Audit and feedback interventions involving pharmacists to influence prescribing behaviour in general practice: a systematic review and meta-analysis

Abstract Introduction Pharmacists, as experts in medicines, are increasingly employed in general practices and undertake a range of responsibilities. Audit and feedback (A&F) interventions are effective in achieving behaviour change, including prescribing. The extent of pharmacist involvement in A&F interventions to influence prescribing is unknown. This review aimed to assess the effectiveness of A&F interventions involving pharmacists on prescribing in general practice compared with no A&F/usual care and to describe features of A&F interventions and pharmacist characteristics. Methods Electronic databases (MEDLINE, EMBASE, CINAHL, Cochrane Central Register of Controlled Trials, (Social) Science Citation Indexes, ISI Web of Science) were searched (2012, 2019, 2020). Cochrane systematic review methods were applied to trial identification, selection, and risk of bias. Results were summarized descriptively and heterogeneity was assessed. A random-effects meta-analysis was conducted where studies were sufficiently homogenous in design and outcome. Results Eleven cluster-randomized studies from 9 countries were included. Risk of bias across most domains was low. Interventions focussed on older patients, specific clinical area(s), or specific medications. Meta-analysis of 6 studies showed improved prescribing outcomes (pooled risk ratio: 0.78, 95% confidence interval: 0.64–0.94). Interventions including both verbal and written feedback or computerized decision support for prescribers were more effective. Pharmacists who received study-specific training, provided ongoing support to prescribers or reviewed prescribing for individual patients, contributed to more effective interventions. Conclusions A&F interventions involving pharmacists can lead to small improvements in evidence-based prescribing in general practice settings. Future implementation of A&F within general practice should compare different ways of involving pharmacists to determine how to optimize effectiveness. PRISMA-compliant abstract included in Supplementary Material 1.


Introduction
][6][7] Despite extensive guidance to promote evidence-based prescribing, i.e. to optimize the safe, effective, and efficient use of medicines, some unwarranted variation persists. 8,9Some variation may be expected, since evidence-based guidelines do not apply in all scenarios, but previous studies have found that some differences are clinically unjustified and associated with disparities in patient outcomes, 10 medicines waste, 11 and rising costs. 127][18][19][20][21] An examination of pharmacists' involvement in the delivery of a proven method for behaviour change (A&F) may contribute to identifying a role in which pharmacists can fully use and develop their expertise.
A&F interventions seek to influence clinical practice through monitoring and reinforcement of positive behaviours. 22Specifically, data about individual or group practice are collected and compared with a standard, e.g.evidencebased guidelines, professional standards, or peer performance.This information is fed back to the individual/group to encourage change in practice or closer compliance with the standard. 23A 2012 Cochrane review 24 demonstrated A&F interventions to be effective in achieving health professional behaviour change when feedback is provided by a supervisor or colleague; more than once; both verbally and in writing; and includes clear targets and an action plan.Additional characteristics associated with effective A&F include the credibility of the data used in A&F interventions, opportunity for recipients to discuss feedback, and choice of comparator. 25,26his systematic review builds on and forms a discrete part of an ongoing update of the earlier Cochrane review. 24It focussed on the effectiveness of A&F interventions involving pharmacists as key contributors on prescribing in general practice.
The specific objectives of the pharmacist-related review were to:

Methods
The review protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO), registration number CRD42020194355.This report is guided by the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) checklist 27 (Supplementary Material 2).

Scope of the review
Randomized studies, including cluster and step wedge trials, in general practice (or facilities in which general practitioners [GPs] provided medical services) and which met the following eligibility criteria were included:

Information sources
The A&F Systematic Review (A&F SR) Group (see Acknowledgements for membership) conducted searches (without language restrictions): Cochrane Library, clinical trials.gov,MEDLINE (Ovid), EMBASE (Ovid), CINAHL (Ebsco) (from June 2010 to June 2020), and WHO International Clinical Trials Registry (June 2010 to February 2019) to identify studies of A&F interventions (pharmacist and non-pharmacist) 28 for inclusion in the Cochrane update.Studies from before 2010 were identified from the original Cochrane A&F systematic review. 24Details of searches are included in Supplementary Material 3.
Duplicate, independent screening was undertaken (MC, MCW) in May 2020 of all titles and abstracts identified for inclusion in the Cochrane review update by the A&F SR Group, to identify trials that evaluated A&F interventions focussed on prescribing in general practice settings.Reference lists of trials identified for the pharmacist sub-review were searched for additional studies.MC undertook screening of additional trials identified by the 2020 search for inclusion in the Cochrane update in February 2022.

Data extraction and management
Duplicate data extraction was undertaken for all studies included in the Cochrane update 28 by members of the A&F SR Group, using the Cochrane Effective Practice and Organization of Care (EPOC) extraction form.Independent, duplicate extraction was undertaken (MC, NA) of additional data items for the pharmacist sub-review, including the number of pharmacists and their role(s) in the intervention, details of the prescribing topic addressed in intervention, pharmacists' years of experience, and their work situation in relation to participating GPs.Authors of studies for which results data were missing were contacted by email.Data items extracted for the sub-review were added to details concerning study and intervention characteristics extracted for the Cochrane update.

Risk of bias in individual studies
Duplicate, independent evaluation of the risk of bias was undertaken by members of the A&F SR Group and/or MC and NA, using EPOC-recommended risk of bias methods (adapted from the general Cochrane tool 29 ).
Discrepancies between reviewers relating to screening, data extraction, and risk of bias assessment were resolved by exchange of emails and online discussions where further explanations were necessary.

Summary measures
Where possible, risk ratios (RRs) of appropriate prescribing were calculated using a 95% confidence interval (CI).For other continuous outcomes and where data were available, standardized mean differences and standard deviation were calculated.

Key messages
• Audit and feedback (A&F) is effective in changing prescribing behaviour.
• Pharmacist-led A&F influences prescribing in primary care settings.
• Pharmacists in general practice may be ideally situated for delivering A&F.

Data synthesis and meta-analysis
All studies were included in the descriptive analysis.Details about the A&F interventions, including the characteristics of the pharmacist(s) involved, were summarized descriptively and frequencies produced.Only studies deemed sufficiently homogenous in design and outcome were included in a meta-analysis. 30Included outcomes concerned potentially inappropriate or risky prescribing, or prescribing that did not comply with specified guidelines.Cochrane Review Manager (RevMan) v5.4 software was used to produce a random-effects model.Effect sizes were calculated using the Mantel-Haenszel RR and 95% CIs.Heterogeneity was assessed using the I 2 statistic.A funnel plot for assessment of bias across studies was not considered appropriate, due to the low number of studies included in the meta-analysis. 30

Results
Of the 332 studies identified for inclusion in the Cochrane update, 28 11 were included in this pharmacist-focussed review (Fig. 1).The studies were conducted in 9 countries: 2 each from the Netherlands 31,32 and Italy 33 and one each from the United Kingdom, 34 Denmark, 35 Norway, 36 Republic of Ireland, 37 Australia, 38 United States, 39 and Malaysia. 40The article from Italy reported 2 studies, 33 and these were treated as 2 separate studies for the purpose of this review.
The percent agreement between raters (screening, data extraction, and risk of bias assessment) was 84%.

Characteristics of included studies
Nine studies included 2 arms (intervention, control) (Table 1).Two 3-armed studies 35,40 were included with full intervention,   36 ), with 279 clinicians (range 41 35 to 1,737 33 ) and 1,884 patients (range 196 37 to 63,337 34 ).In 3 studies, control group participants received no active intervention 32,35,36 ; in 1 study, control group participants had access to the same prescribing and benchmarking data as intervention group participants but did not implement a team-based care system to optimize this knowledge. 39In all other studies, control group participants received a non-A&F intervention such as access to information technology resources or guidelines, or prescription review only.
GPs were the recipients of the A&F intervention in all studies.The interventions took place in general practices or primary care clinics in all studies apart from one which focussed on GPs' care for patients in residential care facilities. 38ll A&F interventions included outcomes associated with prescribing (Table 2).The median number of prescribing outcomes was 2 (range 1 32,35,36,39,40 to 19 31 ).Eight studies included outcomes which aimed to reduce prescribing errors or inappropriate prescribing.In the 3 other studies, the outcome was an increase in a desired prescription of selected medications for osteoporosis and prostatic hyperplasia, 33 thiazide for hypertension, 36 and lipid-lowering medication. 39The implementation of a guideline for the use of antihypertensive and cholesterol-lowering drugs was used as a specific target for participants in 1 study. 36Clinical and prescribing guidelines were explicitly mentioned in descriptions of interventions, e.g. as the basis for discussions and education sessions, in 6 studies. 31,32,34,35,38,40These included guidelines used internationally, e.g.World Health Organization 41 and British National Formulary 42 and national guidelines, e.g.Dutch College of GPs (NHG) 43 and Norwegian General Practice. 44Two studies (reported together) 33 explicitly stated that clinical guidelines were not selected as a comparator because they were viewed with suspicion by participating clinicians.
In 4 studies 31,33,35 prescribing data were sourced from regional or local databases and in 3 studies the research team extracted computerized data from the practice clinical system. 32,34,368][39][40] An association between the source of the data and the effect of the A&F intervention was not observed.

Risk of bias
Three studies were assigned low risk of bias for all 10 domains evaluated 31,34,37 and a further 5 scored low risk for 7 of the domains 33,35,36,39 (Fig. 2).Blinding of participants and personnel were assigned high risk in 2 studies, 36,40 while in 2 other studies, 32,40 both random sequence generation and allocation concealment were assessed as unclear.Both selective outcome reporting and incorrect analysis were assessed as unclear in 6 studies each ( 32,33,35,38,39 and 32,33,35,38,40 , respectively).

Effectiveness of pharmacist A&F intervention
Six studies (N = 71,092) were included in a meta-analysis (Fig. 3).The purpose of 4 of these studies was to reduce inappropriate prescribing 34,37,38,40 and to increase guidelinecompliant prescribing in the 2 remaining studies. 36,39he pooled RR across these 6 studies was 0.78 (95% CI: 0.64-0.94),demonstrating that the risk of inappropriate/ non-compliant prescribing was 22% lower following an A&F intervention than after usual care or control conditions.High levels of heterogeneity were detected (I 2 = 98%).A funnel plot was not constructed to assess bias due to the small number of studies included in the meta-analysis. 45he 5 studies not represented in the meta-analysis had a range of different outcome measures including: the number of antibiotic prescriptions for urinary tract infection 31 and respiratory tract infection 32 ; and a Medication Appropriate Index 35 score.Two of the studies excluded from the metaanalysis showed improved prescribing in the intervention group 32,35 but this was not demonstrated in a third study. 31No numerical results were available for the remaining 2 studies (reported in 1 paper).

Determinants of A&F effectiveness
The following results are organized under 3 headings which reflect groups of factors which have been identified as determinants of A&F effectiveness [24][25][26] : (i) A&F intervention process, (ii) content of feedback reports, (iii) characteristics of the individual (pharmacist) delivering the A&F intervention (Table 3).

(i) A&F intervention process
The A&F intervention was incorporated into educational sessions led by pharmacists in 5 studies [34][35][36][37][38] ; in 4 of these studies appropriate prescribing in the intervention group improved more than in the control group.This included the 2 studies 35,37 in which the sessions were described as "interactive." In 4 further studies, A&F was incorporated into meetings (lasting up to 3 h) of pre-existing collaborative groups of GPs 31,33 or GPs and pharmacists. 32Meetings included pharmacist-facilitated discussions and/or problem-based learning in interprofessional groups.These studies had mixed results.
The 2 remaining studies included skills training for participants 32,39 and showed more favourable results for prescribing in the intervention groups.
Five studies involved computerized decision support for prescribing, 32,34,36,37,39 all of which showed increased appropriate prescribing in the intervention group compared with control.
Pharmacists provided ongoing prescribing support (12 weeks to 2 years) for individual patients in 3 studies 34,35,39 all of which reported increased appropriate prescribing in the intervention group relative to control.
In 1 study 40 the pharmacist visited participating clinics to collect and screen handwritten prescriptions from participants on a monthly basis.They provided feedback to participants by post for 3 months; results showed increased appropriate prescribing in the intervention group.
Several studies (n = 7) included only 1 episode of feedback [33][34][35][36]38,39 ; in the 2 studies 32,40 which included 3 episodes of feedback, the A&F intervention had a small effect. The nmber of episodes of feedback was given was unclear in the 2 remaining studies. 31,37(ii) Content of feedback reports In 2 studies 36,40 general information about the prescribing topic was included in feedback reports; both studies showed improvements in prescribing.Four studies comprised feedback reports that combined general information about the clinical topic of interest as well as specific plans developed for or with individual participants.31,35,37,39 Three of these studies included action plans for individual participants, 35,37,39 and all achieved positive effects on prescribing in the intervention group.
In 1 study, prescribers in the intervention group received individual plans based upon discussion with research pharmacists, and their prescribing improved compared with control group prescribers who received general information only. 335][36][37][38][39] Whilst evaluations of feedback of individual clinician-level data showed variable effect, most studies of individual patient-level data had positive effects.[34][35][36][37]39 Feedback was provided in both verbal and written formats in 7 studies, 31,[33][34][35]39,40 4 of which achieved more favourable results in the intervention group compared with the control.34,35,39,40 Of the 4 studies which evaluated only verbal feedback, 32,[36][37][38] 3 reported more favourable results in the intervention groups. (iii)

Characteristics of the pharmacist delivering the A&F intervention
The pharmacist was a colleague of participating GPs in 2 studies 39,40 and external to the practices in the 9 remaining studies.In 4 studies, the pharmacist was known to participants from regular interprofessional meetings. 31Whether the pharmacist was internal or external to the general practice did not make a substantial difference to the effectiveness of the intervention in most studies.Two studies which reported improvements in prescribing due to A&F, the pharmacist was a colleague to the prescribers, 34,39 but in another study which demonstrated a positive effect from A&F, the pharmacist was neither a colleague nor interprofessional collaborator. 36Family Practice, 2022, Vol.40, No. 5-6

626
Systematic review of pharmacist-led A&F to prescribing in general practice Four studies reported the contributing pharmacists' years of experience 32,34,36,37 ; the median was 16 (range 0 34 to 30 32 ) years since registration.Pharmacists undertook study-specific training in 7 studies, e.g.communication skills, evidencebased medicine methodology; increased appropriate prescribing in the intervention group was observed in 4 of these studies. 32,36,37,40In 5 studies, the pharmacist reviewed prescriptions and records for individual patients 34,35,37,39,40 and presented feedback to individual participants; all 5 studies showed improved prescribing in their intervention groups.

Discussion
The results of this review indicate that A&F interventions in general practice involving pharmacists tend to be effective at improving prescribing compared with no intervention or non-A&F interventions, such as education only or distribution of guidelines alone.The effect size of these pharmacistrelated A&F interventions were moderate and were similar in magnitude to those reported in earlier reviews of A&F interventions delivered by different healthcare professionals. 24,46urthermore, the findings indicate the effectiveness of the pharmacist-related A&F is associated with specific pharmacist characteristics, e.g.receipt of focussed training and intervention components, e.g.delivery of feedback concerning prescribing for individual patients.

Comparison with existing literature
This review adds to existing evidence of the effectiveness of pharmacist involvement in interventions to improve prescribing in a range of healthcare settings. 17,47Recent systematic reviews 48,49 reported that academic detailing delivered by pharmacists, both singly and as part of a multifaceted intervention, was effective in reducing adverse drug events and medication errors, respectively.In academic detailing, the educator is typically a health professional based outside the participant's practice 50 ; the professional may be a pharmacist. 51This current review included studies of multifaceted interventions, which included pharmacist-led education in addition to pharmacist conduct of prescribing audits and delivery of feedback.Pharmacists in this review included colleagues, interprofessional collaborators and external experts, but the existence of a pre-existing relationship with target prescribers was not associated with the effectiveness of A&F interventions.The results suggest that interventions where pharmacists provide ongoing feedback on individual prescribing decisions may be more effective than those in which their involvement is either fleeting or based on sessions in preexisting collaborations of prescribers.
The results of the current review differed from previous findings 24 which have found that feedback of general information plus tailored action plans are more effective than feedback of general information only.Reports containing individual patient-level data appeared to have greater impact on prescribing than those containing team-or clinician-level data, but given the small number of studies in this review it is not possible to detect statistically significant differences.
Previous reviews have identified other influential features relating to the process of feedback, including the provision of feedback to groups and individuals, 52,53 repeated provision of feedback, 24,52 the use of a range of media used to convey feedback, 53 and the role of clinical decision support systems. 54This current review concurs with previous findings about the effectiveness of providing both verbal and written feedback, [34][35][36][37] but was inconclusive about the impact of providing multiple episodes of feedback. 32,40[37]39,40 The inclusion of computerized decision support at the point of prescribing also contributed to the effectiveness of interventions. 32,34,36,37,396,37 In the light of the small number of studies in this review, and the level of heterogeneity amongst them, comparisons must be treated with caution.

Implications for policy and research
This review demonstrated that A&F interventions involving pharmacists have a moderate positive effect on prescribing in general practice settings.Successful A&F interventions involved pharmacists in providing ongoing support to physicians about their prescribing for individual patients as well as scenarios in which pharmacists partnered physicians in local prescribing groups.It was not possible to identify the optimal working relationship between the pharmacist leading the A&F intervention and participants (i.e.colleague or external contact) from this review.Successful interventions may seek to increase a positive prescribing behaviour or reduce inappropriate prescribing; the direction of change, i.e. increased or decreased prescribing behaviour, does not appear to be a determining factor in an intervention's success.
Although this review suggests that A&F interventions involving pharmacists who have undertaken study-specific training may have a more positive effect on prescribing, information relating to the content of the training and about the pharmacist's general level of experience and expertise was limited.These are topics which warrant further enquiry.

Strengths and limitations
This is the first review to focus specifically on A&F interventions involving pharmacists as key contributors to improve prescribing in general practice settings.A pre-defined study protocol is publicly available.All included studies were cluster-randomized trials which focussed on enhanced roles for pharmacists in general practice settings.The risk of bias in most domains was generally assessed as low.
Although this review adopted a robust search strategy recommended by the Cochrane Information Retrieval Methods group and followed the Cochrane EPOC methodology for duplicate data extraction and risk of bias assessments, screening for pharmacist-led A&F studies was limited to titles and abstracts (from the main Cochrane review) to identify eligible studies.As such, it may not have captured all relevant studies where pharmacists were not mentioned in either the title or abstract.An additional study was identified from examination of the full text of a study already identified for inclusion in the review. 31tudies included in this review reported pharmacist interventions in relatively affluent healthcare settings.Opportunities for pharmacists to influence prescribing in settings with fewer resources may be limited.
Owing to the lack of existing directly comparing A&F against A&F with pharmacist involvement, it was not possible within this review to estimate the relative effects of specifically pharmacist-led feedback.It would be difficult to produce a straightforward hierarchy of the "best" healthcare professionals to deliver A&F, as this would entail examination of the moderating effects of a range of factors, such as training, feedback type, professional role, and team relationships.
Meta-analysis was performed where appropriate, but the level of heterogeneity amongst included studies was high.Owing to the low number of studies included in the metaanalysis, it was not possible to assess publication bias.

Conclusions
By undertaking a range of responsibilities to promote evidence-based prescribing and encourage the judicious use of medicines, pharmacists make an important contribution to improving patient outcomes in general practice.A&F may be particularly well-matched with pharmacists' professional skills and expertise.
Further exploration is needed to optimize their involvement in the provision of A&F interventions.The extent to which pharmacists currently deliver A&F interventions in general practice is unknown but is being explored in the United Kingdom as part of this research programme.The content and focus of training in undergraduate curricula and during foundation years should also be investigated to determine whether pharmacists are equipped to deliver interventions of this type as part of their general practice responsibilities.

1 .
Compare the effectiveness of A&F interventions involving pharmacists on prescribing in general practice with usual care or non-A&F interventions.2. Identify and describe the: • features of A&F interventions involving pharmacists • characteristics of the pharmacists contributing to A&F interventions

Fig. 1 .
Fig. 1.PRISMA flow diagram of A&F intervention studies identified and screened for inclusion in the final review.
in A&F: AUD conducted audit; FRP provided feedback report; GPD facilitated group feedback discussion with GPs; IPT provided feedback on patient-by-patient basis; ED provided education.NR, not reported.

Table 1 .
Characteristics of studies, including description of participants and intervention.

Table 1 .
Continued620Systematic review of pharmacist-led A&F to influence prescribing in general practice

Table 1 .
32ntinued622Systematic review of pharmacist-led A&F to influence prescribing in general practice partial intervention, and control arms.The median number of participating practices/clinics was 47 (range 832to 146

Table 2 .
Effects of A&F interventions on prescribing.

Table 3 .
Details of A&F intervention and pharmacist characteristics.