Social norms interventions to change clinical behaviour in health workers: a systematic review and meta-analysis

Social comparison

‘Draw attention to others’ performance to allow comparison with the person’s own performance. Note: being in a group setting does not necessarily mean that social comparison is actually taking place.’24

‘Example: Show the doctor the proportion of patients who were prescribed antibiotics for a common cold by other doctors and compare with their own data.’24

Information about others’ approval

‘Provide information about what other people think about the behaviour. The information clarifies whether others will like, approve or disapprove of what the person is doing or will do.’24

‘Example: Tell the staff at the hospital ward that staff at all other wards approve of washing their hands according to the guidelines.’24

Credible source

‘Present verbal or visual communication from a credible source in favour of or against the behaviour. Note: code this BCT if source generally agreed on as credible e.g. health professionals, celebrities or words used to indicate expertise or leader in field and if the communication has the aim of persuading.’24

‘Example: Present a speech given by a high status professional to emphasise the importance of not exposing patients to unnecessary radiation by ordering x-rays for back pain.’24

The following two social norm BCTs have been amended slightly to ensure that the definition is sufficiently tight to allow us to identify and delineate interventions. Further details of the reasons for change are available in the protocol. 2

Social reward

The original definition of social reward was ‘Arrange verbal or non-verbal reward if and only if there has been effort and/or progress in performing the behaviour (includes ‘Positive reinforcement’). Examples: Congratulate the person for each day they eat a reduced fat diet’. 24

Our new tighter definition is as follows: arrange praise, commendation, applause or thanks if and only if there has been effort and/or progress in performing the behaviour (includes ‘positive reinforcement’). Example: arrange for a family doctor to be sent a thank you note for each week that they reduce their level of antibiotic prescribing.

Social incentive

The original definition for social incentive was ‘Inform that a verbal or non-verbal reward will be delivered if and only if there has been effort and/or progress in performing the behaviour (includes ‘positive reinforcement’). Examples: Inform that they will be congratulated for each day that they eat a reduced fat diet.’24

Our new tighter definition is as follows: inform that praise, commendation, applause or thanks will be delivered if and only if there has been effort and/or progress in performing the behaviour (includes ‘positive reinforcement’). Example: promise a family doctor in advance that they will be sent a thank you note for each week that they reduce their level of antibiotic prescribing.

Population

The population of interest was health workers (including managers) responsible for patient care in a health-care setting. Health workers in training were included, but only if they were in a health-care setting (i.e. not in campus or laboratory environments). Any health-care setting was eligible, including primary care, secondary care, care homes, nursing homes and patients’ own homes. Interventions taking place in simulated environments were not eligible for inclusion.

Interventions

A social norms intervention seeks to change the clinical behaviour of a target health worker by exposing them to the values, beliefs, attitudes or behaviours of a reference person or group. We looked for the five BCTs that we considered to have a social norms element to them: 6.2. social comparison, 6.3. information about others’ approval, 9.1. credible source, 10.4. social reward and 10.5. social incentive. However, we were open to including studies that met all other inclusion criteria and had a social norms element, even if they did not include any of these five BCTs.

Included studies must have stated a clinical behaviour of health workers that was targeted for change through the use of social norms. If the behaviour was not specified, it was not possible to determine which aspects of an intervention were relevant to the anticipated behaviour change. Indeed, the BCT taxonomy v1 coding guidance states that the target behaviour needs to be specified and BCTs must target that behaviour for BCTs to be coded. 28 Clinical behaviour here is defined as any behaviour that is performed within a (non-simulated) environment that affects patient diagnosis, care, treatment or recovery. We have reported the number of studies identified by our search that met all other inclusion criteria but did not mention a target behaviour.

Comparators

All comparators were eligible for inclusion, including alternative interventions, no intervention or comparison of one social norms BCT with one or more other social norms BCTs.

Study designs

Only randomised controlled trials (RCTs) were included in the review. All designs of RCTs (cluster, factorial, parallel, cross over and stepped wedge) were eligible for inclusion. The justification for restricting the review to RCTs was that the review is concerned with the effectiveness of social norms, and RCTs are the best method for assessing the effectiveness of an intervention.

We included both published and unpublished research. Studies had to be reported in English because the research team had no resource for translation from other languages.

Population

A list of population terms was developed by looking at Cochrane reviews25,29 that included a similar population of health workers, augmented by job roles included in the UK national workforce data set produced by NHS Digital. 30

Interventions

Social norms interventions are not described consistently in the literature, and different terms are used in various academic disciplines. This presented us with the challenge of finding appropriate search terms to make sure that we would discover the full range of literature on this topic. We were aware that many studies involving A&F contain a social comparison element;25 therefore, we looked at the search terms that were used in a previous systematic review of A&F. 25 We omitted anything relating solely to ‘audit’, as this was not relevant for this review.

During the scoping phase, various feedback terms were tried out. The use of ‘feedback’ alone produced many irrelevant papers, such as those relating to educational feedback and electronic feedback. The final search, following extensive trial and error in the piloting phase, included ‘feedback’ when used alongside other relevant terms (audit, monitoring, peer, performance, data, individualised, web, personalised, comparative, team, practitioner, practice and clinical or social). We also included ‘benchmark’. We included some overall terms that are used in the literature on social norms: ‘norm’ used close to ‘social’, ‘descriptive’, ‘peer’ or ‘subjective’; ‘social influence’; ‘benchmarking’; ‘social or peer comparison’; and ‘social competition’. Terms that appeared in behavioural economics literature were included: ‘social proof’, ‘image motivation’ and ‘warm glow’.

Additional search terms were developed for each of the five social norms BCTs by looking at the text used to describe them in the BCT taxonomy v1,24,31 extensive discussion in the team and examining relevant articles. Additional terms for information about others’ approval and credible source included ‘positive reinforcement’, ‘congratulate’, ‘praise’ and ‘commendation’. Terms for social reward and social incentive included ‘social’, ‘verbal’ and ‘non-verbal’ alongside ‘incentive’ or ‘reward’. Finally, the search included terms to describe theories that are used to explain interventions based on social norms: the Theory of Planned Behaviour,32 the Theory of Reasoned Action,33 the Theoretical Domains Framework,34 Social Cognitive Theory,35 and the Theory of Normative Social Behaviour. 22

Study design

The search for RCTs was taken directly from the Cochrane RCT search described in the Cochrane Handbook for Systematic Reviews of Interventions. 36 This was translated into other relevant databases.

The search was developed in MEDLINE and then adapted for other databases. Terms relating to the same concept (e.g. different types of health workers) were combined using the Boolean operator ‘OR’ and different concepts (e.g. health workers and social norms) were combined using ‘AND’. The search strategy was tailored for the different electronic databases using medical subject headings (MeSH) where appropriate, wildcard symbols and truncations (see Appendix 1). Backward- and forward-citation searching was not conducted owing to time and resource constraints.

Published literature was systematically searched on 24 July 2018 in electronic databases relevant to health and social care: Ovid MEDLINE, Ovid EMBASE, Healthcare Databases Advanced Search (HDAS) – Cumulative Index to Nursing and Allied Health Literature (CINAHL), HDAS British Nursing Index (BNI), Cochrane Central Register of Controlled Trials (CENTRAL), Ovid PsycINFO and Web of Science (see Appendix 1 for search strategies and Appendix 1, Table 15, for the results).

Study selection

The process for identifying studies for review followed the stages of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. 37

All references generated from the search were managed in Covidence (Melbourne, VIC, Australia): an online screening and data extraction tool for systematic reviews. All reviewers were provided with instructions for both the title and the abstract, and full-text screening stages (see Appendix 2). At the title and abstract screening stage there was an initial learning phase (305 studies), during which the coders worked steadily through the task, applying the inclusion criteria to the papers and stopping after small batches to discuss any discrepancies as they went along. Disagreements and uncertainties were discussed with the wider research team. This process enabled the main coder to build up a high level of consistency. For the remaining studies, one reviewer (MYT) independently screened all of the titles and abstracts against the inclusion criteria, and another researcher (SC, SR or JW) screened a sample of 20% of the records. These were randomly selected using a random integer generator (www.random.org; accessed 1 September 2020). By the time that 20% of the records (493 studies) had been screened, there was very little difference between the decisions of the two coders, and we were confident that the main coder could make the exclusion decisions with reliability. Inter-rater reliability on these 493 studies was good38 (kappa = 0.68). Where there was any hesitation on her part about whether to include or exclude, she erred on the side of inclusion and continued to discuss any uncertainties with the wider team.

All of the studies at the full-text stage were independently screened by two researchers from the screening team (MYT, SC or SR). The two reviewers screened the papers concurrently using Covidence, and were not aware of the other person’s recommendation until after they had entered their own. The screening involved reading the full-text paper and deciding whether or not the paper met the eligibility criteria (a population of health workers in a health-care setting, a social norms intervention targeted at clinical behaviour change and a RCT). If the study was excluded, the reviewer entered a reason for exclusion. Any disagreements over the recommendation to exclude or the reason for exclusion were flagged up by Covidence and the two reviewers met to discuss. If they were unable to come to a consensus, there was moderation by a third researcher or discussion at a team meeting.

Data extraction and management

For efficiency, data extraction was conducted in three stages: stage 1 involved extraction of all data apart from the details of the intervention, stage 2 was the BCT coding (carried out by a different team concurrently with stage 1) and stage 3 was carried out later, because it relied on data collected during the BCT coding (e.g. we needed to identify which aspects of the intervention were based on social norms to assess the frequency or format of the social norms intervention). Data from included studies were extracted using data extraction forms derived from the Cochrane Effective Practice and Organisation of Care data collection form. 39

Assessment of risk of bias

As part of the first stage of data extraction, risk of bias for each included study was independently assessed by the data extractors (LH, MYT, SC and SR) using The Cochrane Collaboration tool for assessing risk of bias36 across a range of criteria: selection bias, performance bias, detection bias, attrition bias, reporting bias, selective outcome reporting and other biases. Included studies were classified as having a high, low or unclear risk of bias for each criterion. All risk-of-bias criteria were added as part of the data extraction form in Covidence. Where disagreements occurred, a discussion between the two extractors took place to resolve the disagreement or a third data extractor would be brought in when an agreement could not be reached. Percentages of high/low/unclear judgements for each risk-of-bias criterion across included studies were calculated and reported as a bar chart to provide a summary of the risk of bias across criteria domains (see Figure 2). Text summaries across each criteria of bias were produced in line with The Cochrane Collaboration’s guidance for large reviews. 36 Judgements for each risk-of-bias criterion for all included studies were reported (see Appendix 15, Figure 20).

Outcomes and prioritisation

The primary outcome for the review was compliance of the health worker with the desired behaviour at the time point closest to 6 months post intervention. We expected studies to report different behaviours (e.g. prescribing, hand-washing and test ordering) and we expected studies to measure those behaviours in different ways. We converted any observed measure of health worker behaviour into a standardised mean difference (SMD) between groups in terms of compliance with the desired behaviour. Common examples included the mean number of times a behaviour was performed per health worker or the mean rate of behaviour (e.g. percentage of the population for whom antibiotic items were dispensed). At times, compliance was reported as a binary outcome, such as compliance versus non-compliance on a single occasion, and was expressed either in a binary format or using an odds ratio.

We used the methods of Chinn44 to convert binary outcomes to a SMD with associated standard errors (see Appendix 8 for the formula).

If several measures of compliance were reported in sufficient detail to enable the analysis of a trial, we used the following criteria to select the outcome for the primary analysis, in decreasing order of importance: (1) observed measure rather than self-report, (2) appropriate adjustment for clustering, (3) continuous measure, (4) final score rather than percentage change or change from baseline, (5) described as the primary outcome, (6) used to calculate the sample size and (7) reported first.

The secondary outcome for the review was patient health-related outcomes that were likely to result from targeting the health worker behaviour. These were converted to a SMD using a similar approach.

Some trials incorporated baseline measurements into their analyses. This was carried out either by adjusting for baseline values of the outcome measure or of other prognostic variables in an analysis of covariance (ANCOVA), or by reporting outcomes as changes from baseline. We have prioritised ANCOVA-adjusted estimates of the treatment effect where relevant or those from logistic regression, given that these are generally more precise. Change scores cannot be pooled through conversion to SMDs.

Missing data

Our preferred approach to dealing with missing data was to take steps to try to obtain them. We searched for companion papers by author searching and citation searching. Contacting trial authors was not possible owing to limited resources and the large number of studies. We imputed estimates of standard deviations where necessary by using any available information, such as p-values, confidence intervals (CIs), ranges or standard errors of baseline data, by pooling standard deviations from other similar studies that use the same type of outcome or by searching for trials that used the same outcome. Where necessary, for cluster randomised trials we imputed a value of the intraclass correlation coefficient (ICC) by pooling across similar studies.

Unit of analysis issues

Where any of the studies in the review were cluster randomised trials, we extracted both raw summary measures (e.g. means and numbers having had the event) and adjusted standard errors from appropriately analysed trials. Where it was not possible to obtain the adjusted SMD and its standard error directly, the methods that were used to calculate the SMD and standard error are shown in Appendix 9, Table 21.

Several studies had more than two relevant arms (e.g. two different social norms interventions and a control group). In each case, we extracted data on any comparison that was relevant to our primary research question, while avoiding double counting where possible. Where relevant, we combined study arms that contained identical BCTs. In cases with two different social norms interventions and a single control arm, where possible we divided the number of health-care workers in the control arm approximately evenly between the comparisons to avoid double counting, while retaining the correct intervention effect. In studies with more than one candidate control arm, we chose the comparison that provided the more pure test of social norms (e.g. social norms intervention + X vs. X is a more pure test of social norms than social norms intervention + X vs. usual care).

Where a study was a factorial trial analysed appropriately using linear or logistic regression, we extracted the covariate and standard error that best assessed the effect of social norms BCTs, for example a covariate comparing all arms containing a social norms BCT with all arms without.

Analysis of skewed data

If the primary outcome data were heavily skewed, meta-analyses based on SMDs of the untransformed data would be expected to produce biased estimates. In some cases, compliance was reported as ‘mean per cent compliance’ or similar, and there is a likelihood that this outcome is skewed when close to 0% or 100% owing to it being bounded. We removed data likely to be skewed (where mean compliance was close to 0% or 100%) in a sensitivity analysis.

Utilising the behaviour change technique coding in the analysis

The approach we took to utilising the BCTs in the meta-analysis was to create an Excel file of all the trials, listing the intervention and control BCTs on separate rows. We subtracted the control arm BCTs from the intervention arm BCTs to identify the BCTs that would be expected to be responsible for the differences between the two arms.

Using the five types of comparison (extracted during the BCT coding process), listed in Box 1, allowed us to separate out three different tests of social norms:

1. ‘Pure’ test of social norms intervention alone (comparisons 1 and 2, see Box 1).

   This involved trials with social norms BCT(s) in the intervention arm and no BCTs in the control arm (comparison 1). These trials were the purest test of social norms interventions: the BCTs being tested were those found in the intervention arm. For trials in which an intervention arm including a social norms BCT combined with other BCTs was tested against a control arm containing the same other BCTs (comparison type 2), the control arm BCTs were subtracted from the intervention arm to reveal the BCTs that would be expected to account for differences in outcome. For example, if the study tested social comparison and instructions on how to perform the behaviour (intervention arm) against instructions on how to perform the behaviour (control arm), the comparison type would be ‘social comparison’.

2. ‘Complex’ test of a social norms intervention alongside one or more other BCTs (comparison 3, see Box 1)

   This involved trials in which an intervention arm including a social norms BCT combined with other BCTs was tested against a control arm containing none of the same BCTs (comparison 3). The control arm was deducted from the intervention arm to reveal the test involved in the comparison. For example, if the study tested a complex intervention such as credible source, feedback on behaviour, social support unspecified and behavioural practice/rehearsal (intervention arm) against social support unspecified and behavioural practice/rehearsal (control group), the comparison would be ‘credible source’ and ‘feedback on behaviour’ versus control.

3. Social norms intervention occurring in both arms (comparisons 4 and 5; see Box 1)

   In some studies, two different social norms interventions were compared (comparison 4) or the same social norms intervention appeared in both arms (comparison 5). Where social norms interventions occurred in both arms of a trial, the study did not provide useful information for the meta-analysis, because these trials do not test the effect of social norms interventions, but they were potentially useful to the review as follows:

   • Any study that directly compared one social norms BCT against another (e.g. social comparison vs. credible source) could potentially be included in the network meta-analysis.

   • Any study that compared the same social norms BCT in both arms, with the addition of other BCTs (e.g. with the addition of social support in one of the arms) or comparing differing modes of delivery (e.g. social comparison delivered in person or by e-mail) could potentially be included in the metaregression.

   • Any study where social norms BCT(s) were delivered in both arms as a control intervention, for the purpose of testing a separate intervention, in which the social norm was a minor part were not included in any analysis.

In summary, the information extracted for the analysis describes the BCTs that were tested in the study rather than all of the BCTs that make up the intervention. In some cases (comparison 1) the content of the comparison is the same as the content of the intervention arm, but in most cases (comparison 2 and 3) the content of the comparison is what is left of the intervention when the control arm is taken away. We regard this as the part of the intervention that was actively tested in the trial. A limitation of this approach is that we may have missed some interaction effects.

Feedback on behaviour

Early on in our coding, we observed that the BCT ‘feedback on behaviour’ was often found to be presented alongside a social norms BCT. The implementation of three social norms BCTs (social comparison, social incentive and social reward) would seem to be greatly facilitated by combination with ‘feedback on behaviour’. Social comparison, defined as ‘draw attention to others’ performance to allow comparison with the person’s own performance’,24 does not by definition require feedback on the target’s own behaviour to be provided, but providing such feedback (e.g. performance data) would be expected to facilitate comparison. Social reward, ‘arrange verbal or non-verbal reward if and only if there has been effort and/or progress in performing the behaviour’,24 and social incentive, ‘inform that a verbal or non-verbal reward will be delivered if and only if there has been effort and/or progress in performing the behaviour’,24 similarly do not require feedback on the target’s behaviour to be provided (e.g. the behaviour could be monitored by others without feedback to make the reward/incentive process clear to a target), but feedback on the behaviour fits very well with these social norms BCTs and might be expected to facilitate the action of these BCTs.

Because of the high prevalence of feedback on behaviour (present in 88/100 comparisons), we combined ‘feedback on behaviour’ with the social norms BCT with which it appeared for the purpose of primary analyses: in the forest plots we have listed each social norm with or without feedback. However, it was important to unpick the separate effects of feedback on behaviour: this was examined as part of the metaregression. As a sensitivity analysis, we examined the overall effects of social norms interventions with and without feedback on behaviour.

Criteria for study data to be meta-analysed

We included in a meta-analysis those studies that report a primary outcome measure (clinical behaviour of a health worker) or secondary outcome (patient outcome) that can be converted into a SMD.

Planned approach for meta-analysis

Research question (RQ) 1: what is the effect of social norms interventions on the clinical behaviour of health workers, and the resulting patient outcomes?

The comparisons used in the analysis to answer RQ1 are shown in Appendix 10, Table 22. We stratified the studies in the forest plot according to the type of comparison (see Utilising the behaviour change technique coding in the analysis) and the type of target behaviour, and pooled estimates across strata. The aim of this was to provide some initial insight into whether or not, and how, treatment effects vary systematically in trials using different social norms techniques, while remaining aware of the likely confounding by other trial characteristics. We considered I² and tau when interpreting heterogeneity, but did not use it as the basis for analytic decisions. We preferred a fixed-effects approach rather than a random-effects approach to meta-analysis, which we consider to yield a summary of the evidence in these trials (i.e. the average effect), rather than an estimate of a common underlying treatment effect. However, we also reported a random-effects analysis.

Research question 2: which contexts, modes of delivery and BCTs are associated with health worker clinical behaviour change? To address this research question, we followed steps 1 to 3.

Step 1 – we explored sources of variation using forest plots and narrative description. In addition to those comparisons used in RQ1, we included the following types of comparison in a narrative description: (1) social norms intervention A versus social norms intervention B and (2) social norms intervention + X versus social norms intervention + Y, where X and Y are any BCT or combination of BCTs, and A and B are either two different types of social norms BCT or the same social norms BCT delivered by two different methods.

Step 2 – we undertook an exploratory analysis using multivariable metaregression to investigate sources of heterogeneity and explain variation in the results. Metaregression is an appropriate regression method in which weights are assigned to studies/subgroups based on the standard error of the treatment effect. Appendix 11, Table 23, shows the predictor variables together with anticipated parameterisations that we included in the metaregression analyses. Although controlling for multiple predictors at once is desirable, in practice this was governed by the number of trials and the observed distributions of the variables. We allowed for trials from the different comparisons to enter into a single metaregression given that we anticipated we would be able to control for comparators and co-interventions in the regression. We had intended to categorise the control conditions, but were unable to do this robustly.

Step 3 – we used network meta-analysis to explore which social norms BCT, combination of social norms BCTs or combination of social norms BCT with other BCTs, if any, appears most effective. We considered two broad approaches for network meta-analysis, and made the decision to employ type (a) after consultation with our SSC.

• Network meta-analysis

  We examined data from all trials to look at the most commonly occurring combinations of social norms BCTs, either alone or alongside other BCTs. We built and examined a network diagram including social norms BCTs and commonly occurring combinations of social norms BCTs with other BCTs, plus control. Decisions about whether or not to ‘lump together’ BCTs or combinations of BCTs into categories were made after careful discussion by the project team. The justifications were recorded. The geometry of the network diagram was evaluated and no revisions were required to achieve a connected network. Fixed-effects and random-effects network meta-analyses were fitted in Stata.

• Multi-components-based network meta-analysis. 45

  Each intervention in the review would have been considered as a combination of BCT components. We would include all social norms BCTs along with other commonly found BCTs in a components-based network plot. This type of analysis ideally requires all available trials that test the BCT components of interest; our search strategy was not appropriate for this as we were focusing on the social norms components only. We therefore decided not to pursue this approach.

The results from direct and indirect evidence were compared to check for consistency. Trials grouped by comparison were examined to assess transitivity. Metaregression did not identify any clear potential effect modifiers; therefore, although we planned in the protocol to include these in the model, this did not happen.

Additional analyses

We carried out the following sensitivity analyses for the primary outcome:

• include only studies with a low risk of bias (for the key domains of allocation concealment, sequence generation, attrition, selective outcome reporting and other sources of bias)

• exclude continuous outcomes reported as ‘mean percentage’ that were < 20% or > 80%, as these are unlikely to come from a normal distribution

• include only studies in which the standard deviation was not imputed

• using alternative values of imputed ICC

• studies with and without feedback on behaviour.

Study characteristics

A detailed summary of study characteristics is shown in Table 1. Over half of the included trials were conducted in North America (Canada: n = 15, 14.2%; USA: n = 45, 42.5%) and the most common settings were primary care (n = 57, 53.8%) and hospitals (including both inpatient and outpatient: n = 31, 29.3%). GPs were the most frequently targeted type of health worker (n = 45, 42.5%), with many studies also targeting a mixture of health workers (n = 35, 33.0%). In terms of target behaviour, 40 studies (37.7%) aimed to change prescribing behaviours (including vaccinations), with 25 (23.6%) concerned with the overall management of conditions/communications (e.g. being friendly during consultations) and 21 (19.8%) focusing on arranging, conducting or administering tests/assessments (e.g. performing HbA_1c testing). Of the 106 trials that contributed findings to the review, the majority (n = 70, 66%) were cluster RCTs, with 31 RCTs (29.2%), four stepped-wedge designs (3.8%) and one two-arm matched-cluster RCT. The majority of trials (n = 103, 97.2%) did not explicitly target participants with a low target performance. This is surprising, because the literature on social norms strongly suggests that social comparison is more likely to be successful if it is addressed to low performers: telling a high performer that they are already doing more than their peers does not motivate them to improve. 56,57 It is possible that some of the trials took place in contexts where the performance of all the health professionals was generally low at baseline, so they did not need to specifically seek out low performers to target, but no information was provided to support such an assumption. A complete list of all trials and their characteristics is included in Appendix 14, Table 26.

Intervention characteristics

Details of intervention characteristics are shown in Table 1. Of the 117 comparisons, many were delivered using a written (paper) format (n = 29, 24.8%) or utilised a mixed format (e.g. face to face and written) (n = 18, 15.4%). Participants in 45 (38.5%) interventions received the intervention more than twice, whereas 10 comparisons delivered the intervention twice (8.5%) and 35 (29.9%) delivered the intervention only once. In the majority of comparisons, the investigators were the source of the intervention (k = 83, 70.9%) and the intervention was delivered by someone external to the target health worker’s organisation (k = 81, 69.2%). In 97 (82.9%) of the comparisons, the reference group/person was the target health worker’s peer. In terms of the desired direction of change, 85 (72.6%) studies aimed to increase the behaviour. There was a lack of clarity in reporting across many intervention characteristics within the included studies. For example, in 34 (29.1%) interventions, the format was unclear or not reported and the frequency of the intervention was unclear or not reported in 27 (23.1%) comparisons.

Description of the behaviour change techniques

The frequency of specific social norms BCTs occurring within the 100 comparisons that tested social norms interventions against a control are shown in Table 2. We found tests of social comparison (n = 79), credible source (n = 7) and social reward (n = 2) against control. Some studies tested more than one social norms BCT together: social comparison and credible source (n = 6), social comparison and social reward (n = 2) and multiple social norms BCTs (more than two) together (n = 4). The social norms interventions often occurred alongside other BCTs, and 22 different techniques were identified (see Table 2).

The 117 comparisons belonged to the three comparison categories as follows (Table 3):

• Pure comparisons. There were 36 comparisons that offered a ‘pure’ test of social norms interventions. Most of these tested social comparison (33 comparisons). There were far fewer comparisons testing credible source (n = 3), social reward (n = 1), social comparison and credible source together (n = 2), or social comparison and social reward together (n = 2).

• Comparisons involving other BCTs. Social comparison was combined with social support (unspecified) (n = 7), prompts and cues (n = 5), information about health consequences (n = 4) and instruction on how to perform the behaviour and prompts/cues (combined) (n = 5). Combined interventions involving social comparison with more than two other BCTs or where the combination occurred only once in the study were combined into one group: social comparison and other BCTs (n = 25). Credible source did not occur more than once with any one particular BCT, so there is one category of credible source with other BCTs (n = 4) and another of social comparison and credible source with other BCTs (n = 4). There was one example of social reward with other BCTs (n = 1). Where more than two social norms BCTs occurred together, they were combined in a category (n = 4).

• Social norms BCTs in both arms. There were 17 comparisons in which both arms involved social norms interventions (Table 4).

Variation in trial characteristics by type of social norms intervention

Table 5 shows the key trial characteristics by the type of comparison. In total, 33 different comparisons were a pure test of ‘social comparison’ and these were quite varied in terms of type of target behaviour, type of health-care worker and type of setting; similarly, those trials that tested social comparison alongside other BCTs were also quite varied. There were 13 comparisons that tested ‘credible source’ alone or with other BCTs and four that included social reward, and again these were spread over a range of behaviours, contexts and settings. Reassuringly, there is no clear pattern to suggest that the use of BCTs was restricted to particular behaviours, contexts or settings, and this is consistent with the regression results, which suggested that the results were consistent after adjustment.

Allocation

In terms of random sequence generation, methods were deemed sufficient to produce comparable groups for the majority of studies (n = 78, 73.6%) and were, therefore, considered to be at low risk of bias. Only one study was rated to be at high risk of bias, and this was because of the original randomisation being rejected because of a perceived lack of balance between groups. All other studies (n = 27) were rated as unclear because of insufficient information to permit a judgement. The majority of studies were also rated to be at low risk of bias for allocation concealment (n = 78, 73.6%), primarily because of recruitment and consent being conducted before randomisation took place. Three studies were rated to be at high risk of bias because randomisation took place before recruitment and/or obtaining consent. The remaining studies (n = 25) were considered unclear because there was insufficient information available, or because recruitment/consent had occurred post randomisation and it was unclear whether or not participants were aware of their allocation at the time of enrolment/consent.

Blinding

Many of the studies were cluster trials, randomised at the hospital or clinic level, making the blinding of participants and personnel impractical. Owing to this clustering, most studies were rated to be at high risk of bias (n = 85). Fourteen studies were considered to be at low risk of bias owing to participants/personnel being unaware of the study aims and hypothesis, intervention content and existence of other groups/interventions, or being unaware that they were taking part in a trial. Studies rated as unclear (n = 7) lacked clarity in reporting the blinding of participants and whether or not participants were aware of the intervention or study aims and outcomes. For blinding of outcome assessment, most studies (n = 80) obtained data from electronic health records, online reports or databases or routinely collected data, in which the outcome assessors were blind to group allocations and were, therefore, rated to be at low risk of bias. Eight studies were judged to be at high risk of bias where participants selected consecutive patient records to contribute to the outcome assessment (so could have selected groups based on good practice), where participants selected the patients and collected the data or where outcome measures were recorded by participants. The remaining studies (n = 18) were judged to be unclear because of a lack of information or lack of clarity in terms of whether or not outcome assessors were blinded to allocation.

Incomplete outcome data

Over half of the included studies (n = 61) were considered to be at low risk of bias for incomplete outcome data for the following reasons: no drop-outs/low attrition, attrition evenly distributed across groups, drop-out reasons unlikely to be connected to interventions, analysis conducted on intention-to-treat basis and different patients before and after. Seventeen studies were rated to be at high risk of bias as they had large numbers of drop-outs, a lack of discussion of drop-out reasons, unequal (or unreported) attrition across groups or reasons for drop-outs being related to the intervention. The remaining studies (n = 28) were judged to be unclear as a result of insufficient or unclear reporting of attrition.

Selective reporting

Where outcomes appeared to have been reported as stated in the protocol (or where there was only one outcome and this was adequately reported), studies were judged to be at low risk of bias for selective outcome reporting (n = 42). Five studies had made changes from the planned outcomes in the protocol, or had ambiguously reported the primary outcome at registration (enabling multiple interpretations), and were, therefore, considered to be at high risk of bias. Studies that had no available protocol and had multiple outcomes, or several ways of reporting outcomes, were categorised as being unclear (n = 59).

Other potential sources of bias

No potential other sources of bias were identified for over half of the included studies (n = 58). A small number of studies (n = 7) were rated as unclear for reasons including a lack of clarity in whether or not adjustments were made for clustering in the analysis, an unclear unit of analysis for some statistical tests and only summary trial methods and data reported. The remaining studies (n = 41) were judged to be at high risk of bias for the following reasons: no adjustment for clustering in analyses; poor reporting (particularly of outcome data, therefore making interpretation difficult); potential problems with the study design (e.g. stepped-wedge, step-wise regression); concerns over analysis processes (e.g. extreme values replaced or excluded); important baseline differences between groups or large differences in participant numbers across arms; and analysis within rather than between groups.

Overall effect

There were 100 comparisons suitable for meta-analysis. Figure 3 shows the SMD summarised by type of BCT comparison in a fixed-effects meta-analysis. From this plot we can see that, as expected, there is a large amount of heterogeneity with an overall I²-value of 85.4%. Overall, combined data suggest that, on average, interventions that include social norms components were associated with a modest improvement of 0.08 SMD (95% CI 0.07 to 0.10). Forest plots showing each individual study in the review, by the social norm BCT used in the study, are in Appendix 17, Figures 22–26.

FIGURE 3.

Fixed-effects forest plot summarised by type of comparison. CS, credible source; SC, social comparison; SR, social reward. Note that SR (one comparison) and SR and other BCTs (one comparison) have been combined in this graph to improve presentation.

The I² is interpreted as a measure of the proportion of variability owing to heterogeneity between studies. This can be calculated only when two or more studies are included in the same subgroup/meta-analysis, as can the p-value alongside I² that tests the null hypothesis that I² = 0. Note that I² is related to precision and rapidly approaches 100% when the number of studies is large. 58 τ² is an alternative measure of heterogeneity, calculated only during a random-effects meta-analysis, and can be interpreted as the between-study variance.

Note that most trials in this review were randomised at a cluster level and the unit of analysis may be patient, health-care worker or a larger unit, such as clinic or hospital. With that in mind, it is impossible to report ‘N’ for each trial in any consistent way. In the meta-analysis, the weights were calculated based on the standard error of the SMD extracted from the individual trials, which has been adjusted for clustering where necessary. The number of comparisons is reported for each subgroup in brackets on all forest plots. Forest plots showing the effect in every individual study, summarised by social norms intervention, are included in Appendix 1 (see Figures 22–26).

Figure 4 shows the SMD, summarised by the type of BCT comparison in a random-effects meta-analysis. Similar conclusions can be drawn when looking at the random-effects result compared with the fixed-effects result; however, using weights from a random-effects meta-analysis suggests a larger overall SMD (0.16, 95% CI 0.11 to 0.21, I² = 85.4%, τ² = 0.043) and a wider CI, because the random-effects meta-analysis attributes more weight to smaller trials.

FIGURE 4.

Random-effects forest plot summarised by type of comparison. CS, credible source; SC, social comparison; SR, social reward.

Illustration of standardised mean differences

For this review, we converted all measures of the effectiveness of health worker behaviour into a common scale: the SMD (standardised effect size). SMDs can be difficult to interpret. To illustrate how the observed average standardised effect sizes translate into real health-care scenarios, we have converted the SMD into a risk difference (difference in percentage points) for a range of typical baseline compliance rates (Table 6). This was carried out in two steps: (1) transforming the SMD into an odds ratio using a method suggested in the Cochrane handbook36 section 12.6.3, and (2) transforming the odds ratio into a risk difference, using a method proposed by Grant. 59

Investigation of social norms behaviour change techniques

Note that owing to the high prevalence of the BCT feedback on behaviour (present in 88/100 comparisons), in the forest plots we have combined feedback on behaviour with the social norms BCT with which it appeared; that is we have listed each social norms BCT with or without feedback. Later we examine the separate effect of feedback on behaviour as part of the metaregression and as a sensitivity analysis.

We summarised the SMDs by the type of social norms comparison (see Figure 3). There is little consistency in SMDs when looking at the different types of social norms interventions being tested, with subgroup CIs that do not overlap each other and that are inconsistent with the overall effect. Interventions including credible source appear to have larger effect sizes on average than other social norms interventions, and this is true for both credible source on its own (n = 3) (SMD 0.24, 95% CI –0.01 to 0.49) and credible source combined with other BCTs (n = 4) (SMD 0.35, 95% CI 0.12 to 0.59), although the CIs are wide. Credible source combined with social comparison (n = 2) had an average effect of 0.06 SMD (95% CI 0.02 to 0.11), and credible source combined with social comparison and various other BCTs (n = 6) appeared to be the most effective of all, with a SMD of 0.40 (95% CI 0.32 to 0.47). Comparisons that were a ‘pure’ test of social comparison (with or without feedback, n = 33) appeared to have a small effect (SMD 0.05, 95% CI 0.03 to 0.08), and the size of the effect is similar when social comparison is combined with various other BCTs (n = 23) (SMD 0.04, 95% CI –0.00 to 0.08). Social comparison appeared to be very effective when combined with prompts/cues (n = 5) (SMD 0.33, 95% CI 0.22 to 0.44), but ineffective when combined with both prompts/cues and instruction on how to perform the behaviour (n = 5) (SMD 0.01, 95% CI –0.10 to 0.11). Social reward appeared to be very effective when combined with social comparison (n = 2) (SMD 0.39, 95% CI 0.15 to 0.64), but only one study looked at social reward on its own and found a negative effect. We need to interpret these observations cautiously owing to the large amount of heterogeneity and the differences in contexts and settings.

In an attempt to ease interpretation, Figure 5 shows a re-categorisation of Figure 3. In this plot, all comparisons that test each of the social norms BCTs (social comparison, credible source and social reward), whether alone or alongside other BCTs, have been combined. Trials that combine two or more social norms BCTs have been put together in one group. As before, comparisons that test credible source (n = 7), either alone or in combination with other BCTs, appear to be the most effective on average (SMD 0.30, 95% CI 0.13 to 0.47). The effect of social comparison (n = 77) appears to be very small (SMD 0.06, 95% CI 0.04 to 0.8). There is little evidence to suggest that social reward is effective (SMD 0.03, 95% CI –0.08 to 0.13), but this is based on only two trials. Trials involving a mixture of more than one social norms BCT (n = 14) have a larger than average effect (SMD 0.13, 95% CI 0.10 to 0.16).

FIGURE 5.

Fixed-effects forest plot summarised by alternative categorisation of BCTs.

Illustrative case studies

The purpose of this review is to offer a structured summary across all of the 106 studies, but we have included some illustrative case studies (Table 7) to provide a concrete example of each of the three intervention types that were found to be the most effective (credible source and social comparison, social comparison with prompts/cues, social comparison and social reward).

Variation in context and mode of delivery

We have summarised the SMD in various contexts and modes of delivery to examine where, how and with whom social norms interventions are most likely to be effective.

Type of health-care worker

Figure 6 shows the SMD summarised by the type of health-care worker in a fixed-effects meta-analysis. The effect of social norms interventions appears to be quite consistent when comparing GPs with doctors in secondary care as the type of health-care worker targeted (not shown), with an overall effect with doctors (n = 68) of 0.08 SMD (95% CI 0.07 to 0.10). We found no evidence that social norms interventions were effective with nurses or allied health professionals (AHPs) (SMD –0.01, 95% CI –0.12 to 0.11), although the number of comparisons was small (n = 5). The effect with other health workers, many of which were mixed groups such as doctors/nurses or nurses/AHPs (n = 27), was 0.08 SMD (95% CI 0.07 to 0.10).

FIGURE 6.

Fixed-effects forest plot summarised by type of health-care worker.

Target behaviour

Figure 7 shows the SMD summarised by the type of target behaviour in a fixed-effects meta-analysis. Interventions targeting prescribing behaviour (n = 40) appeared to be the most effective, on average (SMD 0.11, 95% CI 0.09 to 0.13). The effect of social norms interventions appeared to be reasonably consistent across other types of target behaviour, including test ordering (n = 21) and management of/communication about conditions (n = 23). Social norms interventions appear to be less effective with hand-washing (n = 3, SMD 0.04, 95% CI –0.05 to 0.13) and referrals (n = 3, SMD –0.08, 95% CI –0.23 to 0.07), but the number of studies is small.

FIGURE 7.

Fixed-effects forest plot summarised by target behaviour.

We have not presented a forest plot summarising the SMD by whether or not the participants were targeted based on low baseline performance, because there were only two trials that did this.

Health-care setting

Figure 8 shows the SMD summarised by health-care setting in a fixed-effects meta-analysis. The effect of social norms interventions appeared to be slightly lower in primary care settings (SMD 0.07, 95% CI 0.05 to 0.09) (n = 56) than in hospital settings (SMD 0.12, 95% CI 0.07 to 0.18) (n = 27), but both are consistent with the overall effect. Trials taking place in community settings (n = 4) and care/nursing home settings (n = 4) appear to be less effective on average; however, both CIs do overlap with the overall effect. Trials conducted in mixed settings (n = 9) appear to be more effective on average (SMD 0.35, 95% CI 0.27 to 0.42).

FIGURE 8.

Fixed-effects forest plot summarised by health-care setting.

Reference group

The reference group is the person or persons that the target is compared with or receives approval from. Figure 9 shows the SMD summarised by the type of reference group within the trials. The effect of social norms interventions appeared to be reasonably consistent across different types of reference group, with most CIs overlapping, and there was general consistency of each group with the overall effect. Most trials (n = 84, 82.9%) had peers as the reference group (SMD 0.08, 95% CI 0.06 to 0.10). Only one trial had patients as the reference group; the effect was consistent with other studies (SMD 0.10, 95% CI –0.17 to 0.37) but the CI was wide because of the low weight in the review.

FIGURE 9.

Fixed-effects forest plot summarised by reference group.

Benchmarks

When social comparison interventions are delivered, they sometimes include a benchmark: this may be a peer-related benchmark, such as the top 10% or 20% of performers among their peer group, or it may be an external benchmark, such as a performance target set by a royal college. If no benchmark is set, the social comparison usually reports the average performance among peers. The downside of the average approach is that the above-average performers will receive feedback suggesting that they are already performing better than their peers, which may lead them to reduce their effort. 56 Figure 10 shows the SMD summarised by the type of benchmark that was used. Only trials involving social comparison have been included, because benchmarking is not relevant to the other social norms interventions. The effect of social norms interventions appeared to be reasonably consistent, regardless of whether a peer benchmark (13 studies: SMD 0.06, 95% CI 0.02 to.011) or the average performance (67 studies: SMD 0.11, 95% CI 0.09 to 0.13) was included: the CIs overlap and there is general consistency of each group with the overall effect.

FIGURE 10.

Fixed-effects forest plot summarised by benchmark.

Source of the intervention

Figure 11 shows the SMD summarised by the source of the intervention (i.e. the person delivering the intervention) in a fixed-effects meta-analysis. The effect of social norms interventions appeared to be consistent across the different sources, with the exception of supervisor/senior colleague (n = 2), which appeared to be, on average, less effective (SMD –0.28, 95% CI –0.56 to 0.01). In most trials, the source of the intervention was the investigator (n = 72) (SMD 0.08, 95% CI 0.06 to 0.10) or a respected source (n = 11) (SMD 0.09, 95% CI 0.03 to 0.16). The credible sources that were found in the literature included:

• nurses in management positions who encouraged change in various behaviours to improve hospital stroke care63

• a ‘highly respected senior clinician’ who persuaded doctors of the harms and limited diagnostic benefit of X-ray for lower back pain64

• maternal–fetal medicine specialists, perinatologists or obstetricians who were influential with colleagues, who championed the use of corticosteroids to colleagues in antenatal care65

• nurse facilitators with master’s degrees and specialist training who promoted changes in preventative care in general practices66

• opinion leaders nominated by a peer for their expertise in obstetric care67 or breast cancer surgery68

• a clinical co-ordinator regarded as a ‘credible role model’ in managing patients with congestive heart failure69

• a letter to poorly performing GPs from the Chief Medical Officer about their rates of antibiotic prescribing. 60

The other categories occurred infrequently and we should interpret these results with caution owing to some wide CIs.

FIGURE 11.

Fixed-effects forest plot summarised by source of intervention.

Direction of change targeted

Figure 12 shows the SMD summarised by the intended direction of change in the behaviour in a fixed-effects meta-analysis. The social norms intervention appeared to be, on average, slightly less effective when the intervention was aimed at increasing a behaviour (n = 70) (e.g. more hand-washing) (SMD 0.06, 95% CI 0.04 to 0.09) than when it was aimed at decreasing a behaviour (n = 28) (e.g. prescription of antibiotics) (SMD 0.10, 95% CI 0.08 to 0.12), but both CIs are consistent with the overall effect.

FIGURE 12.

Fixed-effects forest plot summarised by the direction of change targeted.

Frequency of the intervention

Figure 13 shows the SMD summarised by the frequency/intensity of the interventions in a fixed-effects meta-analysis. The effect of the social norms interventions appeared to be, on average, most effective when the intervention was delivered only once (n = 28) (SMD 0.25, 95% CI 0.21 to 0.30). It appeared to be less effective, on average, when delivered more frequently (n = 47) (SMD 0.06, 95% CI 0.04 to 0.08).

FIGURE 13.

Fixed-effects forest plot summarised by frequency of the intervention.

Format of the intervention

Figure 14 shows the SMD summarised by the format of the intervention. Trials delivered via computerised methods whereby the intervention was posted on a website or other computerised format that was not integrated into the health-care worker’s workflow (n = 8) appeared to be more effective than average (SMD 0.23, 95% CI 0.15 to 0.31). By contrast, interventions delivered face to face (n = 14) appeared to be ineffective, on average, with a SMD of –0.01 (95% CI –0.06 to 0.03). Trials with an e-mailed (n = 9), written (n = 25) or a mixed format (n = 14) appeared to be reasonably consistent with each other and with the overall effect.

FIGURE 14.

Fixed-effects forest plot summarised by format of intervention.

Person delivering the intervention

Figure 15 shows the SMD summarised by whether the person who delivered the intervention was internal or external to the target’s organisation in a fixed-effects meta-analysis. Most (n = 68) interventions were delivered by an external person, often the investigator. The effect of social norms interventions, on average, seemed to be consistent across internal sources (n = 17) (SMD 0.11, 95% CI 0.05 to 0.17) and external sources (n = 68) (SMD 0.08, 95% CI 0.06 to 0.10). However, this should be interpreted cautiously given the wide CIs for internal sources.

FIGURE 15.

Fixed-effects forest plot summarised by person delivering the intervention.

Studies with social norms behaviour change techniques in both arms

There were 17 comparisons in which both arms involved social norms interventions. These are summarised in Appendix 16, Figure 21. Most of these involved social comparison in both arms, with feedback on behaviour20,70–77 or without feedback18 (n = 10), and typically they were studies that tested some other combination of BCTs in the intervention arm, but had A&F offered as part of usual care in both arms. One study tested a combination of BCTs in the intervention arm, but had credible source in both arms as part of usual care. 78 None of these studies offered any interesting insights for the review because they were designed to test the effect of other interventions.

Only two studies had a different social norms BCT in each arm, which offered a head-to-head comparison. One study tested the effect of credible source, social comparison and feedback on behaviour and other BCTs against social comparison and feedback, with an estimated SMD of 1.30 (95% CI 0.50 to 2.10),79 suggesting that credible source with other BCTs has a high effect compared with social comparison and feedback. Another study found no evidence of a difference between social comparison, credible source and feedback, and credible source alone, with an estimated SMD of 0.29 (95% CI –0.39 to 0.98). 80

There were four studies that examined the effect of different variants of social comparison (n = 3) or credible source (n = 1). Wright et al. 81 offered formal educational sessions led by a highly regarded surgeon (credible source) to both arms of the trial, and offered one-to-one ‘academic detailing’ by the highly regarded surgeon to a local opinion leader. The addition of academic detailing was effective compared with the educational sessions alone (SMD 0.32, 95% CI 0.08 to 0.56). Kiefe et al. 82 found that providing social comparison where the average performance for the top 10% of the physicians was reported was more effective than social comparison reporting the mean performance of other physicians (SMD 0.25, 95%CI 0.13 to 1.37). Schneider et al. 83 similarly tested benchmarked social comparison (best 10% of GPs) with the median performance, and found no evidence of an effect (SMD 0.03, 95% CI –0.05 to 0.56). There was no evidence of a difference in outcome when a similar social comparison intervention was provided to each arm, where the intervention arm was given the information in a work book and the control arm received a graphical computer slide show (SMD –0.04, 95% CI –0.52 to 0.44). 84

Investigation of behaviour change techniques, settings and contexts, using metaregression

Investigation of behaviour change techniques, settings and contexts, using network meta-analysis

Network meta-analysis is a suitable method to rank the social norm BCTs from most effective to least effective. For the network meta-analysis, we used all of the 100 comparisons that tested the effect of social norm BCTs with two additional comparisons79,80 that had different social norm BCTs in each arm. Some regrouping was carried out to the social norm categories to reduce the number of categories and avoid small groups: all comparisons testing social reward with or without other BCTs were combined, all comparisons testing credible source with or without other BCTs were combined and all comparisons testing social comparison and credible source with or without other BCTs were combined. The number of comparisons available for each of the social norm BCTs is shown in Table 11, and the diagram of how they are networked in shown in Figure 16.

FIGURE 16.

Network diagram to showing the available comparisons. 0, any control; 1, social comparison; 2 credible source; 3, social reward; 4, social comparison and credible source; 5, social comparison and social reward; 6, social comparison and social support (unspecified); 7, social comparison and prompts/cues; 8, social comparison and information about health consequences; 9, social comparison and instruction on how to perform the behaviour and prompts/cues; 10, social comparison and other BCTs; 11, other multiple social norm BCTs.

Network meta-analysis gives similar effect sizes and CIs to those seen in the meta-analysis (see Figure 3), but also allows us to rank the social norms interventions from best to worst (Table 12). The evidence from 102 tests of social norm BCTs suggests that the most effective interventions contain social comparison and social reward (SMD 0.39, 95% CI 0.15 to 0.64), social comparison and prompts/cues (SMD 0.33, 95% CI 0.22 to 0.44) or credible source (SMD 0.30, 95% CI 0.13 to 0.47). Social comparison on its own (SMD 0.05, 95% CI 0.03 to 0.08) or combined with social support (unspecified) (SMD 0.10, 95% CI 0.04 to 0.16) or other BCTs (SMD 0.04, 95% CI 0.00 to 0.08) all appear to be more effective than control, on average, in improving compliance with the desired behaviour; however, they were associated with a very modest effect size. The use of credible source and social comparison together (SMD 0.16, 95% CI 0.12 to 0.20), and other combinations of two or more social norm BCTs together (SMD 0.07 95% CI 0.03 to 0.12), similarly have a modest effect on behavioural outcomes. There is no evidence to suggest that social reward (SMD 0.03, 95% CI –0.08 to 0.13), social comparison, instruction on how to perform the behaviour and prompts/cues (SMD 0.01, 95% CI –0.10 to 0.11), or social comparison and information about health consequences (SMD –0.14, 95% CI –0.33 to 0.05) offer benefit above control; however, in all cases the CIs were wide and we cannot rule out a modest effect.

Social norm behaviour change techniques

There were three types of social norms intervention that were the most effective, on average, in changing health worker behaviour: social comparison together with social reward (SMD 0.39, 95% CI 0.15 to 0.64), social comparison together with prompts/cues (SMD 0.33, 95% CI 0.22 to 0.44), and credible source, either alone or in combination with other BCTs (SMD 0.30, 95%CI 0.13 to 0.47).

Other interventions had a more modest effect, but still appeared to be better than control: social comparison when delivered on its own (SMD 0.05, 95% CI 0.03 to 0.08), with social support (unspecified) (SMD 0.10, 95% CI 0.04 to 0.16) or with other BCTs (SMD 0.04, 95% CI 0.00 to 0.08). Combining credible source with social comparison (SMD 0.16, 95% CI 0.12 to 0.20), and other combinations of two or more social norm BCTs (SMD 0.07 95% CI 0.03 to 0.12) also had a modest average effect.

Interventions involving social reward or information about others’ approval appeared to be ineffective, but this is based on a very small numbers of trials. There was only one trial of social incentive, which showed no evidence of an effect.

These findings were consistent across forest plots, regression and network meta-analysis.

Concomitant behaviour change techniques

We examined whether or not the addition of other BCTs with the social norm BCTs was associated with any improvement in effect. The most frequently occurring BCT was feedback on behaviour (found in 88/100 comparisons); more often than not, health workers were provided with feedback on their own behaviour at the same time as they received interventions telling them about the behaviour of others or the approval of others. It was not possible in this study for us to completely untangle the effect of social comparison and feedback, but the sensitivity analysis suggests that the effect is similar whether or not feedback on behaviour is delivered alongside social comparison.

Interventions that tested social comparison with the addition of social reward (n = 2) or prompts/cues (n = 5) were more effective than social comparison alone, but when social comparison was delivered alongside prompts/cues plus information about health consequences (n = 4) or instructions on how to perform the behaviour and prompts/cues (n = 5) the intervention was ineffective. It may be that prompts and cues enhance social comparison when the behaviour is well understood by the recipient, but the prompts were ineffective when there was a need for instructions or education. Social comparison with social support (unspecified) (n = 7) was effective in changing health worker behaviour and had a similar size of effect to social comparison alone. However, these additional BCTs occurred in a relatively small number of trials, too small to show any additive effect in the metaregression.

Contexts

We explored the contexts in which social norms interventions were delivered. We found that social norms interventions were effective with different types of health workers. However, they may be less effective with nurses and AHPs than with doctors, although the number of studies with nurses and AHPs was low. Social norms interventions have been successful in targeting a wide range of clinical behaviours, including prescribing, tests, and management and communication around health conditions, but may be ineffective with hand-washing and referrals. They were equally effective in primary and secondary care, but may be less effective in community and care home settings, although the number of studies in those settings was low. The effect of social norms interventions appears to be reasonably consistent across different types of reference group (peers, senior persons, patients and mixed populations). Social norms interventions appear to be, on average, slightly more effective at reducing behaviours (e.g. reducing antibiotic prescriptions) than increasing them (e.g. increasing hand-washing).

Modes of delivery

Interventions appear equally effective regardless of who delivers them; there is some indication that interventions delivered by supervisors were less effective, but the number of such studies was small. The effect is similar regardless of whether the intervention comes from an internal or external source, although the number of interventions from internal sources was low. Delivering the intervention once is sufficient: there was no evidence of an increased effect from more frequent delivery. All methods of delivery of social norms interventions were effective apart from face to face; delivery by website appeared to be the most effective method.

In summary, the review provides robust evidence that social norms interventions were effective in changing the clinical behaviour of health workers: credible source, social comparison and social reward were all found to be more effective than control. Credible source was found to be very effective. Social comparison is very effective when combined with social reward or prompts/cues, but there is still an effect when social comparison is delivered on its own or with social support (unspecified), credible source or other BCTs. Social norms interventions have been implemented in a variety of contexts and we did not find strong evidence that the effectiveness varies by context. In terms of modes of delivery, delivering the intervention once was sufficient and sending the intervention by website or other computerised format was the most effective.

Search strategy

1. (anaesthetist* or anesthetist* or audiologist* or cardiologist* or chiropodist* or clinician* or consultant* or cadet* or counsellor* or dentist* or dermatologist* or dietician* or Doctor* or GP or GPs or gynaecologist* or gynecologist* or matron* or midwife or midwives or neurologist* or nurse* or nutritionist* or obstetrician* or oncologist* or optometrist* or orthodontist* or orthoptist* or orthotist* or osteopath* or paediatrician* or pediatrician* or paramedic* or pathologist* or pharmacist* or phlebotomist* or physician* or physiologist* or physiotherapist* or podiatrist* or practice manager* or practice staff or practitioner* or prosthetist* or psychiatrist* or psychologist* or psychotherapist* or radiographer* or radiologist* or registrar* or rheumatologist* or surgeon* or therapist* or urologist* or anesthesiologist* or prescriber* or sonographer*).ab,ti. (1,380,041)

2. exp Health Personnel/ (462,248)

3. exp CONSULTANTS/ (6476)

4. ((ambulance or associate or audiology or cardiology or chiropody or clinical or dental or dermatology or family or gynaecology or gynecology or health or healthcare or “health care” or hospital or house or medical or midwifery or neurology or nursing or nutrition or obstetrics or oncology or optometry or orthodontic or paediatric* or pediatric* or pathology or pharmacy or physiology or physiotherapy or podiatry or psychiatry or psychology or “public health” or radiolog* or rheumatology or surgical or therapy or trainee or urology or respiratory or magnetic resonance imaging) adj2 (assistant* or cadet* or director* or manager* or officer* or personnel or practice or practitioner* or professional* or provider or receptionist* or resident* or scientist* or secretar* or specialist* or staff or technician* or technologist or visitor* or worker*)).ab,ti. (499,319)

5. 1 or 2 or 3 or 4 (1,895,700)

6. benchmark*.ab,ti. (30,808)

7. ((audit or monitoring or peer or performance or data or individualised or individualized or web or personalised or personalized or compar* or team or practitioner or practice or clinical or social) adj2 feedback).ab,ti. (6421)

8. FEEDBACK, PSYCHOLOGICAL/ (3122)

9. ((social or descriptive or peer or subjective) adj2 (norm or norms)).ab,ti. (5654)

10. social influence.ab,ti. (1497)

11. Social Norms/ (747)

12. BENCHMARKING/ (12,165)

13. ((social or peer*) adj2 comparison*).ab,ti. (1649)

14. social competition.ab,ti. (109)

15. social proof.ab,ti. (17)

16. image motivation.ab,ti. (8)

17. warm glow.ab,ti. (27)

18. ((social or verbal or non-verbal or nonverbal or non verbal) adj2 (incentive or incentives or reward or rewards)).ab,ti. (637)

19. positive reinforcement.ab,ti. (1439)

20. “congratul*”.ab,ti. (505)

21. praise.ab,ti. (1720)

22. commendation.ab,ti. (68)

23. Reinforcement, Social/ (1036)

24. credible source.ab,ti. (100)

25. Peer Influence/ (273)

26. theory of planned behavio?r.ab,ti. (2518)

27. theory of reasoned action.ab,ti. (453)

28. theoretical domains framework.ab,ti. (249)

29. social cognitive theory.ab,ti. (1344)

30. “theory of normative social behavio?r*”.ab,ti. (21)

31. 6 or 7 or 8 or 9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or 21 or 22 or 23 or 24 or 25 or 26 or 27 or 28 or 29 or 30 (64,961)

32. randomized controlled trial.pt. (463,949)

33. controlled clinical trial.pt. (92,495)

34. “randomiz*”.ab,ti. (447,770)

35. “randomis*”.ab,ti. (89,649)

36. placebo.ab. (187,269)

37. Clinical Trials as Topic/ (184,102)

38. randomly.ab,ti. (288,777)

39. trial.ti. (180,410)

40. RCT.ab,ti. (15,807)

41. 32 or 33 or 34 or 35 or 36 or 37 or 38 or 39 or 40 (1,188,749)

42. exp Animals/ (21,630,632)

43. Humans/ (17,157,286)

44. 42 not (42 and 43) (4,473,346)

45. 41 not 44 (1,094,500)

46. 5 and 31 and 45 (1486)

Search strategy

1. (anaesthetist* or anesthetist* or audiologist* or cardiologist* or chiropodist* or clinician* or consultant* or cadet* or counsellor* or dentist* or dermatologist* or dietician* or Doctor* or GP or GPs or gynaecologist* or gynecologist* or matron* or midwife or midwives or neurologist* or nurse* or nutritionist* or obstetrician* or oncologist* or optometrist* or orthodontist* or orthoptist* or orthotist* or osteopath* or paediatrician* or pediatrician* or paramedic* or pathologist* or pharmacist* or phlebotomist* or physician* or physiologist* or physiotherapist* or podiatrist* or practice manager* or practice staff or practitioner* or prosthetist* or psychiatrist* or psychologist* or psychotherapist* or radiographer* or radiologist* or registrar* or rheumatologist* or surgeon* or therapist* or urologist* or anesthesiologist* or prescriber* or sonographer*).ab,ti. (1,960,089)

2. exp health care personnel/ (1,342,608)

3. ((ambulance or associate or audiology or cardiology or chiropody or clinical or dental or dermatology or family or gynaecology or gynecology or health or healthcare or “health care” or hospital or house or medical or midwifery or neurology or nursing or nutrition or obstetrics or oncology or optometry or orthodontic or paediatric* or pediatric* or pathology or pharmacy or physiology or physiotherapy or podiatry or psychiatry or psychology or “public health” or radiolog* or rheumatology or surgical or therapy or trainee or urology or respiratory or magnetic resonance imaging) adj2 (assistant* or cadet* or director* or manager* or officer* or personnel or practice or practitioner* or professional* or provider or receptionist* or resident* or scientist* or secretar* or specialist* or staff or technician* or technologist or visitor* or worker*)).ab,ti. (679,603)

4. 1 or 2 or 3 (2,881,143)

5. “benchmark*”.ab,ti. (36,669)

6. ((audit or monitoring or peer or performance or data or individualised or individualized or web or personalised or personalized or compar* or team or practitioner or practice or clinical or social) adj2 feedback).ab,ti. (8968)

7. psychological feedback/ (148)

8. ((social or descriptive or peer or subjective) adj2 (norm or norms)).ab,ti. (6586)

9. social influence.ab,ti. (1678)

10. social norm/ (1918)

11. benchmarking/ (2281)

12. ((social or peer*) adj2 comparison*).ab,ti. (1938)

13. social competition.ab,ti. (114)

14. social proof.ab,ti. (18)

15. image motivation.ab,ti. (8)

16. warm glow.ab,ti. (36)

17. 17 ((social or verbal or non-verbal or nonverbal or non verbal) adj2 (incentive or incentives or reward or rewards)).ab,ti. (786)

18. positive reinforcement.ab,ti. (1856)

19. “congratul* ”.ab,ti. (592)

20. praise.ab,ti. (1997)

21. commendation.ab,ti. (82)

22. credible source.ab,ti. (110)

23. peer pressure/ (1879)

24. theory of planned behavio?r.ab,ti. (2951)

25. theory of reasoned action.ab,ti. (485)

26. theoretical domains framework.ab,ti. (343)

27. social cognitive theory.ab,ti. (1564)

28. “theory of normative social behavio?r* ”.ab,ti. (19)

29. 5 or 6 or 7 or 8 or 9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or 21 or 22 or 23 or 24 or 25 or 26 or 27 or 28 (67,494)

30. randomized controlled trial/ (511,149)

31. “randomized controlled trial (topic)”/ (149,003)

32. controlled clinical trial/ (460,071)

33. “randomiz*”.ab,ti. (651,632)

34. “randomis*”.ab,ti. (134,394)

35. placebo.ab. (267,600)

36. “clinical trial (topic) ”/ (94,526)

37. randomly.ab,ti. (385,072)

38. trial.ti. (253,539)

39. RCT.ab,ti. (29,215)

40. 30 or 31 or 32 or 33 or 34 or 35 or 36 or 37 or 38 or 39 (1,652,219)

41. exp animal/ (24,606,612)

42. human/ (19,708,224)

43. 41 not (41 and 42) (4,898,388)

44. 40 not 43 (1,521,835)

45. 4 and 29 and 44 (2134)

Search strategy

1. (anaesthetist* OR anesthetist* OR audiologist* OR cardiologist* OR chiropodist* OR clinician* OR consultant* OR cadet* OR counsellor* OR dentist* OR dermatologist* OR dietician* OR Doctor* OR GP OR GPs OR gynaecologist* OR gynecologist* OR matron* OR midwife OR midwives OR neurologist* OR nurse* OR nutritionist* OR obstetrician* OR oncologist* OR optometrist* OR orthodontist* OR orthoptist* OR orthotist* OR osteopath* OR paediatrician* OR pediatrician* OR paramedic* OR pathologist* OR pharmacist* OR phlebotomist* OR physician* OR physiologist* OR physiotherapist* OR podiatrist* OR practice manager* OR practice staff OR practitioner* OR prosthetist* OR psychiatrist* OR psychologist* OR psychotherapist* OR radiographer* OR radiologist* OR registrar* OR rheumatologist* OR surgeon* OR therapist* OR urologist* OR anesthesiologist* OR prescriber* OR sonographer*).ti,ab (685,003)

2. exp “HEALTH PERSONNEL”/ (460,134)

3. exp CONSULTANTS/ (6768)

4. exp COUNSELORS/ (2836)

5. ((ambulance OR associate OR audiology OR cardiology OR chiropody OR clinical OR dental OR dermatology OR family OR gynaecology OR gynecology OR health OR healthcare OR “health care” OR hospital OR house OR medical OR midwifery OR neurology OR nursing OR nutrition OR obstetrics OR oncology OR optometry OR orthodontic OR paediatric* OR pediatric* OR pathology OR pharmacy OR physiology OR physiotherapy OR podiatry OR psychiatry OR psychology OR “public health” OR radiolog* OR rheumatology OR surgical OR therapy OR trainee OR urology OR respiratory OR magnetic resonance imaging) ADJ2 (assistant* OR cadet* OR director* OR manager* OR officer* OR personnel OR practice OR practitioner* OR professional* OR provider OR receptionist* OR resident* OR scientist* OR secretar* OR specialist* OR staff OR technician* OR technologist OR visitor* OR worker*)).ti,ab (306,659)

6. (1 OR 2 OR 3 OR 4 OR 5) (1,102,321)

7. (benchmark*).ti,ab (7103)

8. ((audit OR monitoring OR peer OR performance OR data OR individualised OR individualized OR web OR personalised OR personalized OR compar* OR team OR practitioner OR practice OR clinical OR social) ADJ2 feedback).ti,ab (4258)

9. ((social OR descriptive OR peer OR subjective) ADJ2 (norm OR norms)).ti,ab (3664)

10. (“social influence”).ti,ab (647)

11. “SOCIAL NORMS”/ (777)

12. BENCHMARKING/ (6186)

13. ((social OR peer*) ADJ2 comparison*).ti,ab (1032)

14. (“social competition”).ti,ab (6)

15. (“social proof”).ti,ab (14)

16. (“image motivation”).ti,ab (2)

17. (“warm glow”).ti,ab (8)

18. ((social OR verbal OR non-verbal OR nonverbal OR non verbal) ADJ2 (incentive OR incentives OR reward OR rewards)).ti,ab (214)

19. (“positive reinforcement”).ti,ab (364)

20. (congratul*).ti,ab (754)

21. (praise).ti,ab (1230)

22. (commendation).ti,ab (78)

23. exp “REINFORCEMENT (PSYCHOLOGY)”/ (7087)

24. (“credible source”).ti,ab (35)

25. “PEER PRESSURE”/ (774)

26. (“theory of planned behavio?r”).ti,ab (746)

27. (“theory of reasoned action”).ti,ab (377)

28. (“theoretical domains framework”).ti,ab (186)

29. (“social cognitive theory”).ti,ab  (1035)

30. (“theory of normative social behavio?r*”).ti,ab (0)

31. (“randomized controlled trial”).pt (83,342)

32. (“clinical trial”).pt (86,031)

33. (randomiz*).ti,ab (145,263)

34. (randomis*).ti,ab (36,299)

35. (placebo).ab (40,900)

36. exp “CLINICAL TRIALS”/ (241,787)

37. (randomly).ti,ab (64,596)

38. (trial).ti (80,310)

39. (RCT).ti,ab (13,964)

40. (33 OR 34 OR 35 OR 36 OR 37 OR 38 OR 39 OR 40 OR 41) (373,368)

41. exp ANIMALS/ (79,723)

42. HUMAN/ (1,706,502)

43. 43 NOT (43 AND 44) (72,224)

44. 42 NOT 45 (369,810)

45. (7 OR 8 OR 10 OR 11 OR 12 OR 13 OR 14 OR 15 OR 16 OR 17 OR 18 OR 19 OR 20 OR 21 OR 22 OR 23 OR 24 OR 25 OR 26 OR 27 OR 28 OR 29 OR 30 OR 31) (32,595)

46. (6 AND 46 AND 47) (943)

Search strategy

1. (anaesthetist* OR anesthetist* OR audiologist* OR cardiologist* OR chiropodist* OR clinician* OR consultant* OR cadet* OR counsellor* OR dentist* OR dermatologist* OR dietician* OR Doctor* OR GP OR GPs OR gynaecologist* OR gynecologist* OR matron* OR midwife OR midwives OR neurologist* OR nurse* OR nutritionist* OR obstetrician* OR oncologist* OR optometrist* OR orthodontist* OR orthoptist* OR orthotist* OR osteopath* OR paediatrician* OR pediatrician* OR paramedic* OR pathologist* OR pharmacist* OR phlebotomist* OR physician* OR physiologist* OR physiotherapist* OR podiatrist* OR practice manager* OR practice staff OR practitioner* OR prosthetist* OR psychiatrist* OR psychologist* OR psychotherapist* OR radiographer* OR radiologist* OR registrar* OR rheumatologist* OR surgeon* OR therapist* OR urologist* OR anesthesiologist* OR prescriber* OR sonographer*).ti,ab (249,905)

2. “MEDICAL PERSONNEL”/ OR DENTISTS/ OR “NURSE PRACTITIONERS”/ OR NURSES/ OR “NURSING ASSISTANTS”/ OR PARAMEDICS/ OR PHARMACISTS/ OR “PHYSICIAN ASSISTANTS”/ OR PHYSICIANS/ OR “VISITING NURSES”/ OR “SCHOOL NURSES”/ OR “FAMILY PHYSICIANS”/ OR HOSPITALISTS/ OR SURGEONS/ OR “SURGEONS GENERAL”/ (90,414)

3. ((ambulance OR associate OR audiology OR cardiology OR chiropody OR clinical OR dental OR dermatology OR family OR gynaecology OR gynecology OR health OR healthcare OR “health care” OR hospital OR house OR medical OR midwifery OR neurology OR nursing OR nutrition OR obstetrics OR oncology OR optometry OR orthodontic OR paediatric* OR pediatric* OR pathology OR pharmacy OR physiology OR physiotherapy OR podiatry OR psychiatry OR psychology OR “public health” OR radiolog* OR rheumatology OR surgical OR therapy OR trainee OR urology OR respiratory OR ‘magnetic resonance imaging’) ADJ2 (assistant* OR cadet* OR director* OR manager* OR officer* OR personnel OR practice OR practitioner* OR professional* OR provider OR receptionist* OR resident* OR scientist* OR secretar* OR specialist* OR staff OR technician* OR technologist OR visitor* OR worker*)).ti,ab (104,652)

4. (1 OR 2 OR 3) (315,078)

5. (benchmark*).ti,ab (1473)

6. ((audit OR monitoring OR peer OR performance OR data OR individualised OR individualized OR web OR personalised OR personalized OR compar* OR team OR practitioner OR practice OR clinical OR social) ADJ2 feedback).ti,ab (788)

7. ((social OR descriptive OR peer OR subjective) ADJ2 (norm OR norms)).ti,ab (554)

8. (“social influence”).ti,ab (80)

9. BENCHMARKS/ (124)

10. ((social OR peer*) ADJ2 comparison*).ti,ab (120)

11. (“social competition”).ti,ab (0)

12. (“social proof”).ti,ab (5)

13. (“image motivation”).ti,ab (0)

14. (“warm glow”).ti,ab (1)

15. ((social OR verbal OR non-verbal OR nonverbal OR ‘non verbal’) ADJ2 (incentive OR incentives OR reward OR rewards)).ti,ab (24)

16. (“positive reinforcement”).ti,ab (50)

17. (congratul*).ti,ab (146)

18. (praise).ti,ab (367)

19. (commendation).ti,ab (16)

20. (” credible source”).ti,ab (14)

21. “PEER RELATIONSHIPS”/ (111)

22. (“theory of planned behavio?r”).ti,ab (194)

23. (“theory of reasoned action”).ti,ab (63)

24. (“theoretical domains framework”).ti,ab (26)

25. (“social cognitive theory”).ti,ab (238)

26. (“theory of normative social behavio?r*”).ti,ab (0)

27. (“controlled clinical trial”).pt (103)

28. (“clinical trial”).pt (1993)

29. (randomiz*).ti,ab (10,533)

30. (randomis*).ti,ab (13,753)

31. (placebo).ab (4521)

32. “CLINICAL TRIALS”/ (11,795)

33. (randomly).ti,ab (7058)

34. (trial).ti (10,477)

35. (RCT).ti,ab (854)

36. (30 OR 31 OR 32 OR 33 OR 34 OR 35 OR 36 OR 37 OR 38) (36,001)

37. (5 OR 6 OR 8 OR 9 OR 11 OR 12 OR 13 OR 14 OR 15 OR 16 OR 17 OR 18 OR 19 OR 20 OR 21 OR 22 OR 23 OR 24 OR 25 OR 26 OR 27 OR 28) (4166)

38. (4 AND 39 AND 40) (0) (Will not run on the hdas interface. Error occurs.)

39. (39 AND 40) (286) (As a compromise have run the ‘BCT’ terms and the RCT filter instead)

Search strategy

1. #1. (anaesthetist* or anesthetist* or audiologist* or cardiologist* or chiropodist* or clinician* or consultant* or cadet* or counsellor* or dentist* or dermatologist* or dietician* or Doctor* or GP or GPs or gynaecologist* or gynecologist* or matron* or midwife or midwives or neurologist* or nurse* or nutritionist* or obstetrician* or oncologist* or optometrist* or orthodontist* or orthoptist* or orthotist* or osteopath* or paediatrician* or pediatrician* or paramedic* or pathologist* or pharmacist* or phlebotomist* or physician* or physiologist* or physiotherapist* or podiatrist* or practice manager* or practice staff or practitioner* or prosthetist* or psychiatrist* or psychologist* or psychotherapist* or radiographer* or radiologist* or registrar* or rheumatologist* or surgeon* or therapist* or urologist* or anesthesiologist* or prescriber* or sonographer*) (151,905)

2. #2. MeSH descriptor: [Health Personnel] explode all trees (8572)

3. #3. MeSH descriptor: [Consultants] explode all trees (38)

4. #4. ((ambulance or associate or audiology or cardiology or chiropody or clinical or dental or dermatology or family or gynaecology or gynecology or health or healthcare or “health care” or hospital or house or medical or midwifery or neurology or nursing or nutrition or obstetrics or oncology or optometry or orthodontic or paediatric* or pediatric* or pathology or pharmacy or physiology or physiotherapy or podiatry or psychiatry or psychology or “public health” or radiolog* or rheumatology or surgical or therapy or trainee or urology or respiratory or “magnetic resonance imaging”) adj2 (assistant* or cadet* or director* or manager* or officer* or personnel or practice or practitioner* or professional* or provider or receptionist* or resident* or scientist* or secretar* or specialist* or staff or technician* or technologist or visitor* or worker*)) (1579)

5. #5. #1 or #2 or #3 or #4 (154,716)

6. #6. benchmark* (1170)

7. #7. ((audit or monitoring or peer or performance or data or individualised or individualized or web or personalised or personalized or compar* or team or practitioner or practice or clinical or social) adj2 feedback) (466)

8. #8. MeSH descriptor: [Feedback, Psychological] this term only (471)

9. #9. ((social or descriptive or peer or subjective) adj2 (norm or norms)) (89)

10. #10. “social influence” (219)

11. #11. MeSH descriptor: [Social Norms] this term only (53)

12. #12. MeSH descriptor: [Benchmarking] this term only (122)

13. #13. ((social or peer*) adj2 comparison*) (866)

14. #14. “social competition” (3)

15. #15. “social proof” (3)

16. #16. “image motivation” (1)

17. #17. “warm glow” (1)

18. #18. ((social or verbal or non-verbal or nonverbal or “non verbal”) adj2 (incentive or incentives or reward or rewards)) (167)

19. #19. “positive reinforcement” (214)

20. #20. congratul* (39)

21. #21. praise (189)

22. #22. commendation (2)

23. #23. MeSH descriptor: [Reinforcement, Social] this term only (71)

24. #24. “credible source” (10)

25. #25. MeSH descriptor: [Peer Influence] this term only (18)

26. #26. “theory of planned behavio?r” (196)

27. #27. “theory of reasoned action” (70)

28. #28. “theoretical domains framework” (33)

29. #29. “social cognitive theory” (673)

30. #30. “theory of normative social behavio?r*” (0)

31. #31. #6 or #7 or #8 or #9 or #10 or #11 or #12 or #13 or #14 or #15 or #16 or #17 or #18 or #19 or #20 or #21 or #22 or #23 or #24 or #25 or #26 or #27 or #28 or #29 or #30 (4154)

32. #32. #5 and #31 (1999)

Search strategy

1. (anaesthetist* or anesthetist* or audiologist* or cardiologist* or chiropodist* or clinician* or consultant* or cadet* or counsellor* or dentist* or dermatologist* or dietician* or Doctor* or GP or GPs or gynaecologist* or gynecologist* or matron* or midwife or midwives or neurologist* or nurse* or nutritionist* or obstetrician* or oncologist* or optometrist* or orthodontist* or orthoptist* or orthotist* or osteopath* or paediatrician* or pediatrician* or paramedic* or pathologist* or pharmacist* or phlebotomist* or physician* or physiologist* or physiotherapist* or podiatrist* or practice manager* or practice staff or practitioner* or prosthetist* or psychiatrist* or psychologist* or psychotherapist* or radiographer* or radiologist* or registrar* or rheumatologist* or surgeon* or therapist* or urologist* or anesthesiologist* or prescriber* or sonographer*).ab,ti. (475,811)

2. exp Health Personnel/ (125,953)

3. exp CLINICIANS/ (9108)

4. exp psychologists/ (30,765)

5. exp therapists/ (30,979)

6. exp counselors/ (12,843)

7. ((ambulance or associate or audiology or cardiology or chiropody or clinical or dental or dermatology or family or gynaecology or gynecology or health or healthcare or “health care” or hospital or house or medical or midwifery or neurology or nursing or nutrition or obstetrics or oncology or optometry or orthodontic or paediatric* or pediatric* or pathology or pharmacy or physiology or physiotherapy or podiatry or psychiatry or psychology or “public health” or radiolog* or rheumatology or surgical or therapy or trainee or urology or respiratory or magnetic resonance imaging) adj2 (assistant* or cadet* or director* or manager* or officer* or personnel or practice or practitioner* or professional* or provider or receptionist* or resident* or scientist* or secretar* or specialist* or staff or technician* or technologist or visitor* or worker*)).ab,ti. (158,838)

8. 1 or 2 or 3 or 4 or 5 or 6 or 7 (599,269)

9. “benchmark*”.ab,ti. (7399)

10. ((audit or monitoring or peer or performance or data or individualised or individualized or web or personalised or personalized or compar* or team or practitioner or practice or clinical or social) adj2 feedback).ab,ti. (6239)

11. ((social or descriptive or peer or subjective) adj2 (norm or norms)).ab,ti. (10,209)

12. social influence.ab,ti. (4465)

13. Social Norms/ (7479)

14. ((social or peer*) adj2 comparison*).ab,ti. (4744)

15. social competition.ab,ti. (198)

16. social proof.ab,ti. (57)

17. image motivation.ab,ti. (10)

18. warm glow.ab,ti. (74)

19. ((social or verbal or non-verbal or nonverbal or non verbal) adj2 (incentive or incentives or reward or rewards)).ab,ti. (1033)

20. positive reinforcement.ab,ti. (2681)

21. “congratul* ”.ab,ti. (1056)

22. praise.ab,ti. (3944)

23. commendation.ab,ti. (134)

24. Social Reinforcement/ (1531)

25. credible source.ab,ti. (96)

26. Peer Pressure/ (713)

27. theory of planned behavio?r.ab,ti. (4094)

28. theory of reasoned action.ab,ti. (1226)

29. theoretical domains framework.ab,ti. (48)

30. social cognitive theory.ab,ti. (2727)

31. “theory of normative social behavio?r* ”.ab,ti. (31)

32. 9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or 21 or 22 or 23 or 24 or 25 or 26 or 27 or 28 or 29 or 30 or 31 (53,015)

33. clinical trial.md. (21,174)

34. “randomiz* ”.ab,ti. (65,985)

35. “randomis* ”.ab,ti. (8914)

36. placebo.ab. (36,895)

37. Clinical Trials/ (10,978)

38. randomly.ab,ti. (66,420)

39. trial.ti. (26,281)

40. RCT.ab,ti. (3557)

41. 33 or 34 or 35 or 36 or 37 or 38 or 39 or 40 (167,864)

42. 8 and 32 and 41 (470)

Search strategy