Non‐medical prescribing versus medical prescribing for acute and chronic disease management in primary and secondary care

Summary of findings for the main comparison. Non‐medical prescribing compared to medical prescribing for acute and chronic disease management in primary and secondary care

Non‐medical prescribing compared to medical prescribing for acute and chronic disease management in primary and secondary care
Patient or population: patients with acute and chronic disease Settings: secondary care and ambulatory/primary care in low‐, middle‐ and high‐income counties Intervention: non‐medical prescribing Comparison: medical prescribing
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Certainty of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Medical prescribing	Non‐medical prescribing
Systolic blood pressure (mmHg) at 12 months	The mean systolic blood pressure in the control group ranged from 124 mmHg to 149 mmHg	The mean systolic blood pressure in the intervention group was 5.31 mmHg lower (‐6.46 lower to ‐4.16 lower)	‐	4229 (12 RCTs)	⊕⊕⊕⊝ Moderate ^1,2,3	Random‐effects analysis: MD ‐5.91 mmHg lower (95% CI ‐7.71 lower to ‐4.10 lower)
Glycated haemoglobin (HbA1c, %) at 12 months	The mean change in glycated haemoglobin in the control group ranged from ‐0.90% to 9.7%	The mean change in glycated haemoglobin in the intervention group was 0.62% lower (‐0.85 lower to ‐0.38 lower)	‐	775 (6 RCTs)	⊕⊕⊕⊕ High^2,3	Random‐effects analysis: MD ‐0.62 (95% CI ‐0.85 to ‐0.38)
Low‐density lipoprotein (mmol/L) at 12 months	The mean low‐density lipoprotein in the control group ranged from ‐0.26 to 3.41 mmol/L	The mean low‐density lipoprotein in the intervention group was 0.21 mmol/L lower (‐0.29 lower to ‐0.14 lower)	‐	1469 (7 RCTs)	⊕⊕⊕⊝ Moderate^1,2,3	Random‐effects analysis: MD ‐0.30 (95% CI ‐0.62 to 0.02)
Adherence (continuous) 6 months follow‐up	The mean adherence (continuous) in the control group was 0.79	The mean adherence in the intervention group was 0.15 higher (0.00 higher to 0.30 higher)	‐	700 (4 RCTs)	⊕⊕⊕⊝ Moderate^4,5
Patient satisfaction	Patient satisfaction was reported in 14 studies (Table 4). The majority of surveys were either not referenced or developed locally. Validated questionnaires assessing overall non‐medical practitioner satisfaction with care were reported in six studies rather than patient satisfaction with prescribing. An exception was the study by Bruhn 2013, which found for the prescribing intervention, patients were generally positive about the pharmacist prescribing service, 85% (39/46) were totally satisfied, while 9% (4/44) would have preferred to see their GP		Not estimable	7514 (14 RCTs)	⊕⊕⊕⊝ Moderate^8,9
Adverse events	There was little or no difference in adverse events between treatment groups in nine studies. Two studies reported higher rates of adverse events in the usual care group. It was difficult to determine effects in the remaining studies because limited data were reported		Not estimable	18,400 (18 RCTs)	⊕⊕⊝⊝ Low^6,7
Health‐related quality of life measured with SF‐12/36	The mean health‐related quality of life in the control group was 0	The mean health‐related quality of life in the intervention group: physical component was 1.17 higher (0.16 to 2.17) mental component was 0.58 higher (‐0.40 to 1.55)	‐	4631 (8 RCTs)	⊕⊕⊕⊝ Moderate¹⁰
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% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; GP: general practitioner; MD: mean difference; RCT: randomised controlled trial.
GRADE Working Group grades of evidence High‐certainty: Further research is very unlikely to change our confidence in the estimate of effect. Moderate‐certainty: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low‐certainty: 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‐certainty: We are very uncertain about the estimate.
¹Downgraded one level due to serious inconsistency (considerable heterogeneity was found). ²Multifaceted interventions. ³Variable prescribing autonomy. ⁴Downgraded one level due to serious risk of bias (high risk of performance bias). ⁵Variable reporting measures of adherence. ⁶Downgraded one level due to indirectness (range of adverse events; may not be related to the intervention). ⁷Downgraded one level due to selective outcome reporting (adverse events not reported in many studies). ⁸Downgraded one level due to indirectness (prescribing component not adequately assessed across studies). ⁹Variability in satisfaction measures. ¹⁰Downgraded one level due to indirectness (prescribing component effect on quality of life difficult to determine).

Background

Description of the healthcare challenge

A range of health workforce strategies are needed to address issues of health service access and efficiency. In low‐, middle‐ and high‐income countries, the increasing demand for health services arises from an ageing population and the resultant increasing burden of chronic disease (Bhanbhro 2011; Duckett 2005; Phillips 2008; WHO 2012).

Increased health demands can be met in part by task substitution within the health workforce. One health workforce strategy for task substitution is to permit prescribing by healthcare providers other than medical doctors. Non‐medical prescribers may include nurses, pharmacists, allied health professionals, and physician assistants. In some low‐ and middle‐income countries, lay health workers have been used to distribute medications with preventive or curative intent, including contraceptives, iron or vitamin supplements, vaccinations, and agents for tuberculosis management (Glenton 2013).

Extending a health provider's scope of practice, including the right to prescribe, has been supported in a number of countries as a means of benefiting patient care by the effective use of health professionals' skills, improving patient access to timely care, improving patient choice, and enhancing teamwork and the better use of resources (Department of Health 1999; Ellis 2006; Hooker 2006; Stewart 2010).

The devolution of prescribing rights in high‐income countries has continued from a historical base in the United States of America (USA) in the 1970s through to more recent government‐led reforms in the United Kingdom (UK), Canada, the Netherlands, New Zealand, and Australia. While the definition of prescribing may vary between countries, for the purpose of our review, prescribing was defined as: "an iterative process involving the steps of information gathering, clinical decision making, communication and evaluation which results in the initiation, continuation or cessation of a medicine" (Health Workforce Australia 2013). The term 'medical prescribing' refers to prescribing by medically qualified doctors. The supply of non‐prescription (over‐the‐counter) medicines by pharmacists or pharmacy assistants working in community pharmacies is excluded from our definition of prescribing, as is the supply of medicines by lay health workers.

The term 'non‐medical prescribing' originates from the UK, where it is defined as: "prescribing by specially trained nurses, optometrists, pharmacists, physiotherapists, podiatrists, and radiographers, working within their clinical competence as either independent or supplementary prescribers" (NPC 2012).

Supplementary prescribing which was introduced in the UK in 2003, is defined as 'a voluntary partnership between an independent prescriber (a doctor or dentist) and a supplementary prescriber (e.g. nurse, optometrist, pharmacist, physiotherapist, chiropodist/podiatrist, or radiographer) to implement an agreed patient‐specific clinical management plan with the patient's agreement' (Department of Health 2003). Non‐medical prescribing rights were extended in 2006 with the introduction of independent prescribing. The UK Department of Health defines independent prescribing as 'prescribing by a practitioner (e.g. doctor, dentist, nurse, pharmacist, optometrist) responsible and accountable for the assessment of patients with undiagnosed or diagnosed conditions, and for decisions about the clinical management required, including prescribing'. Independent prescribing is one element of the clinical management of a patient and occurs in partnership with the patient. It requires an initial patient assessment, interpretation of that assessment, a decision on safe and appropriate therapy, and a process for ongoing monitoring. The independent prescriber is responsible and accountable for at least this element of a patient's care (Department of Health 2006). Independent prescribing does not require a clinical management plan. From 1 May 2006, nurse and pharmacist independent prescribers who completed the appropriate training could prescribe, with a few exceptions, any licensed medicine for any medical condition within their competence. In 2009, independent prescribing rights were extended to include unlicensed medicines. While prescribing of controlled drugs was restricted, this limitation was removed through legislative change in April 2012 (Home Office 2012).

In the USA, devolution of prescribing authority varies from state to state. Collaborative Practice Agreements in 46 States allow a pharmacist to partner with a physician to manage a number of patient services, including medication management (Law 2013; Thomas 2006). Physician assistants and nurse practitioners were introduced in 1967 to support medical care. These practitioners undertake a range of clinical functions, including prescribing (Hooker 2006).

Within Canada, a pharmacist's scope of prescribing practice varies between the provinces from independently prescribing to adapting (modifying) or continuing prescriptions (Law 2012).

A collaborative prescribing model has emerged as the preferred model of practice within New Zealand and Australia. Collaborative prescribing is undertaken within a multidisciplinary team and can include the continuum of prescribing from transcription of orders (with or without medical signature), prescribing specified drugs and doses by protocol, prescribing by clinical management plan (allowing choice of drugs and doses) to independent prescribing, where a prescribing consultation with a medical practitioner is not required (Weeks 2008; Wheeler 2012).

The Health Professionals Prescribing Pathway developed by Health Workforce Australia (HWA) includes five steps to safely and competently prescribe, and covers: education and training, recognition by the profession's national registering board, authorisation to prescribe by legislation, prescribing within the scope of practice, and maintaining and enhancing competence to prescribe. The prescribing models suggested by HWA emphasise team communication and are divided into autonomous prescribing, prescribing under supervision, and prescribing via a structured prescribing arrangement (HWA 2013). The reforms started by HWA have been transferred to a working group of the Australian Health Practitioner Regulation Agency. As part of the reform process, health agencies in Australia, Canada, New Zealand, and the UK have developed prescribing competency frameworks for non‐medical health professionals (NPC 2012; NPS 2012; Pharmacy Council NZ 2013; Yuksel 2008).

Description of the intervention

For the purpose of our review the term 'non‐medical prescribing' was used to cover prescribing of medicines by a broad range of healthcare providers other than medical doctors, prescribing in primary or secondary care. No limitation was set on the type of non‐medical healthcare provider undertaking prescribing. Frequently, non‐medical prescribing is done in collaboration or partnership with doctors, and within this practice there are different models of prescribing practice. However, for this review the non‐medical prescriber was required to have a high degree of autonomy in their prescribing practice. We excluded studies reporting prescribing practices requiring medical review, consultation, and approval requiring a medical signature on medication orders. Our review focused on prescribing, which as per our definition is much broader than issuing a prescription.

The role of non‐medical prescribers in secondary care settings may involve supporting acute or chronic care by prescribing in a timely way medication on admission, discharge, or where there is a specialist need, e.g. total parenteral nutrition. Specialist outpatient clinics managed by non‐medical health professionals may exist in either the primary or secondary care setting, e.g. for the management of hypertension, lipids, diabetes, and pain. In primary care settings, prescribing may be undertaken for acute or chronic conditions by nurses or other healthcare providers caring for patients in their homes or through involvement with general practice teams, community health centres, mental health teams, or community pharmacies.

How the intervention might work

Non‐medical prescribing has developed as an accepted healthcare practice in a number of countries to improve access to healthcare, to better use the skills of doctors who can focus on more acute patient needs, to better use the skills of pharmacists, nurses and other health providers, to potentially reduce costs for achieving at least equivalent, if not better health outcomes for consumers, and to retain health workers by increasing job satisfaction (Department of Health 1999; Tonna 2007). While qualitative studies support non‐medical prescribing from a patient and practitioner perspective, robust evidence is still required for clinical, patient‐reported, and resource use outcomes. It is noted that where non‐medical prescribers are practising in collaborative teams, it may be difficult to apportion the impact of the non‐medical prescribing component to the primary and secondary outcomes of this review. Wider adoption of non‐medical prescribing practice in high‐income countries frequently faces local regulatory hurdles and opposition from the medical community which has raised concerns about professional autonomy, patient safety, the diagnostic competency of non‐medical prescribers, and costs (Cooper 2008). Evidence that patient outcomes arising from non‐medical prescribing are as effective as those from medical prescribing would provide a basis for policy‐makers to support wider implementation of this practice.

Why it is important to do this review

It is important for health practitioners and policy‐makers to understand the evidence existing for non‐medical prescribing in order to address access or health workforce needs. This information will also guide future decision making with regards to implementing or expanding non‐medical prescribing.

Potential beneficiaries of the findings include:

policy‐makers seeking to use workforce resources more efficiently;
policy‐makers seeking to meet a clinical need;
consumers seeking greater choice and easier access to medicines;
non‐medical health professionals seeking to better utilise their skills and/or extend their scope of practice; and
medical staff seeking to focus on patients with the greatest medical need.

Despite a gradual rolling out of reforms, the evidence for the potential benefits of non‐medical prescribing from well‐controlled trials involving a wide range of health professionals requires identification, synthesis, and evaluation. Several narrative reviews of the non‐medical prescribing literature have been undertaken (Kay 2004; Tonna 2007), and the British government commissioned two evaluations covering supplementary and independent prescribing (Bissell 2008; Latter 2010).

A Cochrane Review on substitution of doctors by suitably trained nurses in primary care found that trained nurses can produce as high a quality of care and as good health outcomes with no appreciable difference between doctors and nurses in resource utilisation outcomes associated with prescribing (Laurant 2005). The review was limited to nurses in the primary care setting as first contact or ongoing care for undifferentiated patients.

A further Cochrane Review found a single RCT of pharmacist‐managed drug therapy (Nkansah 2010), including the prescribing of drugs versus physician medication management (Hawkins 1979). However, we assessed the study to be of low‐quality, leaving open the question of whether the delivery of patient‐targeted services by pharmacists improves patient outcomes compared to other health professionals.

The Driscoll 2015 Cochrane Review of nurse‐led titration of drug therapy for people with heart failure, found that participants in the nurse‐led group were less likely to be admitted to hospital or to die. More participants reached the maximum drug dose in the nurse‐led titration group compared to titration of doses by primary care physicians. The certainty of evidence that nurse‐led titration reduced hospitalisations was graded as high and the certainty of evidence regarding the proportion of participants reaching optimal dose was graded as low. However, in the majority of studies the influence of medical supervision on nurse‐dose titration (prescribing) was unclear.

Against this background, we systematically identified, reviewed, and updated the evidence from controlled studies and uncontrolled studies on the clinical, patient‐reported and resource use outcomes of non‐medical prescribing in primary and secondary care settings. This review considered any adverse effects of non‐medical prescribing which may be clinical (e.g. deterioration in care or incidence of adverse drugs reactions), patient‐reported (e.g. decreased satisfaction), or resource‐related (e.g. increased treatment costs).

The review covered healthcare providers undertaking non‐medical prescribing, spanning primary and secondary care settings, and considered acute and chronic prescribing situations.

Objectives

To assess the clinical, patient‐reported, and resource use outcomes of non‐medical prescribing for managing acute and chronic health conditions in primary and secondary care settings compared with medical prescribing (usual care).

Methods

Criteria for considering studies for this review

Types of studies

We included studies of patients or health professionals or healthcare settings using the definitions of designs outlined in the Cochrane Effective Practice and Organisation of Care (EPOC) Group checklist (Cochrane EPOC Group 2013a). We included randomised controlled trials (RCTs) and cluster‐RCTs, one controlled trial where investigators had allocated participants to the different groups that were being compared using a method that is not random, but where at least two groups with interventions were followed, and one controlled before‐and‐after (CBA) study with at least two intervention sites and two control sites. We did not find either interrupted time series (ITS) studies nor qualitative studies linked to quantitative studies using qualitative analysis methods.

Types of participants

Healthcare providers who are not medical doctors, undertaking prescribing including, nurses, optometrists, pharmacists, physician assistants, and other allied health professionals or categories not specifically mentioned whose roles meet our definition of non‐medical prescribing.

Setting

We included studies based in any primary or secondary care setting where non‐medical prescribing occurred.

Types of interventions

We included studies involving health providers other than medical doctors undertaking prescribing according to our definition of prescribing. We excluded studies limited to the supply function of pharmacists, including over‐the‐counter products and studies involving the supply function of lay health workers.

We included the following six comparisons for non‐medical prescribing.

Non‐medical prescribing versus medical prescribing in acute care.
Non‐medical prescribing versus medical prescribing in chronic care.
Non‐medical prescribing versus medical prescribing in secondary care.
Non‐medical prescribing versus medical prescribing in primary care.
Comparisons between different non‐medical prescriber groups.
Non‐medical healthcare providers with formal prescribing training versus those without formal prescribing training.

Types of outcome measures

The studies included in the review reported a wide variety of outcome measures. We only included studies with objective measures of patient clinical outcomes. Non‐inferiority was regarded as a positive outcome where a non‐medical prescribing outcome was at least as good as the comparator. We excluded studies with only a qualitative component in order to maintain the clinical focus of the review.

Primary outcomes

Clinical outcomes

Patient outcomes

We used standard outcome measures covering health and well‐being, including physiological measures of treatment such as systolic blood pressure, glycated haemoglobin, and low‐density lipoprotein. Outcomes were divided into dichotomous and continuous outcomes.

We also considered the following outcomes.

Proportion of prescribers, medical and non‐medical, appropriately adhering to practice guidelines.
Proportion of patients demonstrating medication adherence.
Proportion of patients and items appropriately prescribed or deprescribed.
Patient satisfaction, where measured by a validated tool as part of an effectiveness study.
Non‐medical prescriber versus medical prescriber waiting time to care.
Non‐medical prescribers adversely affecting the health outcomes of patients through medication errors, prescribing errors, adverse events, wrong diagnoses or treatment, increased hospitalisations, or representations for medical care.

Secondary outcomes

Patient‐reported outcomes

We considered patient‐reported outcomes without clinician interpretation of their knowledge requirements, daily functioning, and health‐related quality of life.

Non‐medical prescriber outcomes

Where present, we also reported non‐medical prescriber outcomes of job satisfaction, skills utilisation, education needs, and workload effects.

Resource use outcomes

Medical time saved by non‐medical prescribers.
Non‐medical prescriber versus medical prescriber prescription volume and cost, patient out‐of‐pocket expenses, service costs, and deprescribing rate and cost.
Increased resource use for providing the intervention and for providing subsequent care such as hospitalisations, emergency department visits, and outpatient visits.

Search methods for identification of studies

Electronic searches

We searched the following databases.

Cochrane Central Register of Controlled Trials (CENTRAL, including the Effective Practice and Organisation of Care (EPOC) Group Specialised Register; 2016, Issue 6), in the Cochrane Library (Wiley).
Cochrane Methodology Register, the Cochrane Library; 2012, Issue 3 (Wiley).
Cochrane Database of Systematic Reviews (CDSR), the Cochrane Library; 2016, Issue 7.
Database of Abstracts and Reviews of Effects (DARE), the Cochrane Library; 2015, Issue 2 (Wiley).
Health Technology Assessment Database, the Cochrane Library; 2016, Issue 2 (Wiley).
NHS Economic Evaluation Database, the Cochrane Library; 2015, Issue 2 (Wiley).
MEDLINE (1946 to 19 July 2016), (Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R) 1946 to 19 July 2016) (OvidSP).
Embase (OvidSP) (1980 to 18 July 2016).
PsycINFO (OVID) (1806 to July Week 2, 2016).
Cumulative Index to Nursing and Allied Health Literature (CINAHL) (EBSCOhost) (1980 to 19 July 2016).

The MEDLINE search strategy as illustrated in Appendix 1 was developed by the Cochrane Effective Practice and Organisation of Care (EPOC) Group Information Specialist in consultation with the authors. We translated it for other databases using appropriate syntax and vocabulary for those databases. We employed the Cochrane Highly Sensitive Search Strategy (sensitivity‐ and precision‐maximizing version, 2008 revision) to identify randomised trials, and the Cochrane EPOC Group methodology filter to identify non‐randomised studies. We managed search results using reference management software and removed duplicates before screening was undertaken. We also searched the Cochrane Database of Systematic Reviews (CDSR) and the Database of Abstracts of Reviews of Effects (DARE) for related systematic reviews.

Searching other resources

Grey literature

We conducted a grey literature search to identify studies not indexed in the databases listed above. We used the following sources.

OpenGrey (www.opengrey.eu).
Grey Literature Report by the New York Academy of Medicine (www.greylit.org).
Agency for Healthcare Research and Quality (AHRQ) (www.ahrq.gov).

Trial registries

We searched the following registries.

The Word Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) search portal (apps.who.int/trialsearch).
ClinicalTrials.gov (clinicaltrials.gov).

The corresponding search terms and numbers of results are reported.

Other resources

We screened individual journals and conference proceedings (via handsearching).
We reviewed reference lists of all included studies, relevant systematic reviews; reference lists of other publications.
We contacted authors of relevant studies or reviews when necessary to clarify reported published information or to seek unpublished results or data.
We contacted researchers with expertise relevant to the review topic/EPOC interventions.
We conducted cited reference searches for all included studies in citations indices.

Data collection and analysis

Selection of studies

We merged the search results through the use of a reference management software and removed duplicate records. Two review authors (GW, JG) then independently assessed the titles and abstracts of the search results to evaluate their potential eligibility, and discussed the relevance of articles to the topic. The two review authors were not responsible for the selection of studies they were involved in or associated with. Neutral members of the review team were responsible for assessing the eligibility of each study for inclusion in the review. We retrieved the full‐text of all remaining relevant papers and the two review authors assessed these full‐text articles independently, based on the review's inclusion criteria.

We included a 'Characteristics of excluded studies' table in the review. This table included studies that appear to meet the inclusion criteria but were eventually excluded, and we reported the reasons for exclusion (e.g. not a RCT, only one intervention and/or control site for a CBA study, absence of non‐medical prescriber autonomy). If there was uncertainty or disagreement, consensus was reached by discussion with other review authors. We corresponded with authors of included studies if necessary to obtain further information in order to assess compliance with eligibility and confirm data. Within the review, we mapped the flow of information of identified, included, and excluded studies by depicting them in a PRISMA flow diagram (Moher 2009) (Figure 1).

Figure 1

Study flow diagram.

Data extraction and management

We adapted a standard data extraction form based on the Cochrane EPOC Group's data collection checklist (Cochrane EPOC Group 2013a). We designed and assessed the form to suitably extract data on the characteristics of each study, including study design, study participants, the interventions and comparators, outcomes and follow‐up periods, funding source, and interest declarations. Four review authors (GW, JG, DS, KM) independently extracted study characteristics and the outcome data outlined above. We checked the data against each other. If there was uncertainty or disagreement, we reached consensus by discussion or in the presence of an adjudicating third review author, if necessary. We contacted study authors to obtain any missing information. If a study was reported in more than one publication, we extracted the data from all publications into separate data collection forms before combining them.

Assessment of risk of bias in included studies

Two review authors (GW, JG) independently assessed the risk of bias of included studies, with any disagreements resolved by consensus with a third review author (KM). We used the Cochrane EPOC Group nine‐point criteria for RCTs, non‐RCTs, and CBA studies (Cochrane EPOC Group 2015).

Allocation sequence generation.
Allocation concealment.
Baseline outcome measurements.
Baseline characteristics.
Incomplete outcome data.
Knowledge of allocated interventions.
Protection against contamination.
Selective outcome reporting.
Other risks of bias.

We did not find any ITS studies, but we will assess future studies using the seven standard Cochrane EPOC Group criteria for ITS studies (Cochrane EPOC Group 2015).

Intervention independent of other changes.
Prespecified effect shape.
Intervention unlikely to affect data collection.
Blinding.
Incomplete outcome data.
Selective outcome reporting.
Other bias.

We rated each component and categorised it in a 'Risk of bias' table as 'low risk', 'unclear risk', or 'high risk', as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We documented for each included study a summary assessment of the risk of bias.

Measures of treatment effect

We recorded and reported measures of effect in the same way investigators reported them. We performed all analyses using Cochrane's statistical software, Review Manager 5 (RevMan 2014), and recorded data in the form of a table included in the Cochrane EPOC Group's data extraction template (Cochrane EPOC Group 2013b). For continuous variables, we reported mean differences (MDs) with 95% confidence intervals (CIs) between the intervention and comparison groups. We used a standardised mean difference (SMD) with 95% CI for the same continuous variable measured with different scales. For dichotomous outcomes, we calculated the risk difference (RD) with 95% CI. We planned to calculate the risk ratio (RR), again with 95% CI.

Unit of analysis issues

We assessed whether an appropriate adjustment had been made for clustering in RCTs and CBA studies to avoid unit of analysis errors. If there were insufficient data for re‐analysis, we attempted to correct such errors by contacting study authors to obtain additional data. Determining the intracluster correlation coefficient from additional data or like studies allows adjustment of clustering by inflating the standard error. Where re‐analysis was not possible we reported the point estimate without a standard error or CI and the P value was annotated 're‐analysed'.

Dealing with missing data

We applied the 'Risk of bias' criteria to exclude studies with a high risk of missing data, as they pose serious threats to validity (Higgins 2011). Where appropriate, we contacted study authors for further information. If this was not possible, we reported the number of participants lost to follow‐up. Imputing missing data was only considered when continuous outcomes were reported without measures of variance. We followed the principles of intention‐to‐treat analysis as far as possible.

Assessment of heterogeneity

We found that the range of healthcare settings, differing non‐medical prescribers, differing clinical conditions being managed, and variation in study designs lead to clinical, methodological, and statistical heterogeneity. Assessment of these differences informed the analysis and determined whether results could be statistically combined in a meta‐analysis. The review team made this decision on a consensus basis. We assessed statistical heterogeneity by using the Chi² test to assess if differences in results are compatible with chance alone using P < 0.10. We quantified statistical heterogeneity using the I² statistic, as appropriate. We determined that heterogeneity might not be important between 0% and 40%, 30% to 60% represented moderate heterogeneity, 50% to 90% substantial heterogeneity, and 75% to 100% considerable heterogeneity (Higgins 2011).

Assessment of reporting biases

We assessed the risk of publication bias based on the information in the 'Risk of bias' tables and constructed funnel plots for the outcomes of systolic blood pressure and low‐density lipoprotein.

Data synthesis

We used a structured synthesis approach to analyses. After consideration of the small‐study effects of many included studies we used a fixed‐effect model for meta‐analysis and compared outcomes with a random‐effects model. For quantitative synthesis we used Review Manager 5 for statistical analysis (RevMan 2014). Where we could not combine data for a meta‐analysis due to inconsistency of reporting measures, or when it was not applicable to use the average effect across studies of an intervention, we reported in this plain language summaries as appropriate. We included key data elements such as explanatory factors, results, effects, and certainty of evidence in a table for each category of interventions.

Summary of findings

We used a 'Summary of findings' table and GRADEpro GDT software to record results, outcomes, and outcome risks in our structured synthesis (GRADEpro GDT 2014). In addition, we used the five GRADE study considerations (study limitations, consistency of effort, imprecision, indirectness, and publication bias) to assess the certainty of the body of evidence and summarise our confidence in the effects of the interventions by outcome across studies (Atkins 2004). We included the following outcomes in the 'Summary of findings' table: systolic blood pressure, glycated haemoglobin, low‐density lipoprotein, medication adherence, patient satisfaction, adverse events, and health‐related quality of life. We justified all decisions to down‐ or upgrade the certainty of evidence using footnotes, and we made comments to aid the reader's understanding of the review where necessary.

Subgroup analysis and investigation of heterogeneity

Differences in healthcare settings, non‐medical prescriber types, clinical conditions being managed, and study designs informed the assessments of methodological and statistical heterogeneity.

Explanatory variables or effect modifiers which may have influenced the size of intervention effects included the level of prescriber education, study location, patient condition being treated, and adherence to therapy and practice guidelines. The degree of non‐medical prescribing autonomy within and across subgroups may have explained differences in outcomes and limited the applicability of findings.

For consistency across studies, we presented data as subgroups for the clinical outcomes of systolic blood pressure, glycated haemoglobin, and low‐density lipoprotein at six and 12 months. We presented quality of life measures (SF‐36 and SF‐12) as subgroups of physical component and mental component scores.

In considering the type of intervention, we did not undertake a meta‐analysis comparing algorithm prescribing to more autonomous prescribing on clinical outcomes due to considerable heterogeneity.

There were insufficient studies to compare outcomes from different non‐medical prescriber settings e.g. secondary care versus primary care.

Variability in education standards made it difficult to compare non‐medical prescriber subgroups.

Sensitivity analysis

We undertook a sensitivity analysis comparing meta‐analyses outcomes using fixed‐effect and random‐effects analyses for the three clinical surrogate markers of disease: systolic blood pressure; glycated haemoglobin; and low‐density lipoprotein (Table 1). The effect modifier of clustering in RCTs on systolic blood pressure at six months was tested by removing these trials from the meta‐analysis (Margolis 2013 at six months; Khunti 2007 and Margolis 2013 at 12 months; Analysis 1.2; Analysis 1.3). We did not undertake a sensitivity analysis excluding unclear or high risk of bias studies due to the similar risk of bias elements existing within the outcome categories.

See: Characteristics of included studies; Characteristics of excluded studies; Characteristics of studies awaiting classification; Characteristics of ongoing studies.

Table 1. Fixed‐effect outcomes versus random‐effects for clinical surrogate markers

Outcome or subgroup	Fixed‐effect estimate	Random‐effects estimate
1.1 Systolic blood pressure (mmHg)	‐5.85 (‐6.76 to ‐4.94)	‐6.59 (‐8.48 to ‐4.71)
1.1.1 6 months	‐6.76 (‐8.24 to ‐5.27)	‐7.34 (‐11.09 to ‐3.60)
1.1.2 12 months	‐5.31 (‐6.46 to ‐4.16)	‐5.91 (‐7.71 to ‐4.10)
1.2 HbA1c (%)	‐0.55 (‐0.74 to ‐0.36)	‐0.55 (‐0.76 to ‐0.35)
1.2.1 HbA1c (6 months)	‐0.42 (‐0.75 to ‐0.09)	‐0.45 (‐0.09 to ‐0.01)
1.2.2 HbA1c (12 months)	‐0.62 (‐0.85 to ‐0.38)	‐0.62 (‐0.85 to ‐0.38)
1.3 LDL (mmol/L)	‐0.23 (‐0.28 to ‐0.17)	‐0.22 (‐0.42 to ‐0.02)
1.3.1 LDL (6 months)	‐0.25 (‐0.34 to ‐0.17)	‐0.13 (‐0.39 to 0.12)
1.3.2 LDL (12 months)	‐0.21 (‐0.29 to ‐0.14)	‐0.3 (‐0.62 to 0.02)

LDL: low‐density lipoprotein

Results

Description of studies

Results of the search

The database search yielded 13,220 titles. We found 51 additional studies through handsearching. After removing duplicates, we screened 9335 studies and reviewed 162 full‐text articles. We excluded 112 studies that did not meet the inclusion criteria and recorded our reasons for exclusion. We included 46 studies (37,337 participants). Of these, 44 were randomised controlled trials (RCTs), including six cluster‐RCTs (Fairall 2008; Heisler 2012; Khunti 2007; Margolis 2013; Moher 2001; Pagaiya 2005), one controlled trial (Denver 2003), and one controlled before‐and‐after (CBA) study (Thompson 1984). Three studies are awaiting classification (Barton 2013; Neilson 2015; Tsuyuki 2014), and one study is ongoing (Mikuls 2015). Refer to Figure 1 for PRISMA diagram.

Included studies

Non‐medical prescribing studies were included where the health professional (other than a medical practitioner) undertook a high level of autonomous prescribing. This included medication initiation, dosage change, or cessation of medication (with or without guidance from established protocols and guidelines).

Participants

Non‐medical prescribing versus medical prescribing was practised by nurses in 26 studies with 28,621 participants (Ansari 2003; Aubert 1998; Barr Taylor 2003; Becker 2005; DeBusk 1994; Denver 2003; Einhorn 1978; Fairall 2008; Fischer 2012; Hill 2003; Houweling 2009; Houweling 2011; Ishani 2011; Khunti 2007; Klingberg‐Allvin 2015; Kuethe 2011; Litaker 2003; Logan 1979; MacMahon Tone 2009; Moher 2001; New 2003; Pagaiya 2005; Rudd 2004; Spitzer 1974; Tobe 2006; Wallymahmed 2011), and by pharmacists in 20 studies with 8716 participants (Bruhn 2013; Chenella 1983; Choe 2005; Cohen 2011; Ellis 2000; Finley 2003; Heisler 2012; Hirsch 2014; Hunt 2008; Jaber 1996; Magid 2013; Margolis 2013; Marotti 2011; McAlister 2014; Taveira 2010; Taveira 2011; Thompson 1984; Tsuyuki 2015; Tsuyuki 2016; Vivian 2002).

The health professionals delivering the interventions were pharmacists or nurses with varying degrees of formal or informal training. We did not find any studies where other non‐medical health professionals, such as physician assistants undertook prescribing roles. Nurse prescribing was undertaken in the majority of studies by reference to algorithms. While nurses exercised independence in prescribing by algorithm, physicians were usually available for consultation for issues beyond the scope of the algorithm, or for more complex cases or for periodic review.

Pharmacist prescribing was generally undertaken in a more autonomous way, with more reliance on clinical judgement and guidelines rather than restrictive algorithms. This broader practice scope was supported through collaborative practice agreements in the USA and independent or supplementary prescribing in the UK. In addition to their defined prescribing autonomy, non‐medical prescribers in several studies had limits placed on additional prescribing, and required medical prescribing or approval for dose acceleration (Tobe 2006), management of conditions outside the focus of care (Finley 2003; Litaker 2003; New 2003; Taveira 2011; Vivian 2002), and initiation of new drugs (Barr Taylor 2003; DeBusk 1994; New 2003; Rudd 2004).

Excluding the cluster‐RCTs, nine studies had less than 100 patients, seven studies had more than 100 and less than 200 patients, 16 studies had more than 200 and less than 500 patients, five studies had between 500 to 800 patients, and three studies included over 1000 patients. There were six cluster‐RCTs: Fairall 2008, 31 clinics, cohort one 9252 patients, cohort two 6231 patients; Heisler 2012, 16 primary care teams at five medical centres, 4100 patients; Khunti 2007, 20 primary care practices, 1316 patients; Margolis 2013, 16 primary care clinics, 450 patients; Moher 2001, 21 general practices, 1906 patients; Pagaiya 2005, 18 nurse‐led health centres, 3960 patients.

Setting

Four nurse prescribing studies (14,921 participants) were undertaken in low‐ and middle‐income settings within Colombia, South Africa, Uganda, and Thailand (Einhorn 1978; Fairall 2008; Klingberg‐Allvin 2015; Pagaiya 2005). The remainder of studies were undertaken in the high‐income countries, of Australia (1), Canada (6), Ireland (1), Netherlands (3), UK (6), and USA (25). Forty‐two studies were based in ambulatory care settings, including primary care clinics, medical centres, general practices, community pharmacies, and hospital outpatient clinics. Two studies were located in secondary care settings (Chenella 1983; Marotti 2011). One study was set in the workplace (Logan 1979), and one in an aged care setting (Thompson 1984).

Interventions

Pharmacist and nurse interventions were often multifaceted, with prescribing being one element of a complex management approach. For example, in diabetes care, patient education, self‐care, diet, exercise, and follow‐up were factors influencing outcomes, as well as the prescribing of medications.

Outcomes

The majority of studies involved the management of one or more chronic diseases (heart failure, hypertension, diabetes, dyslipidaemias) and risk factors for disease recurrence such as stroke (McAlister 2014), and acute myocardial infarction or heart failure (DeBusk 1994; Khunti 2007). Studies outside of these areas included the management of chronic pain (Bruhn 2013), family planning (Einhorn 1978), HIV treatment (Fairall 2008), incomplete abortion (Klingberg‐Allvin 2015), depression (Finley 2003), and asthma in children, which was the only paediatric study (Kuethe 2011).

Non‐medical clinician collaborative care approaches with physicians (Litaker 2003), or community health workers (Becker 2005; Hill 2003), and interventions with telemonitoring (Magid 2013; Margolis 2013), added to the complexity of determining specific non‐medical prescribing outcomes.

The following 21 studies had a more direct relationship between non‐medical prescribing and the outcome markers of the disease or condition: Ansari 2003 (heart failure); Bruhn 2013 (chronic pain); Chenella 1983 (anticoagulation); Denver 2003 (blood pressure); Fairall 2008 (HIV medications); Hirsch 2014 (blood pressure); Houweling 2009 and Houweling 2011 (glycaemia, blood pressure, lipids); Hunt 2008 (blood pressure); Ishani 2011 (glycaemia, blood pressure, lipids); Jaber 1996 (glycaemia, blood pressure, lipids); Klingberg‐Allvin 2015 (incomplete abortion); Logan 1979 (blood pressure); MacMahon Tone 2009 (glycaemia, blood pressure, lipids); McAlister 2014 (blood pressure, lipids); Marotti 2011 (regular medications); Thompson 1984 (medications in the geriatric setting); Tsuyuki 2015 (blood pressure); Tsuyuki 2016 (glycaemia, blood pressure, lipids); Vivian 2002 (blood pressure); and Wallymahmed 2011 (glycaemia, blood pressure, lipids).

Excluded studies

We excluded studies if the study design did not meet the EPOC criteria for a RCT, controlled clinical trial, CBA or ITS. We excluded studies where we judged that the non‐medical health professional did not have a significant degree of autonomy in their prescribing practice, and prescribing required medical review, consultation, or authorisation.

Risk of bias in included studies

The risk of bias assessment for included studies is presented in the 'Risk of bias' tables, under each study in the section Characteristics of included studies. The risk of bias results are presented in a graphical form in Figure 2.

Figure 2

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Thirty‐three of 46 studies (72%) adequately described the random sequence generation and we considered them to be at low risk of bias. Allocation concealment was undertaken in 13 studies (28%), unclear in 31 studies (67%) and with no concealment in two studies (Margolis 2013; Thompson 1984).

Blinding

Blinding of both participants and personnel could not be achieved through the study design in 44 of the 46 included studies. In the Chenella 1983 study it was unclear whether patients would be aware that the pharmacist had undertaken anticoagulation dose determinations, and in the Pagaiya 2005 study, whether the intervention group nurses had undertaken additional training and were using guidelines. Objective clinical outcomes in studies requiring laboratory measures such as glycated haemoglobin and low‐density lipoprotein were coded as blinded outcome assessment. In seven studies, blinded assessment of blood pressure was undertaken (Hill 2003; Hunt 2008; Logan 1979; Magid 2013; McAlister 2014; Moher 2001; Rudd 2004). Where blood pressure assessment was not clear or undertaken by study investigators, we judged this to be an unclear outcome assessment. Ansari 2003 used an independent research assistant to assess β‐blocker use in heart failure.

Incomplete outcome data

Loss to follow‐up of 20% or more in either the intervention or control arms occurred in 14 studies (Aubert 1998; Becker 2005; Bruhn 2013; Choe 2005; Einhorn 1978; Finley 2003; Heisler 2012; Hirsch 2014; Hunt 2008; Ishani 2011; Jaber 1996; McAlister 2014; Moher 2001; New 2003).

Selective reporting

The funnel plots of systolic blood pressure revealed a degree of asymmetry, demonstrating a possible publication bias from an absence of published negative intervention studies. The funnel plot of low‐density lipoprotein studies was asymmetrical, with heterogeneity a consideration.

Other potential sources of bias

The majority of studies had a degree of confounding either by the multifactorial intervention (which made it difficult to distinguish the influence of non‐medical prescribing on outcomes) or by unclear prescribing autonomy or medical influence. The six cluster‐RCTs appropriately accounted for the cluster design.

Effects of interventions

See: Summary of findings for the main comparison Non‐medical prescribing compared to medical prescribing for acute and chronic disease management in primary and secondary care

See: summary of findings Table for the main comparison for the main comparisons; systolic blood pressure, glycated haemoglobin, low‐density lipoprotein, adherence, adverse events, patient satisfaction, and quality of life.

We had planned to analyse the six comparisons listed in the Types of interventions section, however we only found studies for the following two comparisons: non‐medical prescribing in acute care (secondary care); and non‐medical prescribing in chronic care (primary/ambulatory care).

Non‐medical prescribing in acute care (secondary care)

Primary Outcomes

Studies involving non‐medical prescribing interventions were often characterised by degrees of confounding, including the presence of multiple interventions, patient comorbidities, study duration, differing levels of non‐medical prescriber training, and unclear influences from medical prescribers. However, while recognising these complexities and limitations, care involving non‐medical prescribers resulted in improvements or similar effectiveness to usual care for a range of clinical outcomes and surrogate disease markers.

We found two studies (438 participants) where non‐medical prescribing was practised in an acute/secondary care setting (Chenella 1983; Marotti 2011).

1. Systolic blood pressure

Outcome not reported.

2. Glycated haemoglobin

Outcome not reported.

3. Low‐density lipoprotein

Outcome not reported.

4. Proportion of prescribers, medical and non‐medical, appropriately adhering to practice guidelines

Pharmacist prescribers adjusted anticoagulant therapy, as well as an experienced physician, in the independent management of anticoagulation therapy for inpatients. There were no significant differences between groups for mean heparin and warfarin doses, partial thromboplastin time, days to reach therapeutic levels, or mean prescribed and simulated heparin doses (Chenella 1983; Table 2).

Table 2. Outcomes of studies not included in meta‐analyses

Study	Patient group	Comparison	Outcome
Bruhn 2013	Chronic pain	To compare the effectiveness of pharmacist medication review with or without pharmacist prescribing with standard care	Compared with baseline the Chronic Pain Grade improved in prescribing arm 47.7% (21/44; P = 0.003) and review arm 38.6% (17/44; P = 0.001) but not TAU 31.3% (15/48; ns) SF‐12 mental component score showed no effect for prescribing or review arms and deterioration in TAU arm. Hospital Anxiety and Depression scores improved in prescribing arm for depression (P = 0.022) and anxiety (P = 0.007) and between groups (P = 0.022 and P = 0.045 respectively).
Chenella 1983	Anticoagulation	Pharmacist versus physician independent management of anticoagulant therapy for inpatients	There were no differences between groups for mean heparin and warfarin doses, partial thromboplastin time, days to reach therapeutic levels, mean prescribed and simulated heparin doses.
Choe 2005	Type 2 diabetes	Pharmacist case management versus usual medical care	Patients in the pharmacist case managed group received greater reductions in HbA1c (2.1% vs 0.9%, P = 0.03). Three of five process measures were conducted more frequently in the intervention group than control group. LDL measurement (100% vs 85.7%, P = 0.02), retinal examination (97.3% vs 74.3%, P = 0.004), monofilament foot screening, (92.3% vs 62.9%, P = 0.002).
Einhorn 1978	Family planning	Family planning services provided by nurses versus physicians	Nurses' clients were as equally as successful as physicians in continuing contraceptive use and preventing pregnancy. Nurses were less likely than physicians to provide patients on their first visit with IUDs, prescribe oral contraceptives, or sterilisation. Nurses were more likely to give temporary prescriptions than physicians until the next visit (25% vs 16%, P < 0.001) for reasons including possible pregnancy and patients not menstruating.
Ellis 2000	Dyslipidaemia	Clinical pharmacists providing pharmaceutical care in addition to usual medical care versus usual medical care	The absolute change in total cholesterol (17.7 vs 7.4 mg/dL, P = 0.028) and LDL (23.4 vs 12.8 mg/dL, P = 0.042) was greater in the intervention than control group.
Fairall 2008	HIV	Prescribing of antiretroviral treatment by nurses versus doctors	Cohort 1 ‐ not receiving antiretrovirals. Time to death did not differ (HR 0.94, 95% CI 0.76 to1.15). Cohort 2 ‐ received antiretrovirals for at least six months. Viral load suppression 12 months after enrolment was equivalent in intervention and control. Risk difference 1.1% (95% CI ‐2.4 to 4.6).
Finley 2003	Depression	Collaborative care model of clinical pharmacists providing drug therapy management and treatment follow‐up versus usual care	Clinical improvements noted in both groups but not significant. Intervention group had higher drug adherence at six months (67% vs 48%; OR 2.17, 95% CI 1.04 to 4.51; P = 0.038)
Fischer 2012	Lipid control in diabetes	Algorithm‐driven telephone care by nurses as an adjunct to usual care versus usual care	The percentage of patients with an LDL < 100 mg/dL increased from 52% to 58.5% in the intervention group and decreased from 55.6% to 46.7% in the control group (P < 0.01). The intervention did not affect glycaemic and BP outcomes
Heisler 2012	Blood pressure control in diabetes	A pharmacist‐led intervention (Adherence and Intensification of Medications) in patients with diabetes and poor BP control versus usual care	The mean systolic BP decrease from 6 months before to 6 months after the 14‐month intervention was not different (8.9 mmHg decline in the intervention arm and 9.0 mmHg decline in the control arm). There was no difference in the mean HbA1c and LDL levels between groups after the end of the intervention period (examining 12 months). At the end of the first quarter after activation, there was a significantly greater drop in systolic BP in the intervention group versus control, 9.7 mmHg vs 7.2 mmHg; MD 2.4 mmHg (95% CI 1.5 to 3.4 P < 0.001).
Houweling 2011	Type 2 diabetes	Primary care nurse management of type two diabetes versus management by GPs	After 14 months between‐group differences for reduction in HbA1c, BP, and lipid profile were not significant. Mean systolic and diastolic BPs were lower in both groups. Most process indicators were significantly better in the nurse care group. More patients were satisfied with their care in the nurse group however the physical component of the SF‐26 was better in the GP group.
Ishani 2011	Cardiovascular risk factors in diabetes	Nurse case management versus usual care to improve hypertension, hyperglycaemia, and hyperlipidaemia in veterans with diabetes	A greater number of patients in the nurse case management had all three measures under control (21.9% vs 10.1%, P < 0.01). A greater number of intervention group participants achieved individual treatment goals. HbA1c < 8% (73.7% vs 65.8% P = 0.04), BP < 130/80 mmHg (45% versus 25.4%, P < 0.01) but not for LDL < 100 mg/dL (57.6% vs 55.4%, P = 0.61).
Jaber 1996	Non‐insulin dependent diabetes	Pharmacists providing pharmaceutical care versus physicians	Improvement was seen in glycated haemoglobin in the intervention group at 4 months (9.2% ± 2.1 vs 12.1% ± 3.7, P = 0.003), and fasting plasma glucose (8.5 ± 2.3 vs 11.0 ± 3.9 mmol/L, P = 0.015). There was little or no change within or between groups for BP, lipid profile, renal function, weight, or quality of life measures.
Klingberg‐Allvin 2015	Women with signs of incomplete abortion	Midwives diagnosing and treating incomplete abortion with misoprostol compared to physicians	452 (95.8%) women in the midwife group and 467 (96.7%) in the physician group had complete abortion. The model risk difference for midwife versus physician group was ‐0.8% (95% CI ‐2.9 to 1.4) falling within the predefined equivalence range (‐4% to 4%).
Kuethe 2011	Children with asthma	Non‐inferiority of care provided by a hospital‐based specialised asthma nurse versus a GP or paediatrician	The corrected daily dose of inhaled corticosteroids as well as the percentage of children prescribed long‐acting beta agonists/inhaled corticosteroids was not significantly different between groups at one and two years.
Logan 1979	Hypertension	Treatment of hypertension in the workplace by nurses versus treatment in the community by the family doctor	Patients in the nurse group were more likely to be put on antihypertensive medications (94.7% vs 62.7%, P < 0.001), to reach goal BP in the first six months (48.5 vs 27.5%, P < 0.001) and to take drugs prescribed (67.6 vs 49.1%, P < 0.005).
Marotti 2011	Postoperative patients	Pharmacist medication history and supplementary prescribing versus pharmacist medication history versus usual care	The marginal mean number of missed doses per patient was 3.21 (95% CI 2.89 to 3.52) in the control group, which was reduced in the pharmacist prescribing group 1.07 (95% CI 0.90 to 1.25, P = 0.002) but not in the pharmacist history group 3.30 (95% CI 2.98 to 3.63). The number of medications charted at an incorrect dose or frequency was reduced in the pharmacist history group. The pharmacist prescribing group had less dose errors than the pharmacist history group (P = 0.004).
Moher 2001	Secondary prevention of coronary heart disease in primary care	Audit group verus GP recall group versus nurse recall group (disease register and patient recall to nurse‐led clinic)	Little or no difference occurred in assessment between the nurse and GP recall group. Mean BP, total cholesterol, cotinine levels varied little between groups as did prescribing of hypotensive and lipid‐lowering agents. Prescribing of antiplatelet drugs was higher in the nurse recall group vs GP recall group, MD 8% (95% CI 1% to 15%, P = 0 .031).
Pagaiya 2005	Primary care nurses	Education and implementation of prescribing and clinical guidelines by nurses in rural health centres versus usual nurse care	Antibiotic prescribing in children 0 to 5 years for respiratory tract infections fell, (42% at baseline to 27% at follow‐up, control 27% to 30%, P = 0.022). Guidelines had no effect on prescribing antibiotics for diarrhoea but oral rehydration prescribing increased. Diazepam prescribing for adults fell, (intervention 17% to 10%, control 21% to 18%, P = 0.029).
Spitzer 1974	Patients attending primary care	Nurse practitioners versus physicians plus conventional nurse in primary care	Similar mortality experience, no differences in physical functioning capacity, social or emotional function. Quality of care similar. In 510 prescriptions, an adequate rating was given to 75% of conventional group and 71% in the nurse practitioner group, probably leading to little difference between groups.
Taveira 2010	Type 2 diabetes	A pharmacist‐led Veterans affairs Multidisciplinary Education and Diabetes Intervention for Cardiac risk reduction (VA‐MEDIC) plus usual care versus usual care	After four months there was a difference (P < 0.05) in the percentage of VA‐MEDIC patients versus controls in attaining target goals for systolic BP < 130 mmHg and HbA1c < 7% but not lipid control or tobacco use.
Thompson 1984	Drug therapy in a geriatric setting	Drug therapy prescribing and patient care management by clinical pharmacists versus usual care	The clinical pharmacist group probably had a lower number of deaths (P = 0.05), a higher number of patients being discharged to lower levels of care (P = 0.03) and a lower average number of drugs per patient (P = 0.04).
Tsuyuki 2016	Patients with cardiovascular risk factors associated with hypertension, diabetes, dyslipidaemia and smoking	Community pharmacist care versus usual care	At 3 months the intervention group patients had greater improvements in LDL cholesterol (‐ 0.2 mmol/L, P < 0.001, systolic BP (‐9.37 mmHg, P < 0.001), glycosylated haemoglobin (‐0.92%, P < 0.001) and smoking cessation (20.2%, P < 0.002).

BP: blood pressure
CI: confidence interval
GP: general practitioner
HbA1c: glycated haemoglobin
HR: hazard ratio
IUD: inter uterine device
LDL: low‐density lipoprotein
MD: mean difference
OR: odds ratio
TAU: treatment as usual

5. Proportion of patients demonstrating medication adherence

Outcome not reported.

6. Proportion of patients and items appropriately prescribed or deprescribed

Preoperative medication history taking and prescribing by a pharmacist improved the accuracy of medication documentation and significantly reduced missed doses of regular medication for elective surgical patients. The marginal mean number of missed doses per patient was 3.21 (95% confidence interval (CI) 2.89 to 3.52) in the control group, which was significantly reduced in the pharmacist prescribing group 1.07 (95% CI 0.90 to 1.25; P = 0.002) (Marotti 2011; Table 2).

7. Patient satisfaction, where measured by a validated tool as part of an effectiveness study

Outcome not reported.

8. Non‐medical prescriber versus medical prescriber waiting time to care

Outcome not reported.

9. Non‐medical prescribers adversely affecting the health outcomes of patients through medication errors, prescribing errors, adverse events, wrong diagnoses or treatment, increased hospitalisations, or representations for medical care

Chenella 1983 reported no patients had major bleeding but four patients in the pharmacist prescriber group had minor bleeding (one patient had a bleeding facial laceration on admission but a normal prothrombin time). One patient in the physician prescriber group died, after receiving heparin and warfarin for a stroke in evolution, but there was no evidence of bleeding.

Secondary Outcomes

Patient‐reported outcomes

1. Health‐related quality of life

Outcome not reported.

Non‐medical prescriber outcomes

1. Job satisfaction, skills utilisation, education needs, and workload effects

Outcome not reported.

Resource‐use outcomes

1. Medical time saved by non‐medical prescribers

Outcome not reported.

2. Non‐medical prescriber versus medical prescriber prescription volume and cost, patient out‐of‐pocket expenses, service costs, deprescribing rate, and cost

There was little or no difference in amount of anticoagulant drugs prescribed by pharmacists compared to a physician (Chenella 1983; Table 2).

3. Increased resource use for providing the intervention and for providing subsequent care such as hospitalisations, emergency department visits, and outpatient visits.

Outcome not reported.

Non‐medical prescribing in chronic care (primary/ambulatory care)

We included 40 studies in this comparison. We included ambulatory care clinics for chronic disease management located with secondary care hospitals in this subgroup (Denver 2003; Houweling 2009; Jaber 1996; Kuethe 2011; MacMahon Tone 2009; McAlister 2014; New 2003). Two studies were undertaken in the community pharmacy setting (Tsuyuki 2015; Tsuyuki 2016,).

Meta‐analyses were undertaken for systolic blood pressure, glycated haemoglobin, and low‐density lipoprotein using the fixed‐effect method for outcomes at six and 12 months (Figure 3; Figure 4; Figure 5). These studies were skewed toward either nurse or pharmacist prescribers, namely, systolic blood pressure at six months (3 nurse studies, 8 pharmacist studies), systolic blood pressure at 12 months (10 nurse studies, 2 pharmacist studies), glycated haemoglobin at six months (1 nurse study, 2 pharmacist studies) glycated haemoglobin at 12 months (6 nurse studies, 0 pharmacist studies), low‐density lipoprotein at six months (4 nurse studies, 2 pharmacist studies), low‐density lipoprotein at 12 months (7 nurse studies, 0 pharmacist studies).

Figure 3

Forest plot of comparison: 1 Non‐medical prescribing group versus usual care, Outcome: 1.2 Systolic blood pressure mmHg.

Figure 4

Forest plot of comparison: 1 Non‐medical prescribing group versus usual care, Outcome: 1.1 HbA1c (%).

Figure 5

Forest plot of comparison: 1 Non‐medical prescribing group versus usual care, Outcome: 1.3 Low‐density lipoprotein (LDL) mmol/L.

Moderate or considerable heterogeneity was evident in all subgroups apart from the glycated haemoglobin 12‐month subgroup for which heterogeneity might not be important given an I² = 0%. While the degree of heterogeneity provides a caution, studies which contained non‐medical prescribing as an intervention component showed improvement in three surrogate markers of disease; systolic blood pressure, glycated haemoglobin, and low‐density lipoprotein.

A single study compared pharmacist case management versus the active control of nurse‐led case management and feedback to primary care physicians for medication adjustment in the secondary prevention after minor stroke (McAlister 2014). Improvements in both systolic blood pressure and low‐density lipoprotein guideline targets were observed in both the pharmacist group (43.4%) and nurse‐led group (30.9%) after six months (absolute difference 12.5%, P = 0.03). Multivariable analyses confirmed the greater attainment of targets in the pharmacist group (adjusted odds ratio (OR) 2.31, 95% CI 1.29 to 4.2; P = 0.005, adjusted for age, comorbidities, sex, smoking status, and waist circumference). Both groups had similar reductions in systolic blood pressure during the trial and the overall result was driven by a higher proportion of patients meeting low‐density lipoprotein targets in the pharmacist‐led group versus the nurse‐led group (51.1% versus 33.8%, P = 0.003).

Primary Outcomes

1. Systolic blood pressure

Eleven ambulatory care studies (2076 participants) reporting systolic blood pressure at six months showed a mean difference (MD) favouring the non‐medical prescribing group compared to usual care of ‐6.76 mmHg (95% CI ‐8.24 to ‐5.27; Analysis 1.1), but there was considerable heterogeneity (I² = 82%, overall effect P < 0.00001) (Cohen 2011; Denver 2003; Hirsch 2014; Houweling 2009; Magid 2013; Margolis 2013; McAlister 2014; Rudd 2004; Taveira 2011; Tsuyuki 2015; Vivian 2002). At 12 months, 12 ambulatory care studies (4229 participants) showed a MD favouring the non‐medical prescribing group of ‐5.31 mmHg (95% CI ‐6.46 to ‐4.16; Analysis 1.1) with moderate heterogeneity (I² = 50%, overall effect P < 0.00001) (Aubert 1998; Barr Taylor 2003; Becker 2005; Hill 2003; Houweling 2009; Hunt 2008; Khunti 2007; MacMahon Tone 2009; Margolis 2013; New 2003; Tobe 2006; Wallymahmed 2011). The test for subgroup differences was not significant (I² = 56.3%, P= 0.13) (Figure 3).

The systolic blood pressure effect estimate at six months for the fixed‐effect model was MD ‐6.76 mmHg, 95% CI ‐8.24 to ‐5.27 compared to the random‐effects estimate (MD ‐7.34 mmHg, 95% CI ‐11.09 to ‐3.60). At 12 months the respective comparison was MD ‐5.31 mmHg, 95% CI ‐6.46 to ‐4.16 versus MD ‐5.91 mmHg, 95% CI ‐7.71 to ‐4.10 (Table 1). There was a moderate‐certainty of evidence (summary of findings Table for the main comparison).

Excluding the cluster‐RCT at six months (Margolis 2013), the effect estimate was MD ‐6.13 mmHg, 95% CI ‐7.83 to ‐4.44; 10 studies, 1628 participants (Analysis 1.1.3). Excluding the cluster‐RCTs at 12 months (Khunti 2007; Margolis 2013), the effect estimate was MD ‐4.84 mmHg, 95% CI ‐6.29 to ‐3.39; 10 studies, 2627 participants (Analysis 1.1.4).

The subgroup analysis of four studies (695 participants) where non‐medical prescribers demonstrated a higher level of prescribing autonomy in the control of systolic blood pressure showed: fixed‐effect MD ‐2.98 mmHg, 95% CI ‐5.36 to ‐0.59; P = 0.01, compared with a random‐effects model MD ‐6.78 mmHg, 95% CI ‐15.38 to 1.81; P = 0.12, with considerable heterogeneity I² = 90% (Analysis 1.1.5; Figure 3).

2. Glycated haemoglobin

For glycated haemoglobin, three ambulatory care studies at six months demonstrated a MD favouring the non‐medical prescribing group of ‐0.42% (95% CI ‐0.75 to ‐0.09; 271 participants; Analysis 1.2) with moderate heterogeneity (I² = 44%, overall effect P < 0.01) (Cohen 2011; Houweling 2009; Taveira 2011). At 12 months, six ambulatory care studies managing glycated haemoglobin showed a MD favouring the non‐medical prescribing group of ‐0.62% (95% CI ‐0.85 to ‐0.38; 775 participants) with minimal heterogeneity (I² = 0%, overall effect P < 0.00001; Analysis 1.2) (Aubert 1998; Barr Taylor 2003; Houweling 2009; Litaker 2003; MacMahon Tone 2009; Wallymahmed 2011). The test for subgroup differences was not significant (I² = 0%, P = 0.35; Figure 4). For fixed‐effect versus random‐effects estimates refer to Table 1. There was a high‐certainty of evidence (summary of findings Table for the main comparison; Table 1).

3. Low‐density lipoprotein

Six ambulatory care studies (1213 participants) for low‐density lipoprotein management at six months showed a MD favouring the non‐medical prescribing group of ‐0.25 mmol/L (95% CI ‐0.34 to ‐0.17), but these studies demonstrated considerable heterogeneity (I² = 88%, overall effect P < 0.00001; Analysis 1.3) (Cohen 2011; DeBusk 1994; Hirsch 2014; Houweling 2009; McAlister 2014; Taveira 2011). At 12 months the MD favouring the non‐medical prescribing group in seven ambulatory care studies was ‐0.21 mmol/L (95% CI ‐0.29 to ‐0.14; 7 studies, 1469 participants; Analysis 1.3). The studies demonstrated considerable heterogeneity (I² = 93%; overall effect P < 0.00001) (Aubert 1998; Barr Taylor 2003; Becker 2005; DeBusk 1994; Houweling 2009; MacMahon Tone 2009; Wallymahmed 2011). The test for subgroup differences was not significant (I² = 0%, P = 0.53; Figure 5). There was moderate‐certainty of evidence (summary of findings Table for the main comparison; Table 1).

Further exploration of the high heterogeneity in the six‐month low‐density lipoprotein study was undertaken by examining the differences in pharmacist and nurse prescribing. It was found that heterogeneity might not be important in the four pharmacist studies (629 participants) (MD ‐0.09, 95% CI ‐0.20 to 0.02; I²= 0%; Analysis 1.4), which did not yield a significantly different overall effect (P = 0.1). Considerable heterogeneity existed in the two nursing studies (584 participants) (MD ‐0.52, 95% CI ‐0.67 to ‐0.38; I² = 94%; Analysis 1.4), with a significant overall effect (P < 0.00001). The test for overall effect for both subgroups had considerable heterogeneity and was significant (I² = 88%, P < 0.00001). The subgroup differences showed very high heterogeneity and were significant (I² = 95.6%, P < 0.00001). For fixed‐effect versus random‐effects estimates refer to Table 1.

4. Proportion of prescribers, medical and non‐medical, appropriately adhering to practice guidelines

Adherence to practice guidelines was difficult to quantify across studies. Intervention group prescribing was usually aimed at treating a target based on approved therapeutic guidelines. Usual care prescribing may have been based on supplied guidelines, education, or an assumed knowledge of current guidelines.

5. Proportion of patients demonstrating medication adherence (Analysis 1.5 and 1.6)

Medication adherence was assessed in 10 studies using a number of approaches including Morisky Medication Adherence Scale, medication possession ratio, patient report, pill count, electronic drug event monitoring, and pharmacy medication refill information (Table 3). Medication adherence was reported as high in intervention and usual care groups across studies. There was probably little or no difference between groups in six studies (Bruhn 2013; Cohen 2011; Finley 2003; Hunt 2008; Magid 2013; Vivian 2002), and an improved outcome favouring the intervention group in two studies (Logan 1979; Rudd 2004). The study by Margolis 2013 found an improved outcome favouring the intervention group at six months, but no difference between groups at 12 and 18 months. Medication adherence outcomes could not be assessed in the study by Hirsch 2014.

Table 3. Primary outcome ‐ medication adherence

Study	Medication adherence measure	Outcome
Bruhn 2013	Morisky Medication Adherence Scale	Assessed adherence at baseline with patients in both groups reporting full adherence.
Cohen 2011	Medication possession ratios	The medication possession ratio (total days' supply of medication divided by total number of expected medication intake days) used in this study found little or no difference between the pharmacist prescribing arm and usual care, even though more medications were prescribed in the pharmacist arm. Adherence was high and ranked above 80%.
Finley 2003	Medication possession ratios	Determined the medication possession ratio from computerised prescription refill records. Full drug adherence was defined as a medication possession ratio value of 0.83 or more during the six‐month follow‐up. Medication possession ratios at three and six months were probably not different between intervention and control arms even though patients in the intervention group were more likely to change antidepressants. An additional measure, the Health Plan Employer Data Information Set guidelines for successful antidepressant treatment, showed there was little or no difference between groups in compliance with the early phase of treatment, but there was a significant difference in compliance in the intervention group continuation phase.
Hunt 2008	Morisky Medication Adherence Scale	Reported no differences at study end in the proportions of subjects reporting high medication adherence. There was an improvement in adherence with the groups from baseline to study end. Adherence did not predict goal attainment.
Hirsch 2014	Not described	Non‐adherence was identified in five of 33 patients with drug therapy problems at baseline, one of 12 patients at six months and one of four patients at nine months.
Logan 1979	Patient claim and pill counts	High adherence was judged if patients claimed to be taking their medication as instructed and 80% or more of drugs prescribed were consumed as determined by pill counts. In the nurse intervention group patients were more adherent than the control group.
Magid 2013	Medication possession ratios	Little or no difference between groups in the mean medication possession ratio adherence score over the six‐month study.
Margolis 2013	Morisky Medication Adherence Scale	Reported adherence measured by the Morisky scale modified for blood pressure medications. Adherence to antihypertensive medications at six months increased in the pharmacist intervention telemonitoring group but decreased in the usual care group. There was probably no difference between groups at 12 and 18 months.
Rudd 2004	Electronic drug event monitor	The drug event monitor provided the average number of days on which patients took the correct number of doses prescribed. While adherence was high in both groups, the nurse‐managed patient group had higher adherence than usual care.
Vivian 2002	Patient self‐reporting and drug refill information from the pharmacy	Non‐adherence was judged as missing more than three doses a week or pharmacy records indicated a failure to refill drugs within two weeks after the scheduled refill date. Little or no difference in adherence between or within the two groups at baseline or the end of the study was found. Over 90% of patients in both groups indicated they took their drugs as directed. The study was underpowered to detect a significant difference in adherence.

A meta‐analysis was undertaken for four studies (Cohen 2011; Finley 2003; Magid 2013; Rudd 2004), with adherence data captured as continuous variables with an outcome probably favouring the intervention group, standardised mean difference (SMD) 0.15 (95% CI 0.00 to 0.30; 700 participants, overall effect P = 0.05) and moderate heterogeneity I² = 38% (Analysis 1.5). Four studies (935 participants) with dichotomous adherence data (Hunt 2008; Logan 1979; Margolis 2013; Vivian 2002), showed little adherence difference (risk difference (RD) 0.06, 95% CI ‐0.00 to 0.12; P = 0.05) and moderate heterogeneity I² = 67% (Analysis 1.6). There was a moderate‐certainty of evidence (summary of findings Table for the main comparison; Table 1).

6. Proportion of patients and items appropriately prescribed or deprescribed

In the aged care setting the pharmacist prescribed 2.2 fewer drugs per patient than medical colleagues, comparing before‐and‐after study periods (Thompson 1984). Tsuyuki 2015 reported community pharmacist prescribers discontinued 76 antihypertensive drugs in 181 intervention group patients compared to 15 antihypertensive drugs being discontinued in 67 usual care group patients.

7. Patient satisfaction, where measured by a validated tool as part of an effectiveness study

Patient satisfaction was reported in 14 studies (7514 participants) (Table 4). Validated tools assessing the overall satisfaction with care were included in six studies, namely, diabetes care (Houweling 2009; Houweling 2011), hypertension care (Hunt 2008), clinical pharmacist care (Hirsch 2014), and general care (Litaker 2003; Margolis 2013). The majority of satisfaction surveys were not referenced or were locally developed. Some aspects important in the prescribing process were covered in overall satisfaction assessments, e.g. the quantity and quality of contact (Finley 2003; Houweling 2011; Margolis 2013). The locally developed satisfaction survey by Bruhn 2013 focused on the prescribing intervention. Patients were generally positive about the pharmacist prescribing service, 85% (39/46) were totally satisfied, while 9% (4/44) would have preferred to see their general practitioner (GP). Overall, there was a moderate‐certainty of evidence (summary of findings Table for the main comparison; Table 1). Studies looking at medical provider satisfaction with non‐medical prescribers were limited in number and scope (Barr Taylor 2003; Bruhn 2013), but generally positive.

Table 4. Secondary outcomes ‐ patient and provider satisfaction

Study	Satisfaction tool measure	Outcome
Barr Taylor 2003	Not specified	19/57 respondents stated that the nurse care management programme was moderately helpful. 32/57 found it extremely helpful. 9/13 physicians with two or more patients recommended adoption of the nurse management programme. In other health care settings: 9 physicians felt the programme decreased their time with patients, while 4 thought it increased the time spent.
Bruhn 2013	11 patient satisfaction statements derived from a local prescribing feasibility study	For the prescribing intervention, patients were generally positive about the pharmacist prescribing service ‐ 85% (39/46) were totally satisfied, while 9% (4/44) would have preferred to see their GP. In semi‐structured interviews with GPs and pharmacists, all pharmacists and most GPs were positive about the intervention. Pharmacists found their role satisfying, interesting, and challenging. 17 of 23 GPs were positive about the pharmacists’ role. The cost‐effectiveness of the pharmacists' role, given limited resources, was one issue raised in the GP focus group.
Finley 2003	Not specified	Patients reported greater treatment satisfaction with the collaborative care model than the control group in 6 of 11 measures including the overall treatment for depression, personal nature of the care, listening to concerns, explanations about why antidepressants were prescribed and how to take them, availability for advice, and overall satisfaction with the organisation. 18/37 primary care provider questionnaire respondents were satisfied with workflow, patient welfare. and the pharmacists' abilities.
Houweling 2009	Patient Evaluation of the Quality of Diabetes Care (PEQD)	Patients' evaluations of their satisfaction with diabetes care from the specialist diabetes nurse were significantly more positive than the control group.
Houweling 2011	Patient Evaluation of the Quality of Diabetes Care (PEQD)	The total satisfaction sum score for 14 PEQD measures for practice nurses was 66.4%, compared to 51.7% in the GP group which may be confounded by the amount of time given to each patient. On average GPs spent a total of 28 minutes per patient, whereas practice nurses spent 128 minutes per patient.
Hunt 2008	Satisfation in the SF‐36 healthcare domain	Satisfaction with hypertension care was high in both groups, but with little or no difference in any of the 11 satisfaction measures. Satisfaction was not associated with blood pressure goal attainment.
Hirsch 2014	22‐item Pharmacist Service Questionnaire. 0‐100 scale	Patient satisfaction with the clinical pharmacist were high, with mean scores 92.4 (±10.9) at 6 months (n = 49) and 92.7 (±11) at 9 months (n = 44).
Litaker 2003	Patient Satisfaction Questionaire	Improvements in four areas of satisfaction in the intervention group linked to an increased time spent with patients and an emphasis on patient‐centred education and self‐management (i.e. quality and quantity of contact) from base line to study end. Between‐group comparisons at study end demonstrated little or no significant difference in patient satisfaction measures, including overall care and general satisfaction.
Logan 1979	Not specified	6% of patients were dissatisfied with care provided by nurses but details of the survey instrument were not provided: (assumed 12/206 intervention patients at 6 months but not specified).
McAlister 2014	Not specified	Little or no difference in overall health care satisfaction between pharmacist‐ and nurse‐led care.
Magid 2013	Not specified	Patients at 6 months reporting they were very or completely satisfied with their hypertension care was probably higher in the intervention group than the usual care group.
Margolis 2013	Six items from the Consumer Assessment of Healthcare Providers and Systems adult survey (version 4)	Satisfaction items concerning clinicians listening carefully, explaining things clearly, and respecting what patients said showed larger improvements amongst patients in the telemonitoring intervention group than usual care at 6 months but not at 12 or 18 months.
Spitzer 1974	Not specified	96% of patients in the nurse practitioner group and 97% of patients in the conventional care group were satisfied with the health services received in the experimental period.
Vivian 2002	Not specified	Little or no significant differences in patient satisfaction between groups. More patients in the intervention group felt that the pharmacist spent more time with them than did control patients, although there was little difference. There was no difference in satisfaction with pharmacy services or changes in patient satisfaction in either group from baseline to study end. This study was underpowered to detect a significant difference in patient satisfaction.

GP: general practitioner

8. Non‐medical prescriber versus medical prescriber waiting time to care

Outcome not reported.

9. Non‐medical prescribers adversely affecting the health outcomes of patients through medication errors, prescribing errors, adverse effects, wrong diagnoses or treatment, increased hospitalisations, or representations for medical care

Adverse events were reported in 18 of the 46 studies (18,400 participants) (Table 5). There was probably little or no difference in adverse events between the intervention and usual care groups in nine studies (Ansari 2003; Aubert 1998; Fairall 2008; Ishani 2011; Klingberg‐Allvin 2015; Kuethe 2011; Spitzer 1974; Taveira 2011; Tobe 2006), with a probable increase in adverse events in the usual care group in two studies (New 2003; Thompson 1984). We are uncertain whether the intervention has an effect on adverse events in the remaining studies due to limited data reporting. The relationship between increased medication use in intervention groups and adverse events remains uncertain. Overall, there was a low‐certainty of evidence between the intervention and adverse events (summary of findings Table for the main comparison; Table 1).

Table 5. Primary outcome ‐ adverse events

Study	Adverse event
Ansari 2003	There was little or no difference in the proportions of patients between control (provider education), nurse facilitator and provider/patient notification for hospitalisations and emergency room visits. There were few deaths with the higher number (7) in the control group which had more patients on haemodialysis, two of whom died.
Aubert 1998	There appeared little or no difference between intervention and usual care groups for severe low blood glucose events at baseline and during the study period. Mean weight gain differences from insulin treatment in each group or mean weight loss differences with oral agents showed little or no difference.
Chenella 1983	Reported no patients had major bleeding, but four patients in the pharmacist prescriber group had minor bleeding (one laceration before hospital). One patient in physician prescriber group died, after receiving heparin and warfarin for a stroke in evolution but there was no evidence of bleeding.
DeBusk 1994	The first year mortality was 3.4% in usual care and 4.1% in the intervention group. However, a longer study is required to show a difference, namely, 2 years plus a 5‐ to 10‐year follow‐up.
Fairall 2008	The time to death did not differ between primary care nurses and doctors initiating therapy.
Hirsch 2014	Pharmacists identified two adverse drug reactions from 33 drug therapy problems at baseline, two from 12 at six months and none at nine months.
Ishani 2011	Adverse events were similar between groups, with no participants withdrawing from the study due to an adverse event, and there was no difference in the rate of hospitalisation or death between the groups.
Jaber 1996	Reported 17 hypoglycaemic reactions in the intervention group and two in the control group. All were considered mild to moderate. The difference was possibly related to increased training in recognition, documentation, and questioning in the intervention group. Three patients were hospitalised, two in the control and one in the intervention group, and these appear unrelated to treatment.
Klingberg‐Allvin 2015	In treating incomplete abortion bleeding, the same or less than normal menstrual cycle was probably not different between the intervention midwife and usual care physician groups. There was little difference in pain after treatment as assessed by a visual analogue scale. 30 (6%) of women reported unscheduled visits in the midwife group and 18 (4%) in the physician group. Reasons included vaginal bleeding and abdominal pain. Reported side‐effects after treatment were similar in both groups (nausea, vomiting, abdominal pain, chills, and fever).
Kuethe 2011	There were no differences between groups (general practitioner, paediatrician, asthma nurse) with respect to the number of severe asthma exacerbations as expressed by the number of prednisolone courses.
MacMahon Tone 2009	Forty drug‐related adverse events occurred in the intensive intervention group as compared to 10 in the standard group. While the adverse events are known for the drugs in question no further comment was offered.
McAlister 2014	Reported few clinical events at six months in a pharmacist‐led intervention for secondary prevention after ischaemic stroke. There were nine cardiovascular events and no deaths in the pharmacist group versus eight cardiovascular events and one death in the nurse‐led group.
Margolis 2013	There were 60 adverse events in usual care and 49 in the telemonitoring group; most events were non‐cardiac hospitalisations. There were two allergic reactions to blood pressure medication in the usual care group, six events in the telemonitoring group related to hypotension, dizziness, loss of consciousness which compared to one in the usual care group, four events in usual care related to hypertension versus one in the intervention group.
New 2003	In patients randomised to specialist nurse‐led clinics for blood pressure control, lipid control or both, there were less deaths in the intervention group (25, (3.2%) versus 36 (5.7%) in the usual care group) odds ratio 0.55 (95% confidence interval 0.32 to 0.92) P = 0.02.
Spitzer 1974	During the 12‐month experimental period, there were four deaths in the nurse practitioner group and 18 in the conventional care group. There was probably little or no difference in the crude death rate between groups.
Taveira 2011	There were no diabetes‐related admissions or deaths for either group during the six‐month study.
Thompson 1984	The pharmacist prescribing group in a geriatric setting may have had a slightly lower 12‐month mortality than usual care (3/67 versus 10/72, P = 0.05).
Tobe 2006	The incidence of adverse events probably did not differ between the intervention (home care nurse group) and control (primary care physician group) in First Nations people with diabetes and hypertension. Ten patients in the intervention group and seven in the control group required admission to hospital for adverse events.

10. Other surrogate outcome markers

Studies of surrogate outcome markers not included in the meta‐analyses reported either probable improvements favouring the intervention over usual care (Choe 2005; Ellis 2000; Fischer 2012; Logan 1979); little difference in outcome (Houweling 2011; Moher 2001); or uncertainty of outcome, with surrogate markers showing a combination of probable improvements or little difference in outcomes (Heisler 2012; Ishani 2011; Taveira 2010; Table 2).

Secondary outcomes

Patient‐reported outcomes

1. Health‐related quality of life

Quality of life measures reflected general non‐medical prescriber care compared to usual care. We combined physical and mental component scores for the Short Form‐12 (SF‐12) and Short Form‐36 (SF‐36) in a meta‐analysis. Eight studies (2385 participants) were included in the physical component meta‐analysis (Bruhn 2013; Cohen 2011; Houweling 2011; Hunt 2008; Khunti 2007; Litaker 2003; Margolis 2013; Vivian 2002); six studies (2246 participants) contributed to the mental component meta‐analysis (Cohen 2011; Houweling 2011; Hunt 2008; Khunti 2007; Litaker 2003; Margolis 2013). The physical subgroups showed a small effect (MD 1.17, 95% CI 0.16 to 2.17, P = 0.02) favouring intervention, with low heterogeneity, I² = 17% (Analysis 1.7). The mental component subgroup did not show an effect difference (P = 0.25) with a MD of 0.58 (95% CI ‐0.40 to 1.55) with moderate heterogeneity, I² = 66% (Analysis 1.7). There was no significant difference between the subgroups (P = 0.41) where heterogeneity might not be a factor, I² = 0%.

Across studies, various quality of life measures generally demonstrated little difference between intervention and control groups (Table 6). There was a moderate‐certainty of evidence (summary of findings Table for the main comparison; Table 1).

Table 6. Secondary outcome ‐ quality of life

Study	Measures	Outcome
Aubert 1998	Four generic quality of life measures from the Behavioural Risk Factor Surveillance System	Intervention and control groups reported improved perception of health status after 12 months, but intervention patients were twice as likely to report this.
Barr Taylor 2003	SF‐36, the Duke Activity Status Index for QoL, and the BDI for depression	Little or no differences for any of the variables, but an improved mood for both groups was found.
Bruhn 2013	SF‐12, HUI, CPG, and HADS‐D	No one measure was seen as the primary outcome. In the prescribing arm there was a within‐arm improvement for CPG intensity and disability effect size subscales and between arms on the intensity subscale but not the disability subscale. There was a within‐arm improvement in overall CPG in the prescribing and review arms but not the TAU arm. The SF‐12 and HADS‐D showed deterioration in the TAU arm. Compared with baseline, patients had an improved CPG in the prescribing and review arms but not the TAU arm. The SF‐12 physical score difference showed no effect in prescribing or review arms but improvement in the TAU arm. SF‐12 mental score showed no effect in prescribing or review arms and deterioration in the TAU arm. HADS‐D scores within the prescribing arm showed improvement for depression and anxiety which were also significant between groups.
Cohen 2011	SF‐36 for Veterans	Little or no change in quality of life scores over 6 months.
Finley 2003	The Brief Inventory for depressive symptoms and Work and Social Disability Scale	Liitle or no difference at 6 months between intervention and control groups.
Houweling 2009	SF‐36 and the revised version of the Type 2 Diabetes Symptom Checklist to measure the presence and perceived burden of diabetes‐related symptoms	Little or no differences over 12 months between groups in either survey.
Houweling 2011	SF‐36 and the revised version of the Type 2 Diabetes Symptom Checklist to measure the presence and perceived burden of diabetes‐related symptoms	In the control group there were little or no differences between baseline and follow‐up SF‐36 measures, however in the practice nurse intervention group there were differences in physical functioning, role physical, vitality, and the physical component score. Little or no differences were seen in the QoL results over time between the two groups except for the physical component score which was lower in the intervention group. After 14 months responses to the revised Type 2 Diabetes Symptom Checklist revealed little or no differences between groups.
Hunt 2008	SF‐36	Little or no difference except in the general health domain with scores higher in the control group.
Jaber 1996	Health Status Questionnaire version 2 derived from the SF‐36	Little or no difference between or within groups.
Khunti 2007	SF‐36, Seattle Angina Questionnaire and LVD‐36 questionnaire	Differences favouring the intervention group were found in the SF‐36 for physical functioning, general health, vitality, social functioning, and mental health. Seattle Angina Questionnaire scores in patients with angina were significantly better for intervention patients compared to controls for exertional capacity and borderline differences were found for angina frequency and QoL. There was little or no difference in any of the SF‐36 health status domains or LVD‐36 scores for patients with a confirmed diagnosis of left ventricular diastolic dysfunction.
Litaker 2003	SF‐12 Diabetes Quality of Life	Little or no difference between groups in either measure at study end.
McAlister 2014	Self‐related health using a Likert scale The EQ‐5D as an index of health	Little or no difference between the pharmacist‐ and nurse‐led groups in participants overall self‐related health.
Margolis 2013	SF‐12	Little or no differences between groups.
Moher 2001	Dartmouth COOP charts EuroQol scores	Little or no or clinically important differences between groups for any dimension.
Spitzer 1974	Not described	Patients in the nurse practitioner and usual care groups had similar values at baseline and study end for physical, emotional, and social function.
Taveira 2011	Change from baseline in depression symptoms by the PHQ‐9	Even though no pharmacologic treatments for depression symptoms were offered as part of the intervention, the mean change in PHQ‐9 scores was probably not different for intervention and standard care participants.
Vivian 2002	SF‐36	Little or no significant differences either between or within the two groups from baseline to study end, although patients in the control group reported more bodily pain .

BDI: Beck Depression Index
CPG: Chronic Pain Grade
EQ‐5D: EuroQol five dimensions questionnaire
HADS‐D: Hospital Anxiety and Depression Scale
HUI: Health Utilities Index
LVD‐36: Left Ventricular Dysfunction
PHQ‐9: Patient Health Questionnaire‐9
QoL: quality of life
SF‐12: Short‐Form‐12
SF‐36: Short‐Form‐36
TAU: treatment as usual

Non‐medical prescriber outcomes

1. Job satisfaction, skills utilisation, education needs, and workload effects

Outcome not reported.

Resource use outcomes

1. Medical time saved by non‐medical prescribers

Outcome not reported.

2. Non‐medical prescriber versus medical prescriber prescription volume and cost, patient out‐of‐pocket expenses, service costs, deprescribing rate, and cost

Medication use, including medication amount, medication type, medication dosing, medication frequency, and medication cost was higher in 14 non‐medical prescribing groups (7092 participants) compared to usual care (Ansari 2003; Cohen 2011; Denver 2003; Heisler 2012; Houweling 2009; Hunt 2008; Logan 1979; MacMahon Tone 2009; Magid 2013; Margolis 2013; Rudd 2004; Taveira 2010; Taveira 2011; Tsuyuki 2015). Little difference in medication use was reported in two studies (Chenella 1983; Vivian 2002) (137 participants), and a variable outcome was reported in six studies (7924 participants) (Einhorn 1978; Hirsch 2014; McAlister 2014; Moher 2001; Pagaiya 2005; Wallymahmed 2011). (Table 7).

Table 7. Secondary outcome ‐ resource use

Medication and related therapy
Study	Outcome
Ansari 2003	β‐blocker use was higher in the nurse facilitator group with two‐thirds of patients either initiated or up‐titrated on β‐blockers versus fewer than one‐third of patients in the other two study arms (control provider education and provider/patient notification).
Chenella 1983	Little or no difference in amount of anticoagulant drugs prescribed by pharmacists compared to a physician.
Cohen 2011	More patients in the pharmacist prescribing arm were prescribed diuretics and sulphonylureas compared to usual care. Overall there was an increase in the number of medications prescribed by pharmacists for hypertension, diabetes, and cholesterol from baseline to six months, but little or no change in the usual care arm.
Denver 2003	In nurse‐led clinic for hypertension management in diabetics at six months there were increased changes in the proportions of patients receiving new prescriptions for calcium channel blockers and thiazide diuretics as intensification therapy. The median number of drugs per patient increased in the intervention group compared to conventional primary care.
Einhorn 1978	In a family medicine clinic in Bogota, nurses were less likely than physicians to provide intrauterine devices, prescribe oral contraceptives, and sterilisation on the patient's first visit. Nurses were more likely than physicians to provide temporary prescriptions and defer intrauterine devices and contraceptive measures if the patient on their first visit was not menstruating or believed to be pregnant.
Heisler 2012	Observational cohort results taken six months following the quarter start date showed intervention patients had more blood pressure medication changes.
Hirsch 2014	Pharmacists identified at least one hypertension drug therapy problem in 33/73 (45.2%) patients at baseline requiring additional therapy in 14/33 (42.4%) and dosage increases in 11/33 (33.3%).
Houweling 2009	The nurse specialist in diabetes prescribed significantly more antihypertensive agents and the internist (doctor control) prescribed more cholesterol‐lowering agents.
Hunt 2008	The mean number of antihypertensive medications per patient and use of generic antihypertensive agents was higher in the intervention group.
Logan 1979	Patients in the nurse‐managed group were more likely to be put on antihypertensive medications, prescribed more than two pills per day, and to be on more than one antihypertensive medication.
MacMahon Tone 2009	There were more intervention intensive group patients on three or more antihypertensive drugs (at the study beginning more patients in the standard care group were on three or more antihypertensive agents). At the end of the study more patients with dyslipidaemia in the intensive group were receiving statin therapy. More patients in the intervention group were on aspirin antiplatelet therapy at the end of the study.
McAlister 2014	The median number of antihypertensive medications taken at six months was probably not different in the pharmacist‐ and nurse‐led groups. There was a difference favouring pharmacists in maximal dosing of angiotensin‐converting enzyme inhibitors or angiotensin receptor blockers at six months, but not the percentage of patients using these drugs.
Magid 2013	In patients completing the six‐month visit, there were more intervention patients that had an antihypertensive medication added to their regimen and a dose increased for existing medication, than usual care patients. There was an increase in the usage of specific antihypertensive drugs.
Margolis 2013	There were increases in the mean number of antihypertensive medication classes at 6, 12, and 18 months in the intervention group compared to baseline and compared to usual care.
Moher 2001	There was minimal change in prescribing antihypertensive drugs in the three groups. All groups increased prescribing of lipid‐lowering drugs but there was little or no difference between groups. There was an increase of 10% more patients’ prescribed antiplatelet treatment in the nurse recall group versus the audit group and 8% more in the nurse recall group versus the general practitioner recall group.
Pagaiya 2005	In examining the effects of training and guidelines on prescribing by nurses, the mean change in antibiotic prescribing for all patients showed little or no difference. The mean change for antibiotic prescribing for respiratory infections in children (0 to 5 years) fell. No change was detected in prescribing antibiotics for diarrhoea. There was a mean fall in diazepam prescribing in the intervention group.
Rudd 2004	In the nurse management patient group at six months there was an increased number and variety of antihypertensive medications and an increased number of medication changes than in the usual care group.
Taveira 2010	The intervention arm group (VA‐MEDIC) had greater dose titrations of antihypertensive medications, insulin, statins, and niacin compared to the usual care arm.
Taveira 2011	Intervention arm participants (VA‐MEDIC‐D) had more dose increases or initiation of any antihypertensive agents and more dose increases or initiation of antihyperglycaemic agents. There was little or no difference in the initiation or dose titration of any antihyperlipidaemic agent or antidepressants.
Thompson 1984	The average number of drugs prescribed per patient was lower in the pharmacist group compared to the physician group. The number of drugs was reduced by an average of 2.2 drugs per patient from the pre‐study to the study year. The practice of clinical pharmacists prescribing drug therapy under physician supervision has the potential to save the healthcare system USD 70,000 per 100 skilled nursing facility beds.
Tsuyuki 2015	In the pharmacist prescribing arm proportionally more new antihypertensive agents were initiated, more dose changes occurred, more antihypertensives were discontinued, and more patients were prescribed low‐dose aspirin and a statin than in the usual care group.
Vivian 2002	There was little or no difference in the type of antihypertensives prescribed to intervention and control patients during the study.
Wallymahmed 2011	Compared with baseline there were more patients in both groups taking antihypertensive medications but this difference was probably only important in the nurse‐led intervention group.

Healthcare visits, health resources, and associated costs
Ansari 2003	There was no difference in hospitalisations and emergency room visits between the three groups of control (provider education), nurse facilitator, and provider/patient notification.
Aubert 1998	Hospital admissions were rare and did not differ between the intervention and usual care groups. ED visits did not differ between groups or from baseline. No hospital or ED visits were related to diabetes. The average number of outpatient visits during the study was similar. The nurse managed a case load of 71 patients, but it was estimated that a 300 patient case load could be managed.
Barr Taylor 2003	There was no change in health utilisation (physician visits, ED visits, days of hospitalisation) for the year before and after the intervention and between groups.
Choe 2005	In reporting process measures for the clinical pharmacist’s case management of patients there was a difference between pharmacist intervention and control in the frequency of low‐density lipoprotein measurements, retinal examinations, and monofilament foot examinations but not glycated haemoglobin measurement or urine albumin screen.
Cohen 2011	Over six months there were a higher number of primary care visits in the usual care arm; an average 1.65 visits per patient versus 1.56 in the intervention arm. It was suggested the difference in the higher number of primary care visits may offset the intervention cost.
DeBusk 1994	The nursing time spent in the year after myocardial infarction was nine hours per patient; a per patient cost of USD 500 which included the nurse salary, office costs, and other associated costs. This compared with cardiac rehabilitation programmes in the San Francisco Bay area costing USD 1800 to USD 2700 to participate for three months.
Ellis 2000	In investigating the impact of clinical pharmacist interventions in patients with dyslipidaemia there was little or no difference in physician or nurse visits between control and the intervention patients at 12 months. At 12 months the intervention group had more pharmacist visits than the control group. There were little or no difference in costs for hospitalisations, clinic visits, laboratory costs, drug costs, and costs of lipid therapy between groups. The intervention group had a USD 370 greater difference per patient in total costs which was probably not important and approximately 5% of total costs.
Fairall 2008	In the cohort of patients not yet receiving antiretroviral therapy there was little or no difference in clinic visits with a nurse but clinic visits with a doctor were probably higher in the intervention group. In the cohort of patients who had already received at least six months of antiretroviral therapy clinic visits with a nurse probably higher in the intervention group. Economic data from the study is the subject of further analysis by Barton 2013 (see Studies awaiting classification).
Finley 2003	Although the collaborative care model experienced a decrease in the total number of primary care visits, the between‐group difference was probably not important. ED visits increased more in the usual care group but this was probably not important and neither was the difference in utilisation of psychiatric services. The institutional cost of drugs, the cost of antidepressants and the cost of psychotropic drugs overall was higher in the intervention group, but this was not important.
Fischer 2012	Hospital admissions (while trending to fewer admissions) in the nurse intervention group showed little or no difference to the control group. Nurse case management was not associated with a significant difference in the number of outpatient or ED visits. There was a decrease in total costs in the nurse telephone intervention group comparing the period before and after randomisation. In contrast, there was an increase for the same comparison in the control group. Similar results were seen with hospitalisation and ED costs which were lower in the intervention group. There was probably not an intervention effect on outpatient costs. The difference in average per patient cost between the intervention group (USD 6600) and control group (USD 9033) of USD 2433 was important. The control group had higher baseline hospitalisation rates and total costs cautioning interpretation of the result.
Heisler 2012	Little or no difference in health services utilisation (hospitalisations, primary care visits, ED visits) between intervention and control patients during the 14‐month study of blood pressure control through a clinical pharmacist outreach programme in diabetic patients.
Hirsch 2014	The pharmacist collaborative group (PharmD‐PCP MTM) had fewer primary care physician visits during the intervention period than did the usual care group. The mean total combined visits of primary care physician and pharmacist was not greater in the PharmD‐PCP MTM group than in usual care.
Houweling 2009	There was a lower number of visits in the NSD group compared with standard care but not in the duration of visits. Significantly more patients were referred back to their GP by the NSD when meeting treatment goals. Personnel and laboratory costs were lower in the intervention group than the control group. The average per month increase in medication costs between the groups was probably not important apart from the cholesterol‐lowering medications. The average time saving per internist was 61.4 minutes (meaning the internist could supervise 11 patients with the NSD in the time he/she could treat one patient).
Houweling 2011	The mean number of visits and duration of visits was higher in the practice nurse intervention group than the control group.
Hunt 2008	The total number of clinic visits (physician plus pharmacist) was higher in the intervention arm compared to the control arm. The number of physician visits was lower in the intervention arm.
Ishani 2011	Little or no difference in the hospitalisation rate between intervention and control groups.
Kuethe 2011	In testing the non‐inferiority of asthma care in children with stable asthma provided by a hospital‐based specialised asthma nurse versus a GP or paediatrician, there was little or no differences between the groups for medication, school absence or parental work absence after two years. There was little or no difference in unplanned visits and no hospital admissions during the study.
Litaker 2003	Medium number of outpatient visits were higher for the team based intervention patients. Average personnel costs for one year's treatment were significantly higher in the intervention group (USD 134.68 vs USD 93.70, P < 0.001).
Magid 2013	There was little or no difference in the mean number of outpatient clinic visits, total number of ED visits, and hospitalisations between the two groups. The intervention group probably had a higher number of email and telephone encounters.
Margolis 2013	Over 12 months in the telemonitoring intervention group all 228 patients used a mean of 11.4 ± 3.9 pharmacist visits lasting a mean of 34.2 minutes and 217 used telemonitoring services with a mean of 9.8 ± 2.5 months of use. It was estimated direct programme costs would total USD 1350 per patient.
Spitzer 1974	A reported five per cent drop in gross practice revenue was explained by the absence of billing for services provided by the nurse practitioner. Billing for unsupervised practice was not permitted in Ontario at the time of the study. During the trial year the services rendered by the nurse practitioner were worth approximately USD 16,000 of which almost 50% was for unsupervised practice.
Taveira 2011	There was little or no differences in primary carer visits, use of ED services for all cause visits, diabetes‐related ED visits or hospital admission rates.
Thompson 1984	There was little or no difference in the average length of stay or hospitalisations although the latter trended lower in the pharmacist group. Differences favouring the pharmacist group were found in the rate of discharge to home or to a lower level of care.
Vivian 2002	Little or no differences between intervention and control groups in appointments with the primary care provider during the 6 months of the study.

ED: emergency department
GP: general practitioner
NSD: nurse specialised in diabetes

Costs relating to prescription volume, patient out‐of‐pocket expenses, and deprescribing rate were not reported.

3. Increased resource use for providing the intervention and for providing subsequent care such as hospitalisations, emergency department visits, and outpatient visits

Twenty‐five studies (22,590 participants) reported resource use, including hospital admissions, emergency department visits, outpatient visits, primary care visits, physician visits, pharmacists’ visits, examinations, and staff and laboratory costs (Table 7). Due to the heterogeneity of resource use across studies and the measures used to record resource use, meta‐analysis was confined to a limited number of studies of emergency department visits (RD 0.01, 95% CI ‐0.02 to 0.03) and hospitalisation (RD ‐0.01, 95% CI ‐0.03 to 0.01) comparing the non‐medical prescribing group to usual care. There was no statistical difference between study groups for these parameters (P = 0.52 and P = 0.51, respectively) in the meta‐analysis (Analysis 1.8). There appeared to be little difference in hospitalisations, emergency department visits, and outpatient visits between intervention versus control groups across the studies.

Non‐medical prescribing in other settings

Two studies were undertaken in other settings. Logan 1979 described a study of blood pressure control by nurses in the workplace compared to usual medical care. Patients in the nurse group were more likely to be put on antihypertensive medications (94.7% versus 62.7%, P < 0.001), to reach blood pressure goals in the first six months (48.5 versus 27.5%, P < 0.001) and to take drugs prescribed (67.6 versus 49.1%, P < 0.005). Thompson 1984 reported on pharmacist prescribing in a geriatric setting. The clinical pharmacist group probably had a lower number of deaths (P = 0.05), a higher number of patients being discharged to lower levels of care (P = 0.03) and a lower average number of drugs per patient (P = 0.04) (Table 2).

Four studies were undertaken in low‐ and middle‐income country settings. Einhorn 1978 evaluated nurse management versus usual doctor care of family planning and prescribing oral contraceptives. While differences in patient management occurred, the outcomes of continuing oral contraceptive use and preventing pregnancy were probably not different. As outlined, Fairall 2008 evaluated task shifting of antiretroviral therapy from doctors to primary care nurses. The intervention improved survival slightly in patients not yet taking antiretrovirals with CD4 counts of 201 to 350 cells per µL but resulted in little difference in patients with higher cell counts. There was little or no difference in viral load suppression between patient groups for patients already taking antiretrovirals at enrolment. Klingberg‐Allvin 2015 compared treatment of incomplete abortion with misoprostol by physicians and midwives at district level in Uganda and found the diagnosis and treatment of incomplete abortion by midwives equally safe and effective as when provided by physicians. In the study by Pagaiya 2005, educational intervention with guidelines for nurses probably improved antibiotic prescribing for acute respiratory tract infections and the prescribing of diazepam. There was probably no difference in the prescribing of antibiotics for diarrhoea, and it is uncertain whether diabetes care improved because the certainty of evidence is low.

Discussion

Summary of main results

The overall findings suggest that non‐medical prescribing practised with varying but high degrees of autonomy and with collaborative support, can deliver comparable outcomes to usual medical care prescribing. However, these results must be interpreted with a degree of caution, recognising the variation in non‐medical prescribing practice reported within studies and the complex interplay of factors affecting outcomes. There are a limited number of well‐designed randomised controlled trials (RCTs) evaluating the specific prescribing outcomes of non‐medical prescribers.

Meta‐analyses examining surrogate markers of disease with the fixed‐effect method demonstrated interventions with a non‐medical prescribing component decreased systolic blood pressure at six months by ‐6.76 mmHg, and at 12 months by ‐5.31 mmHg. The fixed‐effect estimates gave a more conservative estimate of effect than the random‐effects estimate for systolic blood pressure ( ‐7.34 mmHg and ‐5.91 mmHg, respectively). There was little difference between fixed‐ and random‐effects outcomes for glycated haemoglobin at six months (‐0.42% versus ‐0.45%, respectively) and at 12 months (‐0.62% versus ‐0.62%, respectively). Reductions in low‐density lipoprotein demonstrated variable results using fixed‐ and random‐effects at six months (‐0.25 mmol/L versus ‐0.13 mmol/L, respectively), and 12 months (‐0.21 mmol/L versus ‐0.30 mmol/L, respectively). However, all studies apart from those assessing glycated haemoglobin at 12 months demonstrated moderate to considerable heterogeneity. Removal of the two cluster‐RCTs for systolic blood pressure reduced the fixed‐effect difference by 0.63 mmHg at six months and 0.47 mmHg at 12 months.

Clinical findings of interventions with non‐medical prescribing components outside the meta‐analyses showed equivalence or benefit compared to usual care (Table 2). Medication adherence was measured in less than a quarter (22%) of studies. Where adherence was measured, there was either no difference between study groups or a small improvement in intervention groups. More regular contact by the non‐medical prescriber with intervention patients compared to usual care may be a confounding factor. For example, in the telemonitoring of blood pressure study by Margolis 2013, which demonstrated improved medication adherence at six months, there was regular telephone support from the intervention pharmacist every two weeks until blood pressure control was sustained for six weeks. Contact then reduced to monthly contact for six months which may account for little or no difference in medication adherence at 12 and 18 months. A meta‐analysis of four studies with continuous adherence data favoured the non‐medical prescriber group with minimal heterogeneity.

In studies reporting adverse events, there was either little difference between intervention and usual care groups or insufficient information to determine if differences occurred. In two studies, more deaths were reported in the usual care group versus the intervention group (New 2003; Thompson 1984).

The meta‐analysis of the quality of life measures (SF‐12 and SF‐36 scores at 12 months) probably showed a small improvement favouring the intervention. A variety of other quality of life measures (used in the remaining studies and not included in the meta‐analysis) generally demonstrated little difference between the intervention and usual care groups. In assessing quality of life effects, consideration must be given to the effect of the multifaceted nature of many interventions beyond the non‐medical prescribing component.

Patient satisfaction data were reported in 14/46 (30%) of studies and focused on the care patients received from the non‐medical health professional as a whole, with little specific comparative evidence of satisfaction with the prescribing element of care. Bruhn 2013 obtained a high patient satisfaction rating of 85% (39/46) with the pharmacist service involving prescribing and education in the management of chronic pain. Two studies included results of small samples of medical provider satisfaction with non‐medical providers, which were generally positive, but they raised respective concerns about time commitments to intervention patients and the cost‐effectiveness of non‐medical prescribers (Barr Taylor 2003; Bruhn 2013).

A wide variety of measures of resource use were reported in 37/46 (80%) of studies. In the majority of studies reporting medication use, non‐medical prescribers initiated and prescribed more drugs, titrated drugs to a higher dose, and used a greater variety of drugs than usual care medical prescribers in treating chronic disease. In the aged care setting, the pharmacist prescribed fewer drugs than medical colleagues (Thompson 1984). There was little difference in hospitalisations, emergency department visits, and outpatient visits between intervention versus usual care groups across the studies.

Non‐medical prescribers had varying levels of prescriber training, determined by country or setting, and no studies were found comparing different levels of non‐medical prescriber training and outcomes.

Overall completeness and applicability of evidence

The majority of studies were from high‐income countries with the greater proportion, 25 of 46 studies emanating from the USA. While the results of this review are more applicable in Western countries, the four studies involving non‐medical prescribing nurses in low‐ and middle‐income countries demonstrated safe and effective outcomes compared to usual care, and provide an opportunity for further study in the application of non‐medical prescribing. It is unclear why more studies meeting inclusion criteria did not originate from the UK where legislative change and formal training requirements have allowed independent prescribing by nurses and pharmacists since 2006. Chronic disease management was the focus of most studies with only two studies undertaken in the acute inpatient secondary care setting (Chenella 1983; Marotti 2011).

No studies reported comparisons between non‐medical prescribers in both arms of the study.

In only 19 studies could a more defined non‐medical prescribing role with less confounding elements provide a clearer effect on outcomes. Pharmacists were judged to have more autonomy in their prescribing roles than nurses, who relied more heavily on algorithms to adjust medications. The degree of prescribing autonomy within study designs was guided by local legislative controls and healthcare organisation policies and practices.

Formal training as a requirement to prescribe was limited. Independent pharmacist prescribers in the Bruhn 2013 UK study were required to complete a course of approved study and have registration with the General Pharmaceutical Council as independent prescribers. In Alberta Canada, pharmacists in the Tsuyuki studies were required to undergo an assessment process when applying for the authorisation to prescribe (Tsuyuki 2015; Tsuyuki 2016). In other studies, prescribing permissions were granted through collaborative practice agreements for pharmacists in the USA, and varying degrees of specific on‐the‐job training for the disease or condition of focus. Prescribers frequently had advanced practice qualifications, for example, in diabetes management, and a number of years of experience in ambulatory chronic disease care. Prescribing of oral contraceptives was within the remit of family planning nurses in Bogota, Colombia (Einhorn 1978). Local training was provided to nurses in South Africa covering antiretroviral drug prescribing, drug effects and side‐effects, and the use of algorithmic clinical practice guidelines (Fairall 2008). Midwives in Uganda underwent a five‐day training programme covering incomplete abortion and treatment with misoprostol (Klingberg‐Allvin 2015). Nurses in health centres in Thailand prescribed antibiotics for children and diazepam for adults without additional education and guideline support, which was the focus of the study (Pagaiya 2005).

The heterogeneity of educational requirements for non‐medical prescribers across studies did not allow a pooled assessment of outcomes, but within individual studies the education level did not appear to influence the outcome.

Local trial protocols, which included additional collaborative medical support for the non‐medical prescriber, were aimed at ensuring safe practice.

Most excluded studies were before‐and‐after studies and there remains a need for further large, well‐controlled trials, where the prescribing component can be clearly associated with an outcome, and the degree of prescribing autonomy is clearly defined.

Mikuls 2015 is an ongoing study (see Characteristics of ongoing studies). We are waiting for further information from one study that is reported as an abstract (Tsuyuki 2014). We have placed this study in Characteristics of studies awaiting classification and we will incorporate this study in a future review update. Two further studies, assessing economic impacts, are awaiting assessment (Barton 2013; Neilson 2015). We made the pragmatic decision that these two studies will be incorporated in the update of this review, so as to avoid delaying the publication of the current version of this review.

Quality of the evidence

We evaluated the certainty of the body of evidence for seven outcomes according to the GRADE system.

We graded the certainty of evidence for systolic blood pressure at 12 months as moderate due to considerations of serious inconsistency (finding considerable heterogeneity), the multifaceted nature of interventions, and variable prescribing autonomy. We found high levels of certainty of evidence for the outcome of glycated haemoglobin at 12 months. There were low levels of certainty of evidence for low‐density lipoprotein due to serious inconsistency (finding considerable heterogeneity), multifaceted interventions, and variable prescribing autonomy. We graded medication adherence at moderate‐certainty of evidence due to serious risk of bias (high risk of performance bias) and variable adherence reporting measures. We graded the certainty of evidence around adverse event reporting as low due to indirectness, as the range of adverse events may not be related to the intervention, and selective outcome reporting with adverse events not being reported in many studies. We graded the certainty of evidence for patient satisfaction as moderate due to indirectness in measuring the prescribing component of care, the variability of measures used, and the consideration that some measures were not validated. We graded the health‐related quality of life measures as moderate, considering that within the quality of life outcomes it is difficult to distinguish the contribution non‐medical prescribing made to the outcome versus the other components of care.

The certainty of the body of evidence provides support that there is probably no difference in outcomes between non‐medical and medical prescribers. Specific outcomes may be improved by non‐medical prescribers working within collaborative care arrangements in a range of settings.

Potential biases in the review process

Differing terminologies for non‐medical prescribing across countries may have limited the number of studies found. In addition, we made judgements on the degree of prescribing autonomy for non‐medical prescribers in included studies.

Agreements and disagreements with other studies or reviews

The findings of this review are generally consistent with the findings of other reviews. Meta‐analyses of studies involving pharmacist and nurse‐led care may include studies involving medication management, medication reconciliation, medication education, treatment monitoring, treatment support, and lifestyle advice. Medication management is a broad term that may or may not include a prescribing component. Subgroup analysis of studies involving either independent prescribing, prescribing or dosage adjustment by protocol or algorithm have demonstrated benefit over usual care. Findings of improvements in clinical markers and heterogeneity accord with our findings. In a meta‐analysis, Santschi 2014 reported pharmacist interventions improved blood pressure compared to usual care, but due to the large heterogeneity between studies the effect size varied widely, and it was difficult to determine the most effective intervention. A range of pharmacist interventions were found to reduce systolic blood pressure, but possibly not diastolic blood pressure (Machado 2007). A limitation for both studies was the quality of the studies included in the analyses. In a systematic review of the effects of nurse prescribing, Gielen 2014 reviewed 35 studies including 10 RCTs and one controlled clinical trial. All but five studies had a high risk of bias, but tentative conclusions were that nurses prescribed in a similar way to doctors with few differences in health outcomes, quality of care, and patient satisfaction. Clark 2010 found nurse‐led interventions required an algorithm to improve blood pressure control compared to usual care, and there was some evidence of improved outcomes by nurse prescribers outside the UK. In reviewing 72 RCTs of interventions to control blood pressure in patients with hypertension, Glynn 2010 included 12 studies of nurse‐led or pharmacist‐led care to improve blood pressure control. While the results were significantly heterogeneous, the effects were favourable and warranted further investigation in larger trials. The Hypertension Detection and Follow‐Up study was cited for providing evidence of the importance of a multifaceted intervention in blood pressure control, which consisted of an organised system of regular review and vigorous antihypertensive drug therapy (Hypertension 1979).

In chronic disease management, nurses successfully titrated medications by protocol for diabetes, hypertension, and hyperlipidaemia within a team approach.There were limited descriptions of the interventions and protocols used for studies in the meta‐analysis (Shaw 2014). Greer 2016 found pharmacist‐led chronic disease management was similar to usual care for resource use, and may improve goals for glycaemia, blood pressure, and cholesterol, but there is uncertainty whether clinical outcomes are improved.

In a review of the effects of pharmacist‐provided non‐dispensing services on patient outcomes, health service utilisation, and costs in low‐ and middle‐income countries, Pande 2013 reported on the outcomes of pharmacist interventions that involved counselling, education, and advice. There were small improvements in clinical outcomes (blood pressure, blood glucose, lipids, peak expiratory flow) and quality of life scores, however, the certainty of the evidence was graded as low. Health service utilisation and medication costs were reduced, but again the certainty of the evidence was graded as low. In a review of the effect of outpatient pharmacists' non‐dispensing roles on patient outcomes and prescribing patterns, Nkansah 2010 found that most of the 43 studies included in their review supported the role of pharmacists in medication/therapeutic management as one of a number of interventions to improve clinical outcomes.

It is often difficult to distinguish the specific outcomes of non‐medical prescribing in reported studies and reviews, and the degree of influence on prescribing by physicians where team care arrangements exist. Driscoll 2015, in a review of nurse‐led titration of drug therapy for people with heart failure, found that participants in the nurse‐led group were less likely to be admitted to hospital or to die. More participants reached the maximum drug dose in the nurse‐led group compared to titration of doses by primary care physicians. However we assessed a high level of autonomy in prescribing in only one of the seven reported studies (Ansari 2003).

In a review of substitution of doctors by nurses in primary care, Laurant 2005 found that the quality of care and health outcomes are similar for nurses and doctors, but it is not known if nurse substitution decreases doctors' workload. Nurses tended to provide more health advice and achieve higher levels of patient satisfaction compared to doctors. Nurses' higher use of resources, for example, ordering more tests, may offset savings in lower salary costs.

Figure 1

Study flow diagram.

Figure 2

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.

Figure 3

Forest plot of comparison: 1 Non‐medical prescribing group versus usual care, Outcome: 1.2 Systolic blood pressure mmHg.

Figure 4

Forest plot of comparison: 1 Non‐medical prescribing group versus usual care, Outcome: 1.1 HbA1c (%).

Figure 5

Forest plot of comparison: 1 Non‐medical prescribing group versus usual care, Outcome: 1.3 Low‐density lipoprotein (LDL) mmol/L.

Analysis 1.1

Comparison 1 Non‐medical prescribing group versus usual care, Outcome 1 Systolic blood pressure mmHg.

Analysis 1.2

Comparison 1 Non‐medical prescribing group versus usual care, Outcome 2 HbA1c (%).

Analysis 1.3

Comparison 1 Non‐medical prescribing group versus usual care, Outcome 3 Low‐density lipoprotein (LDL) mmol/L.

Analysis 1.4

Comparison 1 Non‐medical prescribing group versus usual care, Outcome 4 Low‐density lipoprotein pharmacist vs nurse 6 mths.

Analysis 1.5

Comparison 1 Non‐medical prescribing group versus usual care, Outcome 5 Adherence (continuous).

Analysis 1.6

Comparison 1 Non‐medical prescribing group versus usual care, Outcome 6 Adherence (dichotomous).

Analysis 1.7

Comparison 1 Non‐medical prescribing group versus usual care, Outcome 7 Health‐related quality of life.

Analysis 1.8

Comparison 1 Non‐medical prescribing group versus usual care, Outcome 8 Health facility resource use.

Summary of findings for the main comparison. Non‐medical prescribing compared to medical prescribing for acute and chronic disease management in primary and secondary care

Non‐medical prescribing compared to medical prescribing for acute and chronic disease management in primary and secondary care
Patient or population: patients with acute and chronic disease Settings: secondary care and ambulatory/primary care in low‐, middle‐ and high‐income counties Intervention: non‐medical prescribing Comparison: medical prescribing
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Certainty of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Medical prescribing	Non‐medical prescribing
Systolic blood pressure (mmHg) at 12 months	The mean systolic blood pressure in the control group ranged from 124 mmHg to 149 mmHg	The mean systolic blood pressure in the intervention group was 5.31 mmHg lower (‐6.46 lower to ‐4.16 lower)	‐	4229 (12 RCTs)	⊕⊕⊕⊝ Moderate ^1,2,3	Random‐effects analysis: MD ‐5.91 mmHg lower (95% CI ‐7.71 lower to ‐4.10 lower)
Glycated haemoglobin (HbA1c, %) at 12 months	The mean change in glycated haemoglobin in the control group ranged from ‐0.90% to 9.7%	The mean change in glycated haemoglobin in the intervention group was 0.62% lower (‐0.85 lower to ‐0.38 lower)	‐	775 (6 RCTs)	⊕⊕⊕⊕ High^2,3	Random‐effects analysis: MD ‐0.62 (95% CI ‐0.85 to ‐0.38)
Low‐density lipoprotein (mmol/L) at 12 months	The mean low‐density lipoprotein in the control group ranged from ‐0.26 to 3.41 mmol/L	The mean low‐density lipoprotein in the intervention group was 0.21 mmol/L lower (‐0.29 lower to ‐0.14 lower)	‐	1469 (7 RCTs)	⊕⊕⊕⊝ Moderate^1,2,3	Random‐effects analysis: MD ‐0.30 (95% CI ‐0.62 to 0.02)
Adherence (continuous) 6 months follow‐up	The mean adherence (continuous) in the control group was 0.79	The mean adherence in the intervention group was 0.15 higher (0.00 higher to 0.30 higher)	‐	700 (4 RCTs)	⊕⊕⊕⊝ Moderate^4,5
Patient satisfaction	Patient satisfaction was reported in 14 studies (Table 4). The majority of surveys were either not referenced or developed locally. Validated questionnaires assessing overall non‐medical practitioner satisfaction with care were reported in six studies rather than patient satisfaction with prescribing. An exception was the study by Bruhn 2013, which found for the prescribing intervention, patients were generally positive about the pharmacist prescribing service, 85% (39/46) were totally satisfied, while 9% (4/44) would have preferred to see their GP		Not estimable	7514 (14 RCTs)	⊕⊕⊕⊝ Moderate^8,9
Adverse events	There was little or no difference in adverse events between treatment groups in nine studies. Two studies reported higher rates of adverse events in the usual care group. It was difficult to determine effects in the remaining studies because limited data were reported		Not estimable	18,400 (18 RCTs)	⊕⊕⊝⊝ Low^6,7
Health‐related quality of life measured with SF‐12/36	The mean health‐related quality of life in the control group was 0	The mean health‐related quality of life in the intervention group: physical component was 1.17 higher (0.16 to 2.17) mental component was 0.58 higher (‐0.40 to 1.55)	‐	4631 (8 RCTs)	⊕⊕⊕⊝ Moderate¹⁰
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% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; GP: general practitioner; MD: mean difference; RCT: randomised controlled trial.
GRADE Working Group grades of evidence High‐certainty: Further research is very unlikely to change our confidence in the estimate of effect. Moderate‐certainty: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low‐certainty: 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‐certainty: We are very uncertain about the estimate.
¹Downgraded one level due to serious inconsistency (considerable heterogeneity was found). ²Multifaceted interventions. ³Variable prescribing autonomy. ⁴Downgraded one level due to serious risk of bias (high risk of performance bias). ⁵Variable reporting measures of adherence. ⁶Downgraded one level due to indirectness (range of adverse events; may not be related to the intervention). ⁷Downgraded one level due to selective outcome reporting (adverse events not reported in many studies). ⁸Downgraded one level due to indirectness (prescribing component not adequately assessed across studies). ⁹Variability in satisfaction measures. ¹⁰Downgraded one level due to indirectness (prescribing component effect on quality of life difficult to determine).

Summary of findings for the main comparison. Non‐medical prescribing compared to medical prescribing for acute and chronic disease management in primary and secondary care

Table 1. Fixed‐effect outcomes versus random‐effects for clinical surrogate markers

Outcome or subgroup	Fixed‐effect estimate	Random‐effects estimate
1.1 Systolic blood pressure (mmHg)	‐5.85 (‐6.76 to ‐4.94)	‐6.59 (‐8.48 to ‐4.71)
1.1.1 6 months	‐6.76 (‐8.24 to ‐5.27)	‐7.34 (‐11.09 to ‐3.60)
1.1.2 12 months	‐5.31 (‐6.46 to ‐4.16)	‐5.91 (‐7.71 to ‐4.10)
1.2 HbA1c (%)	‐0.55 (‐0.74 to ‐0.36)	‐0.55 (‐0.76 to ‐0.35)
1.2.1 HbA1c (6 months)	‐0.42 (‐0.75 to ‐0.09)	‐0.45 (‐0.09 to ‐0.01)
1.2.2 HbA1c (12 months)	‐0.62 (‐0.85 to ‐0.38)	‐0.62 (‐0.85 to ‐0.38)
1.3 LDL (mmol/L)	‐0.23 (‐0.28 to ‐0.17)	‐0.22 (‐0.42 to ‐0.02)
1.3.1 LDL (6 months)	‐0.25 (‐0.34 to ‐0.17)	‐0.13 (‐0.39 to 0.12)
1.3.2 LDL (12 months)	‐0.21 (‐0.29 to ‐0.14)	‐0.3 (‐0.62 to 0.02)
LDL: low‐density lipoprotein

Table 1. Fixed‐effect outcomes versus random‐effects for clinical surrogate markers

Table 2. Outcomes of studies not included in meta‐analyses

Study	Patient group	Comparison	Outcome
Bruhn 2013	Chronic pain	To compare the effectiveness of pharmacist medication review with or without pharmacist prescribing with standard care	Compared with baseline the Chronic Pain Grade improved in prescribing arm 47.7% (21/44; P = 0.003) and review arm 38.6% (17/44; P = 0.001) but not TAU 31.3% (15/48; ns) SF‐12 mental component score showed no effect for prescribing or review arms and deterioration in TAU arm. Hospital Anxiety and Depression scores improved in prescribing arm for depression (P = 0.022) and anxiety (P = 0.007) and between groups (P = 0.022 and P = 0.045 respectively).
Chenella 1983	Anticoagulation	Pharmacist versus physician independent management of anticoagulant therapy for inpatients	There were no differences between groups for mean heparin and warfarin doses, partial thromboplastin time, days to reach therapeutic levels, mean prescribed and simulated heparin doses.
Choe 2005	Type 2 diabetes	Pharmacist case management versus usual medical care	Patients in the pharmacist case managed group received greater reductions in HbA1c (2.1% vs 0.9%, P = 0.03). Three of five process measures were conducted more frequently in the intervention group than control group. LDL measurement (100% vs 85.7%, P = 0.02), retinal examination (97.3% vs 74.3%, P = 0.004), monofilament foot screening, (92.3% vs 62.9%, P = 0.002).
Einhorn 1978	Family planning	Family planning services provided by nurses versus physicians	Nurses' clients were as equally as successful as physicians in continuing contraceptive use and preventing pregnancy. Nurses were less likely than physicians to provide patients on their first visit with IUDs, prescribe oral contraceptives, or sterilisation. Nurses were more likely to give temporary prescriptions than physicians until the next visit (25% vs 16%, P < 0.001) for reasons including possible pregnancy and patients not menstruating.
Ellis 2000	Dyslipidaemia	Clinical pharmacists providing pharmaceutical care in addition to usual medical care versus usual medical care	The absolute change in total cholesterol (17.7 vs 7.4 mg/dL, P = 0.028) and LDL (23.4 vs 12.8 mg/dL, P = 0.042) was greater in the intervention than control group.
Fairall 2008	HIV	Prescribing of antiretroviral treatment by nurses versus doctors	Cohort 1 ‐ not receiving antiretrovirals. Time to death did not differ (HR 0.94, 95% CI 0.76 to1.15). Cohort 2 ‐ received antiretrovirals for at least six months. Viral load suppression 12 months after enrolment was equivalent in intervention and control. Risk difference 1.1% (95% CI ‐2.4 to 4.6).
Finley 2003	Depression	Collaborative care model of clinical pharmacists providing drug therapy management and treatment follow‐up versus usual care	Clinical improvements noted in both groups but not significant. Intervention group had higher drug adherence at six months (67% vs 48%; OR 2.17, 95% CI 1.04 to 4.51; P = 0.038)
Fischer 2012	Lipid control in diabetes	Algorithm‐driven telephone care by nurses as an adjunct to usual care versus usual care	The percentage of patients with an LDL < 100 mg/dL increased from 52% to 58.5% in the intervention group and decreased from 55.6% to 46.7% in the control group (P < 0.01). The intervention did not affect glycaemic and BP outcomes
Heisler 2012	Blood pressure control in diabetes	A pharmacist‐led intervention (Adherence and Intensification of Medications) in patients with diabetes and poor BP control versus usual care	The mean systolic BP decrease from 6 months before to 6 months after the 14‐month intervention was not different (8.9 mmHg decline in the intervention arm and 9.0 mmHg decline in the control arm). There was no difference in the mean HbA1c and LDL levels between groups after the end of the intervention period (examining 12 months). At the end of the first quarter after activation, there was a significantly greater drop in systolic BP in the intervention group versus control, 9.7 mmHg vs 7.2 mmHg; MD 2.4 mmHg (95% CI 1.5 to 3.4 P < 0.001).
Houweling 2011	Type 2 diabetes	Primary care nurse management of type two diabetes versus management by GPs	After 14 months between‐group differences for reduction in HbA1c, BP, and lipid profile were not significant. Mean systolic and diastolic BPs were lower in both groups. Most process indicators were significantly better in the nurse care group. More patients were satisfied with their care in the nurse group however the physical component of the SF‐26 was better in the GP group.
Ishani 2011	Cardiovascular risk factors in diabetes	Nurse case management versus usual care to improve hypertension, hyperglycaemia, and hyperlipidaemia in veterans with diabetes	A greater number of patients in the nurse case management had all three measures under control (21.9% vs 10.1%, P < 0.01). A greater number of intervention group participants achieved individual treatment goals. HbA1c < 8% (73.7% vs 65.8% P = 0.04), BP < 130/80 mmHg (45% versus 25.4%, P < 0.01) but not for LDL < 100 mg/dL (57.6% vs 55.4%, P = 0.61).
Jaber 1996	Non‐insulin dependent diabetes	Pharmacists providing pharmaceutical care versus physicians	Improvement was seen in glycated haemoglobin in the intervention group at 4 months (9.2% ± 2.1 vs 12.1% ± 3.7, P = 0.003), and fasting plasma glucose (8.5 ± 2.3 vs 11.0 ± 3.9 mmol/L, P = 0.015). There was little or no change within or between groups for BP, lipid profile, renal function, weight, or quality of life measures.
Klingberg‐Allvin 2015	Women with signs of incomplete abortion	Midwives diagnosing and treating incomplete abortion with misoprostol compared to physicians	452 (95.8%) women in the midwife group and 467 (96.7%) in the physician group had complete abortion. The model risk difference for midwife versus physician group was ‐0.8% (95% CI ‐2.9 to 1.4) falling within the predefined equivalence range (‐4% to 4%).
Kuethe 2011	Children with asthma	Non‐inferiority of care provided by a hospital‐based specialised asthma nurse versus a GP or paediatrician	The corrected daily dose of inhaled corticosteroids as well as the percentage of children prescribed long‐acting beta agonists/inhaled corticosteroids was not significantly different between groups at one and two years.
Logan 1979	Hypertension	Treatment of hypertension in the workplace by nurses versus treatment in the community by the family doctor	Patients in the nurse group were more likely to be put on antihypertensive medications (94.7% vs 62.7%, P < 0.001), to reach goal BP in the first six months (48.5 vs 27.5%, P < 0.001) and to take drugs prescribed (67.6 vs 49.1%, P < 0.005).
Marotti 2011	Postoperative patients	Pharmacist medication history and supplementary prescribing versus pharmacist medication history versus usual care	The marginal mean number of missed doses per patient was 3.21 (95% CI 2.89 to 3.52) in the control group, which was reduced in the pharmacist prescribing group 1.07 (95% CI 0.90 to 1.25, P = 0.002) but not in the pharmacist history group 3.30 (95% CI 2.98 to 3.63). The number of medications charted at an incorrect dose or frequency was reduced in the pharmacist history group. The pharmacist prescribing group had less dose errors than the pharmacist history group (P = 0.004).
Moher 2001	Secondary prevention of coronary heart disease in primary care	Audit group verus GP recall group versus nurse recall group (disease register and patient recall to nurse‐led clinic)	Little or no difference occurred in assessment between the nurse and GP recall group. Mean BP, total cholesterol, cotinine levels varied little between groups as did prescribing of hypotensive and lipid‐lowering agents. Prescribing of antiplatelet drugs was higher in the nurse recall group vs GP recall group, MD 8% (95% CI 1% to 15%, P = 0 .031).
Pagaiya 2005	Primary care nurses	Education and implementation of prescribing and clinical guidelines by nurses in rural health centres versus usual nurse care	Antibiotic prescribing in children 0 to 5 years for respiratory tract infections fell, (42% at baseline to 27% at follow‐up, control 27% to 30%, P = 0.022). Guidelines had no effect on prescribing antibiotics for diarrhoea but oral rehydration prescribing increased. Diazepam prescribing for adults fell, (intervention 17% to 10%, control 21% to 18%, P = 0.029).
Spitzer 1974	Patients attending primary care	Nurse practitioners versus physicians plus conventional nurse in primary care	Similar mortality experience, no differences in physical functioning capacity, social or emotional function. Quality of care similar. In 510 prescriptions, an adequate rating was given to 75% of conventional group and 71% in the nurse practitioner group, probably leading to little difference between groups.
Taveira 2010	Type 2 diabetes	A pharmacist‐led Veterans affairs Multidisciplinary Education and Diabetes Intervention for Cardiac risk reduction (VA‐MEDIC) plus usual care versus usual care	After four months there was a difference (P < 0.05) in the percentage of VA‐MEDIC patients versus controls in attaining target goals for systolic BP < 130 mmHg and HbA1c < 7% but not lipid control or tobacco use.
Thompson 1984	Drug therapy in a geriatric setting	Drug therapy prescribing and patient care management by clinical pharmacists versus usual care	The clinical pharmacist group probably had a lower number of deaths (P = 0.05), a higher number of patients being discharged to lower levels of care (P = 0.03) and a lower average number of drugs per patient (P = 0.04).
Tsuyuki 2016	Patients with cardiovascular risk factors associated with hypertension, diabetes, dyslipidaemia and smoking	Community pharmacist care versus usual care	At 3 months the intervention group patients had greater improvements in LDL cholesterol (‐ 0.2 mmol/L, P < 0.001, systolic BP (‐9.37 mmHg, P < 0.001), glycosylated haemoglobin (‐0.92%, P < 0.001) and smoking cessation (20.2%, P < 0.002).
BP: blood pressure CI: confidence interval GP: general practitioner HbA1c: glycated haemoglobin HR: hazard ratio IUD: inter uterine device LDL: low‐density lipoprotein MD: mean difference OR: odds ratio TAU: treatment as usual

Table 2. Outcomes of studies not included in meta‐analyses

Table 3. Primary outcome ‐ medication adherence

Study	Medication adherence measure	Outcome
Bruhn 2013	Morisky Medication Adherence Scale	Assessed adherence at baseline with patients in both groups reporting full adherence.
Cohen 2011	Medication possession ratios	The medication possession ratio (total days' supply of medication divided by total number of expected medication intake days) used in this study found little or no difference between the pharmacist prescribing arm and usual care, even though more medications were prescribed in the pharmacist arm. Adherence was high and ranked above 80%.
Finley 2003	Medication possession ratios	Determined the medication possession ratio from computerised prescription refill records. Full drug adherence was defined as a medication possession ratio value of 0.83 or more during the six‐month follow‐up. Medication possession ratios at three and six months were probably not different between intervention and control arms even though patients in the intervention group were more likely to change antidepressants. An additional measure, the Health Plan Employer Data Information Set guidelines for successful antidepressant treatment, showed there was little or no difference between groups in compliance with the early phase of treatment, but there was a significant difference in compliance in the intervention group continuation phase.
Hunt 2008	Morisky Medication Adherence Scale	Reported no differences at study end in the proportions of subjects reporting high medication adherence. There was an improvement in adherence with the groups from baseline to study end. Adherence did not predict goal attainment.
Hirsch 2014	Not described	Non‐adherence was identified in five of 33 patients with drug therapy problems at baseline, one of 12 patients at six months and one of four patients at nine months.
Logan 1979	Patient claim and pill counts	High adherence was judged if patients claimed to be taking their medication as instructed and 80% or more of drugs prescribed were consumed as determined by pill counts. In the nurse intervention group patients were more adherent than the control group.
Magid 2013	Medication possession ratios	Little or no difference between groups in the mean medication possession ratio adherence score over the six‐month study.
Margolis 2013	Morisky Medication Adherence Scale	Reported adherence measured by the Morisky scale modified for blood pressure medications. Adherence to antihypertensive medications at six months increased in the pharmacist intervention telemonitoring group but decreased in the usual care group. There was probably no difference between groups at 12 and 18 months.
Rudd 2004	Electronic drug event monitor	The drug event monitor provided the average number of days on which patients took the correct number of doses prescribed. While adherence was high in both groups, the nurse‐managed patient group had higher adherence than usual care.
Vivian 2002	Patient self‐reporting and drug refill information from the pharmacy	Non‐adherence was judged as missing more than three doses a week or pharmacy records indicated a failure to refill drugs within two weeks after the scheduled refill date. Little or no difference in adherence between or within the two groups at baseline or the end of the study was found. Over 90% of patients in both groups indicated they took their drugs as directed. The study was underpowered to detect a significant difference in adherence.

Table 3. Primary outcome ‐ medication adherence

Table 4. Secondary outcomes ‐ patient and provider satisfaction

Study	Satisfaction tool measure	Outcome
Barr Taylor 2003	Not specified	19/57 respondents stated that the nurse care management programme was moderately helpful. 32/57 found it extremely helpful. 9/13 physicians with two or more patients recommended adoption of the nurse management programme. In other health care settings: 9 physicians felt the programme decreased their time with patients, while 4 thought it increased the time spent.
Bruhn 2013	11 patient satisfaction statements derived from a local prescribing feasibility study	For the prescribing intervention, patients were generally positive about the pharmacist prescribing service ‐ 85% (39/46) were totally satisfied, while 9% (4/44) would have preferred to see their GP. In semi‐structured interviews with GPs and pharmacists, all pharmacists and most GPs were positive about the intervention. Pharmacists found their role satisfying, interesting, and challenging. 17 of 23 GPs were positive about the pharmacists’ role. The cost‐effectiveness of the pharmacists' role, given limited resources, was one issue raised in the GP focus group.
Finley 2003	Not specified	Patients reported greater treatment satisfaction with the collaborative care model than the control group in 6 of 11 measures including the overall treatment for depression, personal nature of the care, listening to concerns, explanations about why antidepressants were prescribed and how to take them, availability for advice, and overall satisfaction with the organisation. 18/37 primary care provider questionnaire respondents were satisfied with workflow, patient welfare. and the pharmacists' abilities.
Houweling 2009	Patient Evaluation of the Quality of Diabetes Care (PEQD)	Patients' evaluations of their satisfaction with diabetes care from the specialist diabetes nurse were significantly more positive than the control group.
Houweling 2011	Patient Evaluation of the Quality of Diabetes Care (PEQD)	The total satisfaction sum score for 14 PEQD measures for practice nurses was 66.4%, compared to 51.7% in the GP group which may be confounded by the amount of time given to each patient. On average GPs spent a total of 28 minutes per patient, whereas practice nurses spent 128 minutes per patient.
Hunt 2008	Satisfation in the SF‐36 healthcare domain	Satisfaction with hypertension care was high in both groups, but with little or no difference in any of the 11 satisfaction measures. Satisfaction was not associated with blood pressure goal attainment.
Hirsch 2014	22‐item Pharmacist Service Questionnaire. 0‐100 scale	Patient satisfaction with the clinical pharmacist were high, with mean scores 92.4 (±10.9) at 6 months (n = 49) and 92.7 (±11) at 9 months (n = 44).
Litaker 2003	Patient Satisfaction Questionaire	Improvements in four areas of satisfaction in the intervention group linked to an increased time spent with patients and an emphasis on patient‐centred education and self‐management (i.e. quality and quantity of contact) from base line to study end. Between‐group comparisons at study end demonstrated little or no significant difference in patient satisfaction measures, including overall care and general satisfaction.
Logan 1979	Not specified	6% of patients were dissatisfied with care provided by nurses but details of the survey instrument were not provided: (assumed 12/206 intervention patients at 6 months but not specified).
McAlister 2014	Not specified	Little or no difference in overall health care satisfaction between pharmacist‐ and nurse‐led care.
Magid 2013	Not specified	Patients at 6 months reporting they were very or completely satisfied with their hypertension care was probably higher in the intervention group than the usual care group.
Margolis 2013	Six items from the Consumer Assessment of Healthcare Providers and Systems adult survey (version 4)	Satisfaction items concerning clinicians listening carefully, explaining things clearly, and respecting what patients said showed larger improvements amongst patients in the telemonitoring intervention group than usual care at 6 months but not at 12 or 18 months.
Spitzer 1974	Not specified	96% of patients in the nurse practitioner group and 97% of patients in the conventional care group were satisfied with the health services received in the experimental period.
Vivian 2002	Not specified	Little or no significant differences in patient satisfaction between groups. More patients in the intervention group felt that the pharmacist spent more time with them than did control patients, although there was little difference. There was no difference in satisfaction with pharmacy services or changes in patient satisfaction in either group from baseline to study end. This study was underpowered to detect a significant difference in patient satisfaction.
GP: general practitioner

Table 4. Secondary outcomes ‐ patient and provider satisfaction

Table 5. Primary outcome ‐ adverse events

Study	Adverse event
Ansari 2003	There was little or no difference in the proportions of patients between control (provider education), nurse facilitator and provider/patient notification for hospitalisations and emergency room visits. There were few deaths with the higher number (7) in the control group which had more patients on haemodialysis, two of whom died.
Aubert 1998	There appeared little or no difference between intervention and usual care groups for severe low blood glucose events at baseline and during the study period. Mean weight gain differences from insulin treatment in each group or mean weight loss differences with oral agents showed little or no difference.
Chenella 1983	Reported no patients had major bleeding, but four patients in the pharmacist prescriber group had minor bleeding (one laceration before hospital). One patient in physician prescriber group died, after receiving heparin and warfarin for a stroke in evolution but there was no evidence of bleeding.
DeBusk 1994	The first year mortality was 3.4% in usual care and 4.1% in the intervention group. However, a longer study is required to show a difference, namely, 2 years plus a 5‐ to 10‐year follow‐up.
Fairall 2008	The time to death did not differ between primary care nurses and doctors initiating therapy.
Hirsch 2014	Pharmacists identified two adverse drug reactions from 33 drug therapy problems at baseline, two from 12 at six months and none at nine months.
Ishani 2011	Adverse events were similar between groups, with no participants withdrawing from the study due to an adverse event, and there was no difference in the rate of hospitalisation or death between the groups.
Jaber 1996	Reported 17 hypoglycaemic reactions in the intervention group and two in the control group. All were considered mild to moderate. The difference was possibly related to increased training in recognition, documentation, and questioning in the intervention group. Three patients were hospitalised, two in the control and one in the intervention group, and these appear unrelated to treatment.
Klingberg‐Allvin 2015	In treating incomplete abortion bleeding, the same or less than normal menstrual cycle was probably not different between the intervention midwife and usual care physician groups. There was little difference in pain after treatment as assessed by a visual analogue scale. 30 (6%) of women reported unscheduled visits in the midwife group and 18 (4%) in the physician group. Reasons included vaginal bleeding and abdominal pain. Reported side‐effects after treatment were similar in both groups (nausea, vomiting, abdominal pain, chills, and fever).
Kuethe 2011	There were no differences between groups (general practitioner, paediatrician, asthma nurse) with respect to the number of severe asthma exacerbations as expressed by the number of prednisolone courses.
MacMahon Tone 2009	Forty drug‐related adverse events occurred in the intensive intervention group as compared to 10 in the standard group. While the adverse events are known for the drugs in question no further comment was offered.
McAlister 2014	Reported few clinical events at six months in a pharmacist‐led intervention for secondary prevention after ischaemic stroke. There were nine cardiovascular events and no deaths in the pharmacist group versus eight cardiovascular events and one death in the nurse‐led group.
Margolis 2013	There were 60 adverse events in usual care and 49 in the telemonitoring group; most events were non‐cardiac hospitalisations. There were two allergic reactions to blood pressure medication in the usual care group, six events in the telemonitoring group related to hypotension, dizziness, loss of consciousness which compared to one in the usual care group, four events in usual care related to hypertension versus one in the intervention group.
New 2003	In patients randomised to specialist nurse‐led clinics for blood pressure control, lipid control or both, there were less deaths in the intervention group (25, (3.2%) versus 36 (5.7%) in the usual care group) odds ratio 0.55 (95% confidence interval 0.32 to 0.92) P = 0.02.
Spitzer 1974	During the 12‐month experimental period, there were four deaths in the nurse practitioner group and 18 in the conventional care group. There was probably little or no difference in the crude death rate between groups.
Taveira 2011	There were no diabetes‐related admissions or deaths for either group during the six‐month study.
Thompson 1984	The pharmacist prescribing group in a geriatric setting may have had a slightly lower 12‐month mortality than usual care (3/67 versus 10/72, P = 0.05).
Tobe 2006	The incidence of adverse events probably did not differ between the intervention (home care nurse group) and control (primary care physician group) in First Nations people with diabetes and hypertension. Ten patients in the intervention group and seven in the control group required admission to hospital for adverse events.

Table 5. Primary outcome ‐ adverse events

Table 6. Secondary outcome ‐ quality of life

Study	Measures	Outcome
Aubert 1998	Four generic quality of life measures from the Behavioural Risk Factor Surveillance System	Intervention and control groups reported improved perception of health status after 12 months, but intervention patients were twice as likely to report this.
Barr Taylor 2003	SF‐36, the Duke Activity Status Index for QoL, and the BDI for depression	Little or no differences for any of the variables, but an improved mood for both groups was found.
Bruhn 2013	SF‐12, HUI, CPG, and HADS‐D	No one measure was seen as the primary outcome. In the prescribing arm there was a within‐arm improvement for CPG intensity and disability effect size subscales and between arms on the intensity subscale but not the disability subscale. There was a within‐arm improvement in overall CPG in the prescribing and review arms but not the TAU arm. The SF‐12 and HADS‐D showed deterioration in the TAU arm. Compared with baseline, patients had an improved CPG in the prescribing and review arms but not the TAU arm. The SF‐12 physical score difference showed no effect in prescribing or review arms but improvement in the TAU arm. SF‐12 mental score showed no effect in prescribing or review arms and deterioration in the TAU arm. HADS‐D scores within the prescribing arm showed improvement for depression and anxiety which were also significant between groups.
Cohen 2011	SF‐36 for Veterans	Little or no change in quality of life scores over 6 months.
Finley 2003	The Brief Inventory for depressive symptoms and Work and Social Disability Scale	Liitle or no difference at 6 months between intervention and control groups.
Houweling 2009	SF‐36 and the revised version of the Type 2 Diabetes Symptom Checklist to measure the presence and perceived burden of diabetes‐related symptoms	Little or no differences over 12 months between groups in either survey.
Houweling 2011	SF‐36 and the revised version of the Type 2 Diabetes Symptom Checklist to measure the presence and perceived burden of diabetes‐related symptoms	In the control group there were little or no differences between baseline and follow‐up SF‐36 measures, however in the practice nurse intervention group there were differences in physical functioning, role physical, vitality, and the physical component score. Little or no differences were seen in the QoL results over time between the two groups except for the physical component score which was lower in the intervention group. After 14 months responses to the revised Type 2 Diabetes Symptom Checklist revealed little or no differences between groups.
Hunt 2008	SF‐36	Little or no difference except in the general health domain with scores higher in the control group.
Jaber 1996	Health Status Questionnaire version 2 derived from the SF‐36	Little or no difference between or within groups.
Khunti 2007	SF‐36, Seattle Angina Questionnaire and LVD‐36 questionnaire	Differences favouring the intervention group were found in the SF‐36 for physical functioning, general health, vitality, social functioning, and mental health. Seattle Angina Questionnaire scores in patients with angina were significantly better for intervention patients compared to controls for exertional capacity and borderline differences were found for angina frequency and QoL. There was little or no difference in any of the SF‐36 health status domains or LVD‐36 scores for patients with a confirmed diagnosis of left ventricular diastolic dysfunction.
Litaker 2003	SF‐12 Diabetes Quality of Life	Little or no difference between groups in either measure at study end.
McAlister 2014	Self‐related health using a Likert scale The EQ‐5D as an index of health	Little or no difference between the pharmacist‐ and nurse‐led groups in participants overall self‐related health.
Margolis 2013	SF‐12	Little or no differences between groups.
Moher 2001	Dartmouth COOP charts EuroQol scores	Little or no or clinically important differences between groups for any dimension.
Spitzer 1974	Not described	Patients in the nurse practitioner and usual care groups had similar values at baseline and study end for physical, emotional, and social function.
Taveira 2011	Change from baseline in depression symptoms by the PHQ‐9	Even though no pharmacologic treatments for depression symptoms were offered as part of the intervention, the mean change in PHQ‐9 scores was probably not different for intervention and standard care participants.
Vivian 2002	SF‐36	Little or no significant differences either between or within the two groups from baseline to study end, although patients in the control group reported more bodily pain .
BDI: Beck Depression Index CPG: Chronic Pain Grade EQ‐5D: EuroQol five dimensions questionnaire HADS‐D: Hospital Anxiety and Depression Scale HUI: Health Utilities Index LVD‐36: Left Ventricular Dysfunction PHQ‐9: Patient Health Questionnaire‐9 QoL: quality of life SF‐12: Short‐Form‐12 SF‐36: Short‐Form‐36 TAU: treatment as usual

Table 6. Secondary outcome ‐ quality of life

Table 7. Secondary outcome ‐ resource use

Medication and related therapy
Study	Outcome
Ansari 2003	β‐blocker use was higher in the nurse facilitator group with two‐thirds of patients either initiated or up‐titrated on β‐blockers versus fewer than one‐third of patients in the other two study arms (control provider education and provider/patient notification).
Chenella 1983	Little or no difference in amount of anticoagulant drugs prescribed by pharmacists compared to a physician.
Cohen 2011	More patients in the pharmacist prescribing arm were prescribed diuretics and sulphonylureas compared to usual care. Overall there was an increase in the number of medications prescribed by pharmacists for hypertension, diabetes, and cholesterol from baseline to six months, but little or no change in the usual care arm.
Denver 2003	In nurse‐led clinic for hypertension management in diabetics at six months there were increased changes in the proportions of patients receiving new prescriptions for calcium channel blockers and thiazide diuretics as intensification therapy. The median number of drugs per patient increased in the intervention group compared to conventional primary care.
Einhorn 1978	In a family medicine clinic in Bogota, nurses were less likely than physicians to provide intrauterine devices, prescribe oral contraceptives, and sterilisation on the patient's first visit. Nurses were more likely than physicians to provide temporary prescriptions and defer intrauterine devices and contraceptive measures if the patient on their first visit was not menstruating or believed to be pregnant.
Heisler 2012	Observational cohort results taken six months following the quarter start date showed intervention patients had more blood pressure medication changes.
Hirsch 2014	Pharmacists identified at least one hypertension drug therapy problem in 33/73 (45.2%) patients at baseline requiring additional therapy in 14/33 (42.4%) and dosage increases in 11/33 (33.3%).
Houweling 2009	The nurse specialist in diabetes prescribed significantly more antihypertensive agents and the internist (doctor control) prescribed more cholesterol‐lowering agents.
Hunt 2008	The mean number of antihypertensive medications per patient and use of generic antihypertensive agents was higher in the intervention group.
Logan 1979	Patients in the nurse‐managed group were more likely to be put on antihypertensive medications, prescribed more than two pills per day, and to be on more than one antihypertensive medication.
MacMahon Tone 2009	There were more intervention intensive group patients on three or more antihypertensive drugs (at the study beginning more patients in the standard care group were on three or more antihypertensive agents). At the end of the study more patients with dyslipidaemia in the intensive group were receiving statin therapy. More patients in the intervention group were on aspirin antiplatelet therapy at the end of the study.
McAlister 2014	The median number of antihypertensive medications taken at six months was probably not different in the pharmacist‐ and nurse‐led groups. There was a difference favouring pharmacists in maximal dosing of angiotensin‐converting enzyme inhibitors or angiotensin receptor blockers at six months, but not the percentage of patients using these drugs.
Magid 2013	In patients completing the six‐month visit, there were more intervention patients that had an antihypertensive medication added to their regimen and a dose increased for existing medication, than usual care patients. There was an increase in the usage of specific antihypertensive drugs.
Margolis 2013	There were increases in the mean number of antihypertensive medication classes at 6, 12, and 18 months in the intervention group compared to baseline and compared to usual care.
Moher 2001	There was minimal change in prescribing antihypertensive drugs in the three groups. All groups increased prescribing of lipid‐lowering drugs but there was little or no difference between groups. There was an increase of 10% more patients’ prescribed antiplatelet treatment in the nurse recall group versus the audit group and 8% more in the nurse recall group versus the general practitioner recall group.
Pagaiya 2005	In examining the effects of training and guidelines on prescribing by nurses, the mean change in antibiotic prescribing for all patients showed little or no difference. The mean change for antibiotic prescribing for respiratory infections in children (0 to 5 years) fell. No change was detected in prescribing antibiotics for diarrhoea. There was a mean fall in diazepam prescribing in the intervention group.
Rudd 2004	In the nurse management patient group at six months there was an increased number and variety of antihypertensive medications and an increased number of medication changes than in the usual care group.
Taveira 2010	The intervention arm group (VA‐MEDIC) had greater dose titrations of antihypertensive medications, insulin, statins, and niacin compared to the usual care arm.
Taveira 2011	Intervention arm participants (VA‐MEDIC‐D) had more dose increases or initiation of any antihypertensive agents and more dose increases or initiation of antihyperglycaemic agents. There was little or no difference in the initiation or dose titration of any antihyperlipidaemic agent or antidepressants.
Thompson 1984	The average number of drugs prescribed per patient was lower in the pharmacist group compared to the physician group. The number of drugs was reduced by an average of 2.2 drugs per patient from the pre‐study to the study year. The practice of clinical pharmacists prescribing drug therapy under physician supervision has the potential to save the healthcare system USD 70,000 per 100 skilled nursing facility beds.
Tsuyuki 2015	In the pharmacist prescribing arm proportionally more new antihypertensive agents were initiated, more dose changes occurred, more antihypertensives were discontinued, and more patients were prescribed low‐dose aspirin and a statin than in the usual care group.
Vivian 2002	There was little or no difference in the type of antihypertensives prescribed to intervention and control patients during the study.
Wallymahmed 2011	Compared with baseline there were more patients in both groups taking antihypertensive medications but this difference was probably only important in the nurse‐led intervention group.

Healthcare visits, health resources, and associated costs
Ansari 2003	There was no difference in hospitalisations and emergency room visits between the three groups of control (provider education), nurse facilitator, and provider/patient notification.
Aubert 1998	Hospital admissions were rare and did not differ between the intervention and usual care groups. ED visits did not differ between groups or from baseline. No hospital or ED visits were related to diabetes. The average number of outpatient visits during the study was similar. The nurse managed a case load of 71 patients, but it was estimated that a 300 patient case load could be managed.
Barr Taylor 2003	There was no change in health utilisation (physician visits, ED visits, days of hospitalisation) for the year before and after the intervention and between groups.
Choe 2005	In reporting process measures for the clinical pharmacist’s case management of patients there was a difference between pharmacist intervention and control in the frequency of low‐density lipoprotein measurements, retinal examinations, and monofilament foot examinations but not glycated haemoglobin measurement or urine albumin screen.
Cohen 2011	Over six months there were a higher number of primary care visits in the usual care arm; an average 1.65 visits per patient versus 1.56 in the intervention arm. It was suggested the difference in the higher number of primary care visits may offset the intervention cost.
DeBusk 1994	The nursing time spent in the year after myocardial infarction was nine hours per patient; a per patient cost of USD 500 which included the nurse salary, office costs, and other associated costs. This compared with cardiac rehabilitation programmes in the San Francisco Bay area costing USD 1800 to USD 2700 to participate for three months.
Ellis 2000	In investigating the impact of clinical pharmacist interventions in patients with dyslipidaemia there was little or no difference in physician or nurse visits between control and the intervention patients at 12 months. At 12 months the intervention group had more pharmacist visits than the control group. There were little or no difference in costs for hospitalisations, clinic visits, laboratory costs, drug costs, and costs of lipid therapy between groups. The intervention group had a USD 370 greater difference per patient in total costs which was probably not important and approximately 5% of total costs.
Fairall 2008	In the cohort of patients not yet receiving antiretroviral therapy there was little or no difference in clinic visits with a nurse but clinic visits with a doctor were probably higher in the intervention group. In the cohort of patients who had already received at least six months of antiretroviral therapy clinic visits with a nurse probably higher in the intervention group. Economic data from the study is the subject of further analysis by Barton 2013 (see Studies awaiting classification).
Finley 2003	Although the collaborative care model experienced a decrease in the total number of primary care visits, the between‐group difference was probably not important. ED visits increased more in the usual care group but this was probably not important and neither was the difference in utilisation of psychiatric services. The institutional cost of drugs, the cost of antidepressants and the cost of psychotropic drugs overall was higher in the intervention group, but this was not important.
Fischer 2012	Hospital admissions (while trending to fewer admissions) in the nurse intervention group showed little or no difference to the control group. Nurse case management was not associated with a significant difference in the number of outpatient or ED visits. There was a decrease in total costs in the nurse telephone intervention group comparing the period before and after randomisation. In contrast, there was an increase for the same comparison in the control group. Similar results were seen with hospitalisation and ED costs which were lower in the intervention group. There was probably not an intervention effect on outpatient costs. The difference in average per patient cost between the intervention group (USD 6600) and control group (USD 9033) of USD 2433 was important. The control group had higher baseline hospitalisation rates and total costs cautioning interpretation of the result.
Heisler 2012	Little or no difference in health services utilisation (hospitalisations, primary care visits, ED visits) between intervention and control patients during the 14‐month study of blood pressure control through a clinical pharmacist outreach programme in diabetic patients.
Hirsch 2014	The pharmacist collaborative group (PharmD‐PCP MTM) had fewer primary care physician visits during the intervention period than did the usual care group. The mean total combined visits of primary care physician and pharmacist was not greater in the PharmD‐PCP MTM group than in usual care.
Houweling 2009	There was a lower number of visits in the NSD group compared with standard care but not in the duration of visits. Significantly more patients were referred back to their GP by the NSD when meeting treatment goals. Personnel and laboratory costs were lower in the intervention group than the control group. The average per month increase in medication costs between the groups was probably not important apart from the cholesterol‐lowering medications. The average time saving per internist was 61.4 minutes (meaning the internist could supervise 11 patients with the NSD in the time he/she could treat one patient).
Houweling 2011	The mean number of visits and duration of visits was higher in the practice nurse intervention group than the control group.
Hunt 2008	The total number of clinic visits (physician plus pharmacist) was higher in the intervention arm compared to the control arm. The number of physician visits was lower in the intervention arm.
Ishani 2011	Little or no difference in the hospitalisation rate between intervention and control groups.
Kuethe 2011	In testing the non‐inferiority of asthma care in children with stable asthma provided by a hospital‐based specialised asthma nurse versus a GP or paediatrician, there was little or no differences between the groups for medication, school absence or parental work absence after two years. There was little or no difference in unplanned visits and no hospital admissions during the study.
Litaker 2003	Medium number of outpatient visits were higher for the team based intervention patients. Average personnel costs for one year's treatment were significantly higher in the intervention group (USD 134.68 vs USD 93.70, P < 0.001).
Magid 2013	There was little or no difference in the mean number of outpatient clinic visits, total number of ED visits, and hospitalisations between the two groups. The intervention group probably had a higher number of email and telephone encounters.
Margolis 2013	Over 12 months in the telemonitoring intervention group all 228 patients used a mean of 11.4 ± 3.9 pharmacist visits lasting a mean of 34.2 minutes and 217 used telemonitoring services with a mean of 9.8 ± 2.5 months of use. It was estimated direct programme costs would total USD 1350 per patient.
Spitzer 1974	A reported five per cent drop in gross practice revenue was explained by the absence of billing for services provided by the nurse practitioner. Billing for unsupervised practice was not permitted in Ontario at the time of the study. During the trial year the services rendered by the nurse practitioner were worth approximately USD 16,000 of which almost 50% was for unsupervised practice.
Taveira 2011	There was little or no differences in primary carer visits, use of ED services for all cause visits, diabetes‐related ED visits or hospital admission rates.
Thompson 1984	There was little or no difference in the average length of stay or hospitalisations although the latter trended lower in the pharmacist group. Differences favouring the pharmacist group were found in the rate of discharge to home or to a lower level of care.
Vivian 2002	Little or no differences between intervention and control groups in appointments with the primary care provider during the 6 months of the study.
ED: emergency department GP: general practitioner NSD: nurse specialised in diabetes

Table 7. Secondary outcome ‐ resource use

Comparison 1. Non‐medical prescribing group versus usual care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Systolic blood pressure mmHg Show forest plot	21		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

1.1 6 months	11	2076	Mean Difference (IV, Fixed, 95% CI)	‐6.76 [‐8.24, ‐5.27]
1.2 12 months	12	4229	Mean Difference (IV, Fixed, 95% CI)	‐5.31 [‐6.46, ‐4.16]
1.3 6 months systolic blood pressure removing cluster effect (Margolis)	10	1628	Mean Difference (IV, Fixed, 95% CI)	‐6.13 [‐7.83, ‐4.44]
1.4 12 months systolic blood pressure excluding cluster trials (Khunti and Margolis)	10	2627	Mean Difference (IV, Fixed, 95% CI)	‐4.84 [‐6.29, ‐3.39]
1.5 Systolic blood pressure at 6 months (more NMP prescribing autonomy)	4	695	Mean Difference (IV, Fixed, 95% CI)	‐2.98 [‐5.36, ‐0.59]
2 HbA1c (%) Show forest plot	8		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

2.1 HbA1c 6 mths	3	271	Mean Difference (IV, Fixed, 95% CI)	‐0.42 [‐0.75, ‐0.09]
2.2 HbA1c 12 mths	6	775	Mean Difference (IV, Fixed, 95% CI)	‐0.62 [‐0.85, ‐0.38]
3 Low‐density lipoprotein (LDL) mmol/L Show forest plot	11		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

3.1 LDL 6 mths	6	1213	Mean Difference (IV, Fixed, 95% CI)	‐0.25 [‐0.34, ‐0.17]
3.2 LDL 12 mths	7	1469	Mean Difference (IV, Fixed, 95% CI)	‐0.21 [‐0.29, ‐0.14]
4 Low‐density lipoprotein pharmacist vs nurse 6 mths Show forest plot	6	1213	Mean Difference (IV, Fixed, 95% CI)	‐0.25 [‐0.34, ‐0.17]

4.1 Pharmacist	4	629	Mean Difference (IV, Fixed, 95% CI)	‐0.09 [‐0.20, 0.02]
4.2 Nurse	2	584	Mean Difference (IV, Fixed, 95% CI)	‐0.52 [‐0.67, ‐0.38]
5 Adherence (continuous) Show forest plot	4	700	Std. Mean Difference (IV, Fixed, 95% CI)	0.15 [0.00, 0.30]

6 Adherence (dichotomous) Show forest plot	4	935	Risk Difference (M‐H, Fixed, 95% CI)	0.06 [‐0.00, 0.12]

7 Health‐related quality of life Show forest plot	8		Mean Difference (IV, Fixed, 95% CI)	Subtotals only

7.1 Physical component (SF12 or 36)	8	2385	Mean Difference (IV, Fixed, 95% CI)	1.17 [0.16, 2.17]
7.2 Mental component (SF‐12 or 36)	6	2246	Mean Difference (IV, Fixed, 95% CI)	0.58 [‐0.40, 1.55]
8 Health facility resource use Show forest plot	5		Risk Difference (M‐H, Fixed, 95% CI)	Subtotals only

8.1 Emergency Department visits	3	4626	Risk Difference (M‐H, Fixed, 95% CI)	0.01 [‐0.02, 0.03]
8.2 Hospitalisations	5	4870	Risk Difference (M‐H, Fixed, 95% CI)	‐0.01 [‐0.03, 0.01]

Comparison 1. Non‐medical prescribing group versus usual care