analysis 2.5).

2.5.2 By medium term FUP (18 months)

One single study provided data by medium‐term follow‐up (18 months). Data failed to show any significant differences between groups (n = 237, 1 RCT, MD ‐0.1, 95% CI ‐0.6 to 0.4; Analysis 2.5).

2.5.3 By long‐term FUP (8.5 years)

One single study provided data by medium‐term follow‐up (8.5 years). Data failed to show any significant differences between groups (n = 203, 1 RCT, MD 1.0, 95% CI ‐0.25 to 2.25; Analysis 2.5).

2.6 Service use: 3c. Average number of admissions (skewed data) ‐ by medium term

A trial that included fewer than 200 participants presented skewed data that could not be entered into the meta‐analysis. Data were reported for the outcome 'average number of admissions' by medium term. As with previous findings, they failed to show any trend in effect between groups.

2.7 Adverse event: 1a. Death ‐ any cause

We found seven relevant studies for this outcome and categorised data into five subgroups, by different time period.

2.7.1 By short term

Short‐term data on mortality were available from one study (n = 193) reporting no deaths, therefore a measure of effect was not estimable (Analysis 2.7).

2.7.2 By medium term

Medium‐term data were available from three studies, where 1 death occurred in 148 people treated with ICM, compared with no deaths in 146 people treated with non‐ICM. There were no significant differences in mortality between groups (n = 294, 3 RCTs, RR 2.92, 95% CI 0.12 to 69.43; Analysis 2.7).

2.7.3 By long term

We found long‐term data in five studies, reporting 16 deaths occurring in 816 people treated with ICM, compared with 18 deaths in 821 people treated with non‐ICM. These data confirmed medium‐term findings showing no differences in mortality between groups (n = 1637, 5 RCTs, RR 0.90, 95% CI 0.46 to 1.75; Analysis 2.7).

2.7.4 By medium‐term FUP (18 months)

Medium‐term follow‐up data were available from one study, where 6 deaths occurred in 127 people treated with ICM, compared with 6 deaths in 124 people treated with non‐ICM. These data confirmed the above results, showing no differences in mortality between groups (n = 251, 1 RCT, RR 0.98, 95% CI 0.32 to 2.95; Analysis 2.7).

2.7.5 By long‐term FUP (8.5 years)

Long‐term follow‐up data were available from one study, where 20 deaths occurred in 127 people treated with ICM, compared with 17 deaths in 124 people treated with non‐ICM. These data confirmed the above results, showing no differences in mortality between groups (n = 251, 1 RCT, RR 1.15, 95% CI 0.63 to 2.09; Analysis 2.7).

2.8 Adverse event: 1b. Death ‐ suicide

We found eight relevant studies for this outcome and categorised data into four subgroups, by different time period.

2.8.1 By short term

Short‐term data on suicide mortality were available from one study (n = 193), again reporting no deaths, and therefore a measure of effect was not estimable.

2.8.2 By medium term

Medium‐term data were available from six studies, where 5 suicides occurred in 464 people treated with ICM, compared with 3 suicides in 465 people treated with non‐ICM. There were no significant differences in the suicide rate between groups (n = 929, 6 RCTs, RR 1.61, 95% CI 0.26 to 9.85; Analysis 2.8).

2.8.3 By long term

Long‐term data were available from three studies, reporting 6 suicides occurring in 577 people treated with ICM, compared with 7 suicides in 575 people treated with non‐ICM. These data confirmed medium‐term data on overall mortality and on suicide, with no significant differences in the suicide rate between groups (n = 1152, 3 RCTs, RR 0.88, 95% CI 0.27 to 2.84; Analysis 2.8).

2.8.4 By medium‐term FUP (18 months)

Medium‐term follow‐up data on suicide mortality were available from one study, reporting 1 suicide occurring in 127 people treated with ICM, compared with 3 suicides in 124 people treated with non‐ICM. These data confirmed also in the follow‐up period, medium‐ and long‐term data on overall mortality and on suicide, with no significant differences in the suicide rate between groups (n = 251, 1 RCT, RR 0.33, 95% CI 0.03 to 3.09; Analysis 2.8).

2.9 Global state: 1. Leaving the study early

We found nine studies providing medium‐ and long‐term data, but not short‐term data. When pooling data from sub‐groups for two time periods, we found data still significant, showing an advantage for ICM in reducing the number of participants lost to follow‐up (n = 2195, 9 RCTs, RR 0.72, 95% CI 0.52 to 0.99). Although heterogeneity was reduced in comparison to the medium‐term subgroup data, heterogeneity was still substantial (Chi² = 19.58; df = 8.0; P = 0.01; I² = 59%) (Analysis 2.9).

2.9.1 By medium term

Medium‐term data from two trials showed no treatment effect in reducing number of participants lost to follow‐up (n = 225, 2 RCTs, RR 0.64, 95% CI 0.13 to 3.07), but these data presented a substantial level of heterogeneity (Chi² = 6.38; df = 1.0; P = 0.01; I² = 84%). In addition, there was inconsistency in the direction of effect between the two studies (Analysis 2.9).

2.9.2 By long term

Long‐term data were available from seven studies, and we found a significant advantage for ICM in reducing the number of participants lost to follow‐up (n = 1970, 7 RCTs, RR 0.70, 95% CI 0.52 to 0.95). Different from the medium‐term data, long‐term subgroup analyses did not show substantial heterogeneity (Chi² = 9.88; df = 6.0; P = 0.13; I² = 39%) (Analysis 2.9).

2.10 Global state: 2a. Average endpoint score (Health of the Nation Outcome Scale (HoNOS), high = poor) ‐ by long term

One study reported long‐term data on global state assessed with HoNOS. These data were skewed, but as the study sample size was greater than or equal to 200 participants they entered the analysis. Data failed to show a significant difference between interventions (n = 239, 1 RCT, MD ‐0.40, 95% CI ‐1.77 to 0.97; Analysis 2.10).

2.11 Global state: 2b. Average endpoint score (HoNOS, high = poor) ‐ skewed data

We found skewed data describing global state with HoNOS by medium and long term from one trial. These data tended to favour the standard care group (Analysis 2.11).

2.12 Global state: 3a. Not compliant with medication ‐ by medium term

One study reported medium‐term data for the binary outcome 'number of participants not compliant with medication'. There was no significant difference between groups (n = 73, 1 RCT, RR 1.14, 95% CI 0.42 to 3.05; Analysis 2.12).

2.13 Global state: 3b. Compliance with medication ‐ average endpoint subscale score (Rating of Medication Influences (ROMI)) ‐ by long term

Long‐term compliance scores assessed with the ROMI compliance and non‐compliance subscales were not significantly different (compliance subscale: n = 239, 1 RCT, MD 0.60, 95% CI ‐0.05 to 1.25; non‐compliance subscale: n = 239, 1 RCT, MD ‐0.60, 95% CI ‐1.63 to 0.43), although both subscale scores tended to favour ICM. Note that for the compliance subscale (high = good), the right side of the graph favours experimental (ICM).

2.13.1 Compliance subscale (high = good)

There was a single trial in this subgroup, with a total of 239 people. There was no clear difference between ICM and non‐ICM within this subgroup (MD 0.6, 95% CI ‐0.05 to 1.25; Analysis 2.13).

2.13.2 Non‐compliance subscale (high = poor)

There was a single trial in this subgroup, with a total of 239 people. There was no clear difference between ICM and non‐ICM within this subgroup (MD ‐0.6, 95% CI ‐1.63 to 0.43). This subgroup had important levels of heterogeneity (Chi² = 0.0; df = 0.0; P = 0.0; I² = 100%) (Analysis 2.13).

2.14 Global state: 3c. Compliance with medication ‐ average endpoint subscale score (ROMI, score 1 to 3, skewed data)

Medium‐ and long‐term skewed data were available from one study assessing compliance scores with the ROMI compliance and non‐compliance subscales. These data tended to favour standard care, where participants had a higher level of compliance (Analysis 2.14).

2.15 Social functioning: 1. Contact with legal system (various measurements)

We identified three studies relevant to this outcome, providing data on different time periods and different measures.

2.15.1 By medium term ‐ contact with the police

Medium‐term data were available from one study reporting 'contact with the police'. We found no significant difference between groups (n = 73, 1 RCT, RR 0.32, 95% CI 0.04 to 2.97; Analysis 2.15).

2.15.2 By long term ‐ imprisoned

Long‐term data were available on both binary outcomes of 'imprisoned' and 'arrested'. We found two studies reporting the outcome 'imprisoned', and data failed to show any difference in the number of people imprisoned (n = 959, 2 RCTs, RR 1.15, 95% CI 0.64 to 2.08; Analysis 2.15).

2.15.3 By long term ‐ arrested

Only one study provided data on the outcome 'arrested'. Again, there was no clear difference between groups (n = 251, 1 RCT, RR 0.87, 95% CI 0.53 to 1.42; Analysis 2.15).

2.15.4 By medium‐term FUP (18 months) ‐ imprisoned

We found one trial relevant to this subgroup, with a total of 251 participants. There was no clear difference between ICM and non‐ICM within this subgroup (n = 251, 1 RCT, RR 1.07, 95% CI 0.47 to 2.44; Analysis 2.15).

2.15.5 By long‐term FUP (8.5 years) ‐ imprisoned

There was a single trial in this subgroup. There was no clear difference between ICM and non‐ICM within this subgroup (n = 214, 1 RCT, RR 0.70, 95% CI 0.43 to 1.14; Analysis 2.15).

2.16 Social functioning 2. Employment status (various measurements)

We identified two studies relevant to this outcome: one medium‐term study, reporting 'participant who spent more than one day employed' and 'participants on paid employment', and one long‐term follow‐up study, reporting 'unemployed'. Both medium‐term outcomes showed no significant difference between groups. The long‐term follow‐up outcome confirmed medium‐term data, showing no significant difference between ICM and non‐ICM in employment status.

2.16.1 Spent > 1 day employed ‐ by medium term

There was no significant difference between ICM and non‐ICM in increasing number of days of employment (n = 73, 1 RCT, RR 1.46, 95% CI 0.45 to 4.74; Analysis 2.16).

2.16.2 On paid employment ‐ by medium term

There was no significant difference between ICM and non‐ICM in increasing the chance of being on paid employment by medium term (n = 73, 1 RCT, RR 0.97, 95% CI 0.14 to 6.54; Analysis 2.16).

2.16.3 Unemployed ‐ by long‐term FUP (8.5 years)

There was no significant difference between ICM and non‐ICM in decreasing the risk of unemployment (n = 214, 1 RCT, RR 1.10, 95% CI 0.91 to 1.34; Analysis 2.16).

2.17 Social functioning: 3a. Accommodation status (various measurements)

We found two relevant studies for this outcome, one providing data only for the outcome 'living in supported accomodation' by medium term. The second study provided data for all of the other outcomes, using different measures to assess accomodation status. The data failed to show significant differences between groups at any time period.

2.17.1 By medium term ‐ living in supported accommodation

The outcome 'living in supported accommodation' was only available for the medium term from one study. Data failed to show a significant difference between groups (n = 73, 1 RCT, RR 2.59, 95% CI 0.75 to 9.01; Analysis 2.17).

2.17.2 By long term ‐ homelessness

The outcome 'homelessness' was available for the long term from one study. There were no significant differences between treatment groups in the number of people who were homeless (n = 251, 1 RCT, RR 0.69, 95% CI 0.34 to 1.38; Analysis 2.17).

2.17.3 By medium‐term FUP (18 months) ‐ living independently

One study provided data and failed to show a significant difference between groups (n = 251, 1 RCT, RR 0.98, 95% CI 0.84 to 1.13; Analysis 2.17).

2.17.4 By medium‐term FUP (18 months) ‐ living in supported accomodation

One study provided data and failed to show a significant difference between groups (n = 251, 1 RCT, RR 0.83, 95% CI 0.38 to 1.77). This subgroup had important levels of heterogeneity (Chi² = 0.0; df = 0.0; P = 0.0; I² = 100%; Analysis 2.17).

2.17.5 By medium‐term FUP (18 months) ‐ homelessness

One study provided data and failed to show a significant difference between groups (n = 251, 1 RCT, RR 0.84, 95% CI 0.47 to 1.49). Heterogeneity was high for this outcome (Chi² = 0.0; df = 0.0; P = 0.0; I² = 100%; Analysis 2.17).

2.17.6 By long‐term FUP (8.5 years) ‐ living in supported accomodation

One study provided data and failed to show a significant difference between groups (n = 214, 1 RCT, RR 1.05, 95% CI 0.75 to 1.48; Analysis 2.17).

2.17.7 By long‐term FUP (8.5 years) ‐ homelessness

One study provided data and failed to show a significant difference between groups (n = 214, 1 RCT, RR 0.92, 95% CI 0.55 to 1.53; Analysis 2.17).

2.18 Social functioning: 3b. Accommodation status ‐ average days per month in stable accommodation

The continuous outcome 'average days per month in stable accommodation' was available for short, medium, and long term (2 RCTs). Data did not show significant differences between groups at any time period.

2.18.1 By short term

One trial was relevant to this subgroup. We did not find evidence of a clear difference between the two treatments (n = 203, 1 RCT, MD ‐0.2, 95% CI ‐2.48 to 2.08; Analysis 2.18).

2.18.2 By medium term

One trial was relevant to this subgroup. We did not find evidence of a clear difference between the two treatments (n = 203, 1 RCT, MD 0.1, 95% CI ‐2.15 to 2.35). This subgroup had important levels of heterogeneity (Chi² = 0.0; df = 0.0; P = 0.0; I² = 100%; Analysis 2.18).

2.18.3 By long term

We found two trials relevant to this subgroup (total n = 901). We did not find evidence of a clear difference between ICM and non‐ICM within this subgroup (n = 901, 2 RCTs, MD ‐0.19, 95% CI ‐1.37 to 1.0; Analysis 2.18).

2.19 Social functioning: 4a. Substance abuse ‐ by long term

We found two relevant studies for this outcome. Long‐term binary data on 'substance abuse' differentiated between 'alcohol abuse' and 'illicit drug abuse'. Data did not show significant differences between groups with any measures.

2.19.1 Alcohol abuse

There was no difference between ICM and non‐ICM in the number of people abusing alcohol (n = 251, 1 RCT, RR 1.10, 95% CI 0.67 to 1.83; Analysis 2.19).

2.19.2 Illicit drug abuse

There was no difference between ICM and non‐ICM in the number of people abusing illicit drugs (n = 251, 1 RCT, RR 1.08, 95% CI 0.69 to 1.71; Analysis 2.19).

2.19.3 Alcohol ‐ remission from alcohol use disorder (Alcohol Use Scale (AUS) score < 3)

Binary data on 'remission from alcohol use disorder' (defined AUS score < 3) also did not show any significant difference between groups by long term (n = 223, 1 RCT, RR 0.86, 95% CI 0.65 to 1.14; Analysis 2.19).

2.20 Social functioning: 4b. Substance abuse ‐ average endpoint score (Substance Abuse Treatment Scale (SATS), low = poor)

Short‐, medium‐, and long‐term continuous data were available from one study, assessing substance abuse with the SATS. These data failed to show a significant difference between groups at any time period assessment.

2.20.1 By short term

We did not find evidence of a clear difference between ICM and non‐ICM within this subgroup (n = 203, 1 RCT, MD 0.07, 95% CI ‐0.28 to 0.42; Analysis 2.20).

2.20.2 By medium term

We did not find evidence of a clear difference between ICM and non‐ICM within this subgroup (n = 203, 1 RCT, MD ‐0.11, 95% CI ‐0.55 to 0.33; Analysis 2.20).

2.20.3 By long term

We did not find evidence of a clear difference between ICM and non‐ICM within this subgroup (n = 203, 1 RCT, MD 0.11, 95% CI ‐0.41 to 0.63). This subgroup had important levels of heterogeneity (Chi² = 0.0; df = 0.0; P = 0.0; I² = 100%; Analysis 2.20).

2.21 Social functioning: 4c. Alcohol ‐ abuse (various measurements, skewed data)

Skewed data were available from one study assessing 'days using alcohol' and 'AUS score' by short, medium, and long term. Data on 'days using alcohol' showed a trend towards a higher alcohol consumption in the ICM group in each assessed time period. This trend was confirmed by the short‐ and medium‐term findings on the AUS, where the ICM group rating had a worse outcome than the non‐ICM group; however, long‐term data showed a worse outcome in the non‐ICM group (Analysis 2.21).

2.22 Social functioning: 5a. Average endpoint score (Life Skill Profile (LSP), high = poor) ‐ by long term

We found the LSP social functioning score did not favour one group over the other by long term (n = 239, 1 RCT, MD 4.0, 95% CI ‐0.61 to 8.61).

2.23 Social functioning: 5b. Average endpoint score (Social Functioning Questionnaire (SFQ), high = poor) ‐ skewed data

Skewed data on SFQ social functioning scores showed equivocal results by medium term, and a worse outcome in the ICM group by long term. We entered these data as 'other' data (Analysis 2.23).

2.24 Mental state: 1a. General symptoms ‐ average endpoint score (various scales)

We identified two studies relevant to this outcome, assessing 'mental state: general symptoms' on two different scales (BPRS‐24 items and CPRS) at different time periods. We found BPRS mental state scores favoured neither group during the short‐, medium‐, and long‐term analysis. These data were confirmed by long‐term findings on the CPRS, where there was no significant difference between ICM and non‐ICM.

2.24.1 By short term (BPRS‐24 items, high = poor)

The single trial in this subgroup showed no difference between the two treatments (n = 203, 1 RCT, MD ‐0.65, 95% CI ‐3.99 to 2.69; Analysis 2.24).

2.24.2 By medium term (BPRS‐24 items, high = poor)

The single trial in this subgroup showed no difference between the two treatments (n = 203, 1 RCT, MD ‐1.62, 95% CI ‐4.76 to 1.52; Analysis 2.24).

2.24.3 By long term (BPRS‐24 items, high = poor)

The single trial in this subgroup showed no difference between the two treatments (n = 203, 1 RCT, MD ‐0.22, 95% CI ‐3.32 to 2.88; Analysis 2.24).

2.24.4 By long term (CPRS, high = poor)

The single trial in this subgroup showed no difference between the two treatments (n = 595, 1 RCT, MD 0.40, 95% CI ‐1.83 to 2.63; Analysis 2.24).

2.25 Mental state: 1b. General symptoms ‐ average endpoint scores (various scales, skewed data)

Medium‐ and long‐term skewed data from the Krawiecka Scale and long‐term skewed data from the BPRS scale could not be entered into meta‐analysis. These data, provided from two different trials, consistently suggested a better mental state outcome in the non‐ICM group.

2.26 Mental state: 2a. Specific symptoms: negative symptoms ‐ average endpoint score (SANS, high = poor) ‐ by long term

Long‐term continuous data on negative symptoms from one long‐term trial did not favour either group (n = 593, 1 RCT, MD 0.20, 95% CI ‐2.32 to 2.72).

2.27 Mental state: 2b. Specific symptoms ‐ average endpoint scores (various scales, skewed data)

Skewed data were available on anxiety and depression symptoms assessed with the Hospital Anxiety and Depression Scale (HADS) anxiety and depression subscale by medium and long term. These data, provided by the same study, were not consistent between medium and long term, as they showed a better outcome in the ICM group by medium term and a worse outcome in the ICM group by long term, compared with the non‐ICM group (Analysis 2.27).

2.28 Behaviour: 1. Specific behaviour (various measurements)

We found three relevant studies for this outcome, assessing the outcome at different time periods and using different measures. Data failed to show a significant difference between groups at any time period for any outcomes.

2.28.1 By medium term ‐ harm to self or others

Medium‐term data reporting 'harm to self or others' did not show a significant difference between groups (n = 73, 1 RCT, RR 0.88, 95% CI 0.4 to 1.9; Analysis 2.28).

2.28.2 By long term ‐ self‐harm

We found two trials relevant to this subgroup. There was no clear difference between ICM and non‐ICM within this subgroup (n = 959, 2 RCTs, RR 1.00, 95% CI 0.69 to 1.46; Analysis 2.28).

2.28.3 By long term ‐ injury/assault to others

We found two trials relevant to this subgroup. There was no clear difference between ICM and non‐ICM within this subgroup (n = 959, 2 RCTs, RR 1.09, 95% CI 0.85 to 1.4; Analysis 2.28).

2.28.4 By medium‐term FUP (18 months) ‐ self harm

There was a single trial in this subgroup. We did not find evidence of a clear difference between the two treatments (n = 251, RR 0.85, 95% CI 0.44 to 1.67; Analysis 2.28).

2.28.5 By medium‐term FUP (18 months) ‐ injury/assault to others

There was a single trial in this subgroup. We did not find evidence of a clear difference between the two treatments (n = 251, 1 RCT, RR 1.35, 95% CI 0.87 to 2.1; Analysis 2.28).

2.28.6 By long‐term FUP (8.5 years) ‐ self harm

There was a single trial in this subgroup. We did not find evidence of a clear difference between the two treatments (n = 214, 1 RCT, RR 0.81, 95% CI 0.51 to 1.27; Analysis 2.28).

2.28.7 By long‐term FUP (8.5 years) ‐ injury/assault to others

There was a single trial in this subgroup. We did not find evidence of a clear difference between the two treatments (n = 214, 1 RCT, RR 0.95, 95% CI 0.83 to 1.09; Analysis 2.28).

2.29 Quality of life: 1. Average endpoint score (various scales)

We found three studies assessing quality of life with three different scales (LQoLP, MANSA, QOLI) at different time periods. There were no significant differences between ICM and non‐ICM in any of these measures at any time period. Data on QOLI scores were available by short and medium term from one study. Long‐term data were available from three different studies measuring quality of life with three different scales, therefore not entering the analysis together. No results showed any significant difference between groups. Note that for this outcome the right graph label favours ICM (experimental group).

2.29.1 By short term ‐ overall life satisfaction (QOLI, high = better)

One study provided data, showing no difference between groups (n = 203, 1 RCT, MD ‐0.02, 95% CI ‐0.43 to 0.39; Analysis 2.29).

2.29.2 By medium term ‐ overall life satisfaction (QOLI, high = better)

One study provided data, showing no difference between groups (n = 203, 1 RCT, MD ‐0.04, 95% CI ‐0.43 to 0.35; Analysis 2.29).

2.29.3 By long term (LQoLP, high = better)

One study provided data, showing no difference between groups (n = 526, 1 RCT, MD 0.03, 95% CI ‐0.10 to 0.16; Analysis 2.29).

2.29.4 By long term (MANSA, range 1 to 7, high = better)

One study provided data, showing no difference between groups (n = 166, 1 RCT, MD 0.10, 95% CI ‐0.19 to 0.39).

2.29.5 By long term ‐ overall life satisfaction (QOLI, high = better)

One study provided data, showing no difference between groups (n = 203, 1 RCT, MD 0.1, 95% CI ‐0.25 to 0.45; Analysis 2.29).

2.30 Participant satisfaction/need: 1. Average endpoint scores (various scales) ‐ by long term

Long‐term data were available from one study, assessing participant need with CAN, and participant satisfaction with health service scale. There were no significant differences between groups as assessed with the two scales.

2.30.1 Patient need: CAN (high = poor)

One study provided data, showing no difference between groups (n = 585, 1 RCT, MD ‐0.29, 95% CI ‐0.69 to 0.11; Analysis 2.30).

2.30.2 Patient Satisfaction With Health Services (high = poor)

One study provided data, showing no difference between groups (n = 490, 1 RCT, MD ‐0.4, 95% CI ‐1.25 to 0.45; Analysis 2.30).

2.31 Participant need: 1. Average endpoint scores (various scales, skewed data)

One study provided data suggesting a difference between interventions in participant need as assessed with CAN scores, showing a worse outcome in the ICM group in the medium and long term, but these data were skewed. A second study provided data on participant need, assessed with CANSAS by long term. These data failed to show any difference between groups. We have presented these data as 'other' data (Analysis 2.31).

2.32 Participant satisfaction: 1. Average endpoint scores (CSQ‐modified, high = better, skewed data) ‐ by long term

Skewed data from a different study suggested long‐term better satisfaction with treatment for the ICM group (data assessed with CSQ). We have reported these data as 'Other' data, (Analysis 2.32).

2.33 Costs: 1. Direct costs of psychiatric hospital care (skewed data)

Costs were assessed measuring 'direct costs of psychiatric hospital care' and 'direct cost of all care'. No data were available on 'direct healthcare costs'.

Skewed data on direct costs of psychiatric hospital care were available from two studies by medium and long term. Medium‐ and long‐term data from one study found no significant difference between groups, whilst the second long‐term study data found costs were significantly lower for the ICM group. The latter study was small (Quinlivan‐California 1995, n = 60), where the former one was larger (Harrison‐Read‐UK 2000, n = 193). We have presented these data as 'Other' data, (Analysis 2.33).

2.34 Costs: 2a. Direct costs of all care ‐ by long term (2 years) ‐ unit cost GBP, fiscal year 1997/98

Direct costs of all care were available from one study reporting skewed data (UK700‐UK 1999), but with a sample size greater than 200. There was no significant difference between groups for reducing direct costs of all care (n = 667, 1 RCT, MD 77.00, 95% CI ‐66.63 to 220.63), but findings showed a trend suggesting greater cost in the ICM group. This trend suggested a cost increase of GBP 77 per person per month by long term, referred to as fiscal year 1997/98.

2.35 Costs: 2b. Direct costs of all care (skewed data) ‐ by medium term

One study (n = 58) reported skewed data on the same outcome (direct costs of all care) (Johnston‐Australia 1998). This data set showed no significant differences between groups by medium term, substantially confirming data by long term. We have reported these data as 'Other' data, (Analysis 2.35).

2.36 Costs: 3. Total costs of care per patient ‐ unit cost GBP)

Total costs of care were available from two studies reporting skewed data (REACT‐UK 2002;UK700‐UK 1999), but with a sample size greater than 200. Total costs of care included direct and indirect costs (i.e. informal care, prison, court, probation officer, police custody, etc.).

2.36.1 By 24 months, fiscal year 1997/1998

There was no significant difference between groups for reducing total costs of care (n = 667, 1 RCT, MD 1849.00, 95% CI ‐1598.23 to 5296.23), but findings showed a trend suggesting greater cost in the ICM group (Analysis 2.36).

2.36.2 By 18 months, fiscal year 2003/2004 (GBP 1 = USD 1.58)

There was no significant difference between groups for reducing total costs of care (n = 243, 1 RCT, MD 4031.00, 95% CI ‐2724.13 to 10,786.13), but findings showed a trend suggesting greater cost in the ICM group (Analysis 2.36).

3. SENSITIVITY ANALYSES

As anticipated in the Methods section (see Methods, Sensitivity analysis), we performed the following sensitivity analyses.

3.1 Implication of randomisation

3.1.1 COMPARISON 1: INTENSIVE CASE MANAGEMENT versus STANDARD CARE

All of the included studies were described as randomised, and none were described in a way as to imply randomisation, therefore we did not include any trials in a sensitivity analysis.

3.1.2 COMPARISON 2: INTENSIVE CASE MANAGEMENT versus NON‐INTENSIVE CASE MANAGEMENT

All of the included studies were described as randomised, and none were described in a way as to imply randomisation, therefore we did not include any trials in a sensitivity analysis

3.2 Standard care caseload is over 20

3.2.1 COMPARISON 1: INTENSIVE CASE MANAGEMENT versus STANDARD CARE

Among studies reporting on our primary outcomes, only six studies reported the ratio of staff to participants, and for each study it was greater than 1:20 (Bond‐Chicago1 1990; Bond‐Indiana1 1988; Herinckx‐Oregon 1996; Jerrell‐SCarolina1 1991; OPUS‐Denmark 1999; Sytema‐Netherlands 1999).

When we entered these studies (where standard care group caseload was greater than 20) in the analysis, the first primary outcome 'average days per month in hospital' showed a significantly favourable effect in the ICM group in reducing the length of hospitalisation (n = 951, 7 RCTs/study centres, MD ‐2.01, 95% CI ‐3.36 to ‐0.67), but heterogeneity was substantial (I² = 70%, P = 0.003). These findings confirm those obtained when all studies were entered in the analysis, regardless of caseload in the standard care comparison group.

The second primary outcome 'not remaining in contact with psychiatric services' showed that participants in the ICM group were significantly less likely to lose contact with psychiatric services than participants in the standard care group by medium term, long term, and overall (n = 931, 4 RCTs, MD 0.38, 95% CI 0.23 to 0.63), but heterogeneity was substantial (I² = 67%, P = 0.03). These findings confirm those obtained when all studies were entered in the analysis, regardless of caseload in the standard care comparison group, however data in the sensitivity analysis showed a substantial level of heterogeneity not present when all studies were entered in the analysis.

3.3 Assumptions for lost binary data

3.3.1 COMPARISON 1: INTENSIVE CASE MANAGEMENT versus STANDARD CARE

No assumptions needed to be made regarding people lost to follow‐up for the primary binary outcome 'not remaining in contact with psychiatric services' for this comparison, therefore we did not include any trials in a sensitivity analysis.

3.3.2 COMPARISON 2: INTENSIVE CASE MANAGEMENT versus NON‐INTENSIVE CASE MANAGEMENT

For the second comparison, we needed to make assumptions regarding people lost to follow‐up for the primary binary outcome 'not remaining in contact with psychiatric services' for one trial (REACT‐UK 2002). When we compared the findings of the outcome when we used our assumption to when we used completer data only, results did not differ substantially.

3.4 Assumptions for lost continuous data (for meta‐regression)

We undertook sensitivity analysis, removing 13 out of 52 trials originally entering meta‐regression. These were trials where primary outcome standard deviation was imputed (see Table 3). Meta‐regression was therefore run on the remaining pool of 39 non‐imputed studies. Results no longer reached significance for both variables 'organisation fidelity subscore' and 'baseline hospital use' as P is at trend ˜ 0.07 ('organisation fidelity subscore': regression coefficient ‐0.2, 95% CI ‐0.48 to 0.02, P = 0.067; 'baseline hospital use': regression coefficient ‐0.18, 95% CI ‐0.39 to 0.02, P = 0.077). Removing a high‐influence outlier (Rosenheck‐USA‐NP (C)), significance returns for the effect of 'baseline hospital use' (‐0.26, 95% CI ‐0.51 to ‐0.01, P = 0.046), but not for 'organisation fidelity'. When combining the two variables within the model ('organisation fidelity subscore' and 'baseline hospital use'), the pattern of results did not change for the non‐imputed studies.

Discussion

Summary of main results

1. COMPARISON 1: INTENSIVE CASE MANAGEMENT versus STANDARD CARE

1.1 Service use

1.1.1 Service use: Average number of days in hospital per month ‐ at about 24 months

ICM does seem to reduce length of hospitalisation when compared with standard care (by 0.86 day per month over 24 months). However, these data were low quality and from a heterogeneous analysis; when certain studies are removed from the analysis, the now‐homogeneous result suggests that the saving is in the region of 0.6 days per month. We were unsure which of the two figures was the most reliable, but in any event trial findings suggest a saving of time in hospital per person of between about 7 and 10 days in hospital per year.

What is important, however, is that there was a high duration of hospital stay for these people in the two previous years (averaging 6 days per month). This was higher than for those in the ICM versus non‐ICM comparison (averaging 3.4 days per month), which found no difference between groups (see Figure 10). The impressive saving of time in hospital for ICM was greater than that of standard care, but the saving in standard care was also considerable (absolute decline for ICM was about 3 days, standard care was 1.8 days). Well‐organised standard care, for people with what would seem to be a history of considerable periods spent in hospital over the last 24 months, does seem to reduce the time in hospital for the ensuing two years. However, ICM adds more than an extra day to that gain.

The meta‐regression is a tool of limited power, employed on weak data. However, the results were in keeping with the results of one other review (Burns 2007), which suggests that the gain of ICM was not so much linked with the total fidelity score, the staff subscale, or the comparison group. It did suggest a link with the organisational subscore of the Index of Fidelity to Assertive Community Treatment (IFACT) scale (0.36 day per month out of hospital gained by each point increase of this subscore). It also suggested that ICM effect is linked to length of hospitalisation during the previous two years (0.2 day per month out of hospital gained by each day increase in the day in hospital per month during previous two years).

In other words, this means that:

if the ICM team ratio staff:client is less than 1:20 (calculated dividing the number of active clients on the caseload by the number of full‐time equivalents of direct service staff on the team), then it makes no difference how much the caseload is lower than 1:20;
the ICM team size does not matter (calculated as the number of full‐time clinical staff equivalents, as defined earlier);
the availability of a psychiatrist on the team is not pivotal; and/or
the availability of a nurse on the team is not pivotal.

It also suggested that gains in IFACT scores can reduce average hospital stay. McGrew 1994 and McGrew 1995 suggest that gain can be mediated by:

ensuring that the ICM team performed the role of primary therapist for the client (the primary therapist role designates the person within the local mental health system with primary clinical and record keeping (e.g. treatment plans) responsibility for the client);
the ICM team's offices being located in a separate building from the parent agency's main offices (i.e. usually away from the hospital site);
the ICM team sharing caseloads (rated as the degree to which all staff members on the team had contact with all clients on a regular basis, e.g. through rotation), in contrast to individual caseloads in which specific staff workers are responsible for specific clients;
the ICM team meeting as a group each weekday to discuss their entire caseload;
the ICM team’s supervisor devoting at least half time to client contacts, either cojointly during supervision of team members, or individually as part of his or her duties as a member of the team;
the ICM team providing 24‐hour direct access to the team (if access to the ICM team was triaged through the community mental health centre emergency 24‐hour on‐call service, intermediate score of 0.5 is obtained, therefore the advantage on decreasing days in hospital per month would be halved (‐0.2 days in hospital per month)); and/or
the ICM team serving clients without any expectation of transferring them to another programme.

1.1.2 Service use: Average number of days in hospital per month – by follow‐up

When the same outcome ‘average number of days in hospital per month’ was assessed on medium‐ and long‐term follow up, data did not confirm the effect of ICM in reducing length of hospitalisation assessed by 24 months. These data were provided by one large study (n = 547), where during the follow‐up period all participants received the control intervention (standard care). These data are suggestive of a loss of effect on reducing number of days in hospital over time, if ICM intervention is discontinued.

1.1.3 Service use: Not remaining in contact with psychiatric services

Overall, we found ICM to be better than standard care for retaining people in psychiatric services (n = 1633, 9 RCTs, RR 0.43, 95% CI 0.30 to 0.61). However, this effect was not seen for the short‐term analysis (n = 95, 1 RCT, RR 0.54, 95% CI 0.28 to 1.05), although confidence intervals were compatible with an effect favouring ICM (P = 0.07). Medium‐term findings did suggest an overall effect of ICM being better in reducing loss to follow‐up to psychiatric service contact (n = 1063, 3 RCTs, RR 0.51, 95% CI 0.36 to 0.71). In the longer term, this same effect was heterogeneous, due to an outlying finding of a single study (Herinckx‐Oregon). This trial had a peculiar definition for this outcome, which was different from the other studies. In Herinckx‐Oregon 'not remaining in contact with psychiatric services' did not include refusing re‐interview, moving out, and death, whereas for the other trials it did. If we exclude Herinckx‐Oregon, and only five long‐term trials are retained in the analysis, then ICM appears to be effective in preventing loss to follow‐up (n = 475, 5 RCTs, RR 0.27, 95% CI 0.11 to 0.66).

The result of a better retention in care for the ICM group strengthens the relevance of the result for the first primary outcome (i.e. ICM decreasing time in hospital). ICM decreases days of hospitalisation in a severe mentally ill population where patients are kept in close contact with services, therefore the ICM effect on reducing days in hospital might not be dissipated by a higher rate of loss to follow‐up.

1.1.4 Service use: Use of hospital

We found that ICM reduced the number of people admitted to hospital more than standard care, at least in the medium term (n = 1303, 5 RCTs, RR 0.85, 95% CI 0.77 to 0.93). Skewed data for 'average number of admissions in the medium term' were in accordance with these first findings, whilst skewed data for 'average number of admissions in the long term' failed to show any trend in the effect of intervention. Synthesis of short‐ and‐long term studies in this outcome produced heterogeneous findings. The short‐term data were from only two studies, and when the most positive one is removed (Bond‐Indiana1), the finding becomes clearly null. The long‐term data was from 11 trials; removing the three clearly outlying studies, Curtis‐New York, Macias‐Utah, and Test‐Wisconsin, also restores homogeneity and moves the finding squarely towards the null. One small study provided inconclusive data on the unplanned admission to the emergency department, an outcome describing only one option of psychiatric admission; for this reason, and because these data come from a single small study, we considered this finding weak.

Overall, the effect of ICM on admission to hospital is not strong. Whereas the time spent in hospital does seem to be less if allocated ICM ‐ at least for those whose baseline use of hospital was high ‐ the number of admissions is not greatly changed. Admissions are shorter, but not by much, if any less frequent.

1.1.5 Service use: Use of services outside of mental health provision

More outcomes were available describing ‘use of services outside of mental health provision’. Studies did not report a convincing difference in 'use of emergency room', 'rate of use of emergency room', ‘use of day hospital care’, or ‘rate of outpatient visits’. Only ‘rate of home visits’ was higher for the ICM group. However, we did not consider these data very robust, as they were based on single studies providing data per each outcome or on skewed data that were very difficult to interpret.

1.2 Adverse event

1.2.1 Death ‐ all cause and suicide

This review did not find any difference in mortality either due to all causes or to suicide in the short, medium, and long term and in medium‐ and long‐term follow‐up. Although death is a rare adverse event, the duration of the longer studies (˜ 24 months) means that some deaths would have been expected to occur and did (3.2% ICM versus 3.8% standard care). This difference was not statistically significant, but is homogeneous, and suggests a trend in terms of risk reduction (RR 0.84, 95% CI 0.48 to 1.47), partly confirmed in one large study on medium‐term follow‐up (RR 0.59, 95% CI 0.22 to 1.61). The same trend was confirmed in the ICM effect on mortality due to suicide by long term (suicide rate 1.3% ICM versus 2% standard care). Again, this difference was not statistically significant, but is homogeneous, and stronger compared to death due to all causes. If the suicide risk is in actuality so much reduced (RR 0.68, 95% CI 0.31 to 1.51), this would be a very important finding, although the quality of the evidence is low (Summary of findings table 1).

Few things have changed the outcome of death for people with schizophrenia. If, for this group of people, set in the context of a standard care with a high baseline risk of admission, ICM could decrease death, and by so much, this would be a strong argument in favour of ICM. These data on suicide are the findings of nine studies with only a total number of participants of 1456 and low‐quality evidence. A single larger trial might be able to confirm this important suggestion.

1.3 Global state

1.3.1 Global state: Relapse

No data were provided for the important outcome of relapse.

1.3.2 Global state: Leaving the study early

ICM data regarding number of participants lost to follow‐up for the short term were heterogeneous. However, we found ICM to be more advantageous than standard care in reducing rate of lost to follow‐up both in the medium term (n = 1699, 8 RCTs, RR 0.60, 95% CI 0.51 to 0.70) and long term (n = 1798, 13 RCTs, RR 0.68, 95% CI 0.58 to 0.79, low‐quality evidence). The impression remains that ICM, and again with the proviso that these findings may apply most specifically to a group with high baseline admission, holds on to people more tightly across time. This may not significantly lower admission rate, but loss to follow‐up and length of admission may decrease. Overall, these data were not of high quality (Summary of findings table 1), but there is a belief that ICM is advantageous over standard care for the higher‐risk groups. For example, in groups with only 10% loss to follow‐up across the long term, only 3 more people are not lost for every 100 given ICM. However, with more realistic figures of 50% loss to follow‐up, this figure rises to 15 more people out of every 100 who are kept in care compared with those allocated standard care.

1.3.3 Global state: Global Assessment of Functioning Scale (GAF)

Short‐term studies indicated a better improvement in GAF endpoint score for participants in the ICM group compared with those allocated standard care (short term: n = 797, 4 RCTs, MD 2.07, 95% CI 0.28 to 3.86). This effect was confirmed by the long‐term data (n = 818, 5 RCTs, MD 3.41, 95% CI 1.66 to 5.16). Whilst this is favourable for those allocated ICM, we are unsure of the clinical meaning of these data, as changes of two or three points on a scale that runs to 100 does not seem to be much. We have found no reference to the clinical meaning of such small changes.

1.3.4 Global state: Not compliant with medication

Only one long‐term study provided data on compliance with medication, and these indicated a higher compliance level in those allocated to ICM compared with those in the standard care group (n = 71, 1 RCT, RR 0.35, 95% CI 0.15 to 0.81). Again, this is an important finding and should be replicated. We are surprised that such easily recorded data are not reported in more studies.

1.4 Social functioning

1.4.1 Social functioning: Contact with legal system

Studies measuring contact with legal system used different definitions over varying time periods, which makes interpretation difficult. There was no real suggestion that ICM either increases or decreases the measures of this outcome. The ‘arrested’ and ‘imprisoned’ outcomes were the only ones with some consistency. They showed no significant difference in the intervention effect between groups. The ‘contact with the police’ outcome findings were from only one study and were not significant in the short term, but became significantly different in the medium term (favouring the ICM group). Overall, the legal outcomes were not convincing, and there is a need for more consistency in approach to this area of research.

1.4.2 Social functioning: Employment status

This review did not reveal any significant difference between ICM and standard care in employment status, as measured by different outcomes. Medium‐term findings on ‘number of people not competitively employed’ were based on only one trial. Both medium‐ and long‐term data reporting ‘not employed’ tended to favour ICM, although data were heterogeneous, and overall the quality of the evidence was very low (Summary of findings table 1). Data from one large trial on medium‐ and long‐term follow‐up failed to show any difference between interventions. Again, this is an important outcome for which more data are needed before firm conclusions can be drawn.

1.4.3 Social functioning: Accommodation status

Data on the outcome 'homelessness' were not convincing by short, medium, or long term, but were mostly derived from just a few trials.

The outcome 'not living in stable accommodation' was scarcely reported, available from only one long‐term study.

Regarding the 'not living independently’ outcome, we found that people allocated to ICM were more likely to live independently compared with those allocated standard care ‐ in the medium term, and even more so in the long term. This is another important finding of this review. If the risk of not living independently is in actuality substantially reduced by this ICM package (18% ICM versus 26% standard care, long term), at least for people with high baseline risk of admission, and if this is a desired outcome for this particular client group, then, combined with the other moderate but cumulative advantages, this finding further highlights the advantage of ICM over standard care.

Only one large study reported the outcome ‘days in supported house’, assessing it by long term and by medium‐ and long‐term follow‐up. The advantage for standard care reported on medium‐term follow‐up was not confirmed on long term or on long‐term follow‐up, failing to show a significant trend of effect.

1.4.4 Social functioning: Substance abuse

Only one study (n = 547) reported usable binary data for alcohol and drug abuse. A long‐term single study failed to show any advantage for participants treated with ICM compared to standard care. Skewed data were supplied by Sytema‐Netherlands (n = 81), with Dartmouth Assessment of Lifestyle Interviewscores tending to favour the ICM group for both drug use and alcohol consumption. However, we are unsure of the clinical meaning of these scores. Morse‐Missouri3 (n = 103) reported skewed data on days substances were used per month. There was no indication of any difference between groups. There is currently no compelling evidence that ICM affects abuse of substances or alcohol.

1.4.5 Social functioning: Various scales

Few studies reported usable data for social functioning scale, and outcome data were complicated by the use of different scales within single studies, making meta‐analysis impossible. In this confusion of evidence, we see no advantage for participants treated with ICM compared to those treated with standard care in terms of social functioning measures. This does not seem to concur with other findings on independent living. It could be that the fine‐grain measures of functioning are picking up subtle parameters of social function not effected by the package. It could also be that the measures are not sensitive enough to broad and important issues of social function.

1.5 Mental state

Rating of mental state in these trials illustrates the confusion of how such symptoms are recorded in randomised studies. Timings of use of the scales differ, and the findings are so problematic to interpret from the clinical perspective that we are left to make safe but bland conclusions.

No data were provided for the important outcome 'mental state: not improved to an important extent', and there does not seem to be any compelling evidence that, in this group of people, set where the baseline risk of admission is high, ICM in actuality substantially affects a person's mental state.

1.6 Behaviour: self harm

Based on findings from the larger of the studies, self harm was not convincingly reduced by use of the ICM model. The mortality finding discussed above, however, does seem to suggest that ICM reduced the risk of death. These findings are a little at odds with each other, although not entirely, providing all the more reason to continue to research into this area for these, the simplest of outcomes.

1.7 Quality of life

Few studies reported relevant outcomes. Short‐ and medium‐term outcome data were complicated by the use of different scales within single studies, making meta‐analysis impossible. We found that one short‐term study (n = 125) showed a better quality of life in the ICM group on the Lehman's Quality of Life Interview scale, but more medium‐ and long‐term data failed to show any advantage for participants treated in the ICM group compared with standard care. The few skewed data seem to concur with the impression that for quality of life measures used in trials, ICM confers no advantage over standard care.

1.8 Participant satisfaction/need

Participants administered ICM were more satisfied with their treatment compared with those administered standard care in these trials. These findings were based on data that were quite strong (short term, n = 61; medium term, n = 500; long term, n = 423). More satisfaction with care could enhance medication compliance, the will to keep in services, housing status, and a host of other variables. We are left doubting the size and meaning of the overall finding. We are unsure how encouraged we should be that these packages of care deliver an average of a two‐ to three‐point improvement in the Client Satisfaction Questionnaire.

Several of the smaller trials did measure need and unmet need. These skewed data were difficult to interpret, but did not seem to convincingly favour either of the groups.

1.9 Costs

With respect to cost of inpatient psychiatric care, ICM was consistently superior to standard care for the outcome 'direct costs of psychiatric hospital care', suggesting a saving of money per person of about USD 144 per month (fiscal year 1990). Two studies taking part in the same multicentre trial provided these data. Skewed data were contradictory, neither showing a trend confirming nor disputing these data.

We found no difference between groups with respect to direct healthcare costs (where skewed data were contradictory and provided by only two small studies). Results on ‘direct costs of all care’ were inconcludent, as data were skewed and different trials reported contradictory effects.

2. COMPARISON 2: INTENSIVE CASE MANAGEMENT versus NON‐INTENSIVE CASE MANAGEMENT

2.1 Service use

2.1.1 Service use: Average number of days in hospital per month ‐ at about 24 months

Moderate quality evidence from this review showed no significant advantage of ICM in reducing the average length of hospitalisation when compared with non‐ICM. This could be an important finding, and we see no good reason not to trust this result. The implications from this finding could be that if services are already providing non‐ICM, there is no point in investing in further intensiveness. We currently know of no review comparing non‐ICM with standard care and reporting relevant outcomes. This should be undertaken. It is possible that there are other features of ICM that may improve outcome, but we are not stipulating that we should specifically investigate for these. What was different between the two sets of comparisons was the baseline risk of admission in the previous two years (about 6 days per month for Comparison 1 versus about 3.4 days per month for Comparison 2). This was highlighted by the meta‐regression process. This generates further hypotheses. Baseline hospital risk is linked to service provision, service culture, severity of illness, and other issues. We do not have the sophistication of data to investigate these. What we are left with is the possibility that in a situation where people with severe mental illness have a duration in hospital of less than 4 days per month in the two years preceding the ICM package of care, the increased intensity of approach may not be justified.

2.1.2 Service use: Average number of days in hospital per month – at follow‐up

Data on medium‐ and long‐term follow‐up from one study (n = 237) failed to show a significant advantage of ICM in reducing the average length of hospitalisation when compared with non‐ICM. During the follow‐up period participants could remain in the originally allocated intervention or be transferred to the control one. These data on follow‐up confirmed the data at 24 months discussed above.

2.1.3 Service use: Not remaining in contact with psychiatric service

We found ICM to be more effective in increasing the number of people retained in contact with psychiatric service in the medium term, but we did not consider these findings robust as they were based only on one small trial (n = 73). We found no difference between interventions in the long term, but data were heterogeneous. Overall, when pooling medium‐ and long‐term data, we found no advantage for participants treated with ICM compared to non‐ICM for better retention in psychiatric service but, again, these data were heterogeneous, and we found no obvious explanation for the heterogeneity (n = 1255, 4 RCTs). Medium‐term (18 months) follow‐up data showed a trend favouring ICM in increasing the number of people retained in contact with psychiatric service.

2.1.4 Service use: Admissions

We found no difference between groups in the risk of being admitted to hospital in the long term (n = 1132, 3 RCTs, RR 0.91, 95% CI 0.75 to 1.12). These findings were confirmed by data from one long‐term study (n = 678) and one medium‐term study (n = 68) on the average number of admissions, where no advantage was shown between treatments in reducing number of admissions in the long term (moderate‐quality evidence, summary of findings Table 2) or in the medium‐ and long‐term follow‐up. Data on frequency of admission and on length of hospitalisation therefore consistently show no effect of ICM for both outcomes.

2.2 Adverse events

2.2.1 Death due to all causes and to suicide

This review did not find provide strong evidence on mortality rate either due to all causes or to suicide in the short, medium, and long term, or in the medium‐ and long‐term follow‐up. These data are quite informative, especially those from long‐term studies, where the study length might balance the rarity of the event in detecting any difference between intervention effects. Some deaths occurred in the long‐term studies (2.0% ICM versus 2.2% non‐ICM) (n = 1634, 5 RCTs, RR 0.90, 95% CI 0.46 to 1.75). This impression was confirmed for studies reporting suicide only, as shown in summary of findings Table 2, where low‐quality evidence also showed no difference between groups.

2.3 Global state

2.3.1 Global state: Relapse

No data were provided for this important outcome.

2.3.2 Global state: Leaving the study early

No studies were available for the short‐term outcome.

Data showed no difference between interventions by the medium term, but these data were not strong as they came from a small sample of two studies (n = 225) and were heterogeneous with inconsistency of effect.

We found ICM to be more advantageous than non‐ICM in reducing rate of lost to follow‐up by the long term.

Overall, pooling studies from subgroups for two time points, we found heterogeneous data, but substantially confirming homogeneous data obtained by long term. ICM was confirmed to be more advantageous than non‐ICM in reducing rate of lost to follow‐up (n = 2195, 9 RCTs, RR 0.72, 95% CI 0.52 to 0.99). If we consider this outcome as proxy of a better retention in care, it might overcome the inconsistency of data on 'number of people remaining in contact with psychiatric service' (see Discussion ‐ 2.1.3 Service use: Not remaining in contact with psychiatric service). ICM therefore seems to positively reduce number of lost to follow‐up, but does not affect length and frequency of admission. These data were of low quality (summary of findings Table 2), but it appears that ICM has an advantage over non‐ICM for the higher‐risk groups. For example, in groups with only 10% loss to follow‐up across the long term, only three more people are not lost for every 100 given ICM. However, with a more realistic figure of 50% loss to follow‐up, this rises to 14 more people out of every 100 are kept in care compared with those allocated standard care.

2.3.3 Global state: Health of the Nation Outcome Scale (HoNOS)

Not enough studies were available to run a meta‐analysis on data derived from the HoNOS or from any other scales assessing global state. Considering the comprehensiveness of the scale assessment of other outcomes, it appears that global state as an outcome is under‐recorded by trialists, despite being informative and relevant from a clinical point of view.

2.3.4 Global state: Compliance with medication

As for the previous outcome, not enough studies were available to run a meta‐analysis. Again, a very meaningful outcome from a clinical point of view is neglected.

2.4 Social functioning

2.4.1 Social functioning: Contact with legal system

As for studies included in Comparison 1, studies measuring contact with legal system used different definitions over a variety of time periods. This makes interpretation difficult. In addition, only a few studies addressed this outcome (three trials overall). There was no real suggestion that ICM either increases or decreases the measures of this outcome. The 'imprisoned' outcome was the only one with some consistency, showing no significant difference in the intervention effect between groups. The 'contact with the police' and 'arrested' outcome findings were from only one study each and were not significant in the short and long term, respectively. Overall, the legal outcomes were not convincing, and there is a need for more consistency in approach to this area of research.

2.4.2 Social functioning: Employment status

Data available for this outcome were substantially inconclusive, reported by only one small trial (n = 73). This trial measured employment status according to two different definitions: 'spent more than one day employed' and 'on paid employment'. Both findings were not significant in the medium term, and overall quality of evidence was low (summary of findings Table 2). One larger study (n = 214) provided data on long‐term follow‐up (8.5 years after the randomised allocation was broken), showing no difference between the two groups. These data did not add much to the understanding of the impact of ICM on employment status compared to non‐ICM. Again, this is an important outcome underestimated in the current studies, and at this stage more data are needed before firm conclusions can be drawn.

2.4.3 Social functioning: Accommodation status

Available data on this outcome were surprisingly scarce, as this outcome was reported in just four studies. Two studies described this outcome with binary data: one measuring just 'living in supported accommodation' by medium term, the second measuring 'living in supported accommodation', 'homelessness', and ‘living independently’ by long term and on follow‐up. All data on different measures at different time periods were based on only one trial and they were not significant.

Two more studies described this outcome with continuous data on ‘average days per months in stable accomodation’, again failing to show any difference between the two groups. Findings seemed to point at a non‐significant difference in effects of intervention, although these findings were inconclusive due to the scarcity of available data, despite being easily recorded and very relevant from the perspective of a community‐based service.

2.4.4 Social functioning: Substance abuse

Only two studies measured substance abuse, and they described this outcome as binary and continuous measures, and not consistently across studies. This makes interpretation difficult and findings unconvincing, as a single study entered each measurement, and we therefore could not carry out any meta‐analysis. As far as we could assess, there was no indication of any long‐term difference between groups in 'number of people abusing alcohol', 'number of people abusing illicit drug', or 'remission from alcohol use disorder' (defined as Alcohol Use Scale score less than 3). These findings were confirmed by those continuous data, assessing the substance abuse with Substance Abuse Treatment Scale. These data failed to show a significant difference between groups at any time period assessment. As for the first comparison, currently there is no compelling evidence that ICM affects abuse of substances or alcohol.

2.4.5 Social functioning: Scale data

Findings were equivocal on scale data measuring social functioning, as provided by only one study. We do not think that future studies should address this outcome by use of scale measurement. Scales are not sensitive measures of social functioning. More effort should be placed on consistent and wide measurements of the main issues of social functioning (such as accommodation status, employment status, contact with legal system, rate of permanent social benefits).

2.5 Mental state: General symptoms and specific symptoms

Again, outcomes measured on scales were substantially inconclusive, as the data were spread across single studies on different scales and at different time periods, making meta‐analysis impossible. According to the low‐quality data available, there does not seem to be any compelling evidence that ICM substantially affects mental state. No data were available for the significant outcome of 'important improvement in mental health'.

2.6 Behaviour: Self harm and injury to others

This review did not reveal any long‐term significant difference between ICM and non‐ICM in risk of committing self harm or injury to others. These data were based on findings from two studies (n = 959), one of which is the largest study (UK700‐UK 1999), and the other the only study providing data on medium‐ and long‐term follow‐up (REACT‐UK 2002). These findings are consistent with the mortality findings discussed above, where no significant difference was shown in death rate between groups, either for suicide and for all causes. Although these data are suggestive of no difference of effects between interventions, they are still quite weak due to limited sample size. More trials should address this outcome, one of the simplest ones to collect.

2.7 Quality of life

Quality of life rating in these trials illustrates the confusion of how such symptoms are recorded in randomised studies. The scales differ, and the timings of use of the scales also differ. There was such an inconsistency in approach to this area of research to make meta‐analysis impossible. None of the available findings showed any significant difference between interventions. There does not seem to be any compelling evidence that ICM substantially affects the quality of life of a person with severe mental illness.

2.8 Participant satisfaction/need

Findings tended to favour the ICM group in being better satisfied with health services and in reducing need. However, this difference was not significant, and both findings were based on data from the same trial, the largest one (UK700‐UK 1999, n = 585). We therefore cannot draw any conclusions, but highlight a possible favourable effect in the ICM group, which needs to be confirmed.

2.9 Costs

Studies assessed 'direct costs of psychiatric hospital care', 'direct cost of all care', and ‘total cost of care’. Regarding the first outcome, findings were based on skewed data, provided by one small trial (Quinlivan‐California 1995, n = 60) and one larger one (Harrison‐Read‐UK 2000, n = 193). There did not appear to be any compelling evidence that ICM substantially affects 'direct costs of psychiatric hospital care', either by medium or long term. Also, findings on long‐term 'direct cost of all care' did not show any difference between interventions. Again, findings on ‘total cost of care’ failed to show any difference between ICM and non‐ICM.

Overall completeness and applicability of evidence

1. Completeness

1.1 Duration of follow‐up

The majority of studies presented long‐term data, that is over one year of follow‐up. This is a reasonable length of time to sensibly assess any difference in the intervention effects.

Two studies, one in the first comparison (ICM versus standard care) and one in the second comparison (ICM versus non‐ICM) presented long‐term follow‐up (from 18 months to 8 years), assessing outcomes after the active intervention was discontinued or after participants could chose to which arm they were allocated.

1.2 Coverage of outcomes

As the experimental intervention is a service organisation model, its realisation involves health policy and research should account for efficacy and cost evaluation. The outcomes reported were mainly service use and social functioning oriented. No studies reported data on relapse (see summary of findings Table for the main comparison, summary of findings Table 2), carer satisfaction and family burden. Participant satisfaction was scarcely reported and in a fragmented way, therefore available data are only partially informative on the effects of these approaches. Few studies provided cost data.

2. Applicability

2.1 Origin

The origin of the data has changed in the last decade since the two original reviews (Marshall 2000a; Marshall 2000b). There are now more included trials from Europe, whereas in the past the data source was largely North America, with a few trials from Australia. Thirty per cent of the total sample included in the current review comprises randomised people from Europe. These studies add power to the result for the primary outcome, narrowing the confidence intervals, but otherwise not substantially changing the findings. As only one study was from China, and all the remaining included studies were from Europe, North America, and Australia, the findings of this review still lack applicability to low‐income countries and, more generally, to countries where mental health systems are not community based.

2.2 People

Studies included people presenting a variability that we feel is likely to reflect the heterogeneous population a clinician faces in daily practice when treating people affected by severe mental illness. This variability was in terms of diagnosis (where participants were affected by a wide diagnostic group including schizophrenic, affective, and personality disorder); comorbidity (where four studies included dually diagnosed participants) (Drake‐NHamp 1998;Essock‐Connecticut2 2006;Morse‐Missouri3 2005;Muller‐Clemm‐Canada 1996); and social characteristics (where eight trials included homeless participants). On average, studies included people with a long history of illness; only OPUS‐Denmark 1999 included participants with a first episode of psychotic illness. This fits with the concept of severe mental illness, where this label includes certain criteria relating to length of illness.

2.3 Interventions

Some studies showed a greater applicability because the experimental intervention was provided by pre‐existing team, therefore closer to the real world and less contaminated by the experimental setting.

The majority of new included trials from the 2010 update compared ICM with non‐ICM (8 out of 14 trials), and they are all from Europe, Australia, and North America. This confirms the trend of psychiatric services in those particular areas to increasingly include some elements of the original model, but also to dilute and contaminate them with the current organisation. What we call 'standard care' is therefore converging toward non‐ICM. The two studies newly included from the 2015 update compare ICM with standard care: one is from the USA and assesses ICM adapted to the forensic setting, and the other is from China, where only recently community care is catching on. For those of us who practice in Europe, the second of the two comparisons in this review may well be more applicable to everyday care. Importantly, this comparison did not illustrate a substantial difference between ICM and non‐ICM.

Quality of the evidence

As illustrated in Figure 1, it appears there is an overall unclear risk of bias in these trials. This would therefore mean there is a moderate risk of overestimate of positive effect. Also, making difficult judgements about quality has been greatly helped by a discernable improvement in reporting of methodology.

Potential biases in the review process

There were several potential biases. We have worked mainly with published reports, and only in few cases with unpublished material. By doing this we may be perpetuating a reporting and publishing bias. It would have been better to have much more original individual participant data. This review follows from two past Cochrane reviews (Marshall 2000a; Marshall 2000b), as well as much work already published in paper format (Burns 2007). The conduct of these reports has influenced this document, and it is possible that we have failed to identify systematic biases in the way we have conducted the reviews across time.

An author of this review is an active pioneer in the development and implementation of the experimental intervention model across the scientific community and clinical world (MM), and one included study is his (Marshall‐UK 1995). As a team, we tried to ensure that decisions were made by rational consensus, and not to have an expert in the team would have been an inadvisable omission.

In some cases, protocol rules were unclear, and need for subsequent clarification arose and post hoc decisions had to be taken (see Differences between protocol and review). This could have affected the review process in various cases. This has probably lowered the quality of the data included in the review, but to not include so much, for example, skewed data, would have omitted much information. Also, by breaking down studies into their centres, many fell below the 200‐participant cutoff point. We have included these data in the 'less than 200' category, whereas in previous versions of the review they would have been in the 'greater than 200' category. Due to the overall effect of the changes in protocol, it appears that we have a more inclusive review, with data that are more heterogeneous and also more favourable for the experimental interventions than otherwise would have been the case should we have used a more limited data set. Nevertheless, we did feel it important to present all of these data for the reader to consider.

We prespecified what characteristics of studies could be associated with heterogeneity, and therefore we stated in the protocol what variables were to be explored in the meta‐regression before inspecting the results of the studies. Despite this prespecification, we were not blind to what variables were probably more related to heterogeneity, as we were familiar with some study results previously published. The undertaken exploration of heterogeneity might therefore at best lead to generation of hypothesis, but it cannot provide reliable conclusions.

Agreements and disagreements with other studies or reviews

This review merges two older Cochrane reviews, fully bringing up to date how these data should be considered. This version does not disagree with the older reviews; it simply replaces them with a more current viewpoint of the data. A major improvement in this version is data on duration of admission, which were previously lacking from past reviews. Other research in this area do not provide a full summary of available evidence on ICM effects across various outcomes (Burns 2007). This Cochrane review does not disagree with the paper version; it is just much more comprehensive. Regarding the meta‐regression, this review substantially confirms the hypothesis stated elsewhere that baseline hospital use and fidelity to the model affects outcome (Burns 2007).

Figure 1

Methodological quality summary: review authors' judgements about each methodological quality item for each included study.

Figure 2

Study flow diagram 2015 update

Figure 3

Forest plot of comparison: 1 Intensive Case Management versus standard care, outcome: 1.1 Service use: 1. Average number of days in hospital per month ‐ at about 24 months.

Figure 4

Service use: 1. Average number of days in hospital per month ‐ at about 24 months ‐ restoring homogeneity ‐ 4 studies removed from analysis.

Figure 5

Funnel plot of comparison: 1 Intensive Case Management versus standard care, outcome: 1.1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months.

Figure 6

Meta‐regression: Scatterplot of IFACT organisation subscore versus mean days per month in hospital.

Figure 7

Meta‐regression: Scatterplot of mean baseline days in hospital versus mean days per month in hospital.

Figure 8

Weighted thin plate spline regression showing combined effect of baseline days in hospital and organisational fidelity score on treatment effect.

Figure 9

Forest plot of comparison: 1 Intensive Case Management versus standard care, outcome: 1.2 Service use: 2. Not remaining in contact with psychiatric services by short, medium, long term and overall.

Figure 10

Forest plot of comparison: 2 Intensive Case Management versus non‐Intensive Case Management, outcome: 2.1 Service use: 1. Average number of days in hospital per month ‐ at about 24 months.

Figure 11

Funnel plot of comparison: 2 Intensive Case Management versus non‐Intensive Case Management, outcome: 2.1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months.

Analysis 1.1

Comparison 1 Intensive Case Management versus standard care, Outcome 1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months.

Analysis 1.2

Comparison 1 Intensive Case Management versus standard care, Outcome 2 Service use: 1a. Number of days in hospital ‐ by follow ‐up (skewed data, sample size ≧ 200).

Analysis 1.3

Comparison 1 Intensive Case Management versus standard care, Outcome 3 Service use: 2. Not remaining in contact with psychiatric services.

Analysis 1.4

Comparison 1 Intensive Case Management versus standard care, Outcome 4 Service use: 3a. Admitted to hospital.


Study	Intervention	Mean	SD	Total	Note
by medium term
Bond‐Chicago1 1990	1. ICM	0.16	0.15	42
Bond‐Chicago1 1990	2. Standard care	0.26	0.23	40
by long term
Audini‐UK 1994	1. ICM	0.02	0.05*	33
Audini‐UK 1994	2. Standard care	0.03	0.06*	33	* Carried over from Sytema‐Netherlands.
Muller‐Clemm‐Canada 1996	1. ICM	0.09	0.05*	61
Muller‐Clemm‐Canada 1996	2. Standard care	0.08	0.06*	57	* Carried over from Sytema‐Netherlands.
Sytema‐Netherlands 1999	1. ICM	0.05	0.05	58
Sytema‐Netherlands 1999	2. Standard care	0.05	0.06	57

Analysis 1.5

Comparison 1 Intensive Case Management versus standard care, Outcome 5 Service use: 3b. Average number of admissions per month (skewed data).

Analysis 1.6

Comparison 1 Intensive Case Management versus standard care, Outcome 6 Service use: 4a. Admitted to ER ‐ by long term.


Study	Intervention	Mean	SD	Total	Note
Jerrell‐SCarolina1 1991	1. ICM	0.85	1.7*	40
Jerrell‐SCarolina1 1991	2. Standard care	0.73	3.3*	40	* Carried over from Lehman‐Maryland1.
Lehman‐Maryland1 1994	1. ICM	0.9	1.7	77
Lehman‐Maryland1 1994	2. Standard care	2	3.3	75

Analysis 1.7

Comparison 1 Intensive Case Management versus standard care, Outcome 7 Service use: 4b. Average number of admissions to ER (skewed data) ‐ by medium term.

Analysis 1.8

Comparison 1 Intensive Case Management versus standard care, Outcome 8 Service use: 5a. Received day hospital care ‐ by short term FUP.

Analysis 1.9

Comparison 1 Intensive Case Management versus standard care, Outcome 9 Service use: 5b. Outpatient visits ‐ by short term FUP (6 months).


Study	Intervention	Mean	SD	Total	Note
Cusack‐North Carolina	ICM	95.9	57.1	72	‐
Cusack‐North Carolina	Standard care	43.3	47.9	62	‐

Analysis 1.10

Comparison 1 Intensive Case Management versus standard care, Outcome 10 Service use: 5c. Outpatient visits ‐ by medium term (skewed data).

Analysis 1.11

Comparison 1 Intensive Case Management versus standard care, Outcome 11 Service use: 5d. Received home visits ‐ by short term FUP.

Analysis 1.12

Comparison 1 Intensive Case Management versus standard care, Outcome 12 Adverse event: 1a. Death ‐ any cause.

Analysis 1.13

Comparison 1 Intensive Case Management versus standard care, Outcome 13 Adverse event: 1b. Death ‐ suicide.

Analysis 1.14

Comparison 1 Intensive Case Management versus standard care, Outcome 14 Global state: 1. Leaving the study early.

Analysis 1.15

Comparison 1 Intensive Case Management versus standard care, Outcome 15 Global state: 2. Average endpoint score (GAF, high = good).

Analysis 1.16

Comparison 1 Intensive Case Management versus standard care, Outcome 16 Global state: 3. Not compliant with medication ‐ by long term.

Analysis 1.17

Comparison 1 Intensive Case Management versus standard care, Outcome 17 Social functioning: 1a. Contact with legal system (various measurements).


Study	Intervention	Mean	SD	Total
Bookings
Cusack‐North Carolina	1. ICM	0.64	1.2	72
Cusack‐North Carolina	2. Standard care	1.42	1.86	62
Jail days
Cusack‐North Carolina	1. ICM	18.5	45.3	72
Cusack‐North Carolina	2. Standard care	35.3	56.9	62
Convictions
Cusack‐North Carolina	1. ICM	0.75	0.77	72
Cusack‐North Carolina	2. Standard care	0.85	1.03	62

Analysis 1.18

Comparison 1 Intensive Case Management versus standard care, Outcome 18 Social functioning: 1b. Mean contacts with legal system (skewed data) ‐ by medium term.

Analysis 1.19

Comparison 1 Intensive Case Management versus standard care, Outcome 19 Social functioning: 2. Employment status (various measurements).

Analysis 1.20

Comparison 1 Intensive Case Management versus standard care, Outcome 20 Social functioning: 3a. Accommodation status (various measurements).

Analysis 1.21

Comparison 1 Intensive Case Management versus standard care, Outcome 21 Social functioning: 3b. Accomodation status: mean number of days in supported house (skewed data, sample size ≧ 200).


Study	Intervention	Mean	SD	Total
by medium term ‐ average days per month in stable accommodation
Lehman‐Maryland1 1994	1. ICM	17.5	9	77
Lehman‐Maryland1 1994	2. Standard care	13.34	9	75
Morse‐Missouri3 2005	1. ICM	5.77	7.42	54
Morse‐Missouri3 2005	2. Standard care	5.02	8.62	49
by long term ‐ average days per month in sheltered homes
Morse‐Missouri3 2005	1. ICM	17.78	12.68	54
Morse‐Missouri3 2005	2. Standard care	12.59	13.27	49
Sytema‐Netherlands 1999	1. ICM	2.8	7.4	58
Sytema‐Netherlands 1999	2. Standard care	3.6	9.2	57

Analysis 1.22

Comparison 1 Intensive Case Management versus standard care, Outcome 22 Social functioning: 3c. Accommodation status (various measurements, skewed data).

Analysis 1.23

Comparison 1 Intensive Case Management versus standard care, Outcome 23 Social functioning: 4a. Substance abuse.


Study	Intervention	Mean	SD	Total
alcohol abuse (DALI, ‐4 to + 6, high = worse)
Sytema‐Netherlands 1999	1. ICM	‐0.8	2.7	45
Sytema‐Netherlands 1999	2. Standard Care	‐1.2	2.4	36
drug abuse (DALI, ‐ 4 to + 4, high = worse)
Sytema‐Netherlands 1999	1. ICM	‐1.4	1.3	45
Sytema‐Netherlands 1999	2. Standard Care	‐1.8	1.3	36

Analysis 1.24

Comparison 1 Intensive Case Management versus standard care, Outcome 24 Social functioning: 4b. Substance abuse (DALI, skewness not detectable) ‐ by medium term.


Study	Intervention	Mean	SD	Total
by medium term
Morse‐Missouri3 2005	1. ICM	6.25	7.84	54
Morse‐Missouri3 2005	2. Standard care	6.34	7.52	49
by long term
Morse‐Missouri3 2005	1. ICM	6.77	8.86	54
Morse‐Missouri3 2005	2. Standard care	6.42	7.84	49

Analysis 1.25

Comparison 1 Intensive Case Management versus standard care, Outcome 25 Social functioning: 4c. Substance abuse ‐ days used per month (skewed data).

Analysis 1.26

Comparison 1 Intensive Case Management versus standard care, Outcome 26 Social functioning: 5a. Average endpoint score (various scales).


Study	Intervention	Mean	SD	Total
by short term (SAS, high = poor)
Muijen‐UK2 1994	1. ICM	3.9	1.1	35
Muijen‐UK2 1994	2. Standard care	3.9	1.6	29
by medium term (SAS, high = poor)
Muijen‐UK2 1994	1. ICM	4.3	1.6	29
Muijen‐UK2 1994	2. Standard care	3.7	1.5	25
by long term (SAS, high = poor)
Audini‐UK 1994	1. ICM	3.0	1.6	30
Audini‐UK 1994	2. Standard care	2.9	1.1	28
Muijen‐UK2 1994	1. ICM	3.6	1.4	24
Muijen‐UK2 1994	2. Standard care	4.2	1.4	22
by long term (REHAB, high = poor)
Marshall‐UK 1995	1. ICM	31.7	29.3	31
Marshall‐UK 1995	2. Standard care	40.83	19.65	30

Analysis 1.27

Comparison 1 Intensive Case Management versus standard care, Outcome 27 Social functioning: 5b. Average endpoint score (various scales, skewed data).

Analysis 1.28

Comparison 1 Intensive Case Management versus standard care, Outcome 28 Mental state: 1a. General symptoms ‐ average endpoint score (various scales).

Analysis 1.29

Comparison 1 Intensive Case Management versus standard care, Outcome 29 Mental state: 1b. General symptoms ‐ mean change from baseline (CSI, low = poor ) ‐ by long term.


Study	Intervention	Mean	SD	Total
by short term (BPRS‐24 items, high = poor)
Audini‐UK 1994	1. ICM	39.1	10.0	31
Audini‐UK 1994	2. Standard Care	39.5	12.0	30
Muijen‐UK2 1994	1. ICM	42.4	16.8	36
Muijen‐UK2 1994	2. Standard Care	43.1	15.2	32
by short term (PSE, high = poor)
Audini‐UK 1994	1. ICM	7.2	7.2	31
Audini‐UK 1994	2. Standard Care	8.4	9.3	30
Muijen‐UK2 1994	1. ICM	19.9	19.5	35
Muijen‐UK2 1994	2. Standard Care	17.3	15.8	32
by medium term (BPRS‐24 items, high = poor)
Audini‐UK 1994	1. ICM	42.3	14.8	30
Audini‐UK 1994	2. Standard Care	41.4	12.2	28
Muijen‐UK2 1994	1. ICM	45.7	15.2	32
Muijen‐UK2 1994	2. Standard Care	43.1	12.7	26
Sytema‐Netherlands 1999	1. ICM	38	10	45
Sytema‐Netherlands 1999	2. Standard Care	42	11	36
by medium term (CPRS, high = poor)
Holloway‐UK 1996	1. ICM	20.6	12.1	22
Holloway‐UK 1996	2. Standard Care	21.3	14.0	22
by medium term (PSE, high = poor)
Muijen‐UK2 1994	1. ICM	18.7	15.9	35
Muijen‐UK2 1994	2. Standard Care	14.4	15	27
by long term (BPRS‐18 items, high = poor)
Ford‐UK 1995	1. ICM	12.8	9.6	36
Ford‐UK 1995	2. Standard Care	13.5	11.9	32
by long term (BPRS‐24 items, high = poor)
Muijen‐UK2 1994	1. ICM	44.4	13.3	31
Muijen‐UK2 1994	2. Standard Care	51.8	18.8	26
by long term (CPRS, high = poor)
Holloway‐UK 1996	1. ICM	21.6	12.9	21
Holloway‐UK 1996	2. Standard Care	22.4	14.5	19
by long term (PSE, high = poor)
Audini‐UK 1994	1. ICM	7.6	8.2	30
Audini‐UK 1994	2. Standard Care	10.6	12.2	28
Muijen‐UK2 1994	1. ICM	20.3	13.7	28
Muijen‐UK2 1994	2. Standard Care	27.6	23.5	25
by long term (SCL‐90, high = poor)
Bjorkman‐Sweden 2002	1. ICM	102	68.5	27
Bjorkman‐Sweden 2002	2. Standard Care	81.4	55.1	33

Analysis 1.30

Comparison 1 Intensive Case Management versus standard care, Outcome 30 Mental state: 1c. General symptoms ‐ average endpoint score (various scales, skewed data).

Analysis 1.31

Comparison 1 Intensive Case Management versus standard care, Outcome 31 Mental state: 2a. Specific symptoms ‐ depression at follow up interview.

Analysis 1.32

Comparison 1 Intensive Case Management versus standard care, Outcome 32 Mental state: 2b. Specific symptoms ‐ average endpoint score (various scales, skewed data, sample size ≧ 200).


Study	Intervention	Mean	SD	Total
by medium term ‐ depression symptoms (BDI, high = poor)
Holloway‐UK 1996	1. ICM	11.5	8.9	23
Holloway‐UK 1996	2. Standard care	18.5	13.9	19
by medium term ‐ negative symptoms (SANS, high = poor)
Holloway‐UK 1996	1. ICM	7.3	4	26
Holloway‐UK 1996	2. Standard care	6.3	4.4	22
by long term ‐ depression symptoms (BDI, high = poor)
Holloway‐UK 1996	1. ICM	12.8	8.1	25
Holloway‐UK 1996	2. Standard care	14.8	11.5	17
by long term ‐ negative symptoms (SANS, high = poor)
Holloway‐UK 1996	1. ICM	7.3	3.7	26
Holloway‐UK 1996	2. Standard care	7.1	4.1	20

Analysis 1.33

Comparison 1 Intensive Case Management versus standard care, Outcome 33 Mental state: 2c. Specific symptoms ‐ average endpoint score (various scales, skewed data).

Analysis 1.34

Comparison 1 Intensive Case Management versus standard care, Outcome 34 Behaviour: 1. Specific behaviour ‐ self‐harm.


Study	Intervention	Mean	SD	Total
by medium term
Holloway‐UK 1996	1. ICM	3.2	2.8	33
Holloway‐UK 1996	2. Standard Care	2.4	2.9	30
by long term
Holloway‐UK 1996	1. ICM	3.3	2.8	34
Holloway‐UK 1996	2. Standard Care	2.7	2.2	26

Analysis 1.35

Comparison 1 Intensive Case Management versus standard care, Outcome 35 Behaviour: 2. Social behaviour ‐ average endpoint score (SBS, high = poor).

Analysis 1.36

Comparison 1 Intensive Case Management versus standard care, Outcome 36 Quality of life: 1a. Average endpoint score (various scales).


Study	Intervention	Mean	SD	Total
Shern‐USA1 2000	1. ICM	1.19	1.99	91
Shern‐USA1 2000	2. Standard Care	‐0.02	1.65	77

Analysis 1.37

Comparison 1 Intensive Case Management versus standard care, Outcome 37 Quality of life: 1b. Mean change from baseline (QOLI, high = better, skewed data) ‐ by long term.

Analysis 1.38

Comparison 1 Intensive Case Management versus standard care, Outcome 38 Participant satisfaction: 1a. Average endpoint score (CSQ, high = better).


Study	Intervention	Mean	SD	Total	Note
Muijen‐UK2 1994	1. ICM	26	5.4	30
Muijen‐UK2 1994	2. Standard Care*	22	7.5	13	* Attrition >50%.

Analysis 1.39

Comparison 1 Intensive Case Management versus standard care, Outcome 39 Participants satisfaction: 1b. Average endpoint score (CSQ, high = better, skewed data) ‐ by short term.


Study	Intervention	Mean	SD	Total	Note
by medium term ‐ met needs (CANSAS, high = better)
Sytema‐Netherlands 1999	1. ICM	8.5	4.5	45
Sytema‐Netherlands 1999	2. Standard Care	8.6	4.7	36
by medium term ‐ unmet needs (CANSAS, high = poor)
Sytema‐Netherlands 1999	1. ICM	1.4	1.9	45
Sytema‐Netherlands 1999	2. Standard Care	1.6	1.7	36
by long term (CAN, high = poor)
Bjorkman‐Sweden 2002	1. ICM	3.2	1.8	28
Bjorkman‐Sweden 2002	2. Standard Care	4.6	3.8	36

Analysis 1.40

Comparison 1 Intensive Case Management versus standard care, Outcome 40 Participants need: 1. Average endpoint score (various scales, skewed data).

Analysis 1.41

Comparison 1 Intensive Case Management versus standard care, Outcome 41 Costs: 1a. Direct costs of psychiatric hospital care ‐ by medium term (Unit cost = USD, fiscal year 1990).


Study	Intervention	Mean	SD	Total	Note
by medium term
Cusack‐North Carolina	1. ICM*	5,530	12,414	72	* Unit cost US $, Inpatient costs Time period: 12 months.
Cusack‐North Carolina	2. Standard care*	8,827	19,289	62	* Unit cost US $, Inpatient costs Time period: 12 months.
Lehman‐Maryland1 1994	1. ICM*	2,619	4,440	77
Lehman‐Maryland1 1994	2. Standard care*	4,662	6,034	75	* Unit cost US $, fiscal year 1994. t‐value=2.34 Time period: 12 months.
Morse‐Missouri3 2005	1. ICM*	624	2,314	54
Morse‐Missouri3 2005	2. Standard care*	439	1,596	49	* Unit cost US $, fiscal year 2001. "No main effect of of treatment condition for inpatient costs,F(2, 146)=0.10, p=0.9, ɧ²=0.01." Time period: 6 months.
by long term
Ford‐UK 1995	1. ICM*	378	846	39
Ford‐UK 1995	2. Standard care*	237	492	38	* Unit cost £, fiscal year not reported, study base year 1990. ** No statistical analysis available from the paper. Time period: 18 months.
Morse‐Missouri3 2005	1. ICM*	855	2,356	54
Morse‐Missouri3 2005	2. Standard care*	455	1,065	49	* Unit cost US $, fiscal year 2001. ** "No main effect of treatment condition for inpatient costs,F(2, 146)=0.10, p=0.9, ɧ²=0.01." Time period: 6 months.
Quinlivan‐California 1995	1. ICM*	301	397	30
Quinlivan‐California 1995	2. Standard care*	1,636	2,593	30	* Unit cost US $, fiscal year not reported, but study was carried on from April 1990 to March 1992. ** "Costs significantly lower for the ICM group (F=4.32, df=2.87, p=0.02.)" Time period: 24 months.

Analysis 1.42

Comparison 1 Intensive Case Management versus standard care, Outcome 42 Costs: 1b. Direct costs of psychiatric hospital care ‐ skewed data.

Analysis 1.43

Comparison 1 Intensive Case Management versus standard care, Outcome 43 Costs: 2a. Direct healthcare costs ‐ by long term (Unit cost = USD, fiscal year 1988).


Study	Intervention	Mean	SD	Total	Note
by medium term
Lehman‐Maryland1 1994	1. ICM*	4,229	5,058	77
Lehman‐Maryland1 1994	2. Standard care*	5,540	6,368	75	* Unit cost US $, fiscal year 1994. ** 'Total per‐case cost did not reach statistical significance (p = 0.07). Transformation of total costs per case to account for non‐normality (square root of total costs, t‐test=0.77, df=1,134, NS) and non‐parametric analysis (Wilcoxon test for ranks, Z=0.146, NS) also were non‐significant.' Time period 12 months.
by short term FUP
Chan‐Hong Kong 2000	1. ICM	14,833	1,539	31	HK $ (HK$8=US$1, at time of study publication, 2000). Statistically significant difference (P = 0.017).
Chan‐Hong Kong 2000	2. Standard care	11,230	7,979	31

Analysis 1.44

Comparison 1 Intensive Case Management versus standard care, Outcome 44 Costs: 2b. Direct healthcare costs ‐ skewed data.


Study	Intervention	Mean	SD	Total	Note
all care ‐ by short term
Audini‐UK 1994	1.ICM*	4,264	1,768	33
Audini‐UK 1994	2. Standard care*	7,202	5,564	29	* Unit cost £, fiscal year 1996/7. 'Bivariate cost comparisons (after log transformation) revealed significant advantage for ICM group (p=0.001)'. *Time period: missing (it is not the monthly cost per patient).
all care ‐ by medium term
Marshall‐UK 1995	1. ICM*	1,044	425.3	31
Marshall‐UK 1995	2. Standard care*	1,108	530.4	30	* Unit cost £, fiscal year 1994. 'No significant differences between two groups were found.' *Time period: mean weekly cost.
Morse‐Missouri3 2005	1. ICM*	2,946.8	3,219.3	54
Morse‐Missouri3 2005	2. Standard care*	1,899.5	3,629.6	49	* Unit cost US $, fiscal year 2001. 'There was a main effect of treatment condition on total costs, F(2, 146)=4.00, p=0.02, ɧ²=0.05. Standard care condition had significantly lower costs than ICM.' *Time period: 6 months
all care ‐ by long term
Audini‐UK 1994	1. ICM*	10,192	3,900	32
Audini‐UK 1994	2. Standard care*	15,288	17,160	28	* Unit cost £, fiscal year 1996/7. 'Bivariate cost comparisons (after log transformation) did not revealed significant advantage for ICM group (p=0.09)'. *Time period: missing (it is not the monthly cost per patient).
Ford‐UK 1995	1. ICM*	1,813	1,347	39
Ford‐UK 1995	2. Standard care*	717	768	38	* Unit cost £, fiscal year not reported, study base year 1990. 'ANOVA analysis carried on, revealing significant advantage for ICM group (p<0.05).' *Time period: 12 months.
Marshall‐UK 1995	1. ICM*	996	398	31
Marshall‐UK 1995	2. Standard care*	1,088	562.4	30	* Unit cost £, fiscal year 1994. 'No significant differences between two groups were found.' *Time period: mean weekly cost.
Morse‐Missouri3 2005	1. ICM*	3,190	3,441	54
Morse‐Missouri3 2005	2. Standard care*	1,467	2,173	49	* Unit cost US $, fiscal year 2001. 'There was a main effect of treatment condition on total costs, F(2, 146)=4.00, p=0.02, ɧ²=0.05. Standard care condition had significantly lower costs than ICM.' *Time period: 6 months.
OPUS‐Denmark 1999	1. ICM*	111,924	100,862	151
OPUS‐Denmark 1999	2. Standard Care*	137,638	147,570	150	Unit cost Euro €, fiscal year 2009. No significant difference between the groups. **Time period: FUP 3 years.
specific ‐ outpatient care ‐ by medium term
Cusack‐North Carolina	1. ICM*	13,481	9,547	72	* Unit cost US $. **Time period: 12 months.
Cusack‐North Carolina	2. Standard care*	5,118	6,184	62	* Unit cost US $. **Time period: 12 months.
specific ‐ prison ‐ by medium term
Cusack‐North Carolina	1. ICM*	1,848	4,533	72	* Unit cost US $. **Time period: 12 months.
Cusack‐North Carolina	2. Standard care*	3,530	5,690	62	* Unit cost US $. **Time period: 12 months.

Analysis 1.45

Comparison 1 Intensive Case Management versus standard care, Outcome 45 Costs: 3. Direct costs ‐ other data ‐ skewed data.

Analysis 2.1

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months.

Analysis 2.2

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 2 Service use: 1a. Average number of days in hospital per month ‐ by medium/long term follow up (skewed data, sample size ≧ 200).

Analysis 2.3

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 3 Service use: 2. Not remaining in contact with psychiatric services.

Analysis 2.4

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 4 Service use: 3a. Admitted to hospital ‐ by long term.

Analysis 2.5

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 5 Service use: 3b. Average number of admissions (skewed data ‐sample size ≧ 200).


Study	Intervention	Mean	SD	Total
Johnston‐Australia 1998	1. ICM	1.6	2	35
Johnston‐Australia 1998	2. non‐ICM	1.9	2.4	33

Analysis 2.6

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 6 Service use: 3c. Average number of admissions (skewed data) ‐ by medium term.

Analysis 2.7

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 7 Adverse event: 1a. Death ‐ any cause.

Analysis 2.8

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 8 Adverse event: 1b. Death ‐ suicide.

Analysis 2.9

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 9 Global state: 1. Leaving the study early.

Analysis 2.10

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 10 Global state: 2a. Average endpoint score (HoNOS, high = poor) ‐ by long term.


Study	Intervention	Mean	SD	Total
medium term
Harrison‐Read‐UK 2000	1. ICM	12	6.8	54
Harrison‐Read‐UK 2000	2. non‐ICM	11.4	6.4	64
long term
Harrison‐Read‐UK 2000	1. ICM	11.9	5.9	60
Harrison‐Read‐UK 2000	2. non‐ICM	10.4	6.4	59

Analysis 2.11

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 11 Global state: 2b. Average endpoint score (HoNOS, high = poor) ‐ skewed data.

Analysis 2.12

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 12 Global state: 3a. Not compliant with medication ‐ by medium term.

Analysis 2.13

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 13 Global state: 3b. Compliance with medication ‐ average endpoint sub‐scale score (ROMI) ‐ by long term.


Study	Intervention	Mean	SD	Total
medium term ‐ compliance sub‐scale (high = good)
Harrison‐Read‐UK 2000	1. ICM	1.8	0.4	49
Harrison‐Read‐UK 2000	2. non‐ICM	2.0	0.5	61
medium term ‐ non‐compliance sub‐scale (high = poor)
Harrison‐Read‐UK 2000	1. ICM	1.3	0.3	49
Harrison‐Read‐UK 2000	2. non‐ICM	1.2	0.3	61
long term ‐ compliance sub‐scale (high = good)
Harrison‐Read‐UK 2000	1. ICM	1.8	0.4	62
Harrison‐Read‐UK 2000	2. non‐ICM	1.9	0.5	60
long term ‐ non‐compliance sub‐scale (high = poor)
Harrison‐Read‐UK 2000	1. ICM	1.2	0.3	63
Harrison‐Read‐UK 2000	2. non‐ICM	1.2	0.3	61

Analysis 2.14

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 14 Global state: 3c. Compliance with medication ‐ average endpoint sub‐scale score (ROMI, score 1‐3, skewed data).

Analysis 2.15

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 15 Social functioning: 1. Contact with legal system (various measurements).

Analysis 2.16

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 16 Social functioning 2. Employment status (various measurements).

Analysis 2.17

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 17 Social functioning: 3a. Accommodation status (various measurements).

Analysis 2.18

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 18 Social functioning: 3b. Accommodation status ‐ average days per month in stable accommodation.

Analysis 2.19

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 19 Social functioning: 4a. Substance abuse ‐ by long term.

Analysis 2.20

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 20 Social functioning: 4b. Substance abuse ‐ average endpoint score (SATS, low = poor).


Study	Intervention	Mean	SD	Total
short term ‐ days using alcohol during previous 6 months (TLFB)
Drake‐NHamp 1998	1. ICM	56.8	56.4	75
Drake‐NHamp 1998	2. non‐ICM	47.5	58.4	68
short term ‐ average endpoint score (AUS, high = poor)
Drake‐NHamp 1998	1. ICM	3.09	1.02	83
Drake‐NHamp 1998	2. non‐ICM	2.91	1.08	73
medium term ‐ days using alcohol during previous 6 months (TLFB)
Drake‐NHamp 1998	1. ICM	59.1	53.3	75
Drake‐NHamp 1998	2. non‐ICM	42.8	52.9	68
medium term ‐ average endpoint score (AUS, high = poor)
Drake‐NHamp 1998	1. ICM	3.11	1.05	83
Drake‐NHamp 1998	2. non‐ICM	2.8	1.13	73
long term ‐ days using alcohol during previous 6 months (TLFB)
Drake‐NHamp 1998	1. ICM	46.4	53.6	75
Drake‐NHamp 1998	2. non‐ICM	43.6	57.3	68
long term ‐ average endpoint score (AUS, high = poor)
Drake‐NHamp 1998	1. ICM	2.64	1.12	83
Drake‐NHamp 1998	2. non‐ICM	2.77	1.18	73

Analysis 2.21

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 21 Social functioning: 4c. Alcohol ‐ abuse (various measurements, skewed data).

Analysis 2.22

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 22 Social functioning: 5a. Average endpoint score (LSP, high = poor) ‐ by long term.


Study	Intervention	Mean	SD	Tot
by medium term
Harrison‐Read‐UK 2000	1. ICM	7.3	5.3	49
Harrison‐Read‐UK 2000	2. non‐ICM	7.5	5.1	62
by long term
Harrison‐Read‐UK 2000	1. ICM	8.9	4.9	57
Harrison‐Read‐UK 2000	2. non‐ICM	7.9	4.9	58

Analysis 2.23

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 23 Social functioning: 5b. Average endpoint score (SFQ, high = poor) ‐ skewed data.

Analysis 2.24

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 24 Mental state: 1a. General symptoms ‐ average endpoint score (various scales).


Study	Intervention	Mean	SD	Total
by medium term (Krawiecka Scale, high = poor)
Harrison‐Read‐UK 2000	1. ICM	8.8	5.6	57
Harrison‐Read‐UK 2000	2. non‐ICM	8	4.5	65
by long term (Krawiecka Scale, high = poor)
Harrison‐Read‐UK 2000	1. ICM	9.2	5.5	47
Harrison‐Read‐UK 2000	2. non‐ICM	7.9	4.5	57
by long term (BPRS 24‐items, high = good)
REACT‐UK 2002	1. ICM	32.9	9	91
REACT‐UK 2002	2. non‐ICM	33.5	8.6	75

Analysis 2.25

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 25 Mental state: 1b. General symptoms ‐ average endpoint scores (various scales, skewed data).

Analysis 2.26

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 26 Mental state: 2a. Specific symptoms: negative symptoms ‐ average endpoint score (SANS, high = poor) ‐ by long term.


Study	Intervention	Mean	SD	Total
medium term ‐ anxiety (HADS, high = poor)
Harrison‐Read‐UK 2000	1. ICM	6.5	4.9	52
Harrison‐Read‐UK 2000	2. non‐ICM	6.7	4.6	61
medium term ‐ depression (HADS, high = poor)
Harrison‐Read‐UK 2000	1. ICM	6.4	5.4	52
Harrison‐Read‐UK 2000	2. non‐ICM	6.6	4.9	61
long term ‐ anxiety (HADS, high = poor)
Harrison‐Read‐UK 2000	1. ICM	7.5	5.3	56
Harrison‐Read‐UK 2000	2. non‐ICM	6.4	4.6	58
long term ‐ depression (HADS, high = poor)
Harrison‐Read‐UK 2000	1. ICM	7.3	5.4	56
Harrison‐Read‐UK 2000	2. non‐ICM	6.8	5.6	58

Analysis 2.27

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 27 Mental state: 2b. Specific symptoms ‐ average endpoint scores (various scales, skewed data).

Analysis 2.28

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 28 Behaviour: 1. Specific behaviour (various measurements).

Analysis 2.29

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 29 Quality of life: 1. Average endpoint score (various scales).

Analysis 2.30

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 30 Participant satisfaction/need: 1. Average endpoint scores (various scale) ‐ by long term.


Study	Intervention	Mean	SD	Total
by medium term (CAN, high = poor)
Harrison‐Read‐UK 2000	1. ICM	7.3	3.7	49
Harrison‐Read‐UK 2000	2. non‐ICM	6.1	4	60
by long term (CAN, high = poor)
Harrison‐Read‐UK 2000	1. ICM	6.6	3.6	54
Harrison‐Read‐UK 2000	2. non‐ICM	5.6	3.8	59
by long term (CANSAS, high = poor)
REACT‐UK 2002	1. ICM	3.3	2.7	91
REACT‐UK 2002	2. non‐ICM	3.4	2.9	75

Analysis 2.31

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 31 Participants need: 1. Average endpoint scores (various scales, skewed data).


Study	Intervention	Mean	SD	Total
REACT‐UK 2002	1. ICM	77.2	20	91
REACT‐UK 2002	2. non‐ICM	70	20.6	75

Analysis 2.32

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 32 Participant satisfaction: 1. Average endpoint scores (CSQ‐modified, high = better, skewed data) ‐ by long term.


Study	Intervention	Mean	SD	Total	Note
by medium term
Harrison‐Read‐UK 2000	1. ICM*	501	967.4	97
Harrison‐Read‐UK 2000	2. Low ICM*	527	753	96	* Unit cost £, fiscal year 1995/6. 'No significant difference between groups. Statistical analysis on non‐parametric data were performed using bootstrap methods'. *Time period: 12 months.
by long term
Harrison‐Read‐UK 2000	1. ICM*	414	777.8	97
Harrison‐Read‐UK 2000	2. Low ICM*	478	890	96	* Unit cost £, fiscal year 1995/6. 'No significant difference between groups. Statistical analysis on non‐parametric data were performed using bootstrap methods'. *Time period: 12 months.
Quinlivan‐California 1995	1. ICM*	301	396.6	30
Quinlivan‐California 1995	2. Low ICM*	959	1,572.7	30	* Unit cost US $, fiscal year not reported, but study was carried on from April 1990 to March 1992. 'Costs significantly lower for the ICM group (F=4.32, df=2.87, p=0.02.)' *Time period: 24 months

Analysis 2.33

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 33 Costs: 1. Direct costs of psychiatric hospital care (skewed data).

Analysis 2.34

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 34 Costs: 2a. Direct costs of all care ‐ by long term (2 years). Unit cost GBP, fiscal year 1997/98.


Study	Intervention	Mean	SD	Total	Note
Johnston‐Australia 1998	1. ICM*	2,408	2,581.4	33
Johnston‐Australia 1998	2. Non‐ICM*	1,762	1,872	25	* Unit cost Aus $, fiscal year 1991/2. 'The significance test on the cost of care per patient was performed on transformed means. No significant differences were found between groups'. *Time period: 12 months.

Analysis 2.35

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 35 Costs: 2b. Direct costs of all care ‐ by medium term (skewed data).

Analysis 2.36

Comparison 2 Intensive Case Management versus non‐Intensive Case Management, Outcome 36 Costs: 3. Total costs of care per patient ‐ Unit cost GBP.

Summary of findings for the main comparison. Intensive Case Management versus standard care for severe mental illness

Intensive Case Management versus standard care for severe mental illness
Patient or population: people with severe mental illness Settings: community Intervention: Intensive Case Management versus standard care
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control	Intensive Case Managementversus standard care
Service use: 1. Average number of days in hospital per month ‐ by about 24 months	‐	The mean service use: 1. average number of days in hospital per month ‐ by about 24 months in the intervention groups was 0.86 lower (1.37 lower to 0.34 lower)	‐	3595 (24 studies)	⊕⊕⊝⊝ low^1,2
Adverse event: 1b. Death ‐ suicide ‐ by long term	20 per 1000	13 per 1000 (6 to 30)	RR 0.68 (0.31 to 1.51)	1456 (9 studies)	⊕⊕⊝⊝ low^1,4
Global state: 1. Relapse ‐ by long term	‐	‐	‐	‐	‐	No data available
Global state: 1. Leaving the study early ‐ by long term	331 per 1000	225 per 1000 (192 to 262)	RR 0.68 (0.58 to 0.79)	1798 (13 studies)	⊕⊕⊝⊝ low^1,3
Social functioning: 2. Employment status (various measurements) ‐ by long term ‐ not employed at the end of the trial	766 per 1000	536 per 1000 (375 to 766)	RR 0.7 (0.49 to 1)	1129 (4 studies)	⊕⊝⊝⊝ very low^1,5
Mental state: not improved to an important extent ‐ by long term	‐	‐	‐	‐	‐	No data available
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Downgraded one step for risk of bias: randomisation not well described; problematic to blind. ²Downgraded one step for inconsistency: substantial heterogeneity (I² = 74%). ³Downgraded one step for selective reporting bias: only 13 studies reported fully on the flow of participants through the study. ⁴Downgraded one step for imprecision: the 95% CI includes both appreciable benefit and appreciable harm. ⁵Downgraded two steps for inconsistency: considerable heterogeneity (I² = 94%).

Summary of findings for the main comparison. Intensive Case Management versus standard care for severe mental illness

Summary of findings 2. Intensive Case Management versus non‐Intensive Case Management for severe mental illness

Intensive Case Management versus non‐Intensive Case Management for severe mental illness
Patient or population: people with severe mental illness Settings: community Intervention: Intensive Case Management versus non‐Intensive Case Management
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control	Intensive Case Management versus non‐Intensive Case Management
Service use: 1. Average number of days in hospital per month ‐ by about 24 months	‐	The mean service use: 1. average number of days in hospital per month ‐ by about 24 months in the intervention groups was 0.08 lower (0.37 lower to 0.21 higher)	‐	2220 (21 studies)	⊕⊕⊕⊝ moderate¹
Service use: 3b. Average number of admissions (skewed data ‐ sample size ≧ 200) ‐ by long term	‐	The mean service use: 3b. average number of admissions (skewed data ‐ sample size ≧ 200) ‐ by long term in the intervention groups was 0.18 lower (0.41 lower to 0.05 higher)	‐	678 (1 studies)	⊕⊕⊕⊝ moderate¹
Adverse event: 1b. Death ‐ suicide ‐ by long term	12 per 1000	11 per 1000 (3 to 35)	RR 0.88 (0.27 to 2.84)	1152 (3 studies)	⊕⊕⊝⊝ low^1,3
Global state: 1. Relapse ‐ by long term	‐	‐	‐	‐	‐	No data available
Global state: 1. Leaving the study early ‐ by long term	159 per 1000	111 per 1000 (83 to 151)	RR 0.7 (0.52 to 0.95)	1970 (7 studies)	⊕⊕⊝⊝ low^1,2
Social functioning 2. Employment status ‐ by medium term ‐ spent > 1 day employed	111 per 1000	162 per 1000 (50 to 527)	RR 1.46 (0.45 to 4.74)	73 (1 study)	⊕⊝⊝⊝ very low^1,4
Mental state: not improved to an important extent ‐ by long term	‐	‐	‐	‐	‐	No data available
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Downgraded one step for risk of bias: randomisation not well described; problematic to blind. ²Downgraded one step for selective reporting bias: only 7 studies reported fully on the flow of participants through the study. ³Downgraded one step for imprecision: the 95% CI includes both appreciable benefit and appreciable harm. ⁴Downgraded two steps for imprecision: the 95% CI includes both appreciable benefit and appreciable harm, and only 73 participants were included.

Summary of findings 2. Intensive Case Management versus non‐Intensive Case Management for severe mental illness

Table 1. Case Management and Assertive Community Treatment

1. Case Management (CM)

assessing the patient's needs;
developing a care plan;
arranging suitable care from community services;
keeping contact with the patient.

Initially, in its simplest form (referred to as 'brokerage'), case managers were not mental health professionals, did not provide any direct care, and worked independently.

2. Assertive Community Treatment (ACT)

provide all necessary care themselves, rather than arranging for it to be provided by other services;
provide care at home or in workplaces;
carry low caseloads (usually 10 to 15 patients per member);
practice 'assertive outreach', meaning that they persist in attempts to engage unco‐operative clients;
place particular emphasis on medication compliance;
provide 24‐hour emergency cover.

Table 1. Case Management and Assertive Community Treatment

Table 2. Average number of days in hospital per month ‐ at about 24 months ‐ entering meta‐regression

Intensive Case Management versus standard care	ICM	ICM	ICM	SC	SC	SC	Note
Study ID	Mean	SD	Total	Mean	SD	Total	Note
Audini‐UK 1994	0.95	2.84*	33	0.93	2.03*	33	*SD imputed
Bjorkman‐Sweden 2002	0.83	3.13	33	2.15	4.13	44
Bond‐Chicago1 1990	3.22	4.55	42	5.3	5.42	40
Bond‐Indiana1 (A)	1.28	3.17*	29	7.72	8.99*	32	*SD imputed
Bond‐Indiana1 (B)	2.72	4.54*	34	3.62	5.24*	30	*SD imputed
Bond‐Indiana1 (C)	0.05	1.89*	21	3.38	4.98*	21	*SD imputed
Chandler‐California1 (A)	0.47	2.34*	102	0.78	1.84*	101	*SD imputed
Chandler‐California1 (B)	0.67	2.55*	115	0.96	2.07*	114	*SD imputed
Curtis‐New York 1992	1.77	1.79	146	1.02	1.18	143
Ford‐UK 1995	3.07	6.9	39	1.76	3.67	38
Hampton‐Illinois (A)	1.75	3.63*	48	4.83	6.49*	47	*SD imputed
Hampton‐Illinois (B)	3.25	5.01*	34	3.42	5.02*	36	*SD imputed
Holloway‐UK 1996	2.4	5.1	34	1.2	3	26
Jerrell‐SCarolina1 1991	0.53	2.40*	40	0.8	1.86*	40	*SD imputed
Lehman‐Maryland1 1994	3.04	5.15	77	5.41	7	75
Marshall‐UK 1995	1.04	2.18	40	1.56	4.45	40
Muijen‐UK2 1994	2.53	5.55	41	2.45	5.83	41
Muller‐Clemm‐Canada 1996	1.68	3.56*	61	1.63	2.93*	57	*SD imputed
OPUS‐Denmark 1999	5.11	7.7	263	6.57	8.73	244
Quinlivan‐California 1995	1.09	2.65	30	5.53	8.65	30
Rosenheck‐USA‐GMS (A)	3.63	3.89	44	3.71	2.76	35
Rosenheck‐USA‐GMS (B)	6.99	4.85	47	4.23	5.18	47
Rosenheck‐USA‐NP (C)	18.52	11.16	50	19.16	12.19	43
Rosenheck‐USA‐GMS (D)	2.8	3.31	49	3.26	3.98	53
Rosenheck‐USA‐NP (E)	4.13	5.24	34	3.05	4.61	33
Rosenheck‐USA‐GMS (F)	2.39	3.16	43	2.58	2.45	35
Rosenheck‐USA‐NP (G)	7.68	7.72	40	12.2	10.65	31
Rosenheck‐USA‐NP (H)	4.63	8.58	59	11.21	13.38	55
Rosenheck‐USA‐GMS (I)	5.62	4.67	44	7.8	6.63	44
Sytema‐Netherlands 1999	3.4	5.4	58	4.3	7.3	57
Test‐Wisconsin 1985	0.42	2.29*	72	2.13	3.54*	41	*SD imputed
Intensive Case Management versus non‐Intensive Case Management	ICM	ICM	ICM	Non‐ICM	Non‐ICM	Non‐ICM	Note
Study ID	Mean	SD	Total	Mean	SD	Total	Note
Bush‐Georgia 1990	1.58	3.46*	14	2.39	3.85*	14	*SD imputed
Drake‐NHamp (A)	0.5	0.94	7	2.17	3.21	9
Drake‐NHamp (B)	0.85	1.43	16	1.41	2.06	14
Drake‐NHamp (C)	2.28	3.2	10	1.67	3.84	12
Drake‐NHamp (D)	1.04	2.44	13	0.63	0.91	11
Drake‐NHamp (E)	1.08	4.15	30	1.39	2.36	27
Drake‐NHamp (F)	1.66	4.49	10	0.84	2.33	13
Drake‐NHamp 1998 G	2.05	3.06	9	0.87	0.92	8
Essock‐Connecticut1 1995	2.87	7.82	130	4.3	9.52	132
Essock‐Connecticut2 2006	0.64	1.9	99	0.72	1.3	99
Harrison‐Read‐UK 2000	2.94	5.74	97	3.76	5.83	96
Johnston‐Australia 1998	4.0	5.75	35	3.08	4.3	33
McDonel‐Indiana (A)	3.15	7.1	61	1.43	2.91	64
McDonel‐Indiana (B)	1.22	3.66	14	0.58	1.29	17
Quinlivan‐California 1995	1.09	2.65	30	2.8	4.74	30
REACT‐UK 2002	9.0	8.9	124	8.0	7.8	119
Salkever‐SCarolina 1999	1.12	3.01*	91	1.3	2.51*	53	*SD imputed
UK700‐UK (A)	3.08	5.77	94	2.64	3.49	95
UK700‐UK (B)	3.2	4.79	77	3.16	4.97	73
UK700‐UK (C)	3.29	5.41	76	2.48	4.71	75
UK700‐UK (D)	2.74	4.69	91	3.79	5.22	98
ICM: Intensive Case Management SC: standard care SD: standard deviation Study ID: Study identification name

Table 2. Average number of days in hospital per month ‐ at about 24 months ‐ entering meta‐regression

Table 3. Covariates entering meta‐regression

Intensive Case Management versus standard care	Baseline hospital use	Baseline hospital use	IFACT	IFACT	IFACT	Note
Study ID	Mean	Total	Total score	Organisation subscale score	Staff subscale score	Note
Audini‐UK 1994	1.08	66	6.7	3.5	3.2
Bjorkman‐Sweden 2002	5.63	77	7	4.5	2.5
Bond‐Chicago1 1990	7.83	88	6	4	2
Bond‐Indiana1 (A)	14.17	61	9.2	7	2.2
Bond‐Indiana1 (B)	4.95	64	2.2	1	1.2
Bond‐Indiana1 (C)	10.86	42	7.4	5	2.4
Chandler‐California1 (A)	0.5	203	8.5	5	3.5
Chandler‐California1 (B)	1.14	229	6.6	5	1.6
Curtis‐New York 1992	0.95*	289	5.8	3.5	2.3	*Mean imputed
Ford‐UK 1995	2.61	77	4.8	2	2.8
Hampton‐Illinois (A)	5.6	95	6	4	2
Hampton‐Illinois (B)	5.2	70	5	3	2
Holloway‐UK 1996	7.37	70	9.3	6	3.3
Jerrell‐SCarolina1 1991	2.85	80	8.8	5.5	3.3
Lehman‐Maryland1 1994	4.94*	152	11	7	4	*Mean imputed
Marshall‐UK 1995	3.31*	80	4.9	4	0.9	*Mean imputed
Muijen‐UK2 1994	8.43*	82	5.4	3	2.4	*Mean imputed
Muller‐Clemm‐Canada 1996	4.07	123	6.2	4	2.2
OPUS‐Denmark 1999	NA	547	8	4	4	*Baseline hospital use: not applicable as first episode
Quinlivan‐California 1995	4.50*	60	6.4	4	2.4	*Mean imputed
Rosenheck‐USA‐GMS (A)	3.96	79	6	2	4
Rosenheck‐USA‐GMS (B)	5.83	94	3.8	2	1.8
Rosenheck‐USA‐NP (C)	19.8	93	7.7	5	2.7
Rosenheck‐USA‐GMS (D)	4.19	102	7	3	4
Rosenheck‐USA‐NP (E)	5.33	67	6.4	3.5	2.9
Rosenheck‐USA‐GMS (F)	3.22	78	6.6	3	3.6
Rosenheck‐USA‐NP (G)	11.42	71	8.4	5	3.4
Rosenheck‐USA‐NP (H)	11.4	114	6.4	4	2.4
Rosenheck‐USA‐GMS (I)	8.28	88	5.8	2	3.8
Sytema‐Netherlands 1999	12.17*	118	7.6*	5.1*	2.5*	*Mean and IFACT score imputed
Test‐Wisconsin 1985	2.33	122	8.5	5.5	3
Intensive Case Management versus non‐Intensive Case Management	Baseline hospital use	Baseline hospital use	IFACT	IFACT	IFACT	Note
Study ID	Mean	Total	Total score	Organisation subscale score	Staff subscale score	Note
Bush‐Georgia 1990	3.99	28	3.1	2	1.1
Drake‐NHamp (A)	2.88	19	8	5	3
Drake‐NHamp (B)	1.72	33	3.8	3	0.8
Drake‐NHamp (C)	3.02	25	8.8	5.5	3.3
Drake‐NHamp (D)	1.78	26	7.8	4.5	3.3
Drake‐NHamp (E)	2.76	66	8.5	4.5	4
Drake‐NHamp (F)	2.34	22	3.5	3	0.5
Drake‐NHamp 1998 (G)	4.1	19	5	2	3
Essock‐Connecticut1 1995	2.81*	262	8.5	4.5	4	*Mean imputed
Essock‐Connecticut2 2006	1.08*	198	10*	7*	3*	*Mean and IFACT score imputed
Harrison‐Read‐UK 2000	4.11	193	7.6	4	3.6
Johnston‐Australia 1998	3.66	71	7.3	3.5	3.8
McDonel‐Indiana (A)	4.2	152	4.2	3	1.2
McDonel‐Indiana (B)	1.16	39	4.4	3	1.4
Quinlivan‐California 1995	2.96*	60	6.4	4	2.4	*Mean imputed
REACT‐UK 2002	7.3	251	10.3	6.5	3.8
Salkever‐SCarolina 1999	3.06	144	7	5	2
UK700‐UK (A)	4.55	196	8.8	5	3.8
UK700‐UK (B)	4.66	153	4.5	3	1.5
UK700‐UK (C)	4.33	158	4.2	2	2.2
UK700‐UK (D)	4.59	200	8.5	5	3.5
Baseline hospital use: average number of days per month in hospital for all participants in the two years before the study began IFACT: Index of Fidelity to Assertive Community Treatment NA: not applicable Study ID: Study identification name

Table 3. Covariates entering meta‐regression

Table 4. Interventions in Curtis‐New York

Comparison 1. Intensive Case Management versus standard care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months Show forest plot	24	3595	Mean Difference (IV, Random, 95% CI)	‐0.86 [‐1.37, ‐0.34]

1.1 skewed data (sample size ≧ 200)	5	1812	Mean Difference (IV, Random, 95% CI)	‐0.46 [‐0.95, 0.03]
1.2 skewed data (sample size < 200)	19	1783	Mean Difference (IV, Random, 95% CI)	‐1.01 [‐1.74, ‐0.28]
2 Service use: 1a. Number of days in hospital ‐ by follow ‐up (skewed data, sample size ≧ 200) Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

2.1 by medium term FUP (3 years) (previous year)	1	547	Mean Difference (IV, Random, 95% CI)	0.10 [‐10.26, 10.46]
2.2 by long term FUP (8 years) (previous year)	1	547	Mean Difference (IV, Random, 95% CI)	4.30 [‐4.63, 13.23]
3 Service use: 2. Not remaining in contact with psychiatric services Show forest plot	9	1633	Risk Ratio (M‐H, Random, 95% CI)	0.43 [0.30, 0.61]

3.1 by short term	1	95	Risk Ratio (M‐H, Random, 95% CI)	0.54 [0.28, 1.05]
3.2 by medium term	3	1063	Risk Ratio (M‐H, Random, 95% CI)	0.51 [0.36, 0.71]
3.3 by long term	5	475	Risk Ratio (M‐H, Random, 95% CI)	0.27 [0.11, 0.66]
4 Service use: 3a. Admitted to hospital Show forest plot	16		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

4.1 by short term	2	244	Risk Ratio (M‐H, Random, 95% CI)	0.61 [0.22, 1.69]
4.2 by medium term	5	1303	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.77, 0.93]
4.3 by long term	11	1516	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.74, 1.23]
4.4 by long term‐ during previous 12 months	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.67 [0.52, 0.86]
4.5 by short term FUP ‐ unplanned admission through Emergency Department	1	62	Risk Ratio (M‐H, Random, 95% CI)	1.0 [0.07, 15.28]
5 Service use: 3b. Average number of admissions per month (skewed data) Show forest plot			Other data	No numeric data

5.1 by medium term			Other data	No numeric data
5.2 by long term			Other data	No numeric data
6 Service use: 4a. Admitted to ER ‐ by long term Show forest plot	1	178	Risk Ratio (M‐H, Random, 95% CI)	1.13 [0.72, 1.76]

7 Service use: 4b. Average number of admissions to ER (skewed data) ‐ by medium term Show forest plot			Other data	No numeric data

8 Service use: 5a. Received day hospital care ‐ by short term FUP Show forest plot	1	62	Risk Ratio (M‐H, Random, 95% CI)	2.0 [0.19, 20.93]

9 Service use: 5b. Outpatient visits ‐ by short term FUP (6 months) Show forest plot	1	62	Mean Difference (IV, Random, 95% CI)	0.29 [‐0.14, 0.72]

10 Service use: 5c. Outpatient visits ‐ by medium term (skewed data) Show forest plot			Other data	No numeric data

11 Service use: 5d. Received home visits ‐ by short term FUP Show forest plot	1	62	Mean Difference (IV, Random, 95% CI)	4.32 [3.42, 5.22]

12 Adverse event: 1a. Death ‐ any cause Show forest plot	14		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

12.1 by short term	2	161	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.16, 6.91]
12.2 by medium term	6	901	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.23, 2.62]
12.3 by long term	9	1456	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.48, 1.47]
12.4 by medium term FUP (3 years)	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.59 [0.22, 1.61]
12.5 by long term FUP (8 years)	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.45, 1.88]
13 Adverse event: 1b. Death ‐ suicide Show forest plot	12		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

13.1 by short term	2	127	Risk Ratio (M‐H, Random, 95% CI)	0.35 [0.04, 3.27]
13.2 by medium term	4	819	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.17, 5.60]
13.3 by long term	9	1456	Risk Ratio (M‐H, Random, 95% CI)	0.68 [0.31, 1.51]
13.4 by medium term FUP (3 years)	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.74 [0.17, 3.28]
14 Global state: 1. Leaving the study early Show forest plot	21		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

14.1 by short term	5	598	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.44, 1.41]
14.2 by medium term	8	1699	Risk Ratio (M‐H, Random, 95% CI)	0.60 [0.51, 0.70]
14.3 by long term	13	1798	Risk Ratio (M‐H, Random, 95% CI)	0.68 [0.58, 0.79]
14.4 by medium term FUP (3 years)	1	547	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.84, 1.21]
14.5 by long term FUP (8 years)	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.70, 1.09]
15 Global state: 2. Average endpoint score (GAF, high = good) Show forest plot	5		Mean Difference (IV, Random, 95% CI)	Subtotals only

15.1 by short term	4	797	Mean Difference (IV, Random, 95% CI)	2.07 [0.28, 3.86]
15.2 by medium term	3	722	Mean Difference (IV, Random, 95% CI)	0.09 [‐3.11, 3.28]
15.3 by long term	5	818	Mean Difference (IV, Random, 95% CI)	3.41 [1.66, 5.16]
16 Global state: 3. Not compliant with medication ‐ by long term Show forest plot	1	71	Risk Ratio (M‐H, Random, 95% CI)	0.35 [0.15, 0.81]

17 Social functioning: 1a. Contact with legal system (various measurements) Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

17.1 by short term ‐ contact with the police	1	61	Risk Ratio (M‐H, Random, 95% CI)	2.57 [0.73, 9.04]
17.2 by medium term ‐ arrested	3	604	Risk Ratio (M‐H, Random, 95% CI)	1.08 [0.61, 1.90]
17.3 by medium term ‐ contact with the police	1	88	Risk Ratio (M‐H, Random, 95% CI)	0.23 [0.09, 0.55]
17.4 by medium term ‐ imprisoned	4	804	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.39, 1.64]
17.5 by long term ‐ arrested	1	178	Risk Ratio (M‐H, Random, 95% CI)	0.66 [0.32, 1.37]
17.6 by long term ‐ imprisoned	5	908	Risk Ratio (M‐H, Random, 95% CI)	0.86 [0.45, 1.65]
18 Social functioning: 1b. Mean contacts with legal system (skewed data) ‐ by medium term Show forest plot			Other data	No numeric data

18.1 Bookings			Other data	No numeric data
18.2 Jail days			Other data	No numeric data
18.3 Convictions			Other data	No numeric data
19 Social functioning: 2. Employment status (various measurements) Show forest plot	6		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

19.1 by medium term ‐ not competitively employed at the end of the trial	1	88	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.91, 1.10]
19.2 by medium term ‐ not employed at the end of the trial	4	1136	Risk Ratio (M‐H, Random, 95% CI)	0.89 [0.79, 1.00]
19.3 by long term ‐ not employed at the end of the trial	4	1129	Risk Ratio (M‐H, Random, 95% CI)	0.70 [0.49, 1.00]
19.4 by long term ‐ not working/studying in the previous year	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.86 [0.74, 0.99]
19.5 by medium term FUP (3 years) ‐ not working/studying in the previous year	1	547	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.90, 1.16]
19.6 by long term FUP (8 years) ‐ not working/studying in the previous year	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.88, 1.11]
20 Social functioning: 3a. Accommodation status (various measurements) Show forest plot	10		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

20.1 by short term ‐ homelessness	1	95	Risk Ratio (M‐H, Random, 95% CI)	0.04 [0.00, 0.70]
20.2 by medium term ‐ homelessness	1	88	Risk Ratio (M‐H, Random, 95% CI)	0.32 [0.03, 2.95]
20.3 by medium term ‐ not living independently	5	1303	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.66, 0.97]
20.4 by long term ‐ homelessness	3	418	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.34, 1.82]
20.5 by long term ‐ not living independently	4	1185	Risk Ratio (M‐H, Random, 95% CI)	0.65 [0.49, 0.88]
20.6 by long term ‐ not living in stable accommodation	1	168	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.65, 0.98]
21 Social functioning: 3b. Accomodation status: mean number of days in supported house (skewed data, sample size ≧ 200) Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

21.1 by long term (previous year)	1	547	Mean Difference (IV, Random, 95% CI)	0.30 [‐13.98, 14.58]
21.2 by medium term FUP (3 years) (previous year)	1	547	Mean Difference (IV, Random, 95% CI)	‐22.20 [‐38.47, ‐5.93]
21.3 by long term FUP (8 years) (previous year)	1	547	Mean Difference (IV, Random, 95% CI)	‐6.70 [‐19.35, 5.95]
22 Social functioning: 3c. Accommodation status (various measurements, skewed data) Show forest plot			Other data	No numeric data

22.1 by medium term ‐ average days per month in stable accommodation			Other data	No numeric data
22.2 by long term ‐ average days per month in sheltered homes			Other data	No numeric data
23 Social functioning: 4a. Substance abuse Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

23.1 alcohol abuse ‐ by long term	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.55 [0.26, 1.17]
23.2 illicit drug abuse ‐ by long term	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.63, 1.47]
23.3 substance abuse ‐ by medium term FUP (3 years)	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.67, 1.24]
24 Social functioning: 4b. Substance abuse (DALI, skewness not detectable) ‐ by medium term Show forest plot			Other data	No numeric data

24.1 alcohol abuse (DALI, ‐4 to + 6, high = worse)			Other data	No numeric data
24.2 drug abuse (DALI, ‐ 4 to + 4, high = worse)			Other data	No numeric data
25 Social functioning: 4c. Substance abuse ‐ days used per month (skewed data) Show forest plot			Other data	No numeric data

25.1 by medium term			Other data	No numeric data
25.2 by long term			Other data	No numeric data
26 Social functioning: 5a. Average endpoint score (various scales) Show forest plot	3		Mean Difference (IV, Random, 95% CI)	Subtotals only

26.1 by short term (RFS, low = poor)	1	80	Mean Difference (IV, Random, 95% CI)	‐0.62 [‐2.23, 0.99]
26.2 by short term (SAS‐adapted version, low = poor)	1	80	Mean Difference (IV, Random, 95% CI)	‐3.34 [‐7.55, 0.87]
26.3 by medium term ‐ social role performance (DAS, high = poor)	1	55	Mean Difference (IV, Random, 95% CI)	0.10 [‐0.40, 0.60]
26.4 by medium term (RFS, low = poor)	1	80	Mean Difference (IV, Random, 95% CI)	‐0.86 [‐2.72, 1.00]
26.5 by medium term (SAS‐adapted version, low = poor)	1	80	Mean Difference (IV, Random, 95% CI)	‐3.30 [‐7.83, 1.23]
26.6 by long term ‐ social role performance (DAS, high = poor)	1	58	Mean Difference (IV, Random, 95% CI)	‐0.20 [‐0.67, 0.27]
26.7 by long term (ISSI, low = poor)	1	62	Mean Difference (IV, Random, 95% CI)	3.20 [0.11, 6.29]
26.8 by long term (RFS, low = poor)	1	80	Mean Difference (IV, Random, 95% CI)	‐2.35 [‐4.05, ‐0.65]
26.9 by long term (SAS‐adapted version, low = poor)	1	80	Mean Difference (IV, Random, 95% CI)	‐2.75 [‐7.13, 1.63]
26.10 by long term (Strauss‐Carpenter Scale, low = poor)	1	60	Mean Difference (IV, Random, 95% CI)	0.10 [‐1.17, 1.37]
27 Social functioning: 5b. Average endpoint score (various scales, skewed data) Show forest plot			Other data	No numeric data

27.1 by short term (SAS, high = poor)			Other data	No numeric data
27.2 by medium term (SAS, high = poor)			Other data	No numeric data
27.3 by long term (SAS, high = poor)			Other data	No numeric data
27.4 by long term (REHAB, high = poor)			Other data	No numeric data
28 Mental state: 1a. General symptoms ‐ average endpoint score (various scales) Show forest plot	4		Mean Difference (IV, Random, 95% CI)	Subtotals only

28.1 by short term (BPRS‐18 items, high = poor)	2	668	Mean Difference (IV, Random, 95% CI)	‐1.56 [‐6.85, 3.73]
28.2 by short term (BSI, high = poor)	2	668	Mean Difference (IV, Random, 95% CI)	‐0.06 [‐0.19, 0.06]
28.3 by short term (CSI, low = poor)	1	125	Mean Difference (IV, Random, 95% CI)	‐0.56 [‐0.84, ‐0.28]
28.4 by medium term (BPRS‐18 items, high = poor)	2	662	Mean Difference (IV, Random, 95% CI)	‐0.96 [‐2.42, 0.51]
28.5 by medium term (BSI, high = poor)	2	662	Mean Difference (IV, Random, 95% CI)	‐0.02 [‐0.15, 0.10]
28.6 medium term (CSI, low = poor)	1	125	Mean Difference (IV, Random, 95% CI)	‐0.35 [‐0.65, ‐0.05]
28.7 by long term (BPRS‐18 items, high = poor)	3	777	Mean Difference (IV, Random, 95% CI)	‐1.48 [‐3.69, 0.74]
28.8 by long term (BSI, high = poor)	2	647	Mean Difference (IV, Random, 95% CI)	‐0.18 [‐0.31, ‐0.06]
29 Mental state: 1b. General symptoms ‐ mean change from baseline (CSI, low = poor ) ‐ by long term Show forest plot	1	168	Mean Difference (IV, Random, 95% CI)	‐0.32 [‐0.53, ‐0.11]

30 Mental state: 1c. General symptoms ‐ average endpoint score (various scales, skewed data) Show forest plot			Other data	No numeric data

30.1 by short term (BPRS‐24 items, high = poor)			Other data	No numeric data
30.2 by short term (PSE, high = poor)			Other data	No numeric data
30.4 by medium term (BPRS‐24 items, high = poor)			Other data	No numeric data
30.5 by medium term (CPRS, high = poor)			Other data	No numeric data
30.6 by medium term (PSE, high = poor)			Other data	No numeric data
30.8 by long term (BPRS‐18 items, high = poor)			Other data	No numeric data
30.9 by long term (BPRS‐24 items, high = poor)			Other data	No numeric data
30.10 by long term (CPRS, high = poor)			Other data	No numeric data
30.11 by long term (PSE, high = poor)			Other data	No numeric data
30.12 by long term (SCL‐90, high = poor)			Other data	No numeric data
31 Mental state: 2a. Specific symptoms ‐ depression at follow up interview Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

31.1 by medium term	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.56, 1.04]
31.2 by long term	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.57, 1.21]
31.3 by medium term FUP (3 years)	1	547	Risk Ratio (M‐H, Random, 95% CI)	1.25 [0.91, 1.72]
32 Mental state: 2b. Specific symptoms ‐ average endpoint score (various scales, skewed data, sample size ≧ 200) Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

32.1 by long term ‐ positive symptoms (SAPS, high = poor)	1	547	Mean Difference (IV, Random, 95% CI)	‐0.22 [‐0.45, 0.01]
32.2 by long term ‐ negative symptoms (SANS, high = poor)	1	547	Mean Difference (IV, Random, 95% CI)	‐0.42 [‐0.62, ‐0.22]
32.3 by medium term FUP (3 years) ‐ positive symptoms (SAPS, high = poor)	1	547	Mean Difference (IV, Random, 95% CI)	0.12 [‐0.15, 0.39]
32.4 by medium term FUP (3 years) ‐ negative symptoms (SANS, high = poor)	1	547	Mean Difference (IV, Random, 95% CI)	‐0.10 [‐0.33, 0.13]
32.5 by long term FUP (8 years) ‐ positive symptoms (SAPS, high = poor)	1	547	Mean Difference (IV, Random, 95% CI)	0.03 [‐0.21, 0.27]
32.6 by long term FUP (8 years) ‐ negative symptoms (SANS, high = poor)	1	547	Mean Difference (IV, Random, 95% CI)	0.06 [‐0.13, 0.25]
33 Mental state: 2c. Specific symptoms ‐ average endpoint score (various scales, skewed data) Show forest plot			Other data	No numeric data

33.3 by medium term ‐ depression symptoms (BDI, high = poor)			Other data	No numeric data
33.4 by medium term ‐ negative symptoms (SANS, high = poor)			Other data	No numeric data
33.7 by long term ‐ depression symptoms (BDI, high = poor)			Other data	No numeric data
33.11 by long term ‐ negative symptoms (SANS, high = poor)			Other data	No numeric data
34 Behaviour: 1. Specific behaviour ‐ self‐harm Show forest plot	3		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

34.1 by medium term	2	620	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.61, 1.59]
34.2 by long term	1	123	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.14, 6.55]
34.3 attempted suicide ‐ by long term (during last 12 months)	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.81 [0.47, 1.38]
34.4 attempted suicide ‐ by medium term FUP (3 years) (during last 3 years)	1	547	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.56, 1.62]
35 Behaviour: 2. Social behaviour ‐ average endpoint score (SBS, high = poor) Show forest plot			Other data	No numeric data

35.1 by medium term			Other data	No numeric data
35.2 by long term			Other data	No numeric data
36 Quality of life: 1a. Average endpoint score (various scales) Show forest plot	7		Mean Difference (IV, Random, 95% CI)	Subtotals only

36.1 by short term ‐ general well‐being (QOLI, high = better)	1	125	Mean Difference (IV, Random, 95% CI)	0.53 [0.09, 0.97]
36.2 by medium term (LQoLP, high = better)	1	52	Mean Difference (IV, Random, 95% CI)	0.09 [‐0.60, 0.78]
36.3 by medium term (MANSA ‐ range 1‐7, high = better)	1	81	Mean Difference (IV, Random, 95% CI)	0.20 [‐0.29, 0.69]
36.4 by long term (LQoLP, high = better)	3	274	Mean Difference (IV, Random, 95% CI)	‐0.13 [‐0.38, 0.12]
36.5 by long term (QOLI, high = better)	2	132	Mean Difference (IV, Random, 95% CI)	0.09 [‐0.24, 0.42]
37 Quality of life: 1b. Mean change from baseline (QOLI, high = better, skewed data) ‐ by long term Show forest plot			Other data	No numeric data

38 Participant satisfaction: 1a. Average endpoint score (CSQ, high = better) Show forest plot	3		Mean Difference (IV, Random, 95% CI)	Subtotals only

38.1 by short term	1	61	Mean Difference (IV, Random, 95% CI)	6.20 [2.60, 9.80]
38.2 by medium term	2	500	Mean Difference (IV, Random, 95% CI)	1.93 [0.86, 3.01]
38.3 by long term	2	423	Mean Difference (IV, Random, 95% CI)	3.23 [2.31, 4.14]
39 Participants satisfaction: 1b. Average endpoint score (CSQ, high = better, skewed data) ‐ by short term Show forest plot			Other data	No numeric data

40 Participants need: 1. Average endpoint score (various scales, skewed data) Show forest plot			Other data	No numeric data

40.1 by medium term ‐ met needs (CANSAS, high = better)			Other data	No numeric data
40.2 by medium term ‐ unmet needs (CANSAS, high = poor)			Other data	No numeric data
40.4 by long term (CAN, high = poor)			Other data	No numeric data
41 Costs: 1a. Direct costs of psychiatric hospital care ‐ by medium term (Unit cost = USD, fiscal year 1990) Show forest plot	2	426	Mean Difference (IV, Random, 95% CI)	‐143.74 [‐272.40, ‐15.08]

42 Costs: 1b. Direct costs of psychiatric hospital care ‐ skewed data Show forest plot			Other data	No numeric data

42.1 by medium term			Other data	No numeric data
42.2 by long term			Other data	No numeric data
43 Costs: 2a. Direct healthcare costs ‐ by long term (Unit cost = USD, fiscal year 1988) Show forest plot	2	873	Mean Difference (IV, Random, 95% CI)	‐529.24 [‐2143.59, 1085.10]

44 Costs: 2b. Direct healthcare costs ‐ skewed data Show forest plot			Other data	No numeric data

44.1 by medium term			Other data	No numeric data
44.2 by short term FUP			Other data	No numeric data
45 Costs: 3. Direct costs ‐ other data ‐ skewed data Show forest plot			Other data	No numeric data

45.1 all care ‐ by short term			Other data	No numeric data
45.2 all care ‐ by medium term			Other data	No numeric data
45.3 all care ‐ by long term			Other data	No numeric data
45.4 specific ‐ outpatient care ‐ by medium term			Other data	No numeric data
45.5 specific ‐ prison ‐ by medium term			Other data	No numeric data

Comparison 1. Intensive Case Management versus standard care

Comparison 2. Intensive Case Management versus non‐Intensive Case Management

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Service use: 1. Average number of days in hospital per month ‐ by about 24 months Show forest plot	21	2220	Mean Difference (IV, Random, 95% CI)	‐0.08 [‐0.37, 0.21]

1.1 skewed data (sample size ≧ 200)	3	694	Mean Difference (IV, Random, 95% CI)	‐0.58 [‐1.93, 0.76]
1.2 skewed data (sample size < 200)	18	1526	Mean Difference (IV, Random, 95% CI)	‐0.03 [‐0.33, 0.28]
2 Service use: 1a. Average number of days in hospital per month ‐ by medium/long term follow up (skewed data, sample size ≧ 200) Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

2.1 by medium term FUP (18 months)	1	237	Mean Difference (IV, Random, 95% CI)	0.60 [‐1.25, 2.45]
2.2 by long term FUP (8.5 years)	1	203	Mean Difference (IV, Random, 95% CI)	0.80 [‐1.47, 3.07]
3 Service use: 2. Not remaining in contact with psychiatric services Show forest plot	4		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

3.1 by medium term	1	73	Risk Ratio (M‐H, Random, 95% CI)	0.27 [0.08, 0.87]
3.2 by long term	3	1182	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.34, 1.98]
3.3 by medium term FUP (18 months)	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.42 [0.17, 1.05]
4 Service use: 3a. Admitted to hospital ‐ by long term Show forest plot	3	1132	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.75, 1.12]

5 Service use: 3b. Average number of admissions (skewed data ‐sample size ≧ 200) Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

5.1 ‐ by long term (24 months)	1	678	Mean Difference (IV, Random, 95% CI)	‐0.18 [‐0.41, 0.05]
5.2 by medium term FUP (18 months)	1	237	Mean Difference (IV, Random, 95% CI)	‐0.10 [‐0.60, 0.40]
5.3 by long term FUP (8.5 years)	1	203	Mean Difference (IV, Random, 95% CI)	1.0 [‐0.25, 2.25]
6 Service use: 3c. Average number of admissions (skewed data) ‐ by medium term Show forest plot			Other data	No numeric data

7 Adverse event: 1a. Death ‐ any cause Show forest plot	7		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

7.1 by short term	1	193	Risk Ratio (M‐H, Random, 95% CI)	0.0 [0.0, 0.0]
7.2 by medium term	3	294	Risk Ratio (M‐H, Random, 95% CI)	2.92 [0.12, 69.43]
7.3 by long term	5	1637	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.46, 1.75]
7.4 by medium term FUP (18 months)	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.32, 2.95]
7.5 by long term FUP (8.5 years)	1	251	Risk Ratio (M‐H, Random, 95% CI)	1.15 [0.63, 2.09]
8 Adverse event: 1b. Death ‐ suicide Show forest plot	8		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

8.1 by short term	1	193	Risk Ratio (M‐H, Random, 95% CI)	0.0 [0.0, 0.0]
8.2 by medium term	6	929	Risk Ratio (M‐H, Random, 95% CI)	1.61 [0.26, 9.85]
8.3 by long term	3	1152	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.27, 2.84]
8.4 by medium term FUP (18 months)	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.33 [0.03, 3.09]
9 Global state: 1. Leaving the study early Show forest plot	9	2195	Risk Ratio (M‐H, Random, 95% CI)	0.72 [0.52, 0.99]

9.1 by medium term	2	225	Risk Ratio (M‐H, Random, 95% CI)	0.64 [0.13, 3.07]
9.2 by long term	7	1970	Risk Ratio (M‐H, Random, 95% CI)	0.70 [0.52, 0.95]
10 Global state: 2a. Average endpoint score (HoNOS, high = poor) ‐ by long term Show forest plot	1	239	Mean Difference (IV, Random, 95% CI)	‐0.40 [‐1.77, 0.97]

11 Global state: 2b. Average endpoint score (HoNOS, high = poor) ‐ skewed data Show forest plot			Other data	No numeric data

11.1 medium term			Other data	No numeric data
11.2 long term			Other data	No numeric data
12 Global state: 3a. Not compliant with medication ‐ by medium term Show forest plot	1	73	Risk Ratio (M‐H, Random, 95% CI)	1.14 [0.42, 3.05]

13 Global state: 3b. Compliance with medication ‐ average endpoint sub‐scale score (ROMI) ‐ by long term Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

13.1 compliance sub‐scale (high = good)	1	239	Mean Difference (IV, Random, 95% CI)	0.60 [‐0.05, 1.25]
13.2 non‐compliance sub‐scale (high = poor)	1	239	Mean Difference (IV, Random, 95% CI)	‐0.60 [‐1.63, 0.43]
14 Global state: 3c. Compliance with medication ‐ average endpoint sub‐scale score (ROMI, score 1‐3, skewed data) Show forest plot			Other data	No numeric data

14.1 medium term ‐ compliance sub‐scale (high = good)			Other data	No numeric data
14.2 medium term ‐ non‐compliance sub‐scale (high = poor)			Other data	No numeric data
14.3 long term ‐ compliance sub‐scale (high = good)			Other data	No numeric data
14.4 long term ‐ non‐compliance sub‐scale (high = poor)			Other data	No numeric data
15 Social functioning: 1. Contact with legal system (various measurements) Show forest plot	3		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

15.1 by medium term ‐ contact with the police	1	73	Risk Ratio (M‐H, Random, 95% CI)	0.32 [0.04, 2.97]
15.2 by long term ‐ imprisoned	2	959	Risk Ratio (M‐H, Random, 95% CI)	1.15 [0.64, 2.08]
15.3 by long term ‐ arrested	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.53, 1.42]
15.4 by medium term FUP (18 months) ‐ imprisoned	1	251	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.47, 2.44]
15.5 by long term FUP (8.5 years) ‐ imprisoned	1	214	Risk Ratio (M‐H, Random, 95% CI)	0.7 [0.43, 1.14]
16 Social functioning 2. Employment status (various measurements) Show forest plot	2		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

16.1 spent >1 day employed ‐ by medium term	1	73	Risk Ratio (M‐H, Random, 95% CI)	1.46 [0.45, 4.74]
16.2 on paid employment ‐ by medium term	1	73	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.14, 6.54]
16.3 unemployed ‐ by long term FUP (8.5 years)	1	214	Risk Ratio (M‐H, Random, 95% CI)	1.10 [0.91, 1.34]
17 Social functioning: 3a. Accommodation status (various measurements) Show forest plot	2		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

17.1 by medium term ‐ living in supported accommodation	1	73	Risk Ratio (M‐H, Random, 95% CI)	2.59 [0.75, 9.01]
17.2 by long term ‐ homelessness	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.69 [0.34, 1.38]
17.3 by medium term FUP (18 months) ‐ living independently	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.84, 1.13]
17.4 by medium term FUP (18 months) ‐ living in supported accomodation	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.38, 1.77]
17.5 by medium term FUP (18 months) ‐ homelessness	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.47, 1.49]
17.6 by long term FUP (8.5 years) ‐ living in supported accomodation	1	214	Risk Ratio (M‐H, Random, 95% CI)	1.05 [0.75, 1.48]
17.7 by long term FUP (8.5 years) ‐ homelessness	1	214	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.55, 1.53]
18 Social functioning: 3b. Accommodation status ‐ average days per month in stable accommodation Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

18.1 by short term	1	203	Mean Difference (IV, Random, 95% CI)	‐0.20 [‐2.48, 2.08]
18.2 by medium term	1	203	Mean Difference (IV, Random, 95% CI)	0.10 [‐2.15, 2.35]
18.3 by long term	2	901	Mean Difference (IV, Random, 95% CI)	‐0.19 [‐1.37, 1.00]
19 Social functioning: 4a. Substance abuse ‐ by long term Show forest plot	2		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

19.1 alcohol abuse	1	251	Risk Ratio (M‐H, Random, 95% CI)	1.10 [0.67, 1.83]
19.2 illicit drug abuse	1	251	Risk Ratio (M‐H, Random, 95% CI)	1.08 [0.69, 1.71]
19.3 alcohol ‐ remission from alcohol use disorder (AUS score < 3)	1	223	Risk Ratio (M‐H, Random, 95% CI)	0.86 [0.65, 1.14]
20 Social functioning: 4b. Substance abuse ‐ average endpoint score (SATS, low = poor) Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

20.1 by short term	1	203	Mean Difference (IV, Random, 95% CI)	0.07 [‐0.28, 0.42]
20.2 by medium term	1	203	Mean Difference (IV, Random, 95% CI)	‐0.11 [‐0.55, 0.33]
20.3 by long term	1	203	Mean Difference (IV, Random, 95% CI)	0.11 [‐0.41, 0.63]
21 Social functioning: 4c. Alcohol ‐ abuse (various measurements, skewed data) Show forest plot			Other data	No numeric data

21.1 short term ‐ days using alcohol during previous 6 months (TLFB)			Other data	No numeric data
21.2 short term ‐ average endpoint score (AUS, high = poor)			Other data	No numeric data
21.3 medium term ‐ days using alcohol during previous 6 months (TLFB)			Other data	No numeric data
21.4 medium term ‐ average endpoint score (AUS, high = poor)			Other data	No numeric data
21.5 long term ‐ days using alcohol during previous 6 months (TLFB)			Other data	No numeric data
21.6 long term ‐ average endpoint score (AUS, high = poor)			Other data	No numeric data
22 Social functioning: 5a. Average endpoint score (LSP, high = poor) ‐ by long term Show forest plot	1	239	Mean Difference (IV, Random, 95% CI)	4.0 [‐0.61, 8.61]

23 Social functioning: 5b. Average endpoint score (SFQ, high = poor) ‐ skewed data Show forest plot			Other data	No numeric data

23.1 by medium term			Other data	No numeric data
23.2 by long term			Other data	No numeric data
24 Mental state: 1a. General symptoms ‐ average endpoint score (various scales) Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

24.1 by short term (BPRS‐24 items, high = poor)	1	203	Mean Difference (IV, Random, 95% CI)	‐0.65 [‐3.99, 2.69]
24.2 by medium term (BPRS‐24 items, high = poor)	1	203	Mean Difference (IV, Random, 95% CI)	‐1.62 [‐4.76, 1.52]
24.3 by long term (BPRS‐24 items, high = poor)	1	203	Mean Difference (IV, Random, 95% CI)	‐0.22 [‐3.32, 2.88]
24.4 by long term (CPRS, high = poor)	1	595	Mean Difference (IV, Random, 95% CI)	0.40 [‐1.83, 2.63]
25 Mental state: 1b. General symptoms ‐ average endpoint scores (various scales, skewed data) Show forest plot			Other data	No numeric data

25.1 by medium term (Krawiecka Scale, high = poor)			Other data	No numeric data
25.2 by long term (Krawiecka Scale, high = poor)			Other data	No numeric data
25.3 by long term (BPRS 24‐items, high = good)			Other data	No numeric data
26 Mental state: 2a. Specific symptoms: negative symptoms ‐ average endpoint score (SANS, high = poor) ‐ by long term Show forest plot	1	593	Mean Difference (IV, Random, 95% CI)	0.20 [‐2.32, 2.72]

27 Mental state: 2b. Specific symptoms ‐ average endpoint scores (various scales, skewed data) Show forest plot			Other data	No numeric data

27.1 medium term ‐ anxiety (HADS, high = poor)			Other data	No numeric data
27.2 medium term ‐ depression (HADS, high = poor)			Other data	No numeric data
27.3 long term ‐ anxiety (HADS, high = poor)			Other data	No numeric data
27.5 long term ‐ depression (HADS, high = poor)			Other data	No numeric data
28 Behaviour: 1. Specific behaviour (various measurements) Show forest plot	3		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

28.1 by medium term ‐ harm to self or others	1	73	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.40, 1.90]
28.2 by long term ‐ self‐harm	2	959	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.69, 1.46]
28.3 by long term ‐ injury/assault to others	2	959	Risk Ratio (M‐H, Random, 95% CI)	1.09 [0.85, 1.40]
28.4 by medium term FUP (18 months) ‐ self harm	1	251	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.44, 1.67]
28.5 by medium term FUP (18 months) ‐ injury/assualt to others	1	251	Risk Ratio (M‐H, Random, 95% CI)	1.35 [0.87, 2.10]
28.6 by long term FUP (8.5 years) ‐ self harm	1	214	Risk Ratio (M‐H, Random, 95% CI)	0.81 [0.51, 1.27]
28.7 by long term FUP (8.5 years) ‐ injury/assault to others	1	214	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.83, 1.09]
29 Quality of life: 1. Average endpoint score (various scales) Show forest plot	3		Mean Difference (IV, Random, 95% CI)	Subtotals only

29.1 by short term ‐ overall life satisfaction (QOLI, high = better)	1	203	Mean Difference (IV, Random, 95% CI)	‐0.02 [‐0.43, 0.39]
29.2 by medium term ‐ overall life satisfaction (QOLI, high = better)	1	203	Mean Difference (IV, Random, 95% CI)	‐0.04 [‐0.43, 0.35]
29.3 by long term (LQoLP, high = better)	1	526	Mean Difference (IV, Random, 95% CI)	0.03 [‐0.10, 0.16]
29.4 by long term (MANSA, range 1‐7, high = better)	1	166	Mean Difference (IV, Random, 95% CI)	0.10 [‐0.19, 0.39]
29.5 by long term ‐ overall life satisfaction (QOLI, high = better)	1	203	Mean Difference (IV, Random, 95% CI)	0.10 [‐0.25, 0.45]
30 Participant satisfaction/need: 1. Average endpoint scores (various scale) ‐ by long term Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

30.1 Patient need: CAN (high = poor)	1	585	Mean Difference (IV, Random, 95% CI)	‐0.29 [‐0.69, 0.11]
30.2 Patient Satisfaction with Health Services (high = poor)	1	490	Mean Difference (IV, Random, 95% CI)	‐0.40 [‐1.25, 0.45]
31 Participants need: 1. Average endpoint scores (various scales, skewed data) Show forest plot			Other data	No numeric data

31.1 by medium term (CAN, high = poor)			Other data	No numeric data
31.2 by long term (CAN, high = poor)			Other data	No numeric data
31.3 by long term (CANSAS, high = poor)			Other data	No numeric data
32 Participant satisfaction: 1. Average endpoint scores (CSQ‐modified, high = better, skewed data) ‐ by long term Show forest plot			Other data	No numeric data

33 Costs: 1. Direct costs of psychiatric hospital care (skewed data) Show forest plot			Other data	No numeric data

33.3 by medium term			Other data	No numeric data
33.4 by long term			Other data	No numeric data
34 Costs: 2a. Direct costs of all care ‐ by long term (2 years). Unit cost GBP, fiscal year 1997/98 Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

35 Costs: 2b. Direct costs of all care ‐ by medium term (skewed data) Show forest plot			Other data	No numeric data

36 Costs: 3. Total costs of care per patient ‐ Unit cost GBP Show forest plot	2		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

36.1 by 24 months, fiscal year 1997/98	1	667	Mean Difference (IV, Fixed, 95% CI)	1849.0 [‐1598.23, 5296.23]
36.2 by 18 months, fiscal year 2003/2004 ( GBP 1 = USD 1.58)	1	243	Mean Difference (IV, Fixed, 95% CI)	4031.00 [‐2724.13, 10786.13]

Comparison 2. Intensive Case Management versus non‐Intensive Case Management