Psychological preparation and postoperative outcomes for adults undergoing surgery under general anaesthesia

See: Summary of findings for the main comparison Any intervention compared to control for adults undergoing surgery under general anaesthesia

Summary of findings for the main comparison. Any intervention compared to control for adults undergoing surgery under general anaesthesia

Any psychological preparation intervention compared to control for adults undergoing surgery under general anaesthesia
Patient or population: adults undergoing elective surgery under general anaesthesia Setting: pre‐surgical contexts (typically hospitals/preoperative clinic settings); setting was not limited by country/language/type of hospital Intervention: psychological preparation interventions presented to participants preoperatively; interventions contained one or more of the following components: procedural information; sensory information; behavioural instruction; cognitive intervention; relaxation techniques; hypnosis; emotion‐focused intervention Comparison: control group (typically standard care and/or attention control)
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control	Any intervention
Postoperative pain ‐ measured with a range of tools and placed on a standardized scale Higher scores = higher pain	‐	The mean pain in the intervention group was 0.2 (95% confidence interval 0.35 to 0.06) standard deviations lower	‐	2713 (38 RCTs)	⊕⊕⊝⊝ LOW¹	‐
Behavioural recovery ‐ measured with a range of tools	Insufficient data were available to calculate standardized scores	Findings suggested that psychological preparation has potential to improve behavioural recovery outcomes, but no clear conclusions could be reached	‐	1441 participants were randomized (14 RCTs)	⊕⊝⊝⊝ VERY LOW²	Data from studies were not combined in meta‐analysis because of a low number of studies containing suitable data and a wide range of outcome measures
Length of stay in hospital (days)	The mean length of stay for the control groups ranged from 2.11 to 18.6 days	The mean length of stay (days) in the intervention group was 0.52 days fewer (95% confidence interval 0.82 to 0.22)	‐	3313 (36 RCTs)	⊕⊕⊝⊝ LOW³	‐
Negative affect ‐ measured with a range of tools and placed on a standardized scale Higher scores = higher negative affect (e.g. more anxiety)	‐	The mean negative affect in the intervention group was 0.35 (95% confidence interval 0.54 to 0.16) standard deviations lower	‐	2496 (31 RCTs)	⊕⊕⊝⊝ LOW⁴	‐
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RCT: randomized controlled trial
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Many studies reported insufficient methodological details to ascertain risk of bias (rated `serious', see Figure 1), and heterogeneity was high (71%, also rated `serious'). We therefore downgraded the overall quality of evidence by two points. ²We downgraded the quality of evidence as `risk of bias' was rated as `very serious' ‐ there were a high proportion of `uncertain' ratings for risk of bias categories, and the number of studies with robust measures meeting our inclusion criteria and reporting suitable data for meta‐analysis was low. We made a further downgrade for high heterogeneity (treated as `serious'). We therefore downgraded the overall quality of evidence by three points. ³Many studies reported insufficient methodological details to ascertain risk of bias (rated `serious', see Figure 1), and heterogeneity was high (74%, also rated `serious'). We therefore downgraded the overall quality of evidence by two points. ⁴Many studies reported insufficient methodological details to ascertain risk bias (rated `serious', see Figure 1), and heterogeneity was high (81%, also rated `serious'). We therefore downgraded the overall quality of evidence by two points.

Figure 1

`Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Background

Many people experience anxiety and negative cognitions when approaching surgery (Mathews 1981). There is good evidence that how people think and feel before surgery affects their outcomes after surgery. Negative psychological factors such as anxiety, depression and catastrophizing have been found to predict postoperative pain (Arpino 2004; Bruce 2012; Granot 2005; Munafó 2001). Catastrophizing has been defined as "an exaggerated negative orientation toward noxious stimuli" (Sullivan 1995).

A range of mechanisms exist by which psychological variables could affect recovery after surgery. First, negative emotions can enhance pain sensations (Rainville 2005; van Middendorp 2010). Second, cognitions and emotions influence behaviour (for example doing physiotherapy exercises, taking analgesics) and are likely to influence pain and return to usual activities. Third, stress has been linked to the slower healing of wounds through psychoneuroimmunological mechanisms (mechanisms whereby psychology interacts with the nervous and immune systems) (Maple 2015; Marucha 1998; Walburn 2009). It is therefore likely that psychological interventions that reduce negative emotions such as anxiety, worry about surgery and perceptions of stress, or that change patients' recovery‐related behaviour, may lead to positive postoperative outcomes.

Psychological preparation for surgery has been demonstrated to improve outcomes. In a review and meta‐analysis (Johnston 1993), psychological preparation was found to be beneficial for a range of outcome variables that included negative affect, pain, pain medication, length of hospital stay, behavioural recovery, clinical recovery, physiological indices and satisfaction.

Since the 1993 review (Johnston 1993), this research field has continued to develop. Standards of conducting randomized controlled trials have improved, technology has advanced to permit more detailed bibliographic searching and new studies testing psychological preparation procedures have been published. The present review tested, using modern review techniques, analysis methods and a larger research base, a) whether there is evidence for beneficial (or harmful) effects of psychological preparation for surgery, and b) which outcomes of pain, behavioural recovery, length of stay and negative affect are improved (or worsened) following preparation.

Description of the condition

Surgery is carried out for a range of health conditions either as a diagnostic or treatment intervention. While surgery may lead to health improvements, it also negatively impacts on a range of health outcomes including pain, activity limitations and anxiety, at least in the short term (Johnston 1980).

Elective surgery differs from emergency surgery in that patients have time to prepare themselves and to be prepared for surgery. Preparation for emergency surgery is much more difficult to provide in a controlled manner and the effectiveness of such interventions is likely to differ because of that difference in context. Thus, emergency surgery should be considered separately and we only included participants undergoing elective surgery in this review.

Different psychological threats and coping mechanisms can be involved for the patient depending on whether procedures are undertaken using general anaesthetic or local anaesthetic. For example in some procedures that are performed under local anaesthetic the patients are required to be actively involved, and so effective preparation will have different components compared with preparation for a procedure where the patient is unconscious. Therefore, following Johnston 1993, we only included procedures involving general anaesthetic.

Description of the intervention

Psychological preparation incorporates a range of strategies designed to influence how a person feels, thinks or acts (emotions, cognitions or behaviours). Johnston 1993 found that the following types of intervention benefited patients, on at least one postoperative outcome: procedural information, sensory information, behavioural instruction, cognitive intervention, relaxation techniques, hypnosis and emotion‐focused interventions.

Procedural information

Procedural information describes the process the patient will undergo in terms of what will happen, when it will happen and how it will happen.

Sensory information

Sensory information describes the experiential aspects of the procedure, that is, what it will feel like and any other relevant sensations (for example taste, smell).

Behavioural instruction

Behavioural instruction consists of telling patients what they should do to facilitate either the procedure or their recovery from the procedure (Mathews 1984). For example a patient could be told how to use equipment, such as a patient‐controlled analgesia pump.

Cognitive interventions

Cognitive interventions aim to change how an individual thinks, especially about negative aspects of the procedure. Cognitive techniques include cognitive reframing and distraction. Cognitive reframing involves developing a different perspective that enables a positive or neutral rather than negative thought, for example focusing on the number of people who do well after a surgical procedure rather than the number who fare badly. Distraction leads to focusing thoughts on other things (and could include relaxation).

Relaxation techniques

These involve "systematic instruction in physical and cognitive strategies to reduce sympathetic arousal, and to increase muscle relaxation and a feeling of calm" (Michie 2008). Relaxation techniques can be used before surgery to reduce tension and anxiety and include progressive muscle relaxation (where each muscle group is tensed and then relaxed), simple relaxation (each muscle group is relaxed in turn), breathing techniques (for example the practice of diaphragmatic breathing) and guided imagery (for example imagining a pleasant, relaxing environment).

Hypnosis

A range of procedures are used for hypnotic induction, including suggestions to relax. During hypnosis "one person (the subject) is guided by another (the hypnotist) to respond to suggestions for changes in subjective experience, alterations in perception, sensation, emotion, thought or behavior" (APA 2005).

Emotion‐focused interventions

Emotion‐focused interventions aim to enable the person to regulate or manage their feelings or emotions. Emotion‐focused methods include: enabling the discussion, expression or acceptance of emotions; facilitating contextualization (putting emotions into context, e.g. of life, relationships, past experiences); and enabling the understanding of emotions (e.g. giving them meaning). In this review, if the focus of the intervention was to change how someone thinks, we coded it as a ‘cognitive intervention’.

How the intervention might work

Studies have shown that psychological preparation for surgery can have a beneficial effect upon a range of postoperative outcomes (Johnston 1993). Likely mechanisms for these processes vary depending upon the intervention used. Some intervention types focus on reducing negative emotions, such as anxiety, and negative thought processes. Providing procedural information is expected to reduce anxiety because it helps the patient to know what to expect when they undergo surgery. It reduces uncertainty, and ensures that concern is not caused by events that are part of normal hospital procedures (Ridgeway 1982). Similarly to providing procedural information, providing sensory information is expected to reduce anxiety by reducing the discrepancy between the sensation expected by the patient and the sensation actually experienced (Johnson 1973). For example, if a patient expects to experience discomfort after surgery in a particular bodily location, when this discomfort is experienced it is understood as being part of the normal surgical experience rather than an indication that something has gone wrong. Cognitive interventions aim to reduce negative emotions and thoughts related to the surgical process by either changing negative thoughts or refocusing attention elsewhere, and emotion‐focused interventions target an individual's emotions directly. Relaxation and hypnosis interventions aim to make an individual feel more relaxed, both psychologically and physiologically, and may effectively act as distraction techniques, so reducing both negative emotions and negative thoughts. As noted earlier, negative thoughts and emotions influence wound healing (Kiecolt‐Glaser 1998), perceptions of pain and also behaviour. Finally, behavioural instruction aims to directly influence behaviours that are important in enabling the surgical procedure to go well and to enhance recovery, for example teaching people how to manage their own analgesia, or instructing them as to when they should return to usual activities for optimal recovery.

Why it is important to do this review

Improving outcomes after surgery has a range of benefits both for the individual and for the healthcare service. Individuals will benefit from reduced postoperative pain and a quicker return to activity. Economic benefits include shorter stays in hospital, reduced use of pain medication and quicker return to work.

Objectives

To review the effects of psychological preparation on postoperative outcomes in adults undergoing elective surgery under general anaesthetic.

Methods

Criteria for considering studies for this review

Types of studies

We included both published and unpublished randomized controlled trials (RCTs). We excluded quasi‐randomized trials. We included, and narratively described, cluster‐randomized controlled trials but did not include them in the meta‐analyses.

Types of participants

We included studies with adult participants (aged 16 years or older) undergoing elective surgery under general anaesthesia. If information about anaesthesia was not provided we contacted the study authors for confirmation. If no response was received, we took advice from a clinician (either a surgeon or anaesthesiologist) who assessed whether that type of surgery would usually be performed under general anaesthesia. We included or excluded studies on this basis. Some surgical procedures are carried out under either general or local anaesthesia (for example inguinal hernia repair surgery). We included studies containing a mixture of participants undergoing general and local anaesthesia but excluded studies where all participants underwent, or were expected to have undergone, local (or no) anaesthesia (with or without sedation).

We included studies of people who have received premedicative sedative prior to general anaesthesia. Different issues are encountered with children undergoing surgery (for example their developmental stage) and different psychological techniques are used (Johnston 1993). Studies tend to focus either on adults or children. We excluded participants aged less than 16 years from this current review.

We excluded studies focusing on patient groups with clinically diagnosed psychological morbidity. However, we did not exclude studies that included participants with mental disorders or subclinical symptoms co‐existing with the condition that led to the operation.

Types of interventions

Psychological preparation, including:

procedural information;
sensory information;
behavioural instruction;
cognitive interventions;
relaxation techniques;
hypnosis;
emotion‐focused interventions.

‘Psychological preparation’ was defined as interventions where the intervention was entirely provided before surgery (this preparation could include, for example, instructions for the participant for after surgery, but the implementation of the intervention had to be pre‐surgery). We were interested in the psychological content of the intervention in this review rather than how it is delivered. There are studies that compare different formats (e.g. leaflet versus video) or timings, but the actual content of the intervention is the same. We excluded these papers. Where the control group also received an element of preoperative preparation (for example, procedural information), the intervention group was required to receive that element beyond that received by the control group (for example, more detailed procedural information, or procedural information about additional aspects of surgery) to be considered as an `intervention'.

Types of outcome measures

We included studies that collected data on two primary and two secondary outcomes. We only included outcomes measured within 30 days/one month post‐surgery. We excluded studies that did not measure these outcomes for pragmatic reasons: because of the size of the review and available research team resources, including all studies measuring any outcome was not manageable (see Differences between protocol and review). Where repeated measurements of outcomes were taken postoperatively, we used the earliest measure for the main meta‐analysis. This is because, while the longest follow‐up is important for longer‐term recovery, it was likely that most studies would include short‐term outcome data but only a few would also include longer time frames.

Primary outcomes

1. Postoperative pain

1a. Postoperative pain intensity: there are a range of well‐used measures for pain and some studies report pain as an outcome using more than one measure. We extracted all reported postoperative pain outcomes from each study.

We used the following hierarchy when deciding which postoperative pain measure to use in the meta‐analysis:

the pre‐specified postoperative pain outcome (if given);
a visual analogue scale (VAS), for example from 0 to 100 (or 0 to 10);
McGill Pain Questionnaire (MPQ) (Melzack 1975) intensity rating, Present Pain Intensity;
other MPQ ratings: i) Pain Rating Index (weighted or unweighted), ii) Number of Words Counted;
Short Form‐36 (SF‐36) pain (Ware 2000);
Nottingham Health Profile pain (Hunt 1983);
other pain intensity scale.

We analysed pain at rest over pain at movement; moving in bed over pain when standing or walking; average pain over pain at rest or current pain; current pain over retrospective pain; worst pain over least pain or current pain. We prioritized sensory over affective measures, and self‐report over observer‐report pain measures.

1b. Proportion of participants in pain postoperatively as defined by the authors of included studies.

2. Behavioural recovery* (defined as: resumption of performance of tasks and activities).

Where multiple measures were used, we made the following decisions in prioritizing measures:

SF‐36 physical function (Ware 2000);
Nottingham Health Profile: Physical mobility (Hunt 1983);
Barthel Index (Mahoney 1965);
Western Ontario and McMaster Osteoarthritis Index (WOMAC) functional status (Bellamy 1988).

Secondary outcomes

1 Negative affect*

Where multiple measures were used, we used the following hierarchy when deciding which measures to use in meta‐analysis:

State Trait Anxiety Inventory (STAI) state (Spielberger 1983);
STAI trait (Spielberger 1983);
Profile of Mood States (POMS) tension/anxiety (McNair 1971);
POMS global (McNair 1971);
Multiple Affect Adjective Check List (MAACL) Anxiety/fear (Zuckerman 1965);
MAACL total (Zuckerman 1965);
Mood Adjective Checklist (MACL) (Radloff 1968);
Hospital Anxiety and Depression Scale (HADS) anxiety (Zigmond 1983);
HADS depression (Zigmond 1983);
General Health Questionnaire 28 (Goldberg 1978);
Perceived Stress Scale (Cohen 1983);
Hospital Anxiety Scale (Lucente 1972);
SF‐36 mental health (Ware 2000);
Nottingham Health Profile: Emotional Reaction (Hunt 1983);
Psychologic Global Well‐being Scale (Dupuy 1984);
BSKE (EWL) (Befindlichkeitsskalierung durch Kategorien und Eigenschaftswörter): Psychological Global Well‐being/mood (Janke 1994);
Structured interview: Modified Present State Examination schedule (Tait 1982) and the Diagnostic and Statistical Manual of Mental Disorders, third edition (DSM‐III) (APA 1980).

2. Length of stay in hospital (days)

*For the outcomes of behavioural recovery and negative affect we included only studies that used measures with published psychometric properties, including reliability and validity. We recorded the timing of outcome assessment.

Search methods for identification of studies

Electronic searches

We searched the Cochrane Central Register of Controlled Trials (CENTRAL 2014, Issue 5); MEDLINE (Ovid SP) (1950 to 4 May 2014); EMBASE (Ovid SP) (1982 to 4 May 2014); PsycINFO (Ovid SP) (1982 to 4 May 2014); CINAHL (EBSCOhost) (1980 to 4 May 2014); Dissertation Abstracts and ISI Web of Science (1946 to 4 May 2014). We reran the search on 7 July 2015; the additional studies identified (after screening titles and abstracts to exclude any obviously irrelevant studies) are listed under Characteristics of studies awaiting classification.

We used the following subject search terms for searching the databases:

`psychological preparat*', education, information, instruction, cognitive interven*, `cognitive behavio?ral therapy', `cognitive therapy', `behavio*ral therapy', hypnosis, relaxation, guided imagery, surgery, operat*, surgical procedure, general an*esthetic, elective surgery, cholecystectomy, hysterectomy, hernia repair, herniorrhaphy, hernioplasty, joint replacement surgery, arthroplasty.

We combined our subject search terms with the Cochrane highly sensitive search strategy for identifying RCTs as suggested in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). The full search strategies are provided in the Appendices (Appendix 1 for CENTRAL, The Cochrane Library; Appendix 2 for MEDLINE (OvidSP); Appendix 3 for EMBASE (OvidSP); Appendix 4 for CINAHL (EBSCOhost); Appendix 5 for ISI Web of Science).

Searching other resources

We searched the reference lists of relevant papers for additional sources where references were provided in the English language. We contacted the authors of relevant studies to identify unpublished studies and dissertations.

We did not limit the search by language or publication status. Where papers were in a non‐English language, we asked a speaker of that language to screen the paper. A member of the review team went over the screening, checking each decision with the screener's description of what happened in the paper. Where the paper was deemed to fit the review criteria, if a member of the review team spoke the language, that individual extracted the data, with a second member of the review team (RP) then checking, by discussion with the first extractor, that decisions made and data extracted were correct. Where no member of the review team spoke the language of the paper, we gained English translations and extracted data in the same way as for English language papers.

Data collection and analysis

Selection of studies

One review author (RP) checked titles and abstracts of retrieved studies to exclude obviously irrelevant reports. A small, random sample was double‐checked by a second researcher (research assistant Yvonne Cooper, or authors MU and JB). Where the title and abstract indicated that a paper had the potential to fit inclusion criteria, copies of the trial were independently assessed for inclusion by two researchers (RP and one other member of the team: research assistant Louise Pike or authors MU, AM, CV, JB, NS, MJ or LBD). We resolved any disagreements by discussion with a third researcher (a member of the authorship team who had not assessed the paper).

Data extraction and management

Two review authors (RP and either MU, AM, JB, CV, MJ, JB, NS or research assistant Louise Pike) independently carried out data extraction using a data extraction form (see Appendix 6). We resolved any disagreement by discussing the matter with a third author (an author who had not previously extracted data from that paper). We extracted the following data:

Study participants: age, gender, total number of participants, location, setting, surgery type.
Study methods: study design, study duration.
Interventions: theoretical nature of intervention, number of intervention groups, specific intervention, intervention details (including delivery method), integrity of intervention, timing of intervention, control groups, usual care description, adherence to intervention and control, attrition rate, loss to follow‐up rate.
Outcomes: outcomes and time points a) collected, and b) reported; outcome definition, author's definition of outcome; measurement tool details (including, for example, upper and lower limits, whether high or low score is good outcome).
Results: number of participants allocated to each intervention group, missing participants, means, standard deviations, proportions, estimate of effect with confidence interval, P value, subgroup analysis information when appropriate (e.g. monitors and blunters (information seekers or avoiders), see Miller 1983).
Study withdrawals or losses to follow‐up.

We described interventions according to whether they contained procedural information, sensory information, behavioural instruction, cognitive intervention, relaxation techniques, hypnosis or emotion‐focused interventions. We coded preparation received by control group participants in the same way.

We (RP) contacted study authors for additional data. We used a two‐stage approach. A first email asked for key information: whether (if not stated) general anaesthesia was used, whether they measured any outcomes not reported in the paper and whether they knew of other (e.g. unpublished) studies. We also asked the study authors if they would be happy for us to contact them with additional questions. If the study authors replied and were happy for us to ask them for further information, we sent them a more detailed email if further information was required.

Assessment of risk of bias in included studies

Two review authors (RP and either MU, AM, JB, CV, MJ, JB, NS or research assistant Louise Pike) independently assessed studies' risk of bias using the tool described in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). This tool requires the review authors to assess risk of bias in the following domains: sequence generation, allocation concealment, blinding of participants, personnel and outcome assessors, incomplete outcome data, selective outcome reporting and other sources of bias. In addition, the review authors noted whether the study used intention‐to‐treat analysis methods (Hollis 1999) (see Appendix 7 for table). We used a single criterion to classify studies as following the intention‐to‐treat principle: participants needed to be kept in the intervention groups to which they were randomized, regardless of the intervention they received (i.e. analysis was not according to per‐protocol or treatment‐received).

Studies with high or unclear risk of bias were to be given reduced weight in the meta‐analysis compared with studies at low risk of bias. We anticipated that meta‐analysis would be restricted to studies at low (or lower) risk of bias, as per Section 8.8.3.1 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We planned to conduct sensitivity analyses to determine whether excluding studies at a high risk of bias affected the results but we did not do so because of the low number of studies deemed to be at `low risk' of bias (see Risk of bias in included studies). We did not expect blinding of participants or personnel administering the intervention because of the interactive nature of the interventions. We described any blinding that was carried out, and rated the risk of bias following the Cochrane guidelines, but high risk of bias for performance bias was not seen to diminish the quality of the paper. We recorded the adequacy of the blinding of outcome assessors (returning data by post was deemed acceptable).

Measures of treatment effect

We performed meta‐analyses according to the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). For dichotomous variables, we planned to calculate risk ratios (RR) with 95% confidence intervals (CI). For continuous data where each study used the same units (i.e. for length of stay), we calculated mean differences for each study and their 95% CIs.

For the postoperative pain and negative affect outcomes a variety of scales were used so we calculated a standardized effect size ‐ the standardized mean difference (SMD), or Hedges' g. We used final scores as standard. However, some studies only reported mean (SD) change from baseline; for these studies we used the difference in mean change scores as the effect size. If no continuous postoperative pain data were available but dichotomous data were presented, we used the log odds ratio instead as the effect size. It was only necessary to do this for one study (Coslow 1998).

If necessary, we reversed the sign of the effect size so that values below zero always indicated that the intervention group was favoured.

Unit of analysis issues

We included only patient‐randomized studies in the meta‐analyses. We reported the results of cluster‐randomized studies as part of the narrative review.

Dealing with missing data

If any necessary data were missing, when we contacted authors about their studies we specifically asked them about the missing data (see Data extraction and management for procedure taken with contacting authors). Missing standard deviations (SD) was a common situation in this review. We were able to calculate (or estimate) standard deviations in a variety of ways. These included calculating the SD from the standard error of the mean (SEM), 95% confidence intervals or from t or F statistics. If the majority of studies in a meta‐analysis still had missing SDs we did not impute these. Otherwise, we used an unweighted average of SDs from other studies in the review. We used identical imputed values for both intervention and control groups.

Assessment of heterogeneity

We considered and tested heterogeneity between trials, where appropriate. To test for gross statistical heterogeneity between all trials, we used Chi² tests for heterogeneity and quantified heterogeneity using the I² statistic (Higgins 2011).

Assessment of reporting biases

We did not plan to assess reporting biases using, for example, funnel plots, because of the probable heterogenous nature of the studies and probable small number of studies appropriate for comparison. However, there proved sufficient studies to examine funnel plots for the overall, `omnibus' analyses.

Data synthesis

We entered quantitative data into Cochrane RevMan 5.3 software and, where appropriate, statistically aggregated the data. We pooled data for all outcomes using an inverse variance approach. We used random‐effects models for all analyses because of expected heterogeneity in interventions and outcomes.

Where it was not possible to pool data, or if summary measures were medians (with range or interquartile range (IQR)), we presented these details in table format and discussed the results.

For each outcome we performed an initial `omnibus' meta‐analysis. We use the term `omnibus' to describe an overall analysis, including all of the psychological preparation interventions (whatever the types of interventions used) and compared these (any psychological preparation intervention) versus controls.

Many studies in the review contained two or more randomized arms. We classified the interventions in each arm separately. To avoid double counting of control groups, for the omnibus analysis we pooled the data in all intervention arms using the standard pooling formula and classified the study as administering any of the interventions included in any of the pooled arms.

The only non‐standard design (i.e. non individually randomized controlled trial) that met the inclusion criteria was a clustered randomized controlled trial design. We narratively synthesized these studies ‐ they were not included in meta‐analysis.

`Summary of findings' table

We included each outcome (postoperative pain, behavioural recovery, negative affect and length of stay) in a `Summary of findings' table (summary of findings Table for the main comparison). For each outcome, the table indicated the effect for the control group and corresponding effect for the intervention group as appropriate, with the number of studies and participants included in analyses. We assessed the quality of the body of evidence for each outcome (postoperative pain, behavioural recovery, negative affect, length of stay) using the GRADE approach, as described in Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). As we only included RCTs, our start point for grading the evidence was `high quality'. We downgraded by one level for serious factors, and two levels for very serious factors in: limitations in design or implementation of studies (risk of bias); indirectness of evidence; heterogeneity or inconsistency of results; imprecision of results; or high likelihood of publication bias.

Subgroup analysis and investigation of heterogeneity

We planned to carry out subgroup analyses to compare trials of high methodological quality with trials of low methodological quality but did not do so because of the small number of studies judged to be at `low risk' of bias (see Risk of bias in included studies).

Following the omnibus analysis we carried out additional separate meta‐analyses corresponding to the seven intervention categories (procedural information, sensory information, behavioural instruction, cognitive interventions, relaxation techniques, hypnosis and emotion‐focused interventions). We divided studies into those with that intervention category only (referred to as `pure' studies, e.g. procedural information only) and those including that intervention category in combination with other intervention types (referred to as `mixed', e.g. procedural information + sensory information + behavioural instruction) and conducted subgroup analyses so that the effect of both all studies including procedural information and of `pure' procedural information studies could be evaluated. For multi‐arm studies, by including only data from relevant arms we were often able to include different data to those included in the omnibus analysis.

Sensitivity analysis

Jüni recommends consideration of the important quality components of a given meta‐analysis when conducting sensitivity analyses (Jüni 2001). We planned to perform sensitivity analyses to evaluate the effect on the overall result of removing trials with low methodological quality (as identified using the Cochrane tool) (Appendix 6), but did not do so because of the small number of studies judged to be at low risk of bias (see Risk of bias in included studies). Low methodological quality studies were those where: a) sequence generation or allocation concealment was judged as high risk or unclear, b) there was no or unclear blinding of outcome assessors, c) incomplete outcome data were not adequately addressed (assessed as high risk or unclear), d) the study appeared to be at risk of selective outcome reporting (high risk or unclear), d) the study did not appear to have been conducted according to intention‐to‐treat (i.e. it was not clear that participants were kept in the group to which they were allocated, regardless of the intervention they received) (high risk or unclear), d) the study appeared to be at risk of other sources of bias (high risk or unclear).

Results

Description of studies

Results of the search

Electronic searches identified 6781 papers; we identified an additional 151 papers through contact with authors and screening reference lists. We removed 1816 duplicate papers, leaving 5116 whose titles and abstracts we screened for broad relevance. This led to us retrieving 827 papers for full screening. We were unable to locate 24 references (2.9% of papers to be retrieved for full screening). See Figure 2 for the flow chart of studies included and excluded from the review.

Figure 2

Study flow diagram.

Included studies

We included 105 studies (from 115 papers) in which 10,302 participants were randomized (see Characteristics of included studies). Sixty‐one papers measured the outcome postoperative pain, 58 length of stay, 50 negative affect and 14 behavioural recovery. We attempted to contact all authors, with the exception of five studies' authors where the study reports were retrieved late in the review process (Barbalho‐Moulim 2011; Done 1998; McGregor 2004; Rajendran 1998; Rosenfeldt 2011). The publication dates of the included studies ranged from 1970 to 2014 and studies were conducted in a wide range of countries (36 in the USA, 13 in the UK, nine in Canada, seven in China, six in Australia, five in the Netherlands, four in Germany, three in Sweden, and one or two studies in each of: Austria, Brazil, Denmark, Egypt, France, India, Iran, Ireland, Italy, New Zealand, Nigeria, Romania, Serbia, Singapore, Spain, Switzerland, Taiwan and Turkey).

The study participants underwent a wide range of surgical procedures. Twenty‐seven studies investigated participants undergoing cardiothoracic surgery (including 17 exclusively containing participants undergoing coronary artery bypass graft surgery). Hip or knee joint surgery was examined in 22 studies (four knee replacement only, 10 hip replacement only, eight both hip and knee replacement surgery). Seven studies considered cholecystectomy, seven hysterectomy and two breast surgery. The following procedures were considered in a single study each: urinary diversion surgery, colorectal resection, laparoscopic tubal ligation, minimally invasive radio‐guided parathyroidectomy, rectal cancer surgery, periodontal surgery, inguinal hernia and gastric bypass surgery. Thirty‐one studies addressed a mixture of procedures and one study did not state the surgical procedure(s).

The included studies used a range of intervention components, and intervention content was rarely `pure', consisting of a single intervention. Procedural information was reported in 59 interventions (`pure' procedural information content in eight), sensory information in 38 (`pure' sensory information in one), behavioural instruction in 71 (`pure' in 28), cognitive interventions in 27 (`pure' in eight), relaxation techniques in 35 (`pure' in 13), hypnosis in six (`pure' in one) and emotion‐focused interventions in 12 (`pure' in one). Studies generally contained fairly small sample sizes.

We found that control group content was generally poorly reported. Pure procedural information content was reported in 17 control groups, pure behavioural instruction in 11 and combinations of interventions in 23 studies. Fifty‐six studies provided insufficient information for us to categorize control content ‐ for example, authors frequently described the control group as consisting of `usual care' without describing what usual care was. It is highly likely that intervention content is missing from these descriptions because if participants were provided with absolutely no procedural information or behavioural instruction prior to their surgery they would not know when to arrive for their surgery or what to do (e.g. when to fast prior to their anaesthetic).

As per our protocol (Powell 2010), we did not extract funding sources from papers in this review.

Excluded studies

We excluded 674 papers on full screening of retrieved papers. Details of 27 key excluded papers are provided (Anderson 1987; Blay 2005; Boore 1978; Burton 1991; Burton 1994; Croog 1994; Domar 1987; Enqvist 1995; Eremin 2009; Huang 2012; Johnson 1978a; Lengacher 2008; Liu 2013; Manyande 1995; Manyande 1998; Mitchell 2000; Montgomery 2002; Montgomery 2007; Sheard 2006; Shelley 2009; Stergiopoulou 2006; Sugai 2013; Surman 1974; Timmons 1993; Voshall 1980; Wang 2002; Wells 1986). For further details of the excluded studies see the Characteristics of excluded studies.

Ongoing studies

We did not include two papers, Jong 2012 and Hansen 2013, as the research was complete but authors were reluctant to share study details with us prior to publication.

Studies awaiting classification

On full screening of retrieved papers in May 2014, two provided insufficient information to determine whether or not they met the review's inclusion criteria and our attempts to contact the authors for further information were not successful (Johansson 2007; Lookinland 1998).

We reran the searches in July 2015. These searches identified a further 753 papers. On removing duplicates across databases, 614 papers remained. We checked these references for overlap with searches previously conducted and identified a further 96 duplicates. These searches therefore identified 518 new papers. RP screened the titles and abstracts of these papers for relevance (with JB checking a randomly selected 5% of titles and abstracts); 482 papers were excluded. The remaining 36 papers appear to potentially have relevance and should be retrieved and screened in detail when this review is updated (Akinci 2015; Angioli 2014; Attias 2014; Bergin 2014b; Calsinski Assis 2014; Chevillon 2014; Chow 2014; Dathatri 2014; Eckhouse 2014; El Azem 2014; Ellett 2014; Foji 2015; Fraval 2015; Furuya 2015; Gade 2014; Gillis 2014; Gyulaházi 2015; Hansen 2015; Henney 2014; Heras 2014; Hoppe 2014; Huber 2015; Johansson 2007; Kol 2014; Lai Ngor 2014; Louw 2014; Mohammadi 2014; Novick 2014; Paul 2015; Rolving 2014; Saleh 2015; Shahmansouri 2014; Umpierres 2014; Van Acker 2014; West 2014; Würtzen 2015; Xin 2015). Details of these papers can be found in Characteristics of studies awaiting classification.

Risk of bias in included studies

Details of `Risk of bias' assessments for each study are provided in Characteristics of included studies, with summaries across studies being presented in Figure 1 and Figure 3. We did not expect many studies in this review to be rated as `low risk' for performance bias. However, even ignoring this category, only three studies received `low risk' ratings on all other items (Crowe 2003; Goodman 2008; Mahler 1998). We therefore did not carry out the planned sensitivity analyses to compare meta‐analyses including only high quality, `low risk' studies with analyses including all available data, nor the planned subgroup analyses to compare findings of high quality, `low risk' studies with findings of low quality, `high risk' studies.

Figure 3

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

Allocation

As shown in Figure 1, we rated very few studies as ‘high risk’ for random sequence generation. This is because, following our protocol (Powell 2010), we only included RCTs ‐ where a non‐random approach was described (such as alternation), or where there was no mention of randomization in the study description, studies were excluded. This meant that studies that could be rated as `high risk' would usually be excluded from the review. Despite this inclusion criterion, the randomization procedure was sufficiently described to rate the study as ‘low risk of bias’ in only about half of studies (51 of 105, see Figure 3) – giving insufficient information to ascertain the procedure used for allocation was common.

Clear descriptions of allocation concealment were even more rare, with only 16 (of 105) studies being judged as `low risk of bias' (Beaupre 2004; Crowe 2003; Furze 2009; Giraudet 2003; Goodman 2008; Guo 2012; Hoogeboom 2010; Leserman 1989; Mahler 1995; Mahler 1998; Neary 2010; O'Connor 2014; Oosting 2012; Ridgeway 1982; Schwartz‐B'tt 1994; Shuldham 2002) – this was an aspect that was simply not mentioned in most studies. Awarding the designation of `low risk' tended to depend on information that we were able to gain directly from authors themselves.

Blinding

Studies' poorest risk of bias ratings were for performance bias: blinding of participants and personnel. We rated most studies in the review as being at ‘high risk of bias’ in this category. We anticipated this and did not expect to see blinding of participants or of the personnel administering the intervention because many psychological interventions are interactive in nature. It was therefore rare to find a study where the person administering the intervention could be blind to the participant’s group allocation and, if participants were fully informed about the nature of the study, they would also tend not to be blinded to treatment condition. One study did report blinding of both participants and personnel, using an intervention delivered via a website (Neary 2010). Studies rated as `unclear' for performance bias (n = 5: Barlési 2008; DeLong 1970; Enqvist 1997; Goldsmith 1999; Pellino 2005) used a limited range of intervention formats, administered on paper (Barlési 2008), by audiorecording (DeLong 1970; Enqvist 1997), information on paper and tape (Pellino 2005), or via a website (Goldsmith 1999).

Blinding of outcome assessment (to avoid detection bias) was feasible in the types of studies we assessed – by ensuring that the person administering postoperative measures was blind to allocation. However, this was frequently not reported, allowing us to rate 42 (of 105) studies as `low risk of bias' (Beaupre 2004; Bergmann 2001; Bitterli 2011; Broadbent 2012; Crowe 2003; Doering 2000; Ferrara 2008; Fortin 1976; Furze 2009; Gocen 2004; Gonzales 2010; Goodman 2008; Griffin 1998; Guo 2012; Hart 1980; Hoogeboom 2010; Hulzebos 2006a; Hulzebos 2006b; Johnson 1978b; Johnson 1985; Lam 2001; Langer 1975; Lévesque 1984; Lilja 1998; Lin 2005; Mahler 1995; Mahler 1998; McDonald 2001; McDonald 2004; McDonald 2005; Neary 2010; Oetker‐Black 2003; Oosting 2012; Parthum 2006; Reading 1982; Seers 2008; Shuldham 2002; Watt‐Watson 2000; Watt‐Watson 2004; Wilson 1981; Zhang 2012; Ziemer 1982).

Incomplete outcome data

Attrition was frequently poorly reported in the studies, leading to ratings of `unclear risk of bias'. Sufficient information was provided, demonstrating good practice, in 37 `low risk' studies (Barlési 2008; Bergin 2014a; Chaudhri 2005; Chumbley 2004; Coslow 1998; Crowe 2003; Doering 2000; Ferrara 2008; Fortin 1983; Furze 2009; Giraudet 2003; Gocen 2004; Gonzales 2010; Goodman 2008; Greenleaf 1992; Guo 2012; Hart 1980; Hawkins 1993; Hulzebos 2006a; Lam 2001; Langer 1975; Leserman 1989; Lin 2005; Liu 2004; Mahler 1995; Mahler 1998; Miró 1999; Omlor 2000; Osinowo 2003; Ridgeway 1982; Schmitt 1973; Vukomanović 2008; Watt‐Watson 2004; Wells 1982; Wilson 1981; Yang 2012; Zhang 2012).

Selective reporting

The proportion of studies rated as `low risk' for selective reporting was low (20 of 105) (Bergin 2014a; Cheung 2003; Crowe 2003; D'Lima 1996; Doering 2000; Fortin 1976; Goodman 2008; Hoogeboom 2010; Hulzebos 2006a; Hulzebos 2006b; Langer 1975; Leserman 1989; Levesque 1977; Mahler 1998; McDonald 2001; McDonald 2005; Oosting 2012; Ridgeway 1982; Vukomanović 2008; Wilson 1981). Thirty‐three were designated `high risk'. This may reflect our strict application of the Cochrane guidelines (Cochrane Handbook for Systematic Reviews of Interventions, Table 8.5.d.; Higgins 2011), which stated that for a judgement of `low risk' either "the study protocol is available and all of the study's pre‐specified (primary and secondary) outcomes that are of interest in the review have been reported in the pre‐specified way" or "the study protocol is not available but it is clear that the published reports included all expected outcomes, including those that were pre‐specified (convincing text of this nature may be uncommon)". It was extremely rare to find studies with reference to protocol documents, and only a very small number of trials had been registered. To provide a rating of `low risk' we tended to be dependent on authors responding to our queries as to whether any outcomes were measured but not reported.

Other potential sources of bias

We evaluated studies for analysis according to the principles of intention‐to‐treat (whether participants were analysed in the group to which they were allocated, regardless of the intervention they received). This was often not reported, leading to the evaluation of 31 studies as `low risk of bias' (Beaupre 2004; Bergin 2014a; Coslow 1998; Crowe 2003; Doering 2000; Fortin 1976; Furze 2009; Giraudet 2003; Goodman 2008; Greenleaf 1992; Guo 2012; Hoogeboom 2010; Hulzebos 2006a; Kulkarni 2010; Lam 2001; Lauder 1995; Leserman 1989; Mahler 1995; Mahler 1998; Oetker‐Black 2003; Oosting 2012; Parthum 2006; Postlethwaite 1986; Reading 1982; Ridgeway 1982; Schmitt 1973; Shuldham 2002; Vukomanović 2008; Watt‐Watson 2000; Watt‐Watson 2004; Wilson 1981), 12 as `high risk' and the remainder as `unclear'.

Most studies were not found to have additional sources of bias, with concerns being raised for seven studies rated as `high risk' and 27 as `unclear'.

Effects of interventions

A summary of key findings, with quality gradings, is provided in summary of findings Table for the main comparison.

Findings by outcome

Primary outcomes

1.Postoperative pain

Studies included in meta‐analysis

Sixty‐one studies assessed the outcome postoperative pain. It was possible to include data for 38 studies (36% of 105 studies) (Barbalho‐Moulim 2011; Bergin 2014a; Bitterli 2011; Cheung 2003; Coslow 1998; D'Lima 1996; Doering 2000; Fortin 1983; Giraudet 2003; Gocen 2004; Goldsmith 1999; Gonzales 2010; Gräwe 2010; Griffin 1998; Guo 2012; Heidarnia 2005; Lam 2001; Lauder 1995; Leserman 1989; Levin 1987; Lin 2005; Ma 1996; McDonald 2001; McDonald 2004; McDonald 2005; McGregor 2004; Miró 1999; Neary 2010; Omlor 2000; Pellino 2005; Postlethwaite 1986; Reading 1982; Ridgeway 1982; Roman 2012; Schwartz‐B'tt 1994; Seers 2008; Watt‐Watson 2000; Zieren 2007), with analysis of 2713 participants' data (26% of 10,302 participants randomized across all studies), in the omnibus meta‐analysis, which included studies comparing any intervention versus control (Analysis 1.1; Figure 4). As a variety of scales were used to measure postoperative pain, we used standardized scores to pool data using the SMD (Hedges' g). Higher scores indicate higher pain; effect scores below zero indicate that the intervention group had lower pain. Overall, the pooled effect size (SMD) was ‐0.20 (95% confidence interval (CI) ‐0.35 to ‐0.06), suggesting a statistically significant effect in favour of the intervention groups. There were, however, high levels of statistical heterogeneity between studies (I² statistic = 71%).

Figure 4

Pain (any psychological preparation intervention versus control). B: behavioural instruction; C: cognitive interventions; E: emotion‐focused interventions; H: hypnosis; P: procedural information; R: relaxation; S: sensory information.

One study appeared to be an outlier (Ma 1996). We assumed statistics in the paper to represent mean and standard deviation as the notation "x bar +/‐ s" was used but this was not explicitly stated and it is possible that `s' represented standard error. Excluding this study did not affect the interpretation of the outcome, but reduced the observed statistical heterogeneity (I² statistic = 53%). Excluding the single study where an effect size had been derived from categorical data, Coslow 1998, also had no effect on the results.

Subsequent forest plots show the results for the individual types of intervention (Analysis 2.1; Analysis 3.1; Analysis 4.1; Analysis 5.1; Analysis 6.1; Analysis 8.1; no studies used the intervention hypnosis). Most studies included more than one intervention type and, except for behavioural instruction and relaxation, there were no more than two `pure' studies that included just that particular intervention type. This makes it very difficult to separate the effect of a particular intervention category from other types of intervention also administered. For most intervention types the pattern of results was similar to the omnibus analysis and results for `pure' and `mixed' studies were also similar. The analyses for behavioural instruction showed a somewhat different pattern, however, with a relatively consistent effect size for the `pure' behavioural instruction studies suggesting no difference between intervention and control (SMD 0.01, 95% CI ‐0.19 to 0.21, I² statistic = 27%). The meta‐analysis results for individual intervention types were statistically significant for the meta‐analysis of the two studies including cognitive intervention (Cheung 2003; Ridgeway 1982; SMD ‐0.34, 95% CI ‐0.68 to ‐0.01, I² statistic = 0%; Analysis 5.1.1) and the meta‐analysis of seven `pure' relaxation studies (Gonzales 2010; Leserman 1989; Levin 1987; Ma 1996; Miró 1999; Roman 2012; Seers 2008; SMD ‐0.71, 95% CI ‐1.29 to ‐0.13, I² statistic = 87%; Analysis 6.1.1). No data for studies investigating postoperative pain after hypnosis could be included in the meta‐analyses. The funnel plot showed no clear evidence of publication bias.

Studies not included in meta‐analysis

Twenty‐three studies addressing the postoperative pain outcome did not contain data appropriate for meta‐analysis (Chumbley 2004; Daltroy 1998; Dewar 2003; Enqvist 1997; Ferrara 2008; Field 1974; Gilbey 2003; Hawkins 1993; Johnson 1978b; Johnson 1985; Kulkarni 2010; Lilja 1998; Liu 2004; Oetker‐Black 2003; Parthum 2006; Perri 1979; Shelley 2007; Shuldham 2002; Vukomanović 2008; Watt‐Watson 2004; Wells 1982; Wijgman 1994; Ziemer 1982) (Table 1). Three of these were not eligible for meta‐analysis as they reported cluster‐randomized trials (Chumbley 2004; Parthum 2006; Vukomanović 2008). Median scores were provided in two studies (Kulkarni 2010; Wijgman 1994); most studies in this group lacked sufficient detail to be entered into meta‐analysis.

Table 1. Findings of studies that examined the outcome pain but could not be included in meta‐analyses

Author, year	Surgery type and sample size (randomized)	Intervention categories	Pain measure(s) The first measure listed is that prioritized in this review	Pain findings (as available)
Chumbley 2004	Mixed: surgeries that would receive PCA routinely N = 246	Intervention 1: Behavioural instruction (delivered in leaflet) Intervention 2: Behavioural instruction (delivered in interview)	1) Visual analogue scale (VAS) days 1 to 5 post‐surgery 2) Word rating on 5‐point scale; days 1 to 5 post‐surgery	Cluster‐randomized VAS day 1 postoperatively mean (95% CI): Control: 3.7 (2.93 to 4.45); Intervention 1: 2.8 (2.04 to 3.56); Intervention 2: 3.2 (2.43 to 6.21). ANOVA, repeated measures: for VAS pain scores, between‐groups effect: F = 1.88, P value = 0.23
Daltroy 1998	Total hip or knee arthroplasty N = 12	Procedural and sensory information	Day 4 post‐surgery Measure not clearly described, assume same as preoperatively: mean of 3 x 5‐point scales assessing pain at night, resting and when active	Intervention did not affect pain in general linear model (P value = 0.16)
Dewar 2003	Mixed surgeries N = 254	Procedural information, behavioural instruction, cognitive intervention, relaxation	Evening after surgery (day 0) Brief Pain Inventory: numerical rating scale from 0 to 10	Control n = 118; intervention n = 104 No significant difference
Enqvist 1997	Breast reduction N = 50	Relaxation, hypnosis	Days 1 to 5 post‐surgery, measured with `10‐degree VAS’. Not clear exactly what was asked, or if measured once in this period or daily	Control n = 25; intervention n = 23 No significant differences
Ferrara 2008	Total hip replacement N = 23	Behavioural instruction	15 days and 4 weeks post‐surgery: VAS Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) subscale	Control n = 12; intervention n = 11 VAS pain scores: significantly lower in intervention group at 4 weeks (not at 15 days apparently)
Field 1974	Mixed orthopaedic surgery N = 60	Procedural information, hypnosis	Between 2 and 7 days post‐surgery; no further information	Control n = 30; intervention n = 30 No significant difference
Gilbey 2003	Total hip arthroplasty N = 76	Behavioural instruction	3 weeks post‐surgery Pain domain of WOMAC	Control n = 25; intervention n = 32 Significant difference (P value < 0.01) for total WOMAC (pain, physical function and stiffness) and physical function domain. Reports surgery had such beneficial effect on pain that impact of intervention only marginal.
Hawkins 1993	Gynaecological surgery N = 60	Behavioural instruction	48 hours post‐surgery: VAS of average pain; categorical scale (5 categories from no pain to unbearable pain); nurse ratings of pain (collected hourly pain reports when not sleeping for first 48 hours after surgery)	Control n = 40 (standard care and attention control); intervention n = 20 No significant differences (VAS ANOVA F = 0.06, df = 2, P value = 0.93)
Johnson 1978b	Sample 1: cholecystectomy, N = 81 Sample 2: inguinal hernia repair, N = 68	Intervention 1: ‘Instruction’: Behavioural instruction (deep breathing, coughing, leg exercises) Intervention 2: ‘Procedure information’: focus procedural information, also some sensory information and behavioural instruction Intervention 3: `Sensation information’: focus: sensory information, also some procedural information and behavioural instruction 2 x 3 factorial design: no instruction/instruction (Intervention 1; no information/information (Interventions 2 and 3)	Pain: days 1, 2 and 3 post‐surgery: intensity of sensations on 10‐point scale Scores totaled over the 3 days in analysis	Sample 1 No main effect of condition Sample 2 MANOVA with DVs pain and distress of pain sensation: for first postoperative day: significant main effects for information level (F(4, 104) = 2.55, P value < 0.05), trend for an effect for instruction (F(2, 52) = 3.07, P value = 0.055), but only a main effect for distress scores reported (no univariate findings reported for pain – so seems no significant effects)
Johnson 1985	Abdominal hysterectomy N = 199	Intervention 1: Procedural and sensory information Intervention 2: ‘Cognitive‐coping technique’ – cognitive intervention Intervention 3: ‘Behavioural‐coping technique’ – behavioural instruction 2x3 factorial design: no information/information (Intervention 1); no coping technique/coping technique (Interventions 2 and 3)	Day 3 post‐surgery Pain scale from 1 to 10	MANOVA, controlling for covariates, with various outcomes including pain: ‘significant’ at P value < 0.10: coping technique, F (16, 286) = 1.59, P value =0.07. However, pain does not appear to be one of the outcomes responsible for this.
Kulkarni 2010	Major abdominal surgery N = 80	Intervention 1: Behavioural instruction (deep breathing training) Intervention 2: Behavioural instruction (incentive spirometry) Intervention 3: Behavioural instruction (specific inspiratory muscle training)	Pain (no information of how measured/when)	Control n = 17; intervention 1 n = 17; intervention 2 n = 15; intervention 3 n = 17. Median pain score for all groups is 3 (no ranges/IQRs)
Lilja 1998	Breast cancer (BC) surgery N = 46 Total hip replacement (THR) N = 55	Procedural information, behavioural instruction	First 3 days post‐surgery: VAS	Control: n = 22, mode = 1 (BC day 1); intervention n = 22 No significant differences groups for either BC or THR patients (analysed separately)
Liu 2004	Mixed orthopaedic surgery N = 74	Cognitive intervention	Pain: 0 to 10 VAS; timing not stated	Control n = 35, mean (SD)= 2.5 (0.52); intervention n = 39, mean = 2.85 (0.33) Significant difference (t = 2.61, P value < 0.05). Discussion: authors state “patients from the experimental group…had…low scores on pain compared to the control group with statistical significance” (p5). This appears to be at odds with mean scores, suggesting error in paper.
Oetker‐Black 2003	Total abdominal hysterectomy N = 108	Behavioural instruction, cognitive intervention, relaxation	Day 1 post‐surgery: VAS At discharge: bodily pain (Health Status Questionnaire)	No significant differences (VAS: t(1,105) = ‐0.54, P value = 0.591)
Parthum 2006	Cardiac surgery N = 93	Procedural information, sensory information, behavioural instruction	1. Pain intensity: VAS as part of modified McGill Pain Questionnaire (MPQ), day 1 postoperative and retrospective rating of pain while on ICU 2. Proportion of patients in pain postoperatively (cut off: VAS > 3 on above measures)	Cluster‐randomized Control n = 36, median (VAS current, at rest) = 4.0. Intervention: n = 37, median = 3.0 No significant differences between groups
Perri 1979	Vaginal hysterectomy N = 26	Relaxation	Self report. 1 and 3 days postoperation; ‘McGill‐Melzack Pain Questionnaire’ Observed. 1 and 3 days postoperation – observed pain behaviour – Chambers‐Price Rating Scale for Pain	Control n = 13; intervention mean = 13. No significant differences between groups (P value < 0.05)
Shelley 2007	Coronary artery bypass surgery N = 90	Cognitive intervention	At discharge (4 days post‐surgery): 10 cm VAS	Control n = 43; intervention n = 37 Significant interaction between group, self efficacy and external health locus of control (F(1,71) = 4.06, P value < 0.05). Post hoc analysis: trend‐level effects: smaller increase in pain for prepared patients than controls if high external health locus of control and low self efficacy. Matched control appraisal patients: increased pain in intervention group compared with controls (controlling for baseline pain).
Shuldham 2002	Coronary artery bypass surgery N = 356	Procedural information, behavioural instruction	Questionnaires presented on day 3 post‐surgery (or 3rd day after transfer to ward if still in intensive care unit on day 3 post‐surgery) Composite measure (including VAS, body map and categorical rating scale), authors used VAS in analysis	No significant differences (using Mann‐Whitney U): U = 10,197.5; Z = ‐0.72, P value = 0.47
Vukomanović 2008	Total hip arthroplasty N = 45	Procedural information, behavioural instruction	VAS at discharge: pain at rest and movement	Cluster‐randomized Control n = 20, mean (SD) = 6.2 (14.95); Intervention n = 20, mean (SD) = 3.95 (13.08) No significant difference in pain
Watt‐Watson 2004	Coronary artery bypass surgery N = 406	Behavioural instruction, cognitive intervention	Days 1 to 5 post‐surgery: McGill Short‐form. Scores: Present Pain Intensity: most severe pain in previous 24 hours Pain Rating Index (sensory, affective and total); Numerical Rating Scale (on moving and worst pain in previous 24 hours)	No main effect of group
Wells 1982	Cholecystectomy N = 12	No control group Intervention 1: ‘Control’: Sensory information; behavioural instruction Intervention 2: (do not appear to receive ‘control’ intervention) Relaxation	Rated on 10 cm line on evening on day of surgery, and days 1 and 2 post‐surgery	Intervention 1: n = 6, mean (SD) eve of operation = 5.4 (3.39); intervention 2: n = 6, mean (SD) = 5.65 (1.6) No main effect for treatment (F(1,7) = 3.0, P value = 0.13), time (F(7,2) = 3.3, P value = 0.07) or interaction between treatment and time (F(2,4) = 1.0, P value = 0.4)
Wijgman 1994	Total knee arthroplasty N = 64	No control group Intervention 1: Procedural information Intervention 2: Behavioural instruction	2, 5, 7, 10, 14 days post‐surgery and at discharge. VAS where 100 = worst pain	Overall n at day 2 = 63. Medians (IQRs) presented in Figure 1, not clear. No significant differences between groups
Ziemer 1982	Gynaecologic or gastrointestinal N = 111	Intervention 1: Sensory information Intervention 2: Sensory information, behavioural instruction, cognitive intervention, relaxation	2 to 4 days post‐surgery: 5‐point pain intensity rating scale	Control n = 40; intervention 1 n = 34; intervention 2 n = 37 Focus: correlation of pain with coping scales

ANOVA = analysis of variance

BC = breast cancer

F = F statistic (ANOVA)

ICU = intensive care unit

IQR = interquartile range

MANOVA = multivariate analysis of variance

MPQ = McGill Pain Questionnaire (Melzack 1975)

N = number of participants in sample

PCA = patient‐controlled analgesia

SD = standard deviation

THR = total hip replacement

VAS = visual analogue scale

Fourteen of these studies reported no statistically significant differences between intervention and control conditions (Chumbley 2004; Daltroy 1998; Dewar 2003; Enqvist 1997; Field 1974; Gilbey 2003; Hawkins 1993; Lilja 1998; Oetker‐Black 2003; Parthum 2006; Perri 1979; Shuldham 2002; Vukomanović 2008; Watt‐Watson 2004). A further two studies did not clearly report postoperative pain findings, but this appears to be because comparisons were not significant (Johnson 1978b; Johnson 1985). These studies used a range of intervention techniques: procedural and sensory information (one study), procedural information and behavioural instruction (three), procedural information, sensory information, behavioural instruction (two); behavioural instruction (three); procedural information, behavioural instruction, cognitive interventions, relaxation techniques (one), procedural information, hypnosis (one); procedural and sensory information/cognitive interventions/behavioural instruction (one); behavioural instruction, relaxation techniques, cognitive interventions (one); relaxation (one); behavioural instruction, cognitive interventions (one); relaxation techniques, hypnosis (one).

Less clear findings were reported in two studies. Ferrara 2008 reported that postoperative pain scores were significantly lower in the intervention group than the control group at four weeks after surgery, but a comparison at 15 days was not clearly reported – it is possible that authors were choosing to not report non‐significant findings. Shelley 2007 reported a significant interaction between intervention group, self‐efficacy and external health locus of control (EHLC), but post‐hoc analyses revealed only trend level effects, such that intervention participants had a smaller pain increase than controls if they had high EHLC and low self‐efficacy. Participants with high self‐efficacy and high EHLC, or low self‐efficacy and low EHLC, reported increased pain for intervention participants compared with controls.

Five studies' findings were difficult to interpret in the context of our review questions. Kulkarni 2010 reported that median postoperative pain scores for all three groups was `3', but no information was provided as to how pain was measured, and information about range/interquartile range or analyses were reported. Liu 2004's findings were puzzling because the authors stated in their Discussion that intervention participants had significantly lower pain scores compared with the control group, but the mean scores provided in the Results section suggested their findings were in the opposite direction. Wells 1982 and Wijgman 1994 had no control group – in each case two different interventions were compared, meaning that it was not possible to determine what the effect of the intervention was over standard care or an attention control. Ziemer 1982 did not report postoperative pain as an outcome and instead focused on how pain correlated with coping scales.

Summary: postoperative pain

In summary, the pattern of evidence from the meta‐analyses suggests that psychological preparation may reduce postoperative pain in the first month after surgery, although this finding should be treated with caution since it is based on pooling studies with diverse types of psychological interventions and because the size of the pooled effect (‐0.20) would generally be considered of low magnitude (Cohen 1988). Of the narratively synthesized studies, most found no significant difference between intervention and control groups. It is of interest that, while none of these studies contained `pure' behavioural instruction, 12 of the 16 studies reporting non‐significant differences contained behavioural instruction as a component (Chumbley 2004; Dewar 2003; Hawkins 1993; Johnson 1978b; Johnson 1985; Lilja 1998; Oetker‐Black 2003; Parthum 2006; Shuldham 2002; Vukomanović 2008; Watt‐Watson 2004; Wijgman 1994). This would be consistent with the meta‐analysis findings suggesting that behavioural instruction does not impact postoperative pain. However, similarly to the studies in the meta‐analyses, there is a high degree of heterogeneity in these studies in terms of the types of surgery and intervention content. Due to the high heterogeneity, and the high number of studies reporting sufficient methodological details to ascertain risk of bias, we downgraded the overall quality of evidence for the outcome postoperative pain by two points to `low' (see summary of findings Table for the main comparison).

2.Behavioural recovery

Fourteen studies (13% of 105 studies) were included that measured a behavioural recovery outcome, in which 1135 participants were randomized (11% of 10,302 participants randomized across all studies): D'Lima 1996; Ferrara 2008; Fortin 1976; Gilbey 2003; Heidarnia 2005; Hoogeboom 2010; Lévesque 1984; Mahler 1998; McGregor 2004; Oetker‐Black 2003; Oosting 2012; Ridgeway 1982; Watt‐Watson 2004; Zieren 2007. One study was cluster‐randomized and therefore not eligible for inclusion in meta‐analysis (Lévesque 1984). Suitable continuous data for meta‐analysis were available in only three studies (Mahler 1998; McGregor 2004; Zieren 2007), and dichotomous data in two studies (Fortin 1976; Oosting 2012). As there was also a range of different behavioural recovery outcome measures, we decided that a narrative synthesis would be more appropriate for this outcome than meta‐analysis.

Behavioural recovery findings are summarized in Table 2. Behavioural instruction was a common intervention type for these studies (included in all interventions except that of Ridgeway 1982). Statistically significant beneficial effects of the intervention over control conditions were reported in five studies (Fortin 1976; Gilbey 2003; Heidarnia 2005; McGregor 2004; Oetker‐Black 2003). Ridgeway 1982 reported a trend effect, such as that participants in their cognitive intervention group were carrying out more household activities (P value = 0.10), and Watt‐Watson 2004 found mixed results: behaviours of deep breathing and coughing were experienced as being less affected by pain in the intervention group, but no significant differences were seen for other activities (general activities, sleep, walking). Differences between groups were not significant in three studies (Hoogeboom 2010; Lévesque 1984; Mahler 1998), and analyses were not reported in three studies (Ferrara 2008; Oosting 2012; Zieren 2007).

Table 2. Findings of studies that examined the outcome behavioural recovery

Author, year

Surgery type and sample size (randomized)

Intervention categories

Behavioural recovery measure(s)

The first measure listed is that prioritized in this review

Behavioural recovery findings (as available)

D'Lima 1996

Total knee replacement

N = 30

Intervention 1: Behavioural instruction

Intervention 2: Behavioural instruction

3 weeks post‐surgery

Function scale from Hospital for Special Surgery Knee Rating; high score = better function

Control mean = 35, n= 10

Intervention 1 mean = 32, n = 10

Intervention 2 mean = 30.5, n = 10

"in the immediate postoperative period both exercise groups showed a steeper decline in function than the control group"; statistics not provided

Ferrara 2008

Total hip replacement

N = 23

Behavioural instruction

15 days and 4 weeks post‐surgery:

Disability (Barthel Index) (high scores: less disabled)

Functional status (from WOMAC); high scores = worse function

Intervention n = 11, control n = 12

No data/findings reported for these time points (study focus: 3 months postoperation)

Fortin 1976

Herniorraphy, cholecystectomy, intra‐pelvic surgery (primarily hysterectomies)

n = 69

Procedural information, behavioural instruction

Day 2 postoperation: "inpatient ambulatory activity" (IAA). Ability to do physical activities at hospital in immediate postoperative period – e.g. movements in bed, get up, walk. Higher level (max = 3) = can do more.

Day 10 post‐surgery: ‘Activities of Daily Living’ (ADL). Capacity to perform tasks appropriate to normal life at home. Higher level (max = 3) = more independent.

Authors combined levels 1 and 2 in analysis

2 days IAA: Intervention n at level 3/total N = 27/37, control group = 5/32

10 days ADL: Intervention n at level 3/total N = 27/36, control group = 8/31

Better function in intervention than control group with both assessments

Analysing 29 matched pairs, significant difference at 2 and 10 days (P value < 0.01 for each, Wilcoxon matched pairs)

Full sample: also significantly different at both time points (Mann‐Whitney U, P value < 0.05 for each)

Gilbey 2003

Total hip arthroplasty

N = 76

Behavioural instruction

Week 3 post‐surgery:

Physical function domain of WOMAC

Intervention n = 32; control n = 25

Means/SDs presented only for total WOMAC scale, not for physical function domain. Significant difference (P value < 0.01) for physical function domain reported (intervention group scoring better).

Heidarnia 2005

Coronary artery bypass surgery

N = 80

Procedural information, behavioural instruction

1 month post‐surgery:

SF‐36 Physical Function (high scores = more active)

Nottingham Health Profile (NHP) Physical Mobility (high scores = greater dysfunction)

Intervention n = 35; control n = 35

SF‐36 Physical Function: Intervention mean = 25.3, control mean = 21.8

NHP Physical Mobility: Intervention mean = 32.97, control mean = 26.1

Independent t‐tests. Intervention group better than control group on both outcomes: SF‐36 Physical Function (P value < 0.00001); NHP Physical mobility P value < 0.00001)

Hoogeboom 2010

Total hip replacement

N = 21

Behavioural instruction

Iowa Level of Assistance Scale ‐ taken each postoperative day in hospital; authors used this to measure "time needed to reach functional independence": lower scores = more independent

Intervention: time to reach functional independence median 4 days (range 3 to 6, n = 8(?)); control group median 4 days (range 3 to 5, n = 10)

Difference in time to reach functional independence not significant (P value = 0.963)

Cholecystectomy

N = 125

Intervention 1: Procedural information, sensory information, behavioural instruction, emotion‐focused (at pre‐admission, 15 days before surgery)

Intervention 2: Procedural information, sensory information, behavioural instruction, emotion‐focused (afternoon before surgery)

First 2 post‐surgery days:

A postoperative recovery index; dimension "physical functional ability". Believe high scores = better outcome (not clear).

Cluster‐randomized trial. Data = mean (SD).

Intervention 1: day 1: 14.26 (3.4); day 2: 20.7 (2.5), n = 40

Intervention 2: day 1: 15.45 (3.16); day 2: 20.87 (2.43), n = 42

Control: day 1: 14.65 (3.02); day 2: 20.85 (2.17), n = 43

The 2 intervention groups were combined for analyses. Carried out multiple regressions to control for other independent variables (including study group), and used these to select covariates to enter into MANOVAs. For physical function recovery, no covariates entered for day 1; state anxiety on eve of surgery for day 2. Both day 1 and day 2: F ratios not significant.

Mahler 1998

Coronary artery bypass surgery

N = 268

Intervention 1: Procedural and sensory information; behavioural instruction

Intervention 2: Procedural and sensory information; cognitive intervention

Intervention 3: Procedural and sensory information; cognitive intervention

Monitoring of ambulation with device that counts movements using mercury tilt switch. Worn on days 2, 3 and 4 at one hospital; days 3, 4, 5 post‐surgery at second hospital. Worn from morning to late afternoon/early evening.

Intervention 1: mean (SD) = 11.01 (1.02), n = 65

Intervention 2: 10.77 (1.02), n = 65

Intervention 3: 11.41 (1.12), n = 60

Control: 9.69 (0.85), n = 67

ANOVA and planned orthogonal comparisons. No significant effects by study group (P values < 0.60)

Total hip arthroplasty

N = 39

Procedural information, behavioural instruction

Before discharge:

Barthel Index: high score = less limited

WOMAC function (high scores = worse functional limitations)

Intervention n = 15; control n = 20

Barthel index: Intervention mean (SD): 19.8 (.4); Control: 18.7 (1.4)

WOMAC function: Intervention mean (SD): 25.7 (8.3); Control: 28.3 (12.1)

Barthel Index: better improvement in older adults in intervention group (P value < 0.005). Trend to reduction in WOMAC scores for older adults in intervention group. Does not report analysis of a simple comparison by group alone.

Total abdominal hysterectomy

N = 108

Behavioural instruction, cognitive intervention, relaxation

At discharge: Health Status Questionnaire (HSQ): Physical Functioning Subscale: high scores = better outcome

Length of time ambulated on first post‐surgery day

Mean (SD) not reported for HSQ

Ambulation: Intervention mean (SD): 330 (615); control 156 (97)

HSQ analyses are not presented by subscale

Ambulation: intervention participants ambulated longer than controls (F(1,105) = 2.05, P value = 0.043)

Oosting 2012

Total hip arthroplasty

N = 30

Behavioural instruction

4 days post‐surgery:

Iowa Level of Assistance Scale (ILAS), ability to function in daily life. Low scores = more independent. Split scores: < 6 (for "functional mobility" or ≥ 6

Intervention: 10 of n = 12 rated "functionally mobile"; control: 11 of n = 13 rated "functionally mobile"

No reported test of significance for this outcome

Ridgeway 1982

Abdominal hysterectomy

N = 60

Intervention 1: Procedural and sensory information

Intervention 2: Cognitive intervention

Diary record – days when performed 10 household activities over 3 post‐surgery weeks. For score: summed across tasks and no. days each was performed.

Intervention 1 mean = 6.6, n = 20

Intervention 2 mean = 6.9, n = 20

Control mean = 5.9, n = 20

Report trend, Intervention 2 doing most (ANOVA F = 2.2, df = 3.66, P value = 0.10). NOTE: included a 4^th group in ANOVA – patients who refused information (not relevant to review as not randomized)

CABG

N = 406

Intervention: Behavioural instruction, cognitive intervention

Days 3 and 5 post‐surgery: pain interference with general activities, sleep, walking, deep breathing and coughing (modified Interference Subscale of Brief Pain Inventory)

Behavioural recovery: controls: more pain interference related to deep breathing and coughing (mean 3.8 (SD 3.1) versus mean 2.7 (SD 3.1); t(355) = 2.54; P value < 0.01). Other activities not significant.

Zieren 2007

Inguinal hernia surgery

N = 100

Procedural information, behavioural instruction

DAy 1 post‐surgery:

SF‐36 physical functioning (high scores: less disability)

Intervention n = 50; control n = 50

No statistics presented. Observed that differences were visible on first postoperative day, with physical and psychological functions being less affected in intervention than control group.

ADL = activities of daily living

ANOVA = analysis of variance

CABG = coronary artery bypass graft

F = F statistic (ANOVA)

HSQ = Health Status Questionnaire

IAA = inpatient ambulatory activity

ILAS = Iowa Level of Assistance Scale

N = number of participants in sample

NHP = Nottingham Health Profile

SD = standard deviation

SF = Short Form

T = T statistic value (t‐test)

WOMAC = Western Ontario and McMaster Osteoarthritis Index

Finally, D'Lima 1996 reported more negative outcomes for intervention groups, with the control group having the highest function score. It is of concern that an intervention could lead to a worse outcome, so it is helpful to examine this study further. D'Lima 1996 used two intervention groups, both focused on exercise prior to knee replacement surgery: intervention 1 consisted of physical therapy sessions designed to strengthen muscles and improve range of motion; intervention 2 consisted of cardiovascular conditioning to improve fitness. The mean outcome function scores were 35 for the control group, 32 for intervention 1 and 30.5 for intervention 2, with higher scores indicating better function. The authors do not present any direct comparison information across groups, simply stating that the intervention groups showed a decline in function. It is therefore not clear whether these scores are significantly different across groups, and with only 10 participants in each group it is unclear how reliable these findings are. Other studies in this group also used preoperative exercises to strengthen muscles, improve range of motion and/or improve cardiac fitness in people undergoing joint replacement surgery (Ferrara 2008; Gilbey 2003; Hoogeboom 2010; Oosting 2012). Ferrara 2008 and Oosting 2012 did not report significance of findings for this outcome (at time points relevant to the review). Hoogeboom 2010 reported no difference between intervention and control groups in time to reach "functional independence", but Gilbey 2003 reported that the intervention group had significantly better scores than the control group on the physical function domain of the WOMAC at three weeks after the operation. Thus, while variance in measure types and timings makes it difficult to determine whether, or how, this type of behavioural instruction is of benefit to patients undergoing joint replacement surgery, it is not clear that the intervention is harmful.

The five studies that reported statistically significant effects in favour of the intervention group address a range of surgical procedures (one coronary artery bypass surgery, two total hip arthroplasty, one hysterectomy and one mixed surgical types), and a range of interventions (behavioural instruction about pre and postoperative behaviours, procedural information) (Fortin 1976; Gilbey 2003; Heidarnia 2005; McGregor 2004; Oetker‐Black 2003). One study incorporated relaxation and cognitive intervention alongside behavioural instruction (Oetker‐Black 2003). The three studies that found differences to be non‐significant similarly addressed various procedures (total hip replacement, cholecystectomy, coronary artery bypass surgery) and interventions (behavioural instruction about pre‐ and postoperative behaviour, procedural information, sensory information and emotion‐focused interventions) (Hoogeboom 2010; Lévesque 1984; Mahler 1998). There were also no obvious differences in the types of outcome measures used by studies that did, and did not, find effects: the studies with significant differences used measures of inpatient ambulatory activity and activities of daily living, the physical function domain of WOMAC, SF‐36 physical function, the Barthel Index and the Health Status Questionnaire. The studies that did not find significant differences used the Iowa Level of Assistance Scale (Shields 1995), a postoperative recovery index measuring physical functional ability and ambulation monitoring.

Summary: behavioural recovery

Thus, while there were some promising findings suggesting that psychological preparation, in particular behavioural instruction, may improve behavioural recovery outcomes, there is a need for agreement on the outcome measures used to be able to more directly compare findings across studies, and for studies to consistently report findings in sufficient detail to allow data to be pooled across studies in meta‐analysis. We rated the overall quality of evidence as `very low' (downgraded by three points). We rated the risk of bias as `very serious', leading to downgrading by two points, because of the high proportion of `uncertain' ratings and because the number of studies with sufficiently robust measurement to meet our inclusion criteria and reporting suitable data for meta‐analysis was low. We made a further downgrade for high heterogeneity (see summary of findings Table for the main comparison).

Secondary outcomes

1.Negative affect

Studies included in meta‐analysis

Fifty studies reported the outcome negative affect. We included 31 (30% of 105 studies) in the omnibus meta‐analysis (Ali 1989; Ashton 1997; Bergmann 2001; Bitterli 2011; Broadbent 2012; Cheung 2003; Cuñado Barrio 1999; Cupples 1990; Doering 2000; Done 1998; Felton 1976; Fortin 1983; Giraudet 2003; Guo 2012; Hart 1980; Heidarnia 2005; Lamarche 1998; Leserman 1989; Levesque 1977; Lim 2011; Ma 1996; Oliphant 2013; Pellino 2005; Postlethwaite 1986; Reading 1982; Ridgeway 1982; Schwartz‐B'tt 1994; Seers 2008; Yang 2012; Zhang 2012; Zieren 2007), with data from 2496 participants analysed (24% of 10,302 participants randomized across all studies) (Analysis 1.3; Figure 5). As a variety of scales were used to measure negative affect, we used standardized scores (SMD (Hedges' g)) to pool data. Higher scores indicate higher negative affect; effect scores below zero indicate that the intervention group had lower negative affect. Overall, there was evidence of lower negative affect in the intervention groups compared with the control groups (SMD ‐0.35, 95% CI ‐0.54 to ‐0.16). Although once again there were very high levels of statistical heterogeneity (I² statistic = 81%), which suggests extreme caution needs to be taken when interpreting the result of the meta‐analysis, the results of the forest plot show a consistent pattern of results in favour of lower negative affect after psychological preparation.

Figure 5

Negative affect (any psychological preparation intervention versus control). B: behavioural instruction; C: cognitive interventions; E: emotion‐focused interventions; H: hypnosis; P: procedural information; R: relaxation; S: sensory information.

The results for individual intervention types again tended to be similar to the omnibus meta‐analysis. When considering both `pure' and `mixed' studies together, there were statistically significant results for the procedural information (Analysis 2.3), sensory information (Analysis 3.3), relaxation techniques (Analysis 6.3) and hypnosis (Analysis 7.1) analyses. There was no clear evidence of an effect for the behavioural instruction (Analysis 4.3), cognitive (Analysis 5.3) or emotion‐focused (Analysis 8.3) interventions. The funnel plot showed no clear evidence of publication bias.

Studies not included in meta‐analysis

Nineteen studies contained appropriate data for narrative synthesis only: Barlési 2008; Burton 1995; Chumbley 2004; Daltroy 1998; DeLong 1970; Elsass 1987; Gräwe 2010; Hawkins 1993; Johnson 1978b; Johnson 1985; Klos 1980; Lévesque 1984; McGregor 2004; O'Connor 2014; Oetker‐Black 2003; Osinowo 2003; Shelley 2007; Shuldham 2002; Watt‐Watson 2004 (see Table 3). We excluded one of these studies from meta‐analysis because it was a cluster‐randomized trial (Lévesque 1984); we excluded the remainder from meta‐analysis because they provided insufficient information.

Table 3. Findings of studies that examined the outcome negative affect but could not be included in meta‐analyses

Author, year

Surgery type and sample size (randomized)

Intervention categories

Negative affect measure(s)

The first measure listed is that prioritized in this review

Negative affect findings (as available)

Barlési 2008

Thoracic surgery for non‐small cell lung cancer

N = 102

Procedural information

Timing unclear: at time of surgery (postoperative period) or 1 month post‐surgery

Psychologic Global Well‐being Scale; components include Anxiety, Depressed Mood and Positive Well‐being (also self control, general health, vitality)

Control n = 34; intervention n = 41

Mean/SD provided only for total scale (including non‐negative affect components). For the individual elements, no significant differences (no details provided).

Burton 1995

Mastectomy/sector mastectomy for breast cancer

N = 215

Intervention 1: Cognitive intervention and emotion‐focused (preoperative interview)

Intervention 2: Cognitive intervention and emotion‐focused (preoperative interview + 30 minute ‘chat’ on unrelated matters)

Intervention 3. Cognitive intervention and emotion‐focused (preoperative interview + 30‐minute brief psychotherapeutic intervention – additional emotion‐focused content)

Day 4 post‐surgery: Hospital Anxiety and Depression Scale (HADS) Anxiety and Depression.

Also General Health Questionnaire ‐28 and modified Present State Examination schedule and the Diagnostic and Statistical Manual of Mental Disorders, 3^rd Ed (DSM‐III) but results are not reported.

Only report mean HADS scores for the overall sample, not by group at 4 days postoperation. Other negative affect also not reported by group at this time point.

Chumbley 2004

Mixed: surgeries that would receive PCA routinely

N = 246

Intervention 1: Behavioural instruction (leaflet)

Intervention 2: Behavioural instruction (interview)

24‐72 hours post‐surgery:

HADS Anxiety

Profile of Mood States (POMS) Tension/anxiety

Cluster‐randomized trial

HADS Anxiety: Control mean (95% CI) = 6.17 (5.34 to 8.00, n = 73); Intervention 1 mean (95% CI) = 6.03 (4.94 to 7.12, n = 75); Intervention 2 mean (95% CI) = 6.52 (5.59 to 7.45, n = 72)

No significant difference across groups (HADS anxiety, P value = 0.31; POMS tension/anxiety P value = 0.28)

Daltroy 1998

Total hip or knee arthroplasty

N = 222

Procedural and sensory information

Day 4 after surgery:

State Trait Anxiety Inventory (STAI) state anxiety

Intervention did not affect anxiety in general linear model (P value = 0.94). No interaction between intervention and denial, anxiety or desire for information. No main effects mentioned.

DeLong 1970

Gall bladder removal and removal of uterus

N = 70

Procedural information, sensory information, behavioural instruction

Day 5 or 6 after surgery:

STAI (state and trait anxiety)

No differences in anxiety scores across groups (no statistics provided)

Elsass 1987

Inguinal hernia or varicose vein surgery

N = 90

Procedural information

1 ½ hours after surgery and day after surgery

STAI state anxiety

Control n = 40; intervention n = 40. Anxiety scores are presented but unclear whether mean or median: Control score = 52; intervention score = 42 (reading off Figure 1).

Difference in scores between groups "increased significantly" at 1 ½ hrs after operation (P value < 0.05, Mann Whitney); intervention group less anxious

Gräwe 2010

Mixed: abdominal or vascular surgery

N = 96

Sensory information, cognitive intervention

Days 1 to 3 post‐surgery:

STAI state anxiety

BSKE – general psychological well‐being

Comparisons by group not reported for this outcome

Hawkins 1993

Gynaecological surgery (mixed)

N = 60

Behavioural instruction

48 hours after surgery:

Hospital Anxiety Scale

Control n = 40 (combining standard care and attention controls); intervention n = 20

No report of comparisons for this outcome

Johnson 1978b

Sample 1: cholecystectomy, N = 81

Sample 2: inguinal hernia repair, N = 68

Intervention 1: ‘Instruction’: Behavioural instruction (deep breathing, coughing, leg exercises)

Intervention 2: ‘Procedure information’: focus procedural information, also some sensory information and behavioural instruction

Intervention 3: `Sensation information’: focus: sensory information, also some procedural information and behavioural instruction

2 x 3 factorial design: no instruction/instruction (Intervention 1; no information/information (Interventions 2 and 3)

Scores totaled over days 1, 2 and 3: Mood Adjective Checklist (fear, well‐being, happiness, helplessness, anger)

Sample 1

Negative affect: no main effect of interventions but interactions between instruction and preoperative fear (F(5, 61) = 4.69, P value < 0.001) and information and preoperative fear (F(10,122) = 2.07, P value < 0.05)

Low fear group: `instruction' tended to increase negative moods and decrease positive moods compared with no‐instruction, and tendency for ‘procedure information’ to decrease and ‘sensation information’ to increase negative mood compared with no information, but these comparisons were not significant

High fear group: `instruction' tended to decrease negative mood and increase positive mood compared with no instruction; significant for anger and happiness (Dunnett’s t(1,65) = 3.32, P value < 0.001; t(1,65) = 3.35, P value < 0.001). Those receiving ‘procedure information’: higher means for fear and positive moods, and lower means for helplessness and anger, but only anger significant (Dunnett’s t(2,65) = 2.00, P value < 0.05). ‘Sensation information’: positive moods tended to be higher and negative moods lower than no information group; only anger significant (Dunnett’s t(2,65) = 2.43, P value < 0.025).

Sample 2

Interaction between instruction and information (F(10,96) = 1.93, P value < 0.05) but no significant univariate findings, difficult to interpret

Johnson 1985

Abdominal hysterectomy

N = 199

Intervention 1: procedural and sensory information

Intervention 2: ‘cognitive‐coping technique’ – cognitive intervention

Intervention 3: ‘Behavioural‐coping technique’ – behavioural instruction

2 x 3 factorial design: no information/information (Intervention 1); no coping technique/coping technique (Interventions 2 and 3)

Day 3 post‐surgery (and 1^st and 4^th weeks post‐discharge):

Profile of Mood States (POMS: anxiety, confusion, anger, depression, fatigue, vigour). 3^rd postoperative day and 1^st and 4^th week post‐discharge.

Outcomes entered into MANOVA included anxiety. Significant at P value < 0.10: coping technique, F (16, 286) = 1.59, P value = 0.07 (outcomes physical recovery, narcotic doses and length of stay seem to be responsible for this effect). Included race as factor; interaction between race and coping technique (F16, 286) = 1.58, P value = 0.07). For white patients, ‘behavioural coping’ reduced anxiety (Dunnett’s t(3,150) = 3.45, P value < 0.001); ‘cognitive’ and ‘behavioural’ techniques reduced confusion (Dunnett’s t(3,150) = 2.75, P value < 0.025); non‐significant for black participants.

Klos 1980

Cholecystectomy

N = 50

Intervention 1: Procedural information, behavioural instruction (pamphlet)

Intervention 2: Procedural information, behavioural instruction (nurse visit)

Intervention 3: Procedural information, behavioural instruction (pamphlet and nurse visit)

2^nd post‐surgery day:

Mood Adjective Checklist: 15 adjectives describing 5 mood dimensions: fear, well‐being, happiness, helplessness, anger

Authors did not report analyses by whole intervention group; instead, analyses are reported after median split into high‐ preoperative‐fear and low‐preoperative fear groups

2 x 2 factorial design: pamphlet/no pamphlet versus nurse visit/no nurse visit

Significant differences between means of intervention 2 (nurse‐visit) and no‐nurse visit for high‐preoperative‐fear group for well‐being [F(1,20) = 6.57, P value < 0.10] and happiness (F (1,20) = 11.89, P value < 0.05). Patients with the nurse visit scored higher on positive moods than those who did not receive it.

Cholecystectomy

N = 125

Intervention 1: Procedural information, sensory information, behavioural instruction, emotion focused (15 days before surgery)

Intervention 2: Procedural information, sensory information, behavioural instruction, emotion focused (afternoon before surgery)

First 3 days after surgery

STAI (French version) state anxiety

Cluster‐randomized

Day 1 Control mean (SD) = 37.5 (8.51, n = 43); intervention 1 mean (SD) = 35.34 (9.34, n = 40); intervention 2 mean (SD) = 37.38 (8.29, n = 42)

No significant difference between groups for postoperative state anxiety

Total hip arthroplasty

N = 39

Procedural information, behavioural instruction

Positive & Negative Affect Schedule (PANAS)

Control n = 20; intervention n = 15

No mention of findings for analysis by group. May only be presenting positive findings – if so, this would suggest null result.

O'Connor 2014

Surgery for rectal cancer

N = 85

Procedural information

Prior to discharge:

HADS anxiety and depression

Numerical data not reported for this outcome

Control group: slightly higher anxiety score but not significantly different; depression – similar means, not significantly different

Total abdominal hysterectomy

N = 108

Behavioural instruction, cognitive intervention, relaxation

Day 1 post‐surgery day and at discharge:

STAI state anxiety

Only analyses at later time points reported

Osinowo 2003

Not stated – participants from surgical and gynaecological wards

N = 33

Intervention 1: Cognitive intervention (Rational Emotive Therapy)

Intervention 2: Cognitive intervention (Self‐Instructional Training)

24 hours post‐surgery:

STAI state anxiety

HADS Anxiety

HADS Depression

STAI scores: Control mean unclear (2 possible scores), n = 11; intervention 1 mean (SD) = 30.91 (6.61, n = 11); intervention 2 mean (SD) = 33.82 (6.21, n = 11). Intervention 2 (SIT): decrease in anxiety from pre‐intervention to postoperation.

HADS anxiety: decreased for both intervention groups; changes in control group ns (Intervention 1: t(10) = 3.62, P value < 0.01; Intervention 2: t(10) = 2.06, P value < 0.05; control t(10) = 1.13, non‐significant.

HADS depression: no significant changes across time

Paper generally written unclearly

Shelley 2007

Coronary artery bypass surgery

N = 90

Cognitive intervention

Day 4 post‐surgery:

Distress (Depression, Anxiety and Stress Scales, DASS)

Control n = 43; intervention n = 37

Direct effect of group not significant; 3‐way interaction was significant (intervention x external health locus of control x self efficacy, F(1,71) = 6.20, P value < 0.05). Fig 1 suggests, for intervention participants: lower distress than controls if EHLC and self efficacy either both high or both low. If EHLC low and self efficacy high, appears to be little change; if high EHLC and low self efficacy then lower distress for Control group.

Shuldham 2002

Coronary artery bypass surgery

N = 356

Procedural information and behavioural instruction

Day 3 post‐surgery:

Anxiety – HADS

Depression – HADS

‘tense and uptight’ – General Well‐being Questionnaire

‘worn out’ – General Well‐being Questionnaire

Control n = 156; intervention n = 173

No significant differences between variables at 3 days post‐surgery (using Mann‐Whitney U):

Anxiety: U = 11,636, Z = ‐0.28, P value = 0.78

Depression: U = 10,756; Z = ‐1.24, P value = 0.22

Tense and uptight: U = 10,008, Z = ‐1.27, P value = 0.21

Worn out: U = 9,717.5, Z = ‐1.49, P value = 0.14

CABG

N = 406

Behavioural instruction, cognitive intervention

Days 3 and 5 post‐surgery

Pain interference with mood; modified version of Interference Subscale of the Brief Pain Inventory (BPI‐I)

Findings are not reported for this outcome – it would appear that authors are only reporting significant findings so it seems likely that group differences were not significant

BPI‐I = Interference Subscale of the Brief Pain Inventory

BSKE (EWL) = Befindlichkeitsskalierung durch Kategorien und Eigenschaftswörter (measuring general psychological well‐being)

CABG = coronary artery bypass graft

CI = confidence interval

DASS = Depression, Anxiety and Stress Scales

DSM‐III = Diagnostic and Statistical Manual (of Mental Disorders), version 3

EHLC = external health locus of control

F = F statistic (analysis of variance)

HADS = Hospital Anxiety and Depression Scale

N = number of participants in sample

PANAS = Positive and Negative Affect Schedule

PCA = patient‐controlled analgesia

POMS = Profile of Mood States

SD = standard deviation

SIT = Self‐Instructional Training

STAI = State Trait Anxiety Inventory

U = U statistic (Mann‐Whitney test)

A statistically significant impact of the intervention over control was reported by Elsass 1987, who found a procedural information intervention led to less anxiety in the intervention group than the control group 1½ hours after surgery. Unclear findings were reported by Osinowo 2003: it would seem that participants receiving a cognitive intervention experienced a decrease in anxiety while a control group did not, but groups do not appear to have been directly compared.

Mixed findings were reported by authors of four papers that examined interactions in their data. Johnson 1978b reported interactions in their sample of patients undergoing cholecystectomy. In their low preoperative fear group, differences between groups were not significant. For the high fear group, participants receiving behavioural instruction tended to have decreased negative mood and increased positive mood (significant for anger and happiness, not fear, helplessness or well‐being). In the high fear group, those receiving interventions focusing on procedural and sensory information also had significantly lower anger scores. In a second sample (inguinal hernia repair patients), an interaction was discovered between behavioural instruction and procedural/sensory information‐focused groups but no significant comparisons were identified. Klos 1980 also compared high and low fear groups: participants with high preoperative fear receiving procedural information and behavioural instruction via a nurse visit had higher scores for happiness than a control group; other analyses were not reported (including for the outcome fear, which this review would prioritize), suggesting that other findings were not significant. Johnson 1985 reported that a behavioural instruction intervention reduced postoperative anxiety, although only for white participants (intervention effects were not significant for black participants). An interaction of intervention (cognitive intervention) x external health locus of control (EHLC) x self‐efficacy was examined by Shelley 2007. There was no significant direct effect of group allocation, but the interaction between the three factors was significant. It seemed likely that lower distress was reported for intervention than control participants if EHLC and self‐efficacy were either both high or both low. If participants had high EHLC and low self‐ efficacy, then the control group seemed to be less distressed.

Seven studies reported no significant differences between groups (Barlési 2008; Chumbley 2004; Daltroy 1998; DeLong 1970; Lévesque 1984; O'Connor 2014; Shuldham 2002). These studies included the interventions procedural information (two studies), procedural and sensory information (one), behavioural instruction (one), procedural information, sensory information and behavioural recovery (one), procedural information and behavioural instruction (one), procedural information, sensory information, behavioural instruction and emotion‐focused intervention (one). As such, all but one contained the component (procedural information) that was contained in the study that found significant effects (Elsass 1987). In this narrative synthesis, only one study reported findings using a cognitive intervention (Osinowo 2003); it is unfortunate that the findings were not more clearly reported.

Six studies did not report analyses for the negative affect outcome of relevance to the review even though authors reported measuring it (Burton 1995; Gräwe 2010; Hawkins 1993; McGregor 2004; Oetker‐Black 2003; Watt‐Watson 2004). In some cases it may be because studies reported significant findings only, and findings were not significant, but this is not clear.

Summary: negative affect

In summary, there was some evidence from the meta‐analyses of a beneficial effect of psychological preparation techniques on postoperative negative affect, although once again the high levels of unexplained statistical heterogeneity make it difficult to accept this result with confidence. The pooled effect size from the omnibus analysis was ‐0.35, often considered to represent a small effect (Cohen 1988). There did not appear to be evidence that certain techniques performed better than others in reducing negative affect. Overall, it would seem that psychological preparation techniques may have beneficial effects of postoperative outcomes but the high level of heterogeneity in the data makes it difficult to determine the circumstances and intervention content that would consistently improve outcomes. There is also some suggestion that individual characteristics (e.g. level of preoperative fear) may affect the way that psychological preparations impact on postoperative outcomes. Due to the high heterogeneity and the high number of studies reporting sufficient methodological details to ascertain risk of bias, we downgraded the overall quality of evidence for the outcome negative affect by two points to `low' (see summary of findings Table for the main comparison).

2.Length of stay

Studies included in meta‐analysis

Of the 58 studies with length of stay as an outcome, sufficient data were available for meta‐analysis in 36 (34% of 105 studies: Ashton 1997; Barbalho‐Moulim 2011; Beaupre 2004; Bergin 2014a; Bitterli 2011; Chaudhri 2005; Crowe 2003; Cuñado Barrio 1999; D'Lima 1996; Daltroy 1998; Doering 2000; Felton 1976; Fortin 1976; Furze 2009; Giraudet 2003; Hulzebos 2006a; Lam 2001; Langer 1975; Leserman 1989; Levin 1987; Lin 2005; Lindeman 1973; Mahler 1995; Mahler 1998; McGregor 2004; Oosting 2012; Rajendran 1998; Ridgeway 1982; Schmitt 1973; Shuldham 2002; Watt‐Watson 2000; Watt‐Watson 2004; Wilson 1981; Zhang 2012; Ziemer 1982; Zieren 2007), with data from 3313 participants (32% of 10,302 participants randomized across all studies). Overall, when considering all types of psychological intervention, there was evidence of shorter length of stay in the intervention groups compared with the control groups (mean difference (MD) ‐0.52 days, 95% CI ‐0.82 to ‐0.22) (Analysis 1.2; Figure 6). There were, however, high levels of statistical heterogeneity between studies (I² statistic = 74%).

Figure 6

Length of stay (any psychological preparation intervention versus control). B: behavioural instruction; C: cognitive interventions; E: emotion‐focused interventions; H: hypnosis; P: procedural information; R: relaxation; S: sensory information.

The meta‐analysis results for individual intervention types were generally similar. When looking at all studies including a particular intervention type as one of the intervention components there were statistically significant results for the procedural information (Analysis 2.2), sensory information (Analysis 3.2), behavioural instruction (Analysis 4.2) and relaxation (Analysis 6.2) intervention types, although there was not always evidence of an effect when including just the `pure' studies. There was no evidence of an effect on length of stay for cognitive intervention studies (Analysis 5.2). There were few studies evaluating hypnosis (one study, Ashton 1997, MD 1.80, 95% CI ‐0.86 to 4.46, P value = 0.19) or emotion‐focused (Analysis 8.2) interventions. The funnel plot showed no clear evidence of publication bias.

Studies not included in meta‐analysis

Twenty‐two studies contained sufficient data for narrative synthesis only: Coslow 1998; DeLong 1970; Field 1974; Gocen 2004; Goodman 2008; Greenleaf 1992; Guo 2012; Hoogeboom 2010; Hulzebos 2006b; Johnson 1978b; Johnson 1985; Klos 1980; Kulkarni 2010; Letterstål 2004; Levesque 1977; Lévesque 1984; Oetker‐Black 2003; Oliphant 2013; Omlor 2000; Pellino 1998; Rosenfeldt 2011; Vukomanović 2008 (see Table 4).

Table 4. Findings of studies that examined the outcome length of stay but could not be included in meta‐analyses

Author, year	Surgery type and sample size (randomized)	Intervention categories	Length of stay findings (as available)
Coslow 1998	Laparoscopic tubal ligation N = 30	Procedural information, sensory information and behavioural instruction	Intervention n = 15; control n = 15 No significant difference
DeLong 1970	Gall bladder removal and removal of uterus N = 70	Procedural information, sensory information, behavioural instruction	Intervention n = 31; control n = 33 Intervention significantly decreased no. days in hospital (F = 4.70, df = 1/62, P value < 0.05). Intervention mean standardized days 47.06; control mean standardized days 52.32. When analysed by coping style: intervention reduced length of stay for copers (F = 6.43, df =1/20, P value < 0.05), but not avoiders or non‐specific defenders.
Field 1974	Mixed orthopaedic surgery N = 60	Procedural information, hypnosis	Intervention n = 30; control n = 30 No significant difference
Gocen 2004	Total hip replacement N = 59	Behavioural instruction	Intervention n = 29; control n = 30 No significant difference (P value > 0.05)
Goodman 2008	Cardiac bypass surgery N = 188	Behavioural instruction, relaxation, emotion‐focused	Intervention median 8.5 (IQR 3.25, range 4 to 50 days, n = 91) Control median 9 (IQR 3, range 2 to 170 days, n = 90) No significant difference (Mann‐Whitney U = 0.29, P value not provided)
Greenleaf 1992	Coronary artery bypass surgery N = 32	Intervention 1: Hypnosis and relaxation Intervention 2: Hypnosis	No significant difference between the groups
Guo 2012	Cardiac surgery N = 153	Procedural information	Intervention median 14.0 days (IQR 9.3 to 19.8, n = 68) Control median 12.0 days (IQR 10 to 17, n = 67) No significant difference (P value = 0.17)
Hoogeboom 2010	Primary total hip replacement due to osteoarthritis N = 21	Behavioural instruction	Intervention median: 6 days (range 5 to 22, n = 0) Control median: 6 days (range 4 to 7, n = 10) No significant difference (P value = 0.228)
Hulzebos 2006b	CABG N = 279	Procedural information, behavioural instruction	Intervention median 7 days (range 5 to 41, n = 139) Control median 8 days (range 6 to 70, n = 137) Intervention group: significantly shorter stay. Mann‐Whitney U (z = ‐2.42, P value = 0.02).
Johnson 1978b	Sample 1: cholecystectomy, N = 81 Sample 2: inguinal hernia repair, N = 68	Intervention 1: ‘Instruction’: Behavioural instruction (deep breathing, coughing, leg exercises) Intervention 2: ‘Procedure information’: focus procedural information, also some sensory information and behavioural instruction Intervention 3: `Sensation information’: focus: sensory information, also some procedural information and behavioural instruction 2 x 3 factorial design: no instruction/instruction (Intervention 1; no information/information (Interventions 2 and3)	Sample 1 (Cholecystectomy) Length of stay: patients in Intervention 2 (‘Procedure information’) and Intervention procedure and Intervention 3 (‘Sensation information’): shorter postoperative stays than no‐information participants; only significant for sensation information (Dunnett’s t(3,64) = 3.45, P value < 0.001). Control (no instruction or information intervention): mean stay = 6.36, n = 10; Intervention 1 only: mean stay = 6.20, n = 14; Intervention 2 only: mean = 5.97, n = 14; Intervention 3 only: mean = 5.78, n = 12; Intervention 1 and Intervention 2: mean = 5.84, n = 14; Intervention 1 and Intervention 3: mean = 5.29, n = 13 Sample 2 (Hernia repair) No significant effects of interventions for length of stay
Johnson 1985	Abdominal hysterectomy N = 199	Intervention 1: procedural and sensory information Intervention 2: ‘cognitive‐coping technique’ – cognitive intervention Intervention 3: ‘behavioural‐coping technique’ – behavioural instruction 2 x 3 factorial design: no information/information (Intervention 1); no coping technique/coping technique (Interventions 2 and 3)	Outcomes entered into MANOVA included length of stay. Coping technique was significant using a P value < 0.10 criterion (F (16, 286) = 1.59, P value = 0.07). Cognitive‐coping group: longer hospitalization than control group (Dunnett’s t (3,150) = 2.52, P value < 0.025) Adjusted mean scores and sample size according to coping groups: Control mean = 6.56, n = 72; Intervention 2 mean = 6.97, n = 48; Intervention 3 mean = 6.50, n = 47
Klos 1980	Cholecystectomy N = 50	Intervention 1: procedural information, behavioural instruction (pamphlet) Intervention 2: procedural information, behavioural instruction (nurse visit) Intervention 3: procedural information, behavioural instruction (pamphlet and nurse visit)	Authors did not report analyses by whole intervention group; instead, analyses are reported after median split into high‐preoperative‐fear and low‐preoperative‐fear groups. An interaction effect was reported between preoperative fear and receiving the pamphlet (F(1,39) = 4.14, P value < 0.05). If high preoperative fear and received pamphlet, shorter stay than those with high fear who did not receive pamphlet (but difference in means non‐significant: 5.09 versus 5.79 days). If low preoperative fear and pamphlet: significantly longer postoperative stay than those who did not receive pamphlet (F(1,18) = 4.84, P value < 0.05; means = 5.64 and 4.45). Observations are made about length of stay in the nurse visit groups, but no statistical tests are reported. Low preoperative fear: means for stay length for Interventions 1, 2, 3 and Control respectively are: 5.64, 4.61, 5.05, 4.45 High preoperative fear: means for stay length for Interventions 1, 2, 3 and Control respectively are: 5.18, 6.02, 5.33, 5.91
Kulkarni 2010	Major abdominal surgery N = 80	Intervention 1: behavioural instruction (deep breathing training) Intervention 2: behavioural instruction (incentive spirometry) Intervention 3: behavioural instruction (specific inspiratory muscle training)	Intervention 1 (Deep breathing): median stay = 5 days (range 1 to 10, n=17); Intervention 2 (Incentive spirometry): median = 4 (range 2 to 22, n = 15); Intervention 3 (Inspiratory muscle training); median = 4 (range 1 to 13, n = 17) Control median stay = 6 (range 1 to 14, n = 17) No analysis is reported
Letterstål 2004	Abdominal aortic aneurysm open repair N = 52	Procedural and sensory information	Intervention: median = 11 days (range 4 to 34, n = 18) Control: median = 9 days (range 6 to 42, n = 17) Mann‐Whitney: no difference between groups (P value = 0.14)
Levesque 1977	Cholecystectomy (n = 82); hysterectomy (n = 54) Total N = 136	Procedural information, behavioural instruction, relaxation	No significant difference
Lévesque 1984	Cholecystectomy N = 125	Intervention 1: procedural information, sensory information, behavioural instruction, emotion‐focused (at pre‐admission, 15 days before surgery) Intervention 2: procedural information, sensory information, behavioural instruction, emotion‐focused (afternoon before surgery	Cluster‐randomized trial Intervention 1 mean (SD) = 5.85 (1.19), n = 40 Intervention 2 mean (SD) = 5.94 (1.42), n = 42 Control mean (SD) = 5.60 (1.05), n = 43 No analyses are reported for length of stay
Oetker‐Black 2003	Total abdominal hysterectomy N = 108	Behavioural instruction, cognitive intervention, relaxation	No significant difference: t(1,93) = ‐0.77, P value = 0.444)
Oliphant 2013	Pelvic reconstructive and/or urinary incontinence surgery N = 199	Behavioural instruction	Intervention median = 1 day (IQR 0 to 2, n = 93); control median = 1 day (IQR 0 to 2, n = 93) No significant difference (Mann‐Whitney U, P value = 0.63)
Omlor 2000	Inguinal hernia surgery or thyroidectomy N ≥ 211	Procedural information, relaxation	Intervention n = 103; control n = 105 No significant difference. The paper presents medians (ranges) for control and intervention groups, by each type of surgery and combined, but there appears to be an error as these are contradictory: Inguinal hernia, intervention: 7.5 (1 to 11); control: 8 (3 to 22) Thyroidectomy, intervention: 7.2 (2 to 16); control: 7.9 (4 to 13) Groups combined: intervention median 7.95; control median 7.4
Pellino 2005	Orthopaedic surgery procedures. 90 randomized; 83 consented (consent post‐randomization).	Procedural information, behavioural instruction	No significant difference Data reported: expected length of stay minus actual length of stay (days): Intervention mean = ‐0.46 (SD 1.00, n = 39) Control mean = ‐0.29 (SD 1.19, n = 35)
Rosenfeldt 2011	CABG and/or valve surgery N = 119	Behavioural instruction, cognitive intervention, relaxation	Intervention median = 6 days (IQR 5 to 8, n = 60) Control median = 6 days (IQR 5 to 8, n = 57) No significant difference (Wilcoxon, P value = 0.54)
Vukomanović 2008	Total hip arthroplasty N = 45	Procedural information, behavioural instruction	Cluster‐randomized trial Intervention mean (SD) = 9.8 (2.4), n = 20 Control mean (SD) = 10.2 (1.7), n = 20 No significant difference, P value ≤ 0.67

CABG = coronary artery bypass graft

F = F statistic (analysis of variance)

IQR = inter‐quartile range

MANOVA = multivariate analysis of variance

N = number of participants in sample

SD = standard deviation

Two studies were not included in the meta‐analysis because they were cluster‐randomized trials (Lévesque 1984; Vukomanović 2008). Both of these studies found no statistically significant difference in length of stay between groups.

In nine cases, data were unavailable for meta‐analysis because data were presented in the form of medians and interquartile ranges rather than means and standard deviations (Goodman 2008; Guo 2012; Hoogeboom 2010; Hulzebos 2006b; Kulkarni 2010; Letterstål 2004; Oliphant 2013; Omlor 2000; Rosenfeldt 2011). Seven of these studies reported no significant difference between the groups for length of stay, one found a significantly shorter stay in the intervention group (Hulzebos 2006b), and one study did not report any analysis (Kulkarni 2010).

Complex factorial designs were used by Johnson 1978b and Johnson 1985 with mixed results. Johnson 1978b examined two levels of instruction (no instruction versus behavioural instruction) over three information levels (no instruction versus `sensation information' (focus: sensory information, also containing procedural information and behavioural instruction) versus `procedure information' (focus: procedural information, also containing sensory information and behavioural instruction). In a sample of cholecystectomy patients, patients receiving `sensation information' had a shorter stay than `no information' participants, but no significant effects were seen in their sample of patients receiving inguinal hernia repair. Using a similar design, Johnson 1985 compared two levels of information (no information versus procedural and sensory information) and three `coping levels' (no `coping' intervention versus cognitive intervention versus behavioural instruction) in patients undergoing hysterectomy. The cognitive intervention group had a significantly longer stay than the control group.

Of the remaining studies, no statistically significant difference between groups was reported by Coslow 1998, Field 1974, Gocen 2004, Greenleaf 1992, Levesque 1977 and Oetker‐Black 2003. Pellino 1998 also reported no significant difference but conducted an unusual analysis, comparing expected minus actual length of stay across the two study groups. Two studies looked at interaction effects.

DeLong 1970 found that, for their overall sample, the intervention significantly decreased the number of days in hospital (intervention mean standardized days 47.06; control mean standardized days 52.32). Analysing findings by coping style, it was seen that the intervention reduced length of stay for `copers', but not for `avoiders' or `non‐specific defenders'.

Klos 1980 analysed data by level of preoperative fear (low or high) for four groups (control and three intervention groups: pamphlet, nurse, pamphlet and nurse, all containing procedural information and behavioural instruction), and did not report analyses conducted by whole intervention groups. An interaction was found such that those low in fear who received the pamphlet had a significantly longer stay than control participants low in fear who did not receive the pamphlet containing procedural information and behavioural instruction.

A range of intervention types were used within the 16 studies that did not find an effect of intervention on length of stay, including many using procedural information or behavioural instruction ‐ intervention types suggested to be beneficial by the meta‐analysis. The two studies that did report statistically significant benefits for length of stay were consistent with meta‐analysis findings: the components included were procedural information, sensory information and behavioural instruction (DeLong 1970; Hulzebos 2006b).

Summary: length of stay

In summary, the meta‐analyses suggested that psychological preparation led to a reduction in mean length of stay of around half a day. This effect might be considered important to patients and clinicians and to represent savings in healthcare resources. The meta‐analysis, however, had high statistical heterogeneity, which needs to be considered when interpreting this pooled effect. Although no clear explanations for the heterogeneity could be found, there was clearly considerable variation between the studies in the types of interventions administered. The pattern of results did, however, suggest a similar benefit of psychological preparation for all intervention types. The results of the studies included in the narrative review were generally not statistically significant, but would not contradict a pattern of a modest reduction of length of stay in the intervention group. The high heterogeneity and the high number of studies reporting sufficient methodological details to ascertain risk of bias meant that we downgraded the overall quality of evidence for the outcome length of stay by two points to `low' (see summary of findings Table for the main comparison).

Findings by intervention

1. Procedural information

Of the meta‐analysed studies, procedural information was a component in interventions of 12 studies with the outcome pain (1051 participants); 19 studies with the outcome length of stay (1983 participants) and 17 studies with the outcome negative affect (1334 participants). There was no evidence that interventions containing procedural information improved postoperative pain outcomes (pooled effect size (SMD) ‐0.08, 95% CI ‐0.26 to 0.09, Analysis 2.1), but procedural information was statistically significantly beneficial for length of stay (MD ‐0.63 days, 95% CI ‐1.08 to ‐0.18 days, Analysis 2.2). Procedural information was also beneficial for negative affect (SMD ‐0.45 days, 95% CI ‐0.75 to ‐0.16, Analysis 2.3), although analyses examining interventions containing procedural information alone (`pure' studies) did not reach statistical significance. Procedural information was included in four studies that found a benefit of the intervention on behavioural recovery (Fortin 1976; Heidarnia 2005; McGregor 2004; Zieren 2007 (statistics not presented)) but was also in three studies with non‐significant findings for this outcome (Lévesque 1984; Mahler 1998; Ridgeway 1982; Table 2).

2. Sensory information

Of the meta‐analysed studies, sensory information was a component in interventions of 11 studies with the outcome pain (881 participants); 14 studies with the outcome length of stay (1236 participants) and 12 studies with the outcome negative affect (919 participants). No interventions contained purely sensory information – it was always presented with other intervention components. For the outcome postoperative pain, there was no clear evidence that intervention patients benefited when receiving interventions containing sensory information (SMD ‐0.22, 95% CI ‐0.47 to 0.02, Analysis 3.1). Statistically significant beneficial effects of sensory information were seen for length of stay (MD ‐0.71, 95% CI ‐1.15 to ‐0.27, Analysis 3.2) and negative affect outcomes (SMD ‐0.55, 95% CI ‐0.90 to ‐0.19, Analysis 3.3). Sensory information was not included in any studies finding statistically significant effects for behavioural recovery, but was included in two non‐significant studies (Table 2).

3. Behavioural instruction

Of the meta‐analysed studies, behavioural instruction was a component in interventions of 21 studies with the outcome pain (1241 participants); 25 studies with the outcome length of stay (2338 participants) and 13 studies with the outcome negative affect (1183 participants). There was no evidence that behavioural instruction had an effect on postoperative pain or negative affect (SMD ‐0.14, 95% CI ‐0.33 to 0.05, Analysis 4.1; SMD ‐0.22, 95% CI ‐0.46 to 0.02, Analysis 4.3 respectively). A significantly beneficial effect of behavioural instruction was seen for length of stay (MD ‐0.51 days, 95% CI ‐0.84 to ‐0.19, Analysis 4.2), although findings were not statistically significant when only `pure' studies were included. Behavioural instruction appears to be of greatest potential for behavioural recovery outcomes – it featured as a component in all five studies reporting statistically significant benefits of the intervention (in pure form in one), but it also featured in many studies that did not find significant effects, and was the only component in the one study that reported more negative outcomes for intervention groups (D'Lima 1996; Table 2).

4. Cognitive interventions

Of the meta‐analysed studies, cognitive interventions were a component in interventions of six studies with the outcome pain (355 participants); nine studies with the outcome length of stay (1074 participants) and five studies with the outcome negative affect (251 participants). Thus, a relatively small number of studies contributed to the meta‐analyses. Cognitive interventions were not significantly beneficial overall for the outcome postoperative pain (SMD ‐0.02, 95% CI ‐0.29 to 0.25, Analysis 5.1), although combining the two `pure' studies did indicate a benefit for participants receiving the cognitive intervention (Cheung 2003; Ridgeway 1982). There was no evidence for an effect on length of stay (MD ‐0.43, 95% CI ‐1.07 to 0.22, Analysis 5.2) or negative affect (SMD ‐0.20, 95% CI ‐0.52 to 0.12, Analysis 5.3). Cognitive interventions were a component in one statistically significant and two non‐significant interventions for behavioural recovery (Table 2).

5. Relaxation techniques

Of the meta‐analysed studies, relaxation techniques were a component in interventions of 13 studies with the outcome pain (891 participants); seven studies with the outcome length of stay (473 participants) and 11 studies with the outcome negative affect (687 participants). Relaxation techniques had statistically significant beneficial effects on postoperative pain (SMD ‐0.46, 95% CI ‐0.81 to ‐0.11, Analysis 6.1) and negative affect (SMD ‐0.34, 95% CI ‐0.56 to ‐0.12, Analysis 6.3), although the effect on negative affect was not statistically significant when only `pure' studies were meta‐analysed. For length of stay, the mean difference was high (‐0.97 days – almost a day's shorter stay for intervention participants) but the 95% CI was ‐1.94 to ‐0.00 (P value = 0.05, Analysis 6.3), indicating that caution is needed in interpreting this finding. Relaxation was included in one significantly effective behavioural recovery intervention and in no non‐significant studies (Table 2).

6. Hypnosis

Of the meta‐analysed studies, hypnosis was a component in interventions of no studies with the outcome pain; one study with the outcome length of stay (32 participants) and two studies with the outcome negative affect (72 participants). Thus, hypnosis was rarely seen in the studies eligible for inclusion in the meta‐analyses. No studies used hypnosis for the outcome postoperative pain; for length of stay one study used hypnosis (combined with relaxation, non‐significant; Ashton 1997), and two studies addressing negative affect used hypnosis combined with relaxation, with statistically significant benefits (Ashton 1997; Hart 1980; SMD ‐0.77, 95% CI ‐1.25 to ‐0.30, Analysis 7.1). No studies addressing behavioural recovery incorporated hypnosis (Table 2).

7. Emotion‐focused interventions

Of the meta‐analysed studies, emotion‐focused interventions were a component in interventions of three studies with the outcome pain (180 participants); three studies with the outcome length of stay (212 participants) and four studies with the outcome negative affect (201 participants). For postoperative pain, the three studies meta‐analysed suggested potential for a beneficial impact on pain (SMD ‐0.42, 95% CI ‐0.85 to 0.00, P value = 0.05, Analysis 8.1; Giraudet 2003; Lin 2005; Postlethwaite 1986), although the study including emotion‐focused intervention in `pure' form showed no evidence of benefit (Postlethwaite 1986). Emotion‐focused interventions provided no benefit for length of stay (MD 0.14 days, 95% CI ‐0.67 to 0.94, Analysis 8.2) or negative affect (SMD ‐0.24, 95% CI ‐0.55 to 0.07, Analysis 8.3). For behavioural recovery, emotion‐focused interventions were a component of no significantly beneficial studies, but featured in two non‐significant studies (Table 2). The numbers of studies using emotion‐focused techniques were small for all outcomes.

Discussion

Summary of main results

Summary by outcome

1. Postoperative pain

The meta‐analysis suggested that psychological preparation may reduce postoperative pain (SMD ‐0.20, 95% CI ‐0.35 to ‐0.06), although these findings should be treated with caution because of the high heterogeneity (Figure 4). For most intervention types, results were similar to the omnibus analysis over all intervention types, with the exception of the analyses for behavioural instruction – there was no evidence that behavioural instruction reduced pain. Most studies included in the narrative synthesis found no statistically significant difference between intervention and control groups. While none of the narratively synthesized studies contained `pure' behavioural instruction, 12 of the 16 studies reporting non‐significant differences contained behavioural instruction as a component (Chumbley 2004; Dewar 2003; Hawkins 1993; Johnson 1978b; Johnson 1985; Lilja 1998; Oetker‐Black 2003; Parthum 2006; Shuldham 2002; Vukomanović 2008; Watt‐Watson 2004; Wijgman 1994).

2. Behavioural recovery

We did not conduct meta‐analyses for this outcome as there were few studies, there was large variation in reported outcomes and usable data were often not reported. The evidence was promising, suggesting that psychological preparation, in particular behavioural instruction, has potential to improve behavioural recovery outcomes, but no clear conclusions could be reached. We identified a need for more consistent use of outcome measures and clearer reporting so that findings can be compared across studies.

3. Negative affect

In meta‐analysis, there was some evidence of a beneficial effect of psychological preparation techniques on postoperative negative affect (SMD ‐0.35, 95% CI ‐0.54 to ‐0.16), although high statistical heterogeneity reduces the confidence that can be placed in this finding (Figure 5). While the pooled effect size of ‐0.35 would be regarded as a `small' effect (Cohen 1988), it could still be clinically important. There did not appear to be evidence that certain techniques performed better than others in reducing negative affect. In the narrative synthesis, many studies either reported null effects, or did not report analyses for the negative affect outcome of relevance to the review, even though authors reported measuring it. There was some suggestion that individual characteristics (e.g. level of preoperative fear) may affect the way that psychological preparations impact on postoperative outcomes. Overall, psychological preparation may benefit postoperative negative affect but the high level of heterogeneity in the data makes it difficult to determine the circumstances and intervention content that would consistently improve outcomes.

4. Length of stay

The meta‐analyses suggested that psychological preparation led to a reduction in mean length of stay of around half a day (MD ‐0.52 days, 95% CI ‐0.82 to ‐0.22), an effect size that could have considerable impact for patients and clinicians and represent savings in healthcare resources (Figure 6). However, the effect must be interpreted with caution because of high statistical heterogeneity. There appeared to be a similar benefit of psychological preparation for all intervention types. Studies included in the narrative review generally reported findings that were not statistically significant.

Summary by intervention

The number of studies using each intervention for each outcome varied. In general, pooled effect sizes tended to be similar regardless of the intervention types used and there was no clear evidence that results differed according to intervention. However, a different pattern did seem to emerge for behavioural instruction, for which there was no evidence of an effect for the outcomes postoperative pain or negative affect, but which was a component in all studies that successfully improved the outcome behavioural recovery. This difference may relate to the mechanism by which interventions are expected to take effect. Most of the intervention techniques included in the review are anticipated to improve recovery by reducing negative emotions (such as anxiety, worry about surgery, perceptions of stress) or enhancing relaxation, or both. Behavioural instruction is different: its goal is to help people to change their behaviour in such a way that their recovery is facilitated. Thus, it may be that reducing negative emotion before surgery is key to patients experiencing lower pain and lower negative affect after surgery, but when it comes to supporting patients' return to usual activities then behavioural instruction is more important. However, this is a cautious explanation as the behaviours targeted by behavioural instruction vary widely ‐ for example Chumbley 2004 addressed use of patient‐controlled analgesia ‐ such behavioural instruction might be expected to reduce pain, even though Chumbley 2004's findings were not statistically significant.

Overall completeness and applicability of evidence

This review addressed elective surgery where at least some patients underwent general anaesthesia. Findings cannot, therefore, be generalized to non‐elective procedures, or those where local anaesthesia is routinely used. It also cannot be assumed that similar findings would result in research with children rather than adults. However, as we did not limit the surgical procedures further, these findings would potentially be generalizable across elective surgical procedures, although the high degree of heterogeneity causes some concern. In future work, we plan to carry out secondary analyses to examine the impact of psychological preparation by surgery type.

The effect in our meta‐analyses for postoperative pain and negative affect may appear to be small according to Cohen 1988 (SMD = ‐0.20 and ‐0.35 respectively). However, given the high prevalence of surgery, such effect sizes may still be of clinical or cost significance, or both. There were 4.7 million surgical admissions in 2013‐14 in England alone, with common procedures being operations of the type included in the present review (120,198 hernia repairs, 197,348 hip or knee replacements, 76,497 gall bladder removals) (Royal College of Surgeons 2016).

The focus of the present review was the content of intervention – the types of psychological techniques used. We did not examine how the interventions were delivered – whether the timing or the format of the intervention is important. These are issues addressed by Nicholson 2013 (published Cochrane protocol). We would expect Nicholson 2013's findings to complement those of our review; together the reviews will evidence the current state of knowledge for the preparation of surgical patients.

Some types of intervention included as psychological preparation techniques in the review, for example procedural information and behavioural instruction, might be considered to be `common sense' rather than `psychological'. However, the term `psychological' encompasses what we think, what we feel and what we do, so any intervention that is designed to change what we think (for example, changing our expectations about what will happen) or what we do (for example, deep breathing after surgery) are effectively psychological techniques. An approach may appear to be `common sense' but still have a strong theoretical and evidence base to support it. For example, it may seem obvious that providing procedural information will help people to know what to expect and to feel prepared and less anxious about an event. However, it may seem equally obvious that giving someone procedural information in advance of surgery could increase anxiety by increasing thinking and worrying about a procedure. Thus, even such apparently unsophisticated procedures need to be considered and rigorously evaluated in the same way as more complex intervention techniques.

Studies included in the review measured outcomes at various time points, and some studies measured outcomes at multiple time points. In analysing the data, we used the earliest outcome measure reported by each study (see Characteristics of included studies for details of time points). It is possible that in the earliest times after surgery, within 48 hours, an outcome such as pain may be more influenced by biological factors such as analgesia intake and acute postoperative complications than by psychological aspects. Thus, by focusing on the earliest outcomes we may be under‐estimating the impact of psychological preparation. However, it is likely that psychological and biological factors interact ‐ for example, an intervention may include instruction in the use of using patient‐controlled analgesia (e.g. Chumbley 2004).

Quality of the evidence

We graded the quality of evidence as `low' for the outcomes postoperative pain, negative affect and length of stay and `very low' for the outcome behavioural recovery (see summary of findings Table for the main comparison). The two main problems with studies in the review were risk of bias ratings and heterogeneity. As seen in Figure 1, with the exception of performance bias (blinding of participants and personnel), a low proportion of studies received `high' risk of bias ratings. However, a large number of studies in each category received `unclear' ratings resulting from the poor reporting of studies. It is therefore not clear whether such studies were actually poorly conducted ‐ and therefore at high risk of bias ‐ or whether they were well designed and implemented but poorly reported. Improving reporting should be a primary aim for researchers ‐ and journal editors ‐ in this field. For the outcome behavioural recovery there was the further problem of a small number of studies including outcomes that were a) assessed with measures with demonstrated validity and reliability and b) reported in a form that could be included in meta‐analysis, meaning that only narrative synthesis could be conducted. High heterogeneity was also a problem, particularly in the varying content of interventions. Rather than simply label all interventions as being `psychological' we classified them into seven groups, which has enabled us to demonstrate the high level of variation across interventions. There was also heterogeneity in the wide range of surgery types participants underwent. Nevertheless, we did not find evidence of publication bias and all our outcomes were directly measured. For each outcome, there were some studies with small sample sizes and wide confidence intervals but overall imprecision does not appear to have been a problem.

In this review, we did not consider intervention fidelity ‐ whether interventions were delivered in accordance with the study protocol. This is particularly important where a complex psychological preparation technique is delivered by an individual rather than in a standard format such as a DVD or leaflet. It is important that the individual delivering the intervention is fully trained in the content and technique of delivery, and that this is evaluated during the study to ensure that interventions are indeed delivered as intended. If a fidelity check is not conducted then it is possible that important elements are missed, or that individuals add elements and it is unclear exactly what is being evaluated. We did record intervention fidelity processes on our data extraction forms, and observed that many studies do not seem to have addressed the issue. However, in accordance with our protocol, we did not formally include this when assessing risk of bias.

A limitation of our omnibus meta‐analyses is that they assume that diverse interventions have similar effects, whether separately or in combination with other types of interventions. In addition, most of the evidence for the separate meta‐analyses of individual intervention types came from studies judged to comprise at least one other intervention category. The present analyses can therefore only give a broad indication of the effectiveness of individual intervention types and we are unable to comment on how intervention types may interact with each other.

Potential biases in the review process

We conducted this large review in a careful and thorough manner: we not only carried out detailed searches of databases but also systematically sought to contact every included study's author to ask about additional research, and we searched studies' reference lists to ensure as complete coverage as possible. We also set high standards for the review, by only including randomized trials (excluding any work where we knew a random allocation method was not used), and by only including negative affect and behavioural recovery outcomes where measures with published psychometric properties were used. However, while we included papers in non‐English languages our search was limited in that we only conducted the searches in English, and we did not check the reference lists of non‐English papers. In addition, it is possible that unpublished studies exist of which we did not learn, but the funnel plots did not show evidence of asymmetry.

A relatively low proportion of the identified trials could be included in the meta‐analyses, however. This was sometimes because medians were presented, but there were also many studies that did not report any usable data for our outcomes of interest, despite having collected this. Extracting data from the publications was often challenging and we often had to calculate standard deviations from other statistics or use pooling formulae when authors chose to present data for subgroups only.

For the outcome length of stay, it should be considered that a number of studies appeared to regard length of stay not as an outcome, but as a descriptive statistic that was measured, reported and compared across groups. For example, Oliphant 2013 reported length of stay and compared this across groups, as one of many patient characteristics rather than as a specified outcome measure. The standard of reporting in papers was generally low, and it was often not possible to determine which outcomes were intended, a priori, to be treated as study outcomes, therefore we included any measure reports that fitted the definitions of our review. This was a conservative approach, as time to discharge may be short and largely determined by system factors. This is likely to have resulted in our findings erring on the side of over‐reporting non‐significant length of stay data (particularly in the narrative synthesis, where we could not pool studies to increase power).

We have reported meta‐analysis findings despite high levels of heterogeneity, which limits the confidence that can be placed in the findings. We believe that this is, however, helpful, as this is a large review and summarizing data in this way allows the findings to be more easily interpreted than placing so many studies in a table. In addition, as many studies contained small samples and individual results were often not statistically significant, combining studies allows a helpful picture of the potential of interventions. This practice has been followed in other Cochrane reviews, for example Gurusamy 2014 conducted meta‐analyses despite finding I² statistics of 75% and 87%. Although we did not identify specific reasons for this heterogeneity, it is clear that the studies were very diverse in terms of interventions, surgery types and outcomes used.

Combining interventions allows us to compare our findings with the earlier review (Johnston 1993), and we have, in the main, followed the analysis process as outlined in our protocol (Powell 2010). However, secondary analyses would be helpful in unpicking the cause of the heterogeneity and in identifying where benefits to patients may be obtained. A primary source of heterogeneity is in the varied way in which studies combined the intervention components included in the review. While we have conducted subgroup analyses examining `pure' and mixed' interventions for each outcome, we have not unpicked the value of each intervention component further. In future work we plan to carry out secondary analyses to explore how each individual component contributes to variance for each outcome.

A more challenging aspect of intervention content is that, within each intervention component, studies varied widely. For example, procedural information might focus on what will happen before surgery, the surgical process or what will happen after surgery. This issue was particularly pertinent with behavioural instruction – this could target a range of issues: exercises to be carried out before surgery to enhance strength, fitness or lung capacity (e.g. D'Lima 1996), exercises or movements to be carried out after surgery (e.g. deep breathing, how to turn in bed, e.g. Levesque 1977), behaviours to be carried out to gain effective pain relief (e.g. using patient‐controlled analgesia or asking for pain medication, e.g. Chumbley 2004). In future research we plan to carry out a secondary analysis to compare behavioural instruction targeting behaviours to be carried out before surgery with behavioural instruction targeting behaviours to be carried out after surgery. A further challenge in analysing the detail of psychological preparation interventions is the inconsistency between studies in how the interventions were reported. It was rare to find a sufficiently detailed description for the intervention to be replicated. A barrier to this would appear to be the lack of a standard language to describe intervention content in this context. A Behaviour Change Technique Taxonomy has been developed to enable researchers to describe, and code, interventions designed to effect behaviour change (Michie 2013), and the subset of studies in this review that target behaviour change could be recoded according to this taxonomy. However, many of the interventions do not explicitly aim to change behaviour – focusing instead on outcomes of perception (pain), emotion (negative affect) or a complex outcome that results from an interaction between patient clinical status, patient behavioural recovery and hospital strategy (length of stay). A taxonomy is needed that addresses a wider range of psychological interventions than those focused only on changing participant behaviour.

In future work we also plan to incorporate date of publication into secondary analysis (network meta‐synthesis). We did not exclude studies on the basis of date, ensuring the completeness of this review. However, the inclusion of studies over a wide period of time means that early studies may not reflect what would be found should the same interventions be used in modern practice. In particular, length of stay is now typically much shorter than that at the time of early studies such as DeLong 1970. The reporting of control condition content was generally very poor, but it is likely that, with approaches to patient care changing with psychological input into training of health professionals increasing and patient satisfaction gaining prominence, `standard care' in modern studies would contain a higher level of psychological preparation than in earlier studies. Thus, by assuming the absence of psychological techniques in `usual care' interventions unless stated, it is likely that we are over‐estimating the difference in treatment between patient groups within studies, and underestimating effect sizes. Evidence from other types of psychological intervention have indeed found that the prevalence of psychological techniques in the management of patients in the control groups reduced the effect sizes for trialled interventions (de Bruin 2009). In addition, management and clinical practices have changed, for example with hospitals seeking to discharge patients sooner. It might be expected, for example, that more recent studies would be less likely to show differences between groups in length of stay if length of stay has reduced over time.

As per our protocol, we did not extract studies' funding sources from papers. This is a limitation as we cannot comment on the potential impact of funding source on review outcomes. Similarly, we did not extract information about any conflicts of interest reported.

Potential biases resulting from differences between the protocol and the review

In the review, we more tightly defined the intervention types and what we meant by `psychological preparation for surgery'. We believe that this resulted in a stronger review, with interventions more clearly specified, but it may have led to the exclusion of some studies, which is likely to have affected the results.

In the review, we restricted inclusion to studies that reported one of the four outcomes postoperative pain, behavioural recovery, negative affect or length of stay. This was a pragmatic decision given the large size of the review, but excluding other outcomes means that we may have missed important impacts of psychological preparation. Importantly, if harm were identified on an outcome other than those we included, this review would not have detected it. We also refined our search criteria such that we only searched the reference lists of papers published in English. It is therefore possible that we missed studies in other languages, which may have resulted in bias.

We limited the subgroup analyses (as described in Differences between protocol and review). This has led to a more restricted range of findings but did not affect the planned analyses that we conducted.

In the review, we further specified the way we selected outcome measures where multiple measures were reported. While we conducted this process carefully and as objectively as possible, it would have been better to have pre‐specified the process to eliminate any potential for bias in the measures used.

We did not anticipate all the forms in which data might be presented in the protocol. We made the following decisions after seeing the data set: some studies only reported mean (SD) change from baseline (rather than absolute mean (SD)); for these studies we used the difference in mean change scores as the effect size. If no continuous pain data were available but dichotomous data were presented, we used the log odds ratio as the effect size. As we made these decisions after the data were available there is potential for bias.

Agreements and disagreements with other studies or reviews

Our review was based on that of Johnston 1993 but used different, more recently developed methods. We planned to carry out an up‐to‐date review, using modern techniques and standards. We used the same types of surgery and intervention categories, and examined four of the outcomes addressed by Johnston 1993. There are also some differences: Johnston 1993 required that patients have a postoperative night's stay – we did not make this a criterion as length of stay has reduced in recent years, and we adopted a different analysis strategy. We also had the additional criterion that behavioural recovery and negative affect measures needed to have published psychometric information for inclusion in the review. Johnston 1993 used the binomial test to pool data where studies did not provide the details for the calculation of pooled effect sizes, while we did not attempt to mathematically pool findings with insufficient details, according to standard Cochrane practice (Higgins 2011). Similarly to Johnston 1993, in omnibus analysis assessing whether psychological preparation has an impact on outcome (including all types of preparation), we found significant impact of preparation versus control for the outcomes of postoperative pain, length of stay and negative affect. For the outcome pain, Johnston 1993 reported relaxation, procedural information, cognitive interventions and behavioural instruction to be successful intervention components (assessed using the binomial test), whereas we found interventions other than behavioural instruction to generally appear to be beneficial. Effective preparations for negative affect in Johnston 1993 were procedural information, behavioural instruction, cognitive interventions and relaxation; our findings similarly suggested procedural information and relaxation could be beneficial, but also suggested that sensory information and hypnosis might be important, while we found no clear evidence for behavioural instruction or cognitive interventions. Johnston 1993 found all intervention methods other than cognitive interventions and hypnotic methods to be beneficial for the outcome length of stay; similarly, our meta‐analysis found most intervention types to be beneficial other than cognitive interventions, and we identified few studies for hypnosis or emotion‐focused interventions. Finally, Johnston 1993 found procedural information, sensory information and behavioural instruction to benefit behavioural recovery. We did not meta‐analyse data for this outcome, and we were more selective in which studies we included than were Johnston 1993 (we did not include measures without published psychometrics), but behavioural instruction was commonly used in successful interventions, and procedural information and sensory information also featured in some of these.

Other more recent reviews have addressed aspects of preparation that have also been covered in our review. Kekecs 2014 examined whether `suggestive interventions' (e.g. hypnosis and therapeutic suggestions) improved postoperative distress and pain intensity. They found evidence for suggestive interventions to reduce postoperative anxiety and pain intensity. Similarly, Tefikow 2013 found hypnosis to have positive effects on distress and pain in adults undergoing surgery or medical procedures, and found positive effects of the intervention for these outcomes. We also found studies including hypnosis (alongside relaxation) to be effective in reducing postoperative negative affect in our review. However, we did not include any studies with the component `hypnosis' in meta‐analysis for the outcome pain, and in meta‐analysis two studies with a hypnosis component reported non‐significant findings. This differences may be explained by Kekecs 2014 and Tefikow 2013 having different inclusion criteria to our review: surgical procedures under local anaesthesia, as well as general, were included in both reviews, and the search was limited to studies published after 1980 by Kekecs 2014.

Other reviews have focused on specific types of surgery. Louw 2013 and McDonald 2014 examined preparation for patients undergoing knee or hip replacement surgery. Both of these studies examined the effects of `education' interventions, which would appear to include both procedural information and behavioural instruction, on postoperative pain (Louw 2013 and McDonald 2014), and function, anxiety and length of stay (McDonald 2014 only). McDonald 2014 concluded that preoperative education may not be of benefit; Louw 2013 found benefits to pain to be limited (no meta‐analysis was conducted). McDonald 2014 focused on the latest time point studies included, while we used the first outcome time point assessed, Louw 2013 only included studies published between 1990 and 2011, and both included quasi‐randomized studies, unlike our review. However, our review does also include many studies with null findings – further subgroup analysis would help to establish for which interventions, and which types of surgery, preparation has most/least potential for benefit.

Gurusamy 2014 examined `information' interventions, whose content would appear to fit our categories of procedural information and potentially also behavioural instruction, with studies of patients undergoing day‐patient laparoscopic cholecystectomy. They evaluated the outcomes of pain, length of stay and anxiety, but found few studies addressing these outcomes (one for pain, none for length of stay, one for anxiety) and the authors concluded that the evidence had very low quality.

Hulzebos 2012 included both randomized and quasi‐randomized trials in adults undergoing elective cardiac surgery, with interventions described as "preoperative physical therapy with an exercise component" (p1). Such interventions would typically fit our category `behavioural instruction' where participants are instructed to carry out particular behaviours, such as exercise or incentive spirometry. They found that intervention participants had a significantly shorter length of stay but only one study included the outcome physical function (equivalent to `behavioural recovery'), finding a worse outcome for the intervention group.

A common finding across reviews is that studies are frequently small and of poor quality, consistent with our findings. Our review takes a broader approach than most reviews, including many more studies but of a higher quality, as we excluded quasi‐randomized trials. While heterogeneity of interventions is a problem in our review, we have been more specific in categorising and measuring components within interventions than some other reviews (i.e. those assessing the impact of `education' or `information'), allowing future secondary analyses to assess more precisely which components of interventions are effective.

Figure 1

`Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 2

Study flow diagram.

Figure 3

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

Figure 4

Figure 5

Figure 6

Analysis 1.1

Comparison 1 Any psychological preparation intervention versus control, Outcome 1 Pain.

Analysis 1.2

Comparison 1 Any psychological preparation intervention versus control, Outcome 2 Length of stay (days).

Analysis 1.3

Comparison 1 Any psychological preparation intervention versus control, Outcome 3 Negative affect.

Analysis 2.1

Comparison 2 Procedural information versus control, Outcome 1 Pain.

Analysis 2.2

Comparison 2 Procedural information versus control, Outcome 2 Length of stay (days).

Analysis 2.3

Comparison 2 Procedural information versus control, Outcome 3 Negative affect.

Analysis 3.1

Comparison 3 Sensory information versus control, Outcome 1 Pain.

Analysis 3.2

Comparison 3 Sensory information versus control, Outcome 2 Length of stay (days).

Analysis 3.3

Comparison 3 Sensory information versus control, Outcome 3 Negative affect.

Analysis 4.1

Comparison 4 Behavioural instruction versus control, Outcome 1 Pain.

Analysis 4.2

Comparison 4 Behavioural instruction versus control, Outcome 2 Length of stay (days).

Analysis 4.3

Comparison 4 Behavioural instruction versus control, Outcome 3 Negative affect.

Analysis 5.1

Comparison 5 Cognitive interventions versus control, Outcome 1 Pain.

Analysis 5.2

Comparison 5 Cognitive interventions versus control, Outcome 2 Length of stay (days).

Analysis 5.3

Comparison 5 Cognitive interventions versus control, Outcome 3 Negative affect.

Analysis 6.1

Comparison 6 Relaxation versus control, Outcome 1 Pain.

Analysis 6.2

Comparison 6 Relaxation versus control, Outcome 2 Length of stay (days).

Analysis 6.3

Comparison 6 Relaxation versus control, Outcome 3 Negative affect.

Analysis 7.1

Comparison 7 Hypnosis versus control, Outcome 1 Negative affect.

Analysis 8.1

Comparison 8 Emotion‐focused interventions versus control, Outcome 1 Pain.

Analysis 8.2

Comparison 8 Emotion‐focused interventions versus control, Outcome 2 Length of stay (days).

Analysis 8.3

Comparison 8 Emotion‐focused interventions versus control, Outcome 3 Negative affect.

Summary of findings for the main comparison. Any intervention compared to control for adults undergoing surgery under general anaesthesia

Any psychological preparation intervention compared to control for adults undergoing surgery under general anaesthesia
Patient or population: adults undergoing elective surgery under general anaesthesia Setting: pre‐surgical contexts (typically hospitals/preoperative clinic settings); setting was not limited by country/language/type of hospital Intervention: psychological preparation interventions presented to participants preoperatively; interventions contained one or more of the following components: procedural information; sensory information; behavioural instruction; cognitive intervention; relaxation techniques; hypnosis; emotion‐focused intervention Comparison: control group (typically standard care and/or attention control)
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control	Any intervention
Postoperative pain ‐ measured with a range of tools and placed on a standardized scale Higher scores = higher pain	‐	The mean pain in the intervention group was 0.2 (95% confidence interval 0.35 to 0.06) standard deviations lower	‐	2713 (38 RCTs)	⊕⊕⊝⊝ LOW¹	‐
Behavioural recovery ‐ measured with a range of tools	Insufficient data were available to calculate standardized scores	Findings suggested that psychological preparation has potential to improve behavioural recovery outcomes, but no clear conclusions could be reached	‐	1441 participants were randomized (14 RCTs)	⊕⊝⊝⊝ VERY LOW²	Data from studies were not combined in meta‐analysis because of a low number of studies containing suitable data and a wide range of outcome measures
Length of stay in hospital (days)	The mean length of stay for the control groups ranged from 2.11 to 18.6 days	The mean length of stay (days) in the intervention group was 0.52 days fewer (95% confidence interval 0.82 to 0.22)	‐	3313 (36 RCTs)	⊕⊕⊝⊝ LOW³	‐
Negative affect ‐ measured with a range of tools and placed on a standardized scale Higher scores = higher negative affect (e.g. more anxiety)	‐	The mean negative affect in the intervention group was 0.35 (95% confidence interval 0.54 to 0.16) standard deviations lower	‐	2496 (31 RCTs)	⊕⊕⊝⊝ LOW⁴	‐
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RCT: randomized controlled trial
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Many studies reported insufficient methodological details to ascertain risk of bias (rated `serious', see Figure 1), and heterogeneity was high (71%, also rated `serious'). We therefore downgraded the overall quality of evidence by two points. ²We downgraded the quality of evidence as `risk of bias' was rated as `very serious' ‐ there were a high proportion of `uncertain' ratings for risk of bias categories, and the number of studies with robust measures meeting our inclusion criteria and reporting suitable data for meta‐analysis was low. We made a further downgrade for high heterogeneity (treated as `serious'). We therefore downgraded the overall quality of evidence by three points. ³Many studies reported insufficient methodological details to ascertain risk of bias (rated `serious', see Figure 1), and heterogeneity was high (74%, also rated `serious'). We therefore downgraded the overall quality of evidence by two points. ⁴Many studies reported insufficient methodological details to ascertain risk bias (rated `serious', see Figure 1), and heterogeneity was high (81%, also rated `serious'). We therefore downgraded the overall quality of evidence by two points.

Summary of findings for the main comparison. Any intervention compared to control for adults undergoing surgery under general anaesthesia

Table 1. Findings of studies that examined the outcome pain but could not be included in meta‐analyses

Author, year	Surgery type and sample size (randomized)	Intervention categories	Pain measure(s) The first measure listed is that prioritized in this review	Pain findings (as available)
Chumbley 2004	Mixed: surgeries that would receive PCA routinely N = 246	Intervention 1: Behavioural instruction (delivered in leaflet) Intervention 2: Behavioural instruction (delivered in interview)	1) Visual analogue scale (VAS) days 1 to 5 post‐surgery 2) Word rating on 5‐point scale; days 1 to 5 post‐surgery	Cluster‐randomized VAS day 1 postoperatively mean (95% CI): Control: 3.7 (2.93 to 4.45); Intervention 1: 2.8 (2.04 to 3.56); Intervention 2: 3.2 (2.43 to 6.21). ANOVA, repeated measures: for VAS pain scores, between‐groups effect: F = 1.88, P value = 0.23
Daltroy 1998	Total hip or knee arthroplasty N = 12	Procedural and sensory information	Day 4 post‐surgery Measure not clearly described, assume same as preoperatively: mean of 3 x 5‐point scales assessing pain at night, resting and when active	Intervention did not affect pain in general linear model (P value = 0.16)
Dewar 2003	Mixed surgeries N = 254	Procedural information, behavioural instruction, cognitive intervention, relaxation	Evening after surgery (day 0) Brief Pain Inventory: numerical rating scale from 0 to 10	Control n = 118; intervention n = 104 No significant difference
Enqvist 1997	Breast reduction N = 50	Relaxation, hypnosis	Days 1 to 5 post‐surgery, measured with `10‐degree VAS’. Not clear exactly what was asked, or if measured once in this period or daily	Control n = 25; intervention n = 23 No significant differences
Ferrara 2008	Total hip replacement N = 23	Behavioural instruction	15 days and 4 weeks post‐surgery: VAS Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) subscale	Control n = 12; intervention n = 11 VAS pain scores: significantly lower in intervention group at 4 weeks (not at 15 days apparently)
Field 1974	Mixed orthopaedic surgery N = 60	Procedural information, hypnosis	Between 2 and 7 days post‐surgery; no further information	Control n = 30; intervention n = 30 No significant difference
Gilbey 2003	Total hip arthroplasty N = 76	Behavioural instruction	3 weeks post‐surgery Pain domain of WOMAC	Control n = 25; intervention n = 32 Significant difference (P value < 0.01) for total WOMAC (pain, physical function and stiffness) and physical function domain. Reports surgery had such beneficial effect on pain that impact of intervention only marginal.
Hawkins 1993	Gynaecological surgery N = 60	Behavioural instruction	48 hours post‐surgery: VAS of average pain; categorical scale (5 categories from no pain to unbearable pain); nurse ratings of pain (collected hourly pain reports when not sleeping for first 48 hours after surgery)	Control n = 40 (standard care and attention control); intervention n = 20 No significant differences (VAS ANOVA F = 0.06, df = 2, P value = 0.93)
Johnson 1978b	Sample 1: cholecystectomy, N = 81 Sample 2: inguinal hernia repair, N = 68	Intervention 1: ‘Instruction’: Behavioural instruction (deep breathing, coughing, leg exercises) Intervention 2: ‘Procedure information’: focus procedural information, also some sensory information and behavioural instruction Intervention 3: `Sensation information’: focus: sensory information, also some procedural information and behavioural instruction 2 x 3 factorial design: no instruction/instruction (Intervention 1; no information/information (Interventions 2 and 3)	Pain: days 1, 2 and 3 post‐surgery: intensity of sensations on 10‐point scale Scores totaled over the 3 days in analysis	Sample 1 No main effect of condition Sample 2 MANOVA with DVs pain and distress of pain sensation: for first postoperative day: significant main effects for information level (F(4, 104) = 2.55, P value < 0.05), trend for an effect for instruction (F(2, 52) = 3.07, P value = 0.055), but only a main effect for distress scores reported (no univariate findings reported for pain – so seems no significant effects)
Johnson 1985	Abdominal hysterectomy N = 199	Intervention 1: Procedural and sensory information Intervention 2: ‘Cognitive‐coping technique’ – cognitive intervention Intervention 3: ‘Behavioural‐coping technique’ – behavioural instruction 2x3 factorial design: no information/information (Intervention 1); no coping technique/coping technique (Interventions 2 and 3)	Day 3 post‐surgery Pain scale from 1 to 10	MANOVA, controlling for covariates, with various outcomes including pain: ‘significant’ at P value < 0.10: coping technique, F (16, 286) = 1.59, P value =0.07. However, pain does not appear to be one of the outcomes responsible for this.
Kulkarni 2010	Major abdominal surgery N = 80	Intervention 1: Behavioural instruction (deep breathing training) Intervention 2: Behavioural instruction (incentive spirometry) Intervention 3: Behavioural instruction (specific inspiratory muscle training)	Pain (no information of how measured/when)	Control n = 17; intervention 1 n = 17; intervention 2 n = 15; intervention 3 n = 17. Median pain score for all groups is 3 (no ranges/IQRs)
Lilja 1998	Breast cancer (BC) surgery N = 46 Total hip replacement (THR) N = 55	Procedural information, behavioural instruction	First 3 days post‐surgery: VAS	Control: n = 22, mode = 1 (BC day 1); intervention n = 22 No significant differences groups for either BC or THR patients (analysed separately)
Liu 2004	Mixed orthopaedic surgery N = 74	Cognitive intervention	Pain: 0 to 10 VAS; timing not stated	Control n = 35, mean (SD)= 2.5 (0.52); intervention n = 39, mean = 2.85 (0.33) Significant difference (t = 2.61, P value < 0.05). Discussion: authors state “patients from the experimental group…had…low scores on pain compared to the control group with statistical significance” (p5). This appears to be at odds with mean scores, suggesting error in paper.
Oetker‐Black 2003	Total abdominal hysterectomy N = 108	Behavioural instruction, cognitive intervention, relaxation	Day 1 post‐surgery: VAS At discharge: bodily pain (Health Status Questionnaire)	No significant differences (VAS: t(1,105) = ‐0.54, P value = 0.591)
Parthum 2006	Cardiac surgery N = 93	Procedural information, sensory information, behavioural instruction	1. Pain intensity: VAS as part of modified McGill Pain Questionnaire (MPQ), day 1 postoperative and retrospective rating of pain while on ICU 2. Proportion of patients in pain postoperatively (cut off: VAS > 3 on above measures)	Cluster‐randomized Control n = 36, median (VAS current, at rest) = 4.0. Intervention: n = 37, median = 3.0 No significant differences between groups
Perri 1979	Vaginal hysterectomy N = 26	Relaxation	Self report. 1 and 3 days postoperation; ‘McGill‐Melzack Pain Questionnaire’ Observed. 1 and 3 days postoperation – observed pain behaviour – Chambers‐Price Rating Scale for Pain	Control n = 13; intervention mean = 13. No significant differences between groups (P value < 0.05)
Shelley 2007	Coronary artery bypass surgery N = 90	Cognitive intervention	At discharge (4 days post‐surgery): 10 cm VAS	Control n = 43; intervention n = 37 Significant interaction between group, self efficacy and external health locus of control (F(1,71) = 4.06, P value < 0.05). Post hoc analysis: trend‐level effects: smaller increase in pain for prepared patients than controls if high external health locus of control and low self efficacy. Matched control appraisal patients: increased pain in intervention group compared with controls (controlling for baseline pain).
Shuldham 2002	Coronary artery bypass surgery N = 356	Procedural information, behavioural instruction	Questionnaires presented on day 3 post‐surgery (or 3rd day after transfer to ward if still in intensive care unit on day 3 post‐surgery) Composite measure (including VAS, body map and categorical rating scale), authors used VAS in analysis	No significant differences (using Mann‐Whitney U): U = 10,197.5; Z = ‐0.72, P value = 0.47
Vukomanović 2008	Total hip arthroplasty N = 45	Procedural information, behavioural instruction	VAS at discharge: pain at rest and movement	Cluster‐randomized Control n = 20, mean (SD) = 6.2 (14.95); Intervention n = 20, mean (SD) = 3.95 (13.08) No significant difference in pain
Watt‐Watson 2004	Coronary artery bypass surgery N = 406	Behavioural instruction, cognitive intervention	Days 1 to 5 post‐surgery: McGill Short‐form. Scores: Present Pain Intensity: most severe pain in previous 24 hours Pain Rating Index (sensory, affective and total); Numerical Rating Scale (on moving and worst pain in previous 24 hours)	No main effect of group
Wells 1982	Cholecystectomy N = 12	No control group Intervention 1: ‘Control’: Sensory information; behavioural instruction Intervention 2: (do not appear to receive ‘control’ intervention) Relaxation	Rated on 10 cm line on evening on day of surgery, and days 1 and 2 post‐surgery	Intervention 1: n = 6, mean (SD) eve of operation = 5.4 (3.39); intervention 2: n = 6, mean (SD) = 5.65 (1.6) No main effect for treatment (F(1,7) = 3.0, P value = 0.13), time (F(7,2) = 3.3, P value = 0.07) or interaction between treatment and time (F(2,4) = 1.0, P value = 0.4)
Wijgman 1994	Total knee arthroplasty N = 64	No control group Intervention 1: Procedural information Intervention 2: Behavioural instruction	2, 5, 7, 10, 14 days post‐surgery and at discharge. VAS where 100 = worst pain	Overall n at day 2 = 63. Medians (IQRs) presented in Figure 1, not clear. No significant differences between groups
Ziemer 1982	Gynaecologic or gastrointestinal N = 111	Intervention 1: Sensory information Intervention 2: Sensory information, behavioural instruction, cognitive intervention, relaxation	2 to 4 days post‐surgery: 5‐point pain intensity rating scale	Control n = 40; intervention 1 n = 34; intervention 2 n = 37 Focus: correlation of pain with coping scales
ANOVA = analysis of variance BC = breast cancer F = F statistic (ANOVA) ICU = intensive care unit IQR = interquartile range MANOVA = multivariate analysis of variance MPQ = McGill Pain Questionnaire (Melzack 1975) N = number of participants in sample PCA = patient‐controlled analgesia SD = standard deviation THR = total hip replacement VAS = visual analogue scale

Table 1. Findings of studies that examined the outcome pain but could not be included in meta‐analyses

Table 2. Findings of studies that examined the outcome behavioural recovery

Author, year

Surgery type and sample size (randomized)

Intervention categories

Behavioural recovery measure(s)

The first measure listed is that prioritized in this review

Behavioural recovery findings (as available)

D'Lima 1996

Total knee replacement

N = 30

Intervention 1: Behavioural instruction

Intervention 2: Behavioural instruction

3 weeks post‐surgery

Function scale from Hospital for Special Surgery Knee Rating; high score = better function

Control mean = 35, n= 10

Intervention 1 mean = 32, n = 10

Intervention 2 mean = 30.5, n = 10

"in the immediate postoperative period both exercise groups showed a steeper decline in function than the control group"; statistics not provided

Ferrara 2008

Total hip replacement

N = 23

Behavioural instruction

15 days and 4 weeks post‐surgery:

Disability (Barthel Index) (high scores: less disabled)

Functional status (from WOMAC); high scores = worse function

Intervention n = 11, control n = 12

No data/findings reported for these time points (study focus: 3 months postoperation)

Fortin 1976

Herniorraphy, cholecystectomy, intra‐pelvic surgery (primarily hysterectomies)

n = 69

Procedural information, behavioural instruction

Day 10 post‐surgery: ‘Activities of Daily Living’ (ADL). Capacity to perform tasks appropriate to normal life at home. Higher level (max = 3) = more independent.

Authors combined levels 1 and 2 in analysis

2 days IAA: Intervention n at level 3/total N = 27/37, control group = 5/32

10 days ADL: Intervention n at level 3/total N = 27/36, control group = 8/31

Better function in intervention than control group with both assessments

Analysing 29 matched pairs, significant difference at 2 and 10 days (P value < 0.01 for each, Wilcoxon matched pairs)

Full sample: also significantly different at both time points (Mann‐Whitney U, P value < 0.05 for each)

Gilbey 2003

Total hip arthroplasty

N = 76

Behavioural instruction

Week 3 post‐surgery:

Physical function domain of WOMAC

Intervention n = 32; control n = 25

Means/SDs presented only for total WOMAC scale, not for physical function domain. Significant difference (P value < 0.01) for physical function domain reported (intervention group scoring better).

Heidarnia 2005

Coronary artery bypass surgery

N = 80

Procedural information, behavioural instruction

1 month post‐surgery:

SF‐36 Physical Function (high scores = more active)

Nottingham Health Profile (NHP) Physical Mobility (high scores = greater dysfunction)

Intervention n = 35; control n = 35

SF‐36 Physical Function: Intervention mean = 25.3, control mean = 21.8

NHP Physical Mobility: Intervention mean = 32.97, control mean = 26.1

Independent t‐tests. Intervention group better than control group on both outcomes: SF‐36 Physical Function (P value < 0.00001); NHP Physical mobility P value < 0.00001)

Hoogeboom 2010

Total hip replacement

N = 21

Behavioural instruction

Iowa Level of Assistance Scale ‐ taken each postoperative day in hospital; authors used this to measure "time needed to reach functional independence": lower scores = more independent

Intervention: time to reach functional independence median 4 days (range 3 to 6, n = 8(?)); control group median 4 days (range 3 to 5, n = 10)

Difference in time to reach functional independence not significant (P value = 0.963)

Cholecystectomy

N = 125

Intervention 1: Procedural information, sensory information, behavioural instruction, emotion‐focused (at pre‐admission, 15 days before surgery)

Intervention 2: Procedural information, sensory information, behavioural instruction, emotion‐focused (afternoon before surgery)

First 2 post‐surgery days:

A postoperative recovery index; dimension "physical functional ability". Believe high scores = better outcome (not clear).

Cluster‐randomized trial. Data = mean (SD).

Intervention 1: day 1: 14.26 (3.4); day 2: 20.7 (2.5), n = 40

Intervention 2: day 1: 15.45 (3.16); day 2: 20.87 (2.43), n = 42

Control: day 1: 14.65 (3.02); day 2: 20.85 (2.17), n = 43

Mahler 1998

Coronary artery bypass surgery

N = 268

Intervention 1: Procedural and sensory information; behavioural instruction

Intervention 2: Procedural and sensory information; cognitive intervention

Intervention 3: Procedural and sensory information; cognitive intervention

Intervention 1: mean (SD) = 11.01 (1.02), n = 65

Intervention 2: 10.77 (1.02), n = 65

Intervention 3: 11.41 (1.12), n = 60

Control: 9.69 (0.85), n = 67

ANOVA and planned orthogonal comparisons. No significant effects by study group (P values < 0.60)

Total hip arthroplasty

N = 39

Procedural information, behavioural instruction

Before discharge:

Barthel Index: high score = less limited

WOMAC function (high scores = worse functional limitations)

Intervention n = 15; control n = 20

Barthel index: Intervention mean (SD): 19.8 (.4); Control: 18.7 (1.4)

WOMAC function: Intervention mean (SD): 25.7 (8.3); Control: 28.3 (12.1)

Total abdominal hysterectomy

N = 108

Behavioural instruction, cognitive intervention, relaxation

At discharge: Health Status Questionnaire (HSQ): Physical Functioning Subscale: high scores = better outcome

Length of time ambulated on first post‐surgery day

Mean (SD) not reported for HSQ

Ambulation: Intervention mean (SD): 330 (615); control 156 (97)

HSQ analyses are not presented by subscale

Ambulation: intervention participants ambulated longer than controls (F(1,105) = 2.05, P value = 0.043)

Oosting 2012

Total hip arthroplasty

N = 30

Behavioural instruction

4 days post‐surgery:

Iowa Level of Assistance Scale (ILAS), ability to function in daily life. Low scores = more independent. Split scores: < 6 (for "functional mobility" or ≥ 6

Intervention: 10 of n = 12 rated "functionally mobile"; control: 11 of n = 13 rated "functionally mobile"

No reported test of significance for this outcome

Ridgeway 1982

Abdominal hysterectomy

N = 60

Intervention 1: Procedural and sensory information

Intervention 2: Cognitive intervention

Diary record – days when performed 10 household activities over 3 post‐surgery weeks. For score: summed across tasks and no. days each was performed.

Intervention 1 mean = 6.6, n = 20

Intervention 2 mean = 6.9, n = 20

Control mean = 5.9, n = 20

CABG

N = 406

Intervention: Behavioural instruction, cognitive intervention

Days 3 and 5 post‐surgery: pain interference with general activities, sleep, walking, deep breathing and coughing (modified Interference Subscale of Brief Pain Inventory)

Zieren 2007

Inguinal hernia surgery

N = 100

Procedural information, behavioural instruction

DAy 1 post‐surgery:

SF‐36 physical functioning (high scores: less disability)

Intervention n = 50; control n = 50

No statistics presented. Observed that differences were visible on first postoperative day, with physical and psychological functions being less affected in intervention than control group.

ADL = activities of daily living

ANOVA = analysis of variance

CABG = coronary artery bypass graft

F = F statistic (ANOVA)

HSQ = Health Status Questionnaire

IAA = inpatient ambulatory activity

ILAS = Iowa Level of Assistance Scale

N = number of participants in sample

NHP = Nottingham Health Profile

SD = standard deviation

SF = Short Form

T = T statistic value (t‐test)

WOMAC = Western Ontario and McMaster Osteoarthritis Index

Table 2. Findings of studies that examined the outcome behavioural recovery

Table 3. Findings of studies that examined the outcome negative affect but could not be included in meta‐analyses

Author, year

Surgery type and sample size (randomized)

Intervention categories

Negative affect measure(s)

The first measure listed is that prioritized in this review

Negative affect findings (as available)

Barlési 2008

Thoracic surgery for non‐small cell lung cancer

N = 102

Procedural information

Timing unclear: at time of surgery (postoperative period) or 1 month post‐surgery

Psychologic Global Well‐being Scale; components include Anxiety, Depressed Mood and Positive Well‐being (also self control, general health, vitality)

Control n = 34; intervention n = 41

Mean/SD provided only for total scale (including non‐negative affect components). For the individual elements, no significant differences (no details provided).

Burton 1995

Mastectomy/sector mastectomy for breast cancer

N = 215

Intervention 1: Cognitive intervention and emotion‐focused (preoperative interview)

Intervention 2: Cognitive intervention and emotion‐focused (preoperative interview + 30 minute ‘chat’ on unrelated matters)

Intervention 3. Cognitive intervention and emotion‐focused (preoperative interview + 30‐minute brief psychotherapeutic intervention – additional emotion‐focused content)

Day 4 post‐surgery: Hospital Anxiety and Depression Scale (HADS) Anxiety and Depression.

Only report mean HADS scores for the overall sample, not by group at 4 days postoperation. Other negative affect also not reported by group at this time point.

Chumbley 2004

Mixed: surgeries that would receive PCA routinely

N = 246

Intervention 1: Behavioural instruction (leaflet)

Intervention 2: Behavioural instruction (interview)

24‐72 hours post‐surgery:

HADS Anxiety

Profile of Mood States (POMS) Tension/anxiety

Cluster‐randomized trial

HADS Anxiety: Control mean (95% CI) = 6.17 (5.34 to 8.00, n = 73); Intervention 1 mean (95% CI) = 6.03 (4.94 to 7.12, n = 75); Intervention 2 mean (95% CI) = 6.52 (5.59 to 7.45, n = 72)

No significant difference across groups (HADS anxiety, P value = 0.31; POMS tension/anxiety P value = 0.28)

Daltroy 1998

Total hip or knee arthroplasty

N = 222

Procedural and sensory information

Day 4 after surgery:

State Trait Anxiety Inventory (STAI) state anxiety

Intervention did not affect anxiety in general linear model (P value = 0.94). No interaction between intervention and denial, anxiety or desire for information. No main effects mentioned.

DeLong 1970

Gall bladder removal and removal of uterus

N = 70

Procedural information, sensory information, behavioural instruction

Day 5 or 6 after surgery:

STAI (state and trait anxiety)

No differences in anxiety scores across groups (no statistics provided)

Elsass 1987

Inguinal hernia or varicose vein surgery

N = 90

Procedural information

1 ½ hours after surgery and day after surgery

STAI state anxiety

Control n = 40; intervention n = 40. Anxiety scores are presented but unclear whether mean or median: Control score = 52; intervention score = 42 (reading off Figure 1).

Difference in scores between groups "increased significantly" at 1 ½ hrs after operation (P value < 0.05, Mann Whitney); intervention group less anxious

Gräwe 2010

Mixed: abdominal or vascular surgery

N = 96

Sensory information, cognitive intervention

Days 1 to 3 post‐surgery:

STAI state anxiety

BSKE – general psychological well‐being

Comparisons by group not reported for this outcome

Hawkins 1993

Gynaecological surgery (mixed)

N = 60

Behavioural instruction

48 hours after surgery:

Hospital Anxiety Scale

Control n = 40 (combining standard care and attention controls); intervention n = 20

No report of comparisons for this outcome

Johnson 1978b

Sample 1: cholecystectomy, N = 81

Sample 2: inguinal hernia repair, N = 68

Intervention 1: ‘Instruction’: Behavioural instruction (deep breathing, coughing, leg exercises)

Intervention 2: ‘Procedure information’: focus procedural information, also some sensory information and behavioural instruction

Intervention 3: `Sensation information’: focus: sensory information, also some procedural information and behavioural instruction

2 x 3 factorial design: no instruction/instruction (Intervention 1; no information/information (Interventions 2 and 3)

Scores totaled over days 1, 2 and 3: Mood Adjective Checklist (fear, well‐being, happiness, helplessness, anger)

Sample 1

Sample 2

Interaction between instruction and information (F(10,96) = 1.93, P value < 0.05) but no significant univariate findings, difficult to interpret

Johnson 1985

Abdominal hysterectomy

N = 199

Intervention 1: procedural and sensory information

Intervention 2: ‘cognitive‐coping technique’ – cognitive intervention

Intervention 3: ‘Behavioural‐coping technique’ – behavioural instruction

2 x 3 factorial design: no information/information (Intervention 1); no coping technique/coping technique (Interventions 2 and 3)

Day 3 post‐surgery (and 1^st and 4^th weeks post‐discharge):

Profile of Mood States (POMS: anxiety, confusion, anger, depression, fatigue, vigour). 3^rd postoperative day and 1^st and 4^th week post‐discharge.

Klos 1980

Cholecystectomy

N = 50

Intervention 1: Procedural information, behavioural instruction (pamphlet)

Intervention 2: Procedural information, behavioural instruction (nurse visit)

Intervention 3: Procedural information, behavioural instruction (pamphlet and nurse visit)

2^nd post‐surgery day:

Mood Adjective Checklist: 15 adjectives describing 5 mood dimensions: fear, well‐being, happiness, helplessness, anger

Authors did not report analyses by whole intervention group; instead, analyses are reported after median split into high‐ preoperative‐fear and low‐preoperative fear groups

2 x 2 factorial design: pamphlet/no pamphlet versus nurse visit/no nurse visit

Cholecystectomy

N = 125

Intervention 1: Procedural information, sensory information, behavioural instruction, emotion focused (15 days before surgery)

Intervention 2: Procedural information, sensory information, behavioural instruction, emotion focused (afternoon before surgery)

First 3 days after surgery

STAI (French version) state anxiety

Cluster‐randomized

Day 1 Control mean (SD) = 37.5 (8.51, n = 43); intervention 1 mean (SD) = 35.34 (9.34, n = 40); intervention 2 mean (SD) = 37.38 (8.29, n = 42)

No significant difference between groups for postoperative state anxiety

Total hip arthroplasty

N = 39

Procedural information, behavioural instruction

Positive & Negative Affect Schedule (PANAS)

Control n = 20; intervention n = 15

No mention of findings for analysis by group. May only be presenting positive findings – if so, this would suggest null result.

O'Connor 2014

Surgery for rectal cancer

N = 85

Procedural information

Prior to discharge:

HADS anxiety and depression

Numerical data not reported for this outcome

Control group: slightly higher anxiety score but not significantly different; depression – similar means, not significantly different

Total abdominal hysterectomy

N = 108

Behavioural instruction, cognitive intervention, relaxation

Day 1 post‐surgery day and at discharge:

STAI state anxiety

Only analyses at later time points reported

Osinowo 2003

Not stated – participants from surgical and gynaecological wards

N = 33

Intervention 1: Cognitive intervention (Rational Emotive Therapy)

Intervention 2: Cognitive intervention (Self‐Instructional Training)

24 hours post‐surgery:

STAI state anxiety

HADS Anxiety

HADS Depression

HADS depression: no significant changes across time

Paper generally written unclearly

Shelley 2007

Coronary artery bypass surgery

N = 90

Cognitive intervention

Day 4 post‐surgery:

Distress (Depression, Anxiety and Stress Scales, DASS)

Control n = 43; intervention n = 37

Shuldham 2002

Coronary artery bypass surgery

N = 356

Procedural information and behavioural instruction

Day 3 post‐surgery:

Anxiety – HADS

Depression – HADS

‘tense and uptight’ – General Well‐being Questionnaire

‘worn out’ – General Well‐being Questionnaire

Control n = 156; intervention n = 173

No significant differences between variables at 3 days post‐surgery (using Mann‐Whitney U):

Anxiety: U = 11,636, Z = ‐0.28, P value = 0.78

Depression: U = 10,756; Z = ‐1.24, P value = 0.22

Tense and uptight: U = 10,008, Z = ‐1.27, P value = 0.21

Worn out: U = 9,717.5, Z = ‐1.49, P value = 0.14

CABG

N = 406

Behavioural instruction, cognitive intervention

Days 3 and 5 post‐surgery

Pain interference with mood; modified version of Interference Subscale of the Brief Pain Inventory (BPI‐I)

Findings are not reported for this outcome – it would appear that authors are only reporting significant findings so it seems likely that group differences were not significant

BPI‐I = Interference Subscale of the Brief Pain Inventory

BSKE (EWL) = Befindlichkeitsskalierung durch Kategorien und Eigenschaftswörter (measuring general psychological well‐being)

CABG = coronary artery bypass graft

CI = confidence interval

DASS = Depression, Anxiety and Stress Scales

DSM‐III = Diagnostic and Statistical Manual (of Mental Disorders), version 3

EHLC = external health locus of control

F = F statistic (analysis of variance)

HADS = Hospital Anxiety and Depression Scale

N = number of participants in sample

PANAS = Positive and Negative Affect Schedule

PCA = patient‐controlled analgesia

POMS = Profile of Mood States

SD = standard deviation

SIT = Self‐Instructional Training

STAI = State Trait Anxiety Inventory

U = U statistic (Mann‐Whitney test)

Table 3. Findings of studies that examined the outcome negative affect but could not be included in meta‐analyses

Table 4. Findings of studies that examined the outcome length of stay but could not be included in meta‐analyses

Author, year	Surgery type and sample size (randomized)	Intervention categories	Length of stay findings (as available)
Coslow 1998	Laparoscopic tubal ligation N = 30	Procedural information, sensory information and behavioural instruction	Intervention n = 15; control n = 15 No significant difference
DeLong 1970	Gall bladder removal and removal of uterus N = 70	Procedural information, sensory information, behavioural instruction	Intervention n = 31; control n = 33 Intervention significantly decreased no. days in hospital (F = 4.70, df = 1/62, P value < 0.05). Intervention mean standardized days 47.06; control mean standardized days 52.32. When analysed by coping style: intervention reduced length of stay for copers (F = 6.43, df =1/20, P value < 0.05), but not avoiders or non‐specific defenders.
Field 1974	Mixed orthopaedic surgery N = 60	Procedural information, hypnosis	Intervention n = 30; control n = 30 No significant difference
Gocen 2004	Total hip replacement N = 59	Behavioural instruction	Intervention n = 29; control n = 30 No significant difference (P value > 0.05)
Goodman 2008	Cardiac bypass surgery N = 188	Behavioural instruction, relaxation, emotion‐focused	Intervention median 8.5 (IQR 3.25, range 4 to 50 days, n = 91) Control median 9 (IQR 3, range 2 to 170 days, n = 90) No significant difference (Mann‐Whitney U = 0.29, P value not provided)
Greenleaf 1992	Coronary artery bypass surgery N = 32	Intervention 1: Hypnosis and relaxation Intervention 2: Hypnosis	No significant difference between the groups
Guo 2012	Cardiac surgery N = 153	Procedural information	Intervention median 14.0 days (IQR 9.3 to 19.8, n = 68) Control median 12.0 days (IQR 10 to 17, n = 67) No significant difference (P value = 0.17)
Hoogeboom 2010	Primary total hip replacement due to osteoarthritis N = 21	Behavioural instruction	Intervention median: 6 days (range 5 to 22, n = 0) Control median: 6 days (range 4 to 7, n = 10) No significant difference (P value = 0.228)
Hulzebos 2006b	CABG N = 279	Procedural information, behavioural instruction	Intervention median 7 days (range 5 to 41, n = 139) Control median 8 days (range 6 to 70, n = 137) Intervention group: significantly shorter stay. Mann‐Whitney U (z = ‐2.42, P value = 0.02).
Johnson 1978b	Sample 1: cholecystectomy, N = 81 Sample 2: inguinal hernia repair, N = 68	Intervention 1: ‘Instruction’: Behavioural instruction (deep breathing, coughing, leg exercises) Intervention 2: ‘Procedure information’: focus procedural information, also some sensory information and behavioural instruction Intervention 3: `Sensation information’: focus: sensory information, also some procedural information and behavioural instruction 2 x 3 factorial design: no instruction/instruction (Intervention 1; no information/information (Interventions 2 and3)	Sample 1 (Cholecystectomy) Length of stay: patients in Intervention 2 (‘Procedure information’) and Intervention procedure and Intervention 3 (‘Sensation information’): shorter postoperative stays than no‐information participants; only significant for sensation information (Dunnett’s t(3,64) = 3.45, P value < 0.001). Control (no instruction or information intervention): mean stay = 6.36, n = 10; Intervention 1 only: mean stay = 6.20, n = 14; Intervention 2 only: mean = 5.97, n = 14; Intervention 3 only: mean = 5.78, n = 12; Intervention 1 and Intervention 2: mean = 5.84, n = 14; Intervention 1 and Intervention 3: mean = 5.29, n = 13 Sample 2 (Hernia repair) No significant effects of interventions for length of stay
Johnson 1985	Abdominal hysterectomy N = 199	Intervention 1: procedural and sensory information Intervention 2: ‘cognitive‐coping technique’ – cognitive intervention Intervention 3: ‘behavioural‐coping technique’ – behavioural instruction 2 x 3 factorial design: no information/information (Intervention 1); no coping technique/coping technique (Interventions 2 and 3)	Outcomes entered into MANOVA included length of stay. Coping technique was significant using a P value < 0.10 criterion (F (16, 286) = 1.59, P value = 0.07). Cognitive‐coping group: longer hospitalization than control group (Dunnett’s t (3,150) = 2.52, P value < 0.025) Adjusted mean scores and sample size according to coping groups: Control mean = 6.56, n = 72; Intervention 2 mean = 6.97, n = 48; Intervention 3 mean = 6.50, n = 47
Klos 1980	Cholecystectomy N = 50	Intervention 1: procedural information, behavioural instruction (pamphlet) Intervention 2: procedural information, behavioural instruction (nurse visit) Intervention 3: procedural information, behavioural instruction (pamphlet and nurse visit)	Authors did not report analyses by whole intervention group; instead, analyses are reported after median split into high‐preoperative‐fear and low‐preoperative‐fear groups. An interaction effect was reported between preoperative fear and receiving the pamphlet (F(1,39) = 4.14, P value < 0.05). If high preoperative fear and received pamphlet, shorter stay than those with high fear who did not receive pamphlet (but difference in means non‐significant: 5.09 versus 5.79 days). If low preoperative fear and pamphlet: significantly longer postoperative stay than those who did not receive pamphlet (F(1,18) = 4.84, P value < 0.05; means = 5.64 and 4.45). Observations are made about length of stay in the nurse visit groups, but no statistical tests are reported. Low preoperative fear: means for stay length for Interventions 1, 2, 3 and Control respectively are: 5.64, 4.61, 5.05, 4.45 High preoperative fear: means for stay length for Interventions 1, 2, 3 and Control respectively are: 5.18, 6.02, 5.33, 5.91
Kulkarni 2010	Major abdominal surgery N = 80	Intervention 1: behavioural instruction (deep breathing training) Intervention 2: behavioural instruction (incentive spirometry) Intervention 3: behavioural instruction (specific inspiratory muscle training)	Intervention 1 (Deep breathing): median stay = 5 days (range 1 to 10, n=17); Intervention 2 (Incentive spirometry): median = 4 (range 2 to 22, n = 15); Intervention 3 (Inspiratory muscle training); median = 4 (range 1 to 13, n = 17) Control median stay = 6 (range 1 to 14, n = 17) No analysis is reported
Letterstål 2004	Abdominal aortic aneurysm open repair N = 52	Procedural and sensory information	Intervention: median = 11 days (range 4 to 34, n = 18) Control: median = 9 days (range 6 to 42, n = 17) Mann‐Whitney: no difference between groups (P value = 0.14)
Levesque 1977	Cholecystectomy (n = 82); hysterectomy (n = 54) Total N = 136	Procedural information, behavioural instruction, relaxation	No significant difference
Lévesque 1984	Cholecystectomy N = 125	Intervention 1: procedural information, sensory information, behavioural instruction, emotion‐focused (at pre‐admission, 15 days before surgery) Intervention 2: procedural information, sensory information, behavioural instruction, emotion‐focused (afternoon before surgery	Cluster‐randomized trial Intervention 1 mean (SD) = 5.85 (1.19), n = 40 Intervention 2 mean (SD) = 5.94 (1.42), n = 42 Control mean (SD) = 5.60 (1.05), n = 43 No analyses are reported for length of stay
Oetker‐Black 2003	Total abdominal hysterectomy N = 108	Behavioural instruction, cognitive intervention, relaxation	No significant difference: t(1,93) = ‐0.77, P value = 0.444)
Oliphant 2013	Pelvic reconstructive and/or urinary incontinence surgery N = 199	Behavioural instruction	Intervention median = 1 day (IQR 0 to 2, n = 93); control median = 1 day (IQR 0 to 2, n = 93) No significant difference (Mann‐Whitney U, P value = 0.63)
Omlor 2000	Inguinal hernia surgery or thyroidectomy N ≥ 211	Procedural information, relaxation	Intervention n = 103; control n = 105 No significant difference. The paper presents medians (ranges) for control and intervention groups, by each type of surgery and combined, but there appears to be an error as these are contradictory: Inguinal hernia, intervention: 7.5 (1 to 11); control: 8 (3 to 22) Thyroidectomy, intervention: 7.2 (2 to 16); control: 7.9 (4 to 13) Groups combined: intervention median 7.95; control median 7.4
Pellino 2005	Orthopaedic surgery procedures. 90 randomized; 83 consented (consent post‐randomization).	Procedural information, behavioural instruction	No significant difference Data reported: expected length of stay minus actual length of stay (days): Intervention mean = ‐0.46 (SD 1.00, n = 39) Control mean = ‐0.29 (SD 1.19, n = 35)
Rosenfeldt 2011	CABG and/or valve surgery N = 119	Behavioural instruction, cognitive intervention, relaxation	Intervention median = 6 days (IQR 5 to 8, n = 60) Control median = 6 days (IQR 5 to 8, n = 57) No significant difference (Wilcoxon, P value = 0.54)
Vukomanović 2008	Total hip arthroplasty N = 45	Procedural information, behavioural instruction	Cluster‐randomized trial Intervention mean (SD) = 9.8 (2.4), n = 20 Control mean (SD) = 10.2 (1.7), n = 20 No significant difference, P value ≤ 0.67
CABG = coronary artery bypass graft F = F statistic (analysis of variance) IQR = inter‐quartile range MANOVA = multivariate analysis of variance N = number of participants in sample SD = standard deviation

Table 4. Findings of studies that examined the outcome length of stay but could not be included in meta‐analyses

Comparison 1. Any psychological preparation intervention versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain Show forest plot	38	2713	Std. Mean Difference (Random, 95% CI)	‐0.20 [‐0.35, ‐0.06]

2 Length of stay (days) Show forest plot	36	3313	Mean Difference (IV, Random, 95% CI)	‐0.52 [‐0.82, ‐0.22]

3 Negative affect Show forest plot	31	2496	Std. Mean Difference (Random, 95% CI)	‐0.35 [‐0.54, ‐0.16]

Comparison 1. Any psychological preparation intervention versus control

Comparison 2. Procedural information versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain Show forest plot	12	1051	Std. Mean Difference (Random, 95% CI)	‐0.08 [‐0.26, 0.09]

1.1 Procedural information only	2	186	Std. Mean Difference (Random, 95% CI)	‐0.13 [‐0.42, 0.16]
1.2 Procedural information plus other intervention(s)	10	865	Std. Mean Difference (Random, 95% CI)	‐0.08 [‐0.29, 0.13]
2 Length of stay (days) Show forest plot	19	1983	Mean Difference (IV, Random, 95% CI)	‐0.63 [‐1.08, ‐0.18]

2.1 Procedural information only	1	76	Mean Difference (IV, Random, 95% CI)	‐6.0 [‐9.95, ‐2.05]
2.2 Procedural information plus other intervention(s)	18	1907	Mean Difference (IV, Random, 95% CI)	‐0.57 [‐1.01, ‐0.13]
3 Negative affect Show forest plot	17	1334	Std. Mean Difference (Random, 95% CI)	‐0.45 [‐0.75, ‐0.16]

3.1 Procedural information only	3	269	Std. Mean Difference (Random, 95% CI)	‐0.54 [‐1.25, 0.16]
3.2 Procedural information plus other intervention(s)	14	1065	Std. Mean Difference (Random, 95% CI)	‐0.43 [‐0.77, ‐0.10]

Comparison 2. Procedural information versus control

Comparison 3. Sensory information versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain Show forest plot	11	881	Std. Mean Difference (Random, 95% CI)	‐0.22 [‐0.47, 0.02]

1.1 Sensory information plus other intervention(s)	11	881	Std. Mean Difference (Random, 95% CI)	‐0.22 [‐0.47, 0.02]
2 Length of stay (days) Show forest plot	14	1236	Mean Difference (IV, Random, 95% CI)	‐0.71 [‐1.15, ‐0.27]

2.1 Sensory information plus other intervention(s)	14	1236	Mean Difference (IV, Random, 95% CI)	‐0.71 [‐1.15, ‐0.27]
3 Negative affect Show forest plot	12	919	Std. Mean Difference (Random, 95% CI)	‐0.55 [‐0.90, ‐0.19]

3.1 Sensory information plus other intervention(s)	12	919	Std. Mean Difference (Random, 95% CI)	‐0.55 [‐0.90, ‐0.19]

Comparison 3. Sensory information versus control

Comparison 4. Behavioural instruction versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain Show forest plot	21	1241	Std. Mean Difference (Random, 95% CI)	‐0.14 [‐0.33, 0.05]

1.1 Behavioural instruction only	9	523	Std. Mean Difference (Random, 95% CI)	0.01 [‐0.19, 0.21]
1.2 Behavioural instruction plus other intervention(s)	12	718	Std. Mean Difference (Random, 95% CI)	‐0.28 [‐0.57, 0.01]
2 Length of stay (days) Show forest plot	25	2338	Mean Difference (IV, Random, 95% CI)	‐0.51 [‐0.84, ‐0.19]

2.1 Behavioural instruction only	8	445	Mean Difference (IV, Random, 95% CI)	‐0.26 [‐0.55, 0.03]
2.2 Behavioural instruction plus other intervention(s)	17	1893	Mean Difference (IV, Random, 95% CI)	‐0.64 [‐1.12, ‐0.16]
3 Negative affect Show forest plot	13	1183	Std. Mean Difference (Random, 95% CI)	‐0.22 [‐0.46, 0.02]

3.1 Behavioural instruction only	3	472	Std. Mean Difference (Random, 95% CI)	0.18 [‐0.19, 0.55]
3.2 Behavioural instruction plus other intervention(s)	10	711	Std. Mean Difference (Random, 95% CI)	‐0.37 [‐0.65, ‐0.09]

Comparison 4. Behavioural instruction versus control

Comparison 5. Cognitive interventions versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain Show forest plot	6	355	Std. Mean Difference (Random, 95% CI)	‐0.02 [‐0.29, 0.25]

1.1 Cognitive intervention only	2	136	Std. Mean Difference (Random, 95% CI)	‐0.34 [‐0.68, ‐0.01]
1.2 Cognitive intervention plus other intervention(s)	4	219	Std. Mean Difference (Random, 95% CI)	0.17 [‐0.09, 0.44]
2 Length of stay (days) Show forest plot	9	1074	Mean Difference (IV, Random, 95% CI)	‐0.43 [‐1.07, 0.22]

2.1 Cognitive intervention only	2	77	Mean Difference (IV, Random, 95% CI)	0.62 [‐0.74, 1.99]
2.2 Cognitive intervention plus other intervention(s)	7	997	Mean Difference (IV, Random, 95% CI)	‐0.59 [‐1.27, 0.08]
3 Negative affect Show forest plot	5	251	Std. Mean Difference (Random, 95% CI)	‐0.20 [‐0.52, 0.12]

3.1 Cognitive intervention only	3	173	Std. Mean Difference (Random, 95% CI)	‐0.08 [‐0.58, 0.42]
3.2 Cognitive intervention plus other intervention(s)	2	78	Std. Mean Difference (Random, 95% CI)	‐0.39 [‐0.83, 0.05]

Comparison 5. Cognitive interventions versus control

Comparison 6. Relaxation versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain Show forest plot	13	891	Std. Mean Difference (Random, 95% CI)	‐0.46 [‐0.81, ‐0.11]

1.1 Relaxation only	7	417	Std. Mean Difference (Random, 95% CI)	‐0.71 [‐1.29, ‐0.13]
1.2 Relaxation plus other intervention(s)	6	474	Std. Mean Difference (Random, 95% CI)	‐0.19 [‐0.58, 0.21]
2 Length of stay (days) Show forest plot	7	473	Mean Difference (IV, Random, 95% CI)	‐0.97 [‐1.94, ‐0.00]

2.1 Relaxation only	2	60	Mean Difference (IV, Random, 95% CI)	‐0.80 [‐2.25, 0.64]
2.2 Relaxation plus other intervention(s)	5	413	Mean Difference (IV, Random, 95% CI)	‐1.08 [‐2.39, 0.24]
3 Negative affect Show forest plot	11	687	Std. Mean Difference (Random, 95% CI)	‐0.34 [‐0.56, ‐0.12]

3.1 Relaxation only	4	256	Std. Mean Difference (Random, 95% CI)	‐0.26 [‐0.57, 0.04]
3.2 Relaxation plus other intervention(s)	7	431	Std. Mean Difference (Random, 95% CI)	‐0.40 [‐0.73, ‐0.08]

Comparison 6. Relaxation versus control

Comparison 7. Hypnosis versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Negative affect Show forest plot	2	72	Std. Mean Difference (Random, 95% CI)	‐0.77 [‐1.25, ‐0.30]

1.1 Hypnosis plus other intervention(s)	2	72	Std. Mean Difference (Random, 95% CI)	‐0.77 [‐1.25, ‐0.30]

Comparison 7. Hypnosis versus control

Comparison 8. Emotion‐focused interventions versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Pain Show forest plot	3	180	Std. Mean Difference (Random, 95% CI)	‐0.42 [‐0.85, 0.00]

1.1 Emotion‐focused only	1	18	Std. Mean Difference (Random, 95% CI)	0.12 [‐0.76, 1.00]
1.2 Emotion‐focused plus other intervention(s)	2	162	Std. Mean Difference (Random, 95% CI)	‐0.54 [‐0.97, ‐0.10]
2 Length of stay (days) Show forest plot	3	212	Mean Difference (IV, Random, 95% CI)	0.14 [‐0.67, 0.94]

2.1 Emotion‐focused plus other intervention(s)	3	212	Mean Difference (IV, Random, 95% CI)	0.14 [‐0.67, 0.94]
3 Negative affect Show forest plot	4	201	Std. Mean Difference (Random, 95% CI)	‐0.24 [‐0.55, 0.07]

3.1 Emotion‐focused only	1	18	Std. Mean Difference (Random, 95% CI)	0.12 [‐0.76, 1.00]
3.2 Emotion‐focused plus other intervention(s)	3	183	Std. Mean Difference (Random, 95% CI)	‐0.30 [‐0.66, 0.06]

Comparison 8. Emotion‐focused interventions versus control