Tailored interventions to address determinants of practice
Abstract
Background
Tailored intervention strategies are frequently recommended among approaches to the implementation of improvement in health professional performance. Attempts to change the behaviour of health professionals may be impeded by a variety of different barriers, obstacles, or factors (which we collectively refer to as determinants of practice). Change may be more likely if implementation strategies are specifically chosen to address these determinants.
Objectives
To determine whether tailored intervention strategies are effective in improving professional practice and healthcare outcomes. We compared interventions tailored to address the identified determinants of practice with either no intervention or interventions not tailored to the determinants.
Search methods
We conducted searches of The Cochrane Library, MEDLINE, EMBASE, PubMed, CINAHL, and the British Nursing Index to May 2014. We conducted a final search in December 2014 (in MEDLINE only) for more recently published trials. We conducted searches of the metaRegister of Controlled Trials (mRCT) in March 2013. We also handsearched two journals.
Selection criteria
Cluster‐randomised controlled trials (RCTs) of interventions tailored to address prospectively identified determinants of practice, which reported objectively measured professional practice or healthcare outcomes, and where at least one group received an intervention designed to address prospectively identified determinants of practice.
Data collection and analysis
Two review authors independently assessed quality and extracted data. We undertook qualitative and quantitative analyses, the quantitative analysis including two elements: we carried out 1) meta‐regression analyses to compare interventions tailored to address identified determinants with either no interventions or an intervention(s) not tailored to the determinants, and 2) heterogeneity analyses to investigate sources of differences in the effectiveness of interventions. These included the effects of: risk of bias, use of a theory when developing the intervention, whether adjustment was made for local factors, and number of domains addressed with the determinants identified.
Main results
We added nine studies to this review to bring the total number of included studies to 32 comparing an intervention tailored to address identified determinants of practice to no intervention or an intervention(s) not tailored to the determinants. The outcome was implementation of recommended practice, e.g. clinical practice guideline recommendations. Fifteen studies provided enough data to be included in the quantitative analysis. The pooled odds ratio was 1.56 (95% confidence interval (CI) 1.27 to 1.93, P value < 0.001). The 17 studies not included in the meta‐analysis had findings showing variable effectiveness consistent with the findings of the meta‐regression.
Authors' conclusions
Despite the increase in the number of new studies identified, our overall finding is similar to that of the previous review. Tailored implementation can be effective, but the effect is variable and tends to be small to moderate. The number of studies remains small and more research is needed, including trials comparing tailored interventions to no or other interventions, but also studies to develop and investigate the components of tailoring (identification of the most important determinants, selecting interventions to address the determinants). Currently available studies have used different methods to identify determinants of practice and different approaches to selecting interventions to address the determinants. It is not yet clear how best to tailor interventions and therefore not clear what the effect of an optimally tailored intervention would be.
PICOs
Plain language summary
Tailored interventions to address identified determinants of practice
Tailored interventions to change professional practice are interventions planned following an investigation into the factors that explain current professional practice and any reasons for resisting new practice. These factors are referred to using various terms, including barriers, enablers, obstacles, and facilitators; in this review we use the term determinants of practice to include all such factors. The determinants may vary in different healthcare settings, groups of healthcare professionals, or clinical tasks. It is widely assumed that efforts to change professional practice have a lower likelihood of success unless these determinants are identified and taken into account.
In a previous review, we included 26 studies and we concluded that tailoring can change professional practice. However, more studies of tailoring have been published and therefore we have incorporated the new studies into an update of the review.
We have included 32 studies in the new review. The findings continue to indicate that tailored interventions can change professional practice, although they are not always effective and, when they are, the effect is small to moderate. There is insufficient evidence on the most effective approaches to tailoring, including how determinants should be identified, how decisions should be made on which determinants are most important to address, and how interventions should be selected to account for the important determinants. In addition, there is no evidence about the cost‐effectiveness of tailored interventions compared to other interventions to change professional practice. Therefore, future research studies should seek to develop and evaluate more systematic approaches to tailoring.
Authors' conclusions
Summary of findings
| Interventions tailored to address identified determinants of practice compared with no intervention for implementing appropriate clinical practice | ||||||
| Patient or population: healthcare professionals Settings: mostly primary care in the USA and Europe Intervention: tailored interventions to implement practice guidelines Comparison: no intervention or dissemination of guidelines alone | ||||||
| Outcomes | Absolute effect | Relative effect | No of participants | Certainty of the evidence | Comments | |
| Without tailored intervention | With tailored intervention | |||||
| Difference (95% CI) | ||||||
| Implementation of recommended practice, e.g. clinical practice guideline recommendations | Moderate adherence* 60 per 100 patients | 70 per 100 patients | OR 1.56 (95% CI 1.27 to 1.93, P value < 0.001) | 15 studies with at least 7990 health professionals (numbers unclear in 5 studies) | ⊕⊕⊕⊝ | 17 other studies could not be included in the meta‐regression. The effect of tailored interventions in these studies varied from no effect to moderate effect between studies and between outcomes within studies, a finding consistent with the meta‐regression |
| Difference: 10 more patients receiving recommended practice per 100 patient encounters (Margin of error: 6 to 14 more patients) | ||||||
| Low adherence* 20 per 100 patients | 28 per 100 patients | |||||
| Difference: 8 more patients receiving recommended practice per 100 patient encounters (Margin of error: 4 to 13 more patients) | ||||||
| Healthcare outcomes | ‐ | ‐ | ‐ | No studies | ‐ | The studies did not include sufficient evidence to enable an assessment of effect on healthcare outcomes to be made |
| Costs | ‐ | ‐ | ‐ | No studies | ‐ | The studies did not include sufficient evidence to enable an assessment of effect on healthcare outcomes to be made |
| Adverse effects | ‐ | ‐ | ‐ | No studies | ‐ | The studies did not include sufficient evidence to enable an assessment of effect on healthcare outcomes to be made |
| Margin of error = Confidence Interval (95% CI) OR: Odds Ratio *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). † The OR and confidence intervals shown are taken from a meta‐regression. The results of 14 studies not included in the meta‐regression indicated that, on average, tailored interventions improve professional practice. However, the effects were mixed. CI: confidence interval; OR: odds ratio | ||||||
| GRADE Working Group grades of evidence High = This research provides a very good indication of the likely effect. The likelihood that the effect will be substantially different† is low. Moderate = This research provides a good indication of the likely effect. The likelihood that the effect will be substantially different† is moderate. Low = This research provides some indication of the likely effect. However, the likelihood that it will be substantially different† is high. Very low = This research does not provide a reliable indication of the likely effect. The likelihood that the effect will be substantially different† is very high. †Substantially different = a large enough difference that it might affect a decision | ||||||
| 1The assumed risks without a tailored intervention were selected to help interpret the overall odds ratios in situations in which there is a high risk of undesirable professional practice without intervening (20% desired practice) and a medium risk (60% desired practice). | ||||||
Background
Description of the condition
The extent to which recommendations for clinical practice based on good quality research evidence are implemented varies. Gaps between what is recommended and what health professionals do and patients receive are common, and there can be delays before the findings of research are widely adopted (Grol 2005; Oxman 1995). Although the subject of many research studies in recent years, including trials of interventions to implement recommended practice, the problem persists and more work is needed in order to understand the reasons for gaps in clinical practice and to identify interventions to address them (Eccles 2009; Wensing 2012).
Description of the intervention
This review updates a Cochrane review of the effects of tailored interventions that was originally completed in 2005 (Shaw 2005) and subsequently updated in 2010 (Baker 2010). We define tailored interventions as 'strategies to improve professional practice that are planned, taking account of prospectively identified determinants of practice'. Determinants of practice are factors that could influence the effectiveness of an intervention to improve professional practice, and have been previously referred to using alternative terms, including barriers, obstacles, enablers, and facilitators. They have been classified by the Cochrane Effective Practice and Organisation of Care (EPOC) Group into nine categories (information management, clinical uncertainty, sense of competence, perceptions of liability, patient expectations, standards of practice, financial disincentives, administrative constraints, and other) (EPOC 2002). This categorisation has not been used extensively and more research into the determinants of practice has been completed since the classification was proposed. Following a detailed review of studies of determinants, including recent studies, a new checklist of determinants has been devised (Flottorp 2013).
How the intervention might work
Whether considered in the context of models for quality and safety improvement or guideline implementation initiatives (Ashford 1999; Grol 2005; Lomas 1994; Robertson 1996), systematic reviews of improvement interventions (Chaillet 2006; Grimshaw 2004) or guideline adoption (Cabana 1999), determinants are believed to influence the success of improvement strategies. If the determinants of practice are identified using methods that could include brainstorming, interviews or focus groups of health professionals, or questionnaires (Flottorp 2013; Krause 2014a), and strategies are then implemented that have been chosen to address the determinants using methods such as group interviews of implementation practitioners or health professionals (Huntink 2014; Wensing 2014), it would appear reasonable to expect performance to improve. Despite the attractiveness of this argument, however, the effects of attempts to translate research evidence into practice and improve performance remain inconsistent (Grimshaw 2004; Grimshaw 2012; McGlynn 2003).
Why it is important to do this review
We have not identified any reviews evaluating the effects of tailored implementation strategies on professional performance other than the earlier versions of this review, which concluded that tailored interventions were more likely to improve professional practice than no intervention or dissemination of guidelines or educational materials alone.
Although there are a number of reviews in specific clinical fields (Chaillet 2006; Kroenke 2000), which have discussed the possibility that tailored strategies might be more effective than strategies selected without taking account of determinants, these reviews did not address the effect or costs of tailored interventions specifically. Bosch and colleagues undertook a qualitative analysis of 20 quality improvement studies reporting investigation of determinants (Bosch 2007). Individual and group interviews of professionals were the most commonly used method of identifying determinants, but in many studies the reasons for believing a particular strategy would address a particular determinant were not explained. Again, the effectiveness of tailored strategies was not evaluated.
Since the publication of the last revision of this review (Baker 2010), several new studies of tailored intervention strategies have been published. Consequently, there may be additional evidence on the effectiveness of tailoring or on how it can be undertaken most effectively. Since tailoring is regarded as an important step in improvement interventions, we undertook an update of the review.
Objectives
We have addressed the same question considered in the previous versions of the review: are tailored strategies effective in improving professional practice and healthcare outcomes?
To answer this question, we compared interventions tailored to address identified determinants with either no interventions or an intervention(s) not tailored to the determinants. In addition, in this update, but not in the previous version of the review, we separately compared:
-
implementation interventions tailored to address identified determinants of practice compared to no intervention;
-
implementation interventions tailored to address identified determinants of practice compared to non‐tailored implementation interventions.
We anticipated that sufficient numbers of studies would have been published to allow these separate comparisons, and that comparison of tailoring with non‐tailored interventions would tend to indicate less effect than in comparison with no intervention.
Methods
Criteria for considering studies for this review
Types of studies
Cluster‐randomised controlled trials (cluster‐RCTs) with at least two control and two intervention sites.
Types of participants
Healthcare professionals responsible for patient care. We excluded studies that involved only students.
Types of interventions
We defined tailored strategies as strategies to improve professional practice that are planned, taking account of prospectively identified determinants of practice. Determinants may be identified by various methods, including observation, brainstorming, focus group discussions, interviews or surveys of the involved healthcare professionals, and/or through an analysis of the organisation or system in which care is provided. We excluded studies that use gap analysis only (i.e. audits identifying a gap between actual and desired performance), and studies of educational interventions based on an identified lack of knowledge and designed to improve knowledge only. The identification of determinants must have been undertaken before the design and delivery of the intervention. If the timing of the identification of determinants was not clear, we contacted the study authors for clarification.
Studies had to involve a comparison group that did not receive a tailored intervention, or a comparison between an intervention that was targeted at determinants, compared with an intervention not targeted at identified determinants.
Types of outcome measures
For inclusion, study outcomes had to be either objectively measured adherence of health professionals to recommended practice, in a healthcare setting, or patient outcome, or adverse effects (patient outcomes, quality of care, and adverse effects, as defined in the EPOC guidance on outcomes to be reported in EPOC reviews) (EPOC 2013). When costs were reported in studies that included either measures of professional practice, patient outcomes, or adverse effects, we planned to include these, but we excluded studies of costs alone. We did not include measures of knowledge or performance in a test situation as an outcome measure and we excluded studies that included only this outcome.
Search methods for identification of studies
M. Fiander and J. Camosso‐Stefinovic developed and ran the search strategies. We searched the Cochrane Database of Systematic Reviews and the Database of Abstracts of Reviews of Effects (DARE) for related systematic reviews, and the databases listed below for primary studies. The most recent search was conducted in December 2014.
We searched the following databases:
-
The Cochrane Library (2014, Issue 2) (Central Register of Controlled Trials, Cochrane Methodology Register, Health Technology Assessment Database, NHS Economic Evaluations Database);
-
MEDLINE (R) 1946 onwards, and In‐Process and Other Non‐Indexed Citations, OvidSP;
-
EMBASE, 1947 onwards, OvidSP;
-
EPOC Group Specialised Register, Reference Manager;
-
CINAHL (Cumulative Index to Nursing and Allied Health Literature), 1980 onwards, EbscoHost;
-
British Nursing Index (BNI), 1994 onwards, ProQuest;
-
Health Management Information Consortium (HMIC), 1983 to 2009, Department of Health's Library and Information Services, King's Fund Information and Library Services. We were unable to search this database in 2014.
We searched The Cochrane Library, MEDLINE, EMBASE, and PubMed to May 2014. We searched CINAHL and BNI only to March 2013 due to time constraints. The Cochrane EPOC Group Specialised Register has not been updated since 2012 and therefore has not been searched since that date. We were unable to search HMIC because we no longer had access.
We applied neither language nor date restrictions. We used two methodological search filters to limit retrieval to appropriate study designs: the Cochrane Highly Sensitive Search Strategy (sensitivity‐ and precision‐maximising version, 2008 revision) to identify randomised trials (cf. Cochrane Handbook for Systematic Reviews of Interventions 6.4d) (Lefebvre 2011), and a partial EPOC methodological search filter (cf. lines 37‐40 in the MEDLINE strategy). We repeated the MEDLINE search in December 2014, to identify any recently reported studies. Detailed search strategies used for searches from 2009 to 2014 are provided in Appendix 1 to Appendix 8. Search strategies used prior to 2009 are provided in Appendix 9.
Searching other resources
We searched the following trial registers:
-
metaRegister of Controlled Trials (mRCT) up to March 2013 (http://www.controlled‐trials.com/mrct/), which includes: ISRCTN; Action Medical Research; NIH ClinicalTrials.gov; Wellcome Trust; and UK Trials Register (Appendix 8).
We also:
-
handsearched two key journals (Implementation Science (vol. 1 2006 through 2014 vol. 9, August 2014) and Journal of Evaluation in Clinical Practice (October 2009 to end August 2014, vol. 20, Issue 4));
-
reviewed reference lists of all included studies, relevant systematic reviews, and primary studies;
-
contacted authors of relevant studies or reviews to clarify information presented in published articles where necessary or to request further details and unpublished results or data;
-
contacted researchers with expertise relevant to the review topic.
Data collection and analysis
Selection of studies
We loaded the reference details and abstracts of articles identified in the searches into the Early Review Organizing Software (IECS 2009). Two review authors independently assessed studies for inclusion. We resolved disagreements by discussion, involving a third review author if necessary. We obtained all selected articles in full text.
Data extraction and management
Two review authors independently extracted the data from included studies by using a revised version of the data extraction form used in the previous version of the review (Baker 2010). Information collected on the types of patients in studies included whether some or all were disadvantaged or low‐income.
We summarised the determinants of practice identified and if the included papers provided sufficient information, we classified determinants into the seven domains of the Tailored Implementation in Chronic Disease (TICD) checklist: guideline factors, individual health professional factors, patient factors, professional interactions, incentives and resources, capacity for organisational change, and social, political, and legal factors (Flottorp 2013). We also summarised the methods that were used to identify them and qualitatively assessed the processes used to identify and prioritise them and tailor interventions to account for them. Two review authors independently classified the intensity of the methods used to identify determinants using the following three categories: low – a questionnaire survey of health professionals or informal discussion with, for example, a guideline group; moderate – interviews and/or focus groups with samples of health professionals specifically seeking information about determinants, or a survey supplemented by performance data; high – interviews and/or focus groups of health professionals supplemented by additional methods, for example observation.
We recorded the timings of interventions (whether at the start of the programme and whether delivered once or repeated at intervals). We also recorded the rationales for the choice of interventions. This included the behaviour change mechanism if reported in studies including, for example, role modelling. We also recorded the use of theory to inform interventions, for example, the Theoretical Domains Framework (Cane 2012), or Normalisation Process Theory (May 2007).
Two independent review authors classified the extent to which the tailored intervention was adjusted to account for local factors using the following two categories: not adjusted – the intervention was designed in response to the general determinants affecting all or most professionals, and not adjusted to the particular determinants at individual or team level; some adjustment – one or more of the components of the intervention were adjusted at the level of the team or individual to account for local factors.
Assessment of risk of bias in included studies
Two review authors assessed the risk of bias of the included studies using the approach set out in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). The tool includes the following categories of bias: random sequence generation, allocation concealment, baseline outcomes, baseline characteristics, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, contamination, selective reporting, and other sources of bias. We assessed risk of bias as either high risk, low risk, or unclear risk (either lack of information or uncertainty over the potential for bias). The risk of bias of each of the included studies is presented in the Results section.
Measures of treatment effect
We assessed all the included studies for inclusion in a meta‐regression analysis, with the aim of providing an overall assessment of the effectiveness of tailored interventions in comparison with either no intervention or non‐tailored interventions.
When several outcomes were reported in a trial we only extracted results for the variable(s) explicitly described as the primary outcome(s). When the primary outcome was not specified we took the variable described in the sample size calculation as the primary outcome. When the primary outcome was still unclear or when the manuscript described several primary outcomes, we selected the median effect size value across multiple outcomes. If the median fell between two outcomes, we chose the more conservative (smaller effect size) of the two.
We extracted data for the outcome of interest at both baseline and follow‐up, to allow adjustment for baseline differences between the two treatment groups to be made in the analysis.
Unit of analysis issues
As all the trials were cluster‐randomised, studies needed to report results for each cluster or, failing that, provide an estimate of the intra‐class correlation coefficient (ICC) to enable the clustering effect to be accounted for in the overall effect size estimate from each study (Ukoumunne 2002). Five studies included in the analysis reported either an estimate of the ICC or reported data for each cluster, allowing the ICC to be estimated. Where no ICC could be derived from the study, we utilised published ICCs (Campbell 2005). Campbell et al extracted 220 ICCs from 21 data sets and reported the results separately for both primary and secondary care settings. For each of the studies included in our meta‐analysis where no ICC was available, we utilised the median ICC reported by Campbell et al for the relevant setting. We then used the design effect to adjust the estimated effect sizes for clustering, whereby the variances of the odds ratios were increased by multiplying them by the design effect (Rao 1992).
Dealing with missing data
The trials included in the analysis were randomised at the cluster level, for example, at the level of the clinic or general practice. None of the studies described problems of drop‐outs at this level during the trial period. The majority of trials included in this review collected data before and after interventions on different patients, therefore drop‐out at the patient level was not an issue for these trials. Where data were collected on the same group of patients throughout the trial (three studies), no problem with drop‐outs was reported.
Assessment of heterogeneity
We also investigated heterogeneity within the effectiveness of tailored interventions to identify factors that need consideration when designing and implementing a tailored intervention. In the previous review, we addressed the level of tailoring (whether to the determinants at the level of individual health professionals, or determinants at the level of the healthcare team, or at the level of the organisation), but this did not predict the effectiveness of tailored interventions. In the previous review, we also addressed the rigour of the determinants analyses undertaken in the included studies, anticipating that a more rigorous analysis would lead to a better tailored intervention and therefore greater effect on clinical practice. However, it was difficult to judge the rigour of the analysis of determinants since little is yet known of the most useful approaches (Huntink 2014; Wensing 2014), and this variable failed to predict effectiveness. In the previous review, we also investigated the effect of the presence of administrative constraints (EPOC 2002), since these might limit the ability of health professionals to change their behaviour, but this variable was also not predictive of effectiveness. Therefore, we omitted these variables in this update.
In this update we assessed:
-
the risk of bias in the included studies, in case studies with a higher risk of bias were more likely to show tailored interventions were effective;
-
whether tailoring was informed by theories of behaviour or behaviour change, since theories may be expected to aid the selection and design of tailored interventions;
-
whether adjustment was made for local factors. Interventions delivered to different settings, such as different clinical teams or hospitals, may be adjusted to account for factors such as existing policies, the staff available, or other local issues, and this adjustment may be anticipated to improve the effect of tailored interventions; and
-
the number of domains represented by the determinants identified (Flottorp 2013). If a wider range of determinant domains are found to influence practice, it may be more difficult to implement change in professional behaviour.
A summary of the decisions on inclusion of variables in the investigation of heterogeneity is included in Appendix 10. Although there are many factors that may potentially affect implementation interventions (including, for example, the duration of the intervention and whether it is delivered by influential agents, the complexity of the targeted behaviour, and the extent to which the determinants may be amenable to intervention), the appendix indicates those that we considered for inclusion.
We assessed heterogeneity for all meta‐regression models using the I² residual statistic (Higgins 2003), which represents the proportion of residual between‐study variation due to heterogeneity, as opposed to sampling variability. To investigate possible causes of heterogeneity in the effectiveness of tailored interventions between studies we assessed attributes that might have an impact on findings of intervention effectiveness. These were: risk of bias, use of a theory when developing the intervention, whether adjustment was made for local factors, and number of domains addressed with the determinants identified. Classifications for each study by attribute are reported in the table of Characteristics of included studies. We investigated heterogeneity by fitting the meta‐regression analysis separately for each category of the study attribute of interest and comparing odds ratios, and additionally by fitting the study attributes as continuous variables into the meta‐regression models (Habord 2008).
Assessment of reporting biases
We applied no language restrictions in the searches or inclusion of studies. We conducted a sensitive search of major biomedical databases and trial registries (see Search methods for identification of studies). As the mRCT includes randomised trial records held on the National Institutes of Health (NIH) ClinicalTrials.gov website (available at: http://clinicaltrials.gov/), we did not search the latter registry. Furthermore, as the studies included in the review spanned a number of years and were not all recent publications, time‐lag bias is unlikely to be a major problem. In our analyses, we used meta‐regression in order to be able to account for differences between control and intervention groups at study baselines. There is no equivalent of forest or funnel plots for meta‐regression analyses and reliance on only follow‐up results to produce such plots would have provided misleading information. We did, however, produce meta‐regression plots to present the fitted models, the circles representing the estimate from each study.
Data synthesis
For the 15 studies included in the meta‐regression, we combined the estimated odds ratios for each study (adjusted for clustering) using meta‐regression techniques, whereby the baseline odds ratios were included as a covariate to adjust for any baseline differences between the intervention and control groups (Sutton 2000). The codes in Stata 2013 are included in Appendix 11. In addition, separate models were fitted depending on whether the control group received no intervention (seven studies) or a non‐tailored intervention (eight studies). Of the eight studies that received a non‐tailored intervention, seven received relevant guidelines or educational material only. One study delivered a group lecture and distributed the standard protocol to the control group (Beeckman 2013). Due to a more rigorous non‐tailored intervention being delivered in this study, we also repeated the meta‐regression analysis with this study removed.
Sensitivity analysis
In the meta‐regression analyses, we carried out sensitivity analyses assuming a larger clustering effect than had been accounted for in the standard analyses, by using higher ICC estimates (i.e. the reported upper quartile range values) taken from Campbell (Campbell 2005).
Results
Description of studies
Results of the search
We screened 9403 unique citations (Figure 1) and reviewed the full text of 360. Of these, we included 32. Seven are ongoing studies (Ongoing studies) and seven are awaiting assessment (Studies awaiting classification). We excluded 106 with reasons provided in the Characteristics of excluded studies table and we excluded 208 with quick reference to the full text.
PRISMA diagram
Included studies
Twenty‐three of the studies had been included in the previous version of the review, but we excluded three studies from that review because the interventions were not assessed on this occasion as meeting the criteria for tailoring (Davies 2002; Sehgal 2002; Verhoeven 2005). All included studies were cluster‐randomised trials (Characteristics of included studies). We included 15 of 32 studies in the meta‐regression analysis (Table 1; Table 2). The remaining 17 studies were not eligible for meta‐regression because they either did not assess a suitable binary outcome, or they reported no data at baseline.
| Risk of bias | Study ID | Primary outcome(s) | Effect size | Authors' conclusions |
| Tailored intervention compared to no intervention | ||||
| Unclear | Avorn 1983 | 1. Prescribing of targeted drugs (amount and costs) No suitable dichotomous outcome reported | Costs reduced in intervention arm versus control by 14% (P value = 0.0001) | Academic‐based 'detailing' was a useful and cost‐effective way to improve the quality of drug therapy decisions and reduce unnecessary expenditures |
| Unclear | Avorn 1992 | 1. Residents not on psychoactive drugs | 1. Decrease of 27% in intervention arm and 8% in control arm (P value = 0.02) | An educational programme targeted to physicians, nurses, and aides can reduce the use of psychoactive drugs in nursing homes without adversely affecting the overall behaviour and level of functioning of the patient |
| High | Callahan 1994 | 1. Frequency of recording a depression diagnosis 2. Stopping medications associated with depression 3. Initiating antidepressant medication 4. Psychiatry referral | 1. 12% control and 32% intervention arm (P value < 0.01) 2. 22% control and 23% intervention arm 3. 8% control and 26% intervention arm (P value < 0.01) 4. 14% control and 12% intervention arm | Intensive screening and feedback of patient‐specific treatment recommendations increased the recognition and treatment of late life depression by GPs |
| Low | Fairall 2012 | 1. Time to death 2. Proportion with undetectable viral loads | 1. Time to death did not differ (hazard ratio 0.94, 95% CI 0.76 to 1.15) 2. Viral load suppression was similar in each group, 72% in the intervention and 70% in the control groups; risk difference 1.1% (95% CI ‐2.4 to 4.6) | Expansion of primary care nurses' role to include ART initiation and represcription can be done safely, but might not reduce time to ART or mortality |
| Low | Figueiras 2006 | 1.Number of reported adverse drugs reactions (ADRs) 2. Number of serious ADRs 3. Number of high causality ADRs 4. Number of unexpected ADRs 5. Number of new‐drug related ADRs Results not in a suitable format | 1. RR 10.23 (95% CI 3.81 to 27.51) 2. RR 6.32 (95% CI 2.09 to 19.16) 3. RR 8.75 (95% CI 3.05 to 25.07) 4. RR 30.21 (95% CI 4.54 to 200.84) 5. RR 8.04 (95% CI 2.10 to 30.83) | The intervention increased reporting of ADRs, with effect maximal at 4 months, but no longer from 13 months after intervention |
| Low | Flottorp 2002 | 1. Rate of antibiotic use | 1. 3% less likely to receive antibiotics after intervention in sore throat arm (P value = 0.032), no change in UTI arm | Passively delivered, complex interventions targeted at identified barriers to change had little effect in changing practice |
| Unclear | Goodwin 2001 | 1. Rate of up‐to‐date preventative services Results reported as percentages, numbers of patients not given | 1. Intervention: 31% to 42%, control: 35% to 37% (P value = 0.015) | An approach to increasing preventive service delivery that is individualised to meet particular practice needs can increase global preventive service delivery rates |
| High | Hux 1999 | 1. Median antibiotic cost Results reported as percentages, numbers of patients not given | 1. Change of 0.05% intervention versus 3.37% control, P value < 0.002 | A simple programme of confidential feedback and educational materials blunted cost increases, increased the use of first‐line antibiotics, and was highly acceptable to Ontario primary care physicians |
| Unclear | Looijmans 2007 | 1. The proportion of healthcare workers vaccinated against influenza | 1. Uptake was 9% higher than in the control group (RR 1.59, 95% CI 1.08 to 2.34) | The intervention resulted in higher, though moderate, influenza vaccine uptake among healthcare workers in nursing homes |
| High | Matchar 2002 | 1. % time in target range No suitable dichotomous outcome reported | 1. Difference (intervention minus control) adjusted for minor baseline differences was 5% (95% CI ‐5% to 14%), P value = 0.32 | A properly administered anticoagulation service can successfully manage the anticoagulation of most patients with atrial fibrillation; however, these services did not improve anticoagulation compared to usual care |
| Unclear | Murphy 2009 | The proportion of patients at 18‐month follow‐up above target levels for (1) blood pressure, (2) cholesterol, and (3) hospital admissions | 1. intervention (systolic) 27.2% versus 32.8 % in controls, OR 1.52 (0.99 to 2.30); 2. 15.2% versus 16.4%, OR 1.13 (0.63 to 2.03); 325.8% versus 34.0%, OR 1.56 (1.53 to 2.60) | There was a reduction in hospital admissions, but no other clinical benefits, possibly because of a ceiling effect |
| Unclear | Ross‐Degnan 1996 | 1. Sales of oral rehydration salts Results reported as percentages, numbers of patients not given | 1. Increased by 21% in intervention arms compared to controls (P value < 0.05) | Face‐to‐face training of pharmacy attendants, which targets deficits in knowledge and specific problem behaviours, can result in short‐term improvements in product sales and communication with customers |
| High | Santoso 1996 | 1. Prescribing of oral rehydration solution Results reported as percentages, numbers of patients not given | 1. Increase after intervention, but not after both interventions | The small group face‐to‐face intervention did not appear to offer greater impacts over large seminars in improving the appropriate use of drugs in acute diarrhoea |
| Unclear | Schouten 2007 | 1. Guideline‐adherent antibiotic prescription 2. Adjustment of antibiotic to renal function 3. Switches in therapy 4. Streamlining of therapy 5. Gram staining and culture of sputum samples (No primary outcome specified) | 1. Difference between intervention and control hospitals OR 2.63 (95% CI 1.57 to 4.42) 2. OR 12.9 (95% CI 3.64 to 45.8) 3. OR 1.20 (95% CI 0.02 to 76.51) 4. OR 1.94 (95% CI 0.34 to 11.03) 5. OR 1.13 (95% CI 0.64 to 2.00) Baseline: 1.24 (0.43 to 3.56) Follow‐up: 2.21 (0.79 to 6.17) | For some indicators, the intervention led to improvements. Secular trends may have had an effect on indicators that did not improve to a greater extent in the intervention group |
| Low | Scott 2013 | 1. Administration of alteplase in patients with stroke in emergency departments | 1. Increase from 1.25% to 2.79% in intervention hospitals, 1.25% to 2.10% in controls (RR 1.37, 95% CI 0.96 to 1.93) | The increase in use of alteplase was smaller than the effect to which the study was powered |
| Unclear | Van Gaal 2011 | 1. The incidence of adverse events per patient week (the sum of the incidents of pressure ulcers, urinary tract infections, and falls divided by the total number of weeks) | At follow‐up, the rate was 0.06 in the intervention group, 0.09 in the control group (RR 0.57, 95% CI 0.34 to 0.95) | It is possible to implement multiple guidelines simultaneously |
| Unclear | Zwarenstein 2007 | 1. Asthma symptom score | The decline in score on 1‐year follow‐up was 4.08 in the intervention group and 3.24 in the control group (adjusted for baseline, OR 1.48, 95% CI 1.00 to 2.20) | Educational outreach was effective in reducing children's asthma symptoms |
| Tailored intervention compared to non‐tailored intervention | ||||
| Unclear | Baker 2001 | 1. 3 or more symptoms recorded at diagnosis 2. Suicide risk assessed at diagnosis 3. Treated with antidepressant or cognitive therapy 4. Therapeutic dose of antidepressant 5. Reviewed after 3 weeks 6. Suicide risk reassessed 7. 2 or more follow‐up consultations 8. Treated for 4 months or more | 1. OR 1.9 (97% CI 0.9 to 3.8) 2. OR 5.6 (95% CI 2.8 to 11.3) 3. OR 2.5 (95% CI 0.7 to 9.2) 4. OR 1.3 (95% CI 0.6 to 3.2) 5. OR 1.1 (95% CI 0.5 to 2.4) 6. OR 0.7 (95% CI 0.2 to 3.0) 7. OR 2.0 (95% CI 0.9 to 4.0) 8. OR 1.2 (95% CI 0.6 to 2.4) (ORs adjusted for baseline) Baseline: 1.10 (0.74 to 1.64) Follow‐up: 1.57 (0.98 to 2.51) | The findings suggest that this approach to implementation may be effective and should be further investigated |
| Unclear | Beeckman 2013 | 1. Residents receiving fully adequate pressure ulcer prevention in bed 2. Residents receiving fully adequate prevention in a chair | 1. 4.6% in intervention group, 1.5% in control group 2. 60.0% in intervention group, 13.2% in control group, P value = 0.003 | Positive effects were observed when residents were in a chair, but not when in bed |
| Low | Cheater 2006 | 1. Nurse performance assessed by examining patients' nursing records against a list of review criteria (primary outcome) | Mean improvement in aggregate compliance scores in percentage points: 1. ‐2.3 (95% CI ‐1.63 to 1.7) for audit and feedback compared to control 2. 0.9 (‐3.3 to 5.1) for educational outreach compared to control | In comparison with educational materials alone, the implementation methods did not improve care at 6 months follow‐up |
| Unclear | Coenen 2004 | 1. Antibiotic prescribing rate by GPs for adult patients with acute cough | 1. OR 0.56 (95% CI 0.36 to 0.87) Risk of prescribing antibiotics for intervention group versus controls, adjusted for relevant clinical symptoms | Implementing a guideline for acute cough is successful in optimising antibiotic prescribing |
| Unclear | Engers 2005 | 1. Referrals to a therapist 2. Prescription of pain medication on a time‐contingent basis 3. Prescription of paracetamol versus NSAIDs No baseline data reported | Intervention compared to control: 1. OR 0.8 (95% CI 0.5 to 1.4) 2. OR 1.0 (95% CI 0.3 to 3.0) 3. OR 2.0 (95% CI 0.8 to 5.5) | The intervention modestly improved implementation of the Dutch low back pain guideline by GPs |
| High | Evans 1997 | 1. Rate of diagnosis of asthma | 1. 40/1000 versus 16/1000, P value < 0.01 | The intervention substantially increased child health staff's ability to identify children with asthma, involve them in continuing care, and provide them with state‐of‐the‐art care for asthma |
| Unclear | Foy 2004 | 1. Assessment appointment within 5 days 2. Ascertainment of cervical cytology history 3. Screening or antibiotic prophylaxis for genital tract infection 4. Misoprostol used for cervical priming and early and mid‐trimester abortion 5. Supply of contraception at discharge Results reported as percentages | Difference between intervention and control groups 1. OR 0.89 (95% CI 0.50 to 1.58) 2. OR 0.93 (95% CI 0.36 to 2.40) 3. OR 1.70 (95% CI 0.71 to 5.99) 4. OR 1.00 (95% CI 0.27 to 1.77) 5. OR 1.11 (95% CI 0.48 to 2.53) | The intervention was ineffective, possibly because of high pre‐intervention compliance and limited impact of the intervention on barriers outside the control of clinical staff |
| Low | Fretheim 2006 | 1. Proportion of patients prescribed a thiazide among patients prescribed an antihypertensive for the first time 2. Proportion of those started on antihypertensive or cholesterol‐lowering treatment having a cardiovascular risk assessment 3. Proportion satisfying treatment goals for BP or cholesterol | 1. Prescribing thiazides relative risk intervention versus control 1.94 (1.49 to 2.49) 2. Risk assessment done relative risk intervention versus control 1.04 (0.60 to 1.71) 3. Treatment goal achieved, intervention versus control relative risk 0.98 (0.93 to 1.02) | The intervention had an impact on prescribing patterns, but not on other outcomes |
| Unclear | Karuza 1995 | 1. Physician vaccination rates for influenza Results reported as percentages, numbers of patients not given | 1. The intervention arm had a higher adjusted vaccination rate (62.39%) compared to controls (46.46%), P value < 0.001 | Interventions using small groups can be useful in facilitating adoption of guidelines by physicians |
| Unclear | Langham 2002 | 1. Adequate recording of 3 risk factors n/N not reported | 1. Difference of 10.5% (95% CI ‐3.9 to 24.9) between information and no information and 6.6% (95% CI ‐8.9 to 22.0) between evidence and no evidence | Adequate risk factor recording did not differ between the information (versus not information) or the evidence (versus not evidence) intervention groups |
| Unclear | Lakshminarayan 2010 | Adherence to indicators for stoke care for (1) acute care; (2) in‐hospital care; (3) discharge care | 1. OR 1.8 (95% CI 0.44 to 7.6); 2. OR 1.05 (0.83 to 1.3); 3. OR 1.04 (0.64 to 1.7) | No intervention effect was demonstrated, although there was a secular trend |
| Unclear | Leviton 1999 | 1. Use of corticosteroids | 1. Use increased by 108% in active dissemination hospitals and by 75% in usual dissemination hospitals (P value < 0.01) | An active, focused dissemination effort increased the effectiveness of usual dissemination methods when combined with key principles to change physician practices |
| Unclear | Simon 2005 | 1. Proportion of patients with hypertension receiving a diuretic or beta‐blocker | Difference between control and group detailing OR 1.40 (95% CI 1.11 to 1.76) Difference between control and individual detailing OR 1.30 (95% CI 0.95 to 1.79) Difference between group and individual detailing OR 1.10 (95% CI 0.86 to 1.42) | Both detailing interventions resulted in an approximately 13% absolute increase in guideline‐recommended drugs |
| Unclear | Soumerai 1998 | 1. Appropriateness of the prescribing of selected drugs (aspirin in eligible elderly patients) Data reported as percentages, numbers not given | 1. Median change +0.13 in intervention and ‐0.03 in controls, P value = 0.04 | Working with opinion leaders and providing performance feedback can accelerate adoption of some beneficial acute myocardial infarction therapies |
| Unclear | Van Bruggen 2008 | 1. % of patients with poor control achieving HbA1c of < 8% | 70% in the intervention and 58% in the control groups achieved adequate control (not after controlling for baseline value, potential confounders and clustering) | The process of diabetes care did improve, but intermediate outcomes hardly changed |
ADR: adverse drugs reaction
ART: antiretroviral treatment
BP: blood pressure
CI: confidence interval
GP: general practitioner
NSAID: non‐steroidal anti‐inflammatory drug
OR: odds ratio
RR: risk ratio
UTI: urinary tract infection
| Study ID | Outcome | Baseline odds ratios (95% CI) | Follow‐up odds ratios (95% CI) |
| Avorn 1992 | Residents not on antipsychotic drugs | 0.90 (0.42 to 1.90) | 1.08 (0.49 to 2.34) |
| Baker 2001 | Antidepressants in therapeutic dose | 1.10 (0.74 to 1.64) | 1.57 (0.98 to 2.51) |
| Beeckman 2013 | Fully adequate prevention of ulcers | 1.02 (0.35 to 2.94) | 10.59 (3.56 to 31.45) |
| Callahan 1994 | Depression diagnosis | 1.23 (0.57 to 2.63) | 2.65 (1.40 to 5.03) |
| Cheater 2006 | Recording of management criteria | 1.37 (0.85 to 2.22) | 1.65 (0.99 to 2.71) |
| Coenen 2004 | Antibiotics not prescribed | 0.80 (0.49 to 1.32) | 1.07 (0.59 to 1.92) |
| Evans 1997 | Returning asthma patients from previous year | 0.94 (0.48 to 1.83) | 2.88 (1.28 to 6.46) |
| Flottorp 2002 | Antibiotics not prescribed | 1.12 (0.94 to 1.31) | 1.26 (1.06 to 1.50) |
| Fretheim 2006 | Thiazides prescribed for hypertension | 0.63 (0.42 to 0.95) | 1.68 (1.20 to 2.35) |
| Leviton 1999 | Use of antenatal corticosteroids | 1.00 (0.65 to 1.51) | 1.59 (0.88 to 2.83) |
| Looijmans 2010 | Uptake of flu vaccine | 0.98 (0.63 to 1.52) | 1.71 (1.10 to 2.65) |
| Murphy 2009 | Numbers below the target level for BP | 0.99 (0.68 to 1.41) | 1.31 (0.87 to 1.96) |
| Schouten 2007 | % where key quality indicators performed | 1.24 (0.43 to 3.56) | 2.21 (0.79 to 6.17) |
| Scott 2013 | Use of alteplase for stroke | 1.00 (0.14 to 7.05) | 1.34 (0.36 to 4.92) |
| Simon 2005 | Beta‐blockers or diuretics prescribed for hypertension | 1.03 (0.88 to 1.21) | 1.40 (1.18 to 1.65) |
Note: Odds ratio = odds of outcome in treatment group/odds of outcome in control group, calculated at baseline and follow‐up and adjusted for clustering
BP: blood pressure
CI: confidence interval
Healthcare setting and characteristics of healthcare professionals
Twelve trials were undertaken in the USA, five in the Netherlands, four in the United Kingdom, two each in Belgium, Norway, South Africa and Indonesia, and one each in Portugal, Canada, and Ireland. Seventeen studies were based in primary care settings, seven in hospital settings, three in nursing homes, and one each in child health clinics, community pharmacies, the regional health system, and a Medicaid programme. The health professionals included in the studies were: primary care practitioners (family physicians, general practitioners) in 14 studies, mixed professional groups in eight studies, nursing in four, pharmacy in two, and unclear, geriatric teams, gynaecology teams, and physicians in one study each. The studies did not give particular attention to disadvantaged groups, although two studies were undertaken in a low/middle‐income country (Indonesia).
Targeted behaviours
Twelve studies targeted use of drugs including, for example, the prescribing of antibiotics in the community, medication advised for acute diarrhoea, and drugs used to treat hypertension. Eleven studies targeted the management of disease, including diagnosis, assessment, and treatment. Six studies targeted preventive care, including secondary prevention in coronary heart disease. Two studies targeted influenza vaccination and one study the reporting of adverse drug reactions.
Prospective identification of determinants of practice
We categorised the investigation of determinants as low intensity in 10, moderate in 18, and high in four. The studies using high intensity methods employed a mix of methods. For example, Scott 2013 used both focus groups and interviews, Flottorp 2002 used a literature search, discussion with the guideline development group, brainstorming, focus group interviews with patients and health professionals, discussion groups, and informal interviews, and Murphy 2009 used focus groups with practitioners and patients as well as piloting. Another used an in‐depth practice assessment (Goodwin 2001). In 13 studies, more than one method was used to identify determinants. Interviews with health professionals and occasionally patients were used in 11 studies, focus group interviews in 10 studies, questionnaire surveys in six, review of the literature in four, review of performance data in two, a meeting or workshop in two, and other methods in four (including observation and consultation with an expert group).
Determinants of practice
Four studies did not include information on the determinants identified (Avorn 1992; Hux 1999; Karuza 1995). Individual health professional factors (knowledge, motivation, perceptions of likely benefits, and risks) were the more commonly reported determinants, being noted in 25 studies. Patient factors (patient expectations and preferences) were reported in eight studies, incentives and resources in eight (including lack of staff and time, and financial disincentives to adopt new practices), guideline factors were noted in four studies (lack of clarity or lack of recommendations), organisational capacity (recording information, tools, workload, systems) in nine studies, professional interactions in three studies, and social, political, and legal factors in two studies.
Influence of prospective identification of determinants on intervention design
In 12 studies the rationale used to associate determinants with interventions thought likely to address them was either not clear or not stated (Coenen 2004; Engers 2005; Fairall 2012; Hux 1999; Langham 2002; Leviton 1999; Matchar 2002; Santoso 1996; Soumerai 1998; van Bruggen 2008; van Gaal 2010; Zwarenstein 2007). Behaviour change theories were explicitly referred to in seven studies (Baker 2001; Evans 1997; Foy 2004; Karuza 1995; Lakshminarayan 2010; Murphy 2009; Scott 2013). In five others the principles of academic detailing or persuasive strategies, or a framework of professional attitudes, were referred to (Avorn 1983; Avorn 1992; Figueiras 2006; Ross‐Degnan 1996; Simon 2005). In eight studies, implementation models, existing evidence on intervention effectiveness, intervention mapping, or a statement on the logic of tailoring were given as the rationale for selection of interventions (Beeckman 2013; Callahan 1994; Cheater 2006; Flottorp 2002; Fretheim 2006; Goodwin 2001; Looijmans 2010; Schouten 2007).
Characteristics of the intervention
The interventions applied in the included studies were generally multifaceted. Educational materials, in the form of guidelines, copies of articles, summary documents or abstracts, were the most common intervention, being used in 16 studies. In 15 studies, educational outreach was used, either on a one‐to‐one basis or with groups. Educational group sessions were used in 14 studies and involved different formats, varying from lecture formats to facilitated interactive group discussions. Audit with feedback was also a common intervention, being used in eight studies. Decision support and other tools to aid health professionals in consultations with patients were used in eight studies, and role changes, including the selection of local opinion leaders or co‐ordinators, were included in the interventions in eight studies. In six studies, reminders were used, either in consultations or in mailings or meetings, and practical assistance or organisational changes were included in four studies. In 20 studies, there was some adjustment to local factors, at the level of individual or team level; in five studies there was no adjustment; and in seven it was unclear whether there had been any adjustment.
Excluded studies
We excluded 106 studies for not meeting our eligibility criteria. For details see: Characteristics of excluded studies.
Risk of bias in included studies
For each study, the risk of bias is indicated in Table 1 (see Figure 2 and Figure 3). We assessed the overall risk of bias as high in four studies (Evans 1997; Hux 1999; Matchar 2002; Santoso 1996), unclear in 21 (Avorn 1983; Avorn 1992; Baker 2001; Beeckman 2013; Coenen 2004; Engers 2005; Foy 2004; Goodwin 2001; Karuza 1995; Lakshminarayan 2010; Langham 2002; Leviton 1999; Looijmans 2010; Murphy 2009; Ross‐Degnan 1996; Schouten 2007; Simon 2005; Soumerai 1998; van Bruggen 2008; van Gaal 2010; Zwarenstein 2007), and low in six (Cheater 2006; Fairall 2012; Figueiras 2006; Flottorp 2002; Fretheim 2006; Scott 2013).
'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.
'Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Effects of interventions
See: Summary of findings for the main comparison
The findings are summarised in the GRADE table (summary of findings Table for the main comparison), the summary of findings worksheets being included in Appendix 12 and Appendix 13.
In Table 2, for the 15 studies included in the meta‐regression analyses, we report the data utilised in the models. The effect sizes reported have been adjusted for the clustering effect induced by the study designs. The odds ratios at follow‐up, adjusted for clustering, ranged from 1.08 to 10.59. When combined using meta‐regression techniques and adjusting for baseline odds ratios, the pooled odds ratio was 1.56 (95% confidence interval (CI) 1.27 to 1.93, P value < 0.001). In Figure 4, the log odds ratios at follow‐up are plotted against the log odds ratios at baseline, with each circle representing one study in the analysis, and the straight line indicating the pooled estimated follow‐up log odds ratio for each value of the baseline log odds ratio. Circle size is relative to the standard error of the log odds ratio. The 17 studies not included had findings showing variable effectiveness consistent with the findings of the meta‐regression.
Meta‐regression plot for the 15 studies in the analysis
In addition to the main comparison in this review, we also undertook comparisons with no intervention or non‐tailored interventions separately. Seventeen studies compared tailored interventions to no intervention (Avorn 1983; Avorn 1992; Callahan 1994; Fairall 2012; Figueiras 2006; Flottorp 2002; Goodwin 2001; Hux 1999; Looijmans 2010; Matchar 2002; Murphy 2009; Ross‐Degnan 1996; Santoso 1996; Scott 2013; Schouten 2007; van Gaal 2010; Zwarenstein 2007) (see Table 1). Of these, seven were suitable for inclusion in a meta‐regression, and the pooled odds ratio for the seven studies that received no control intervention was 1.36 (95% CI 0.92 to 1.99, P value = 0.099) (Avorn 1992; Callahan 1994; Flottorp 2002; Looijmans 2010; Murphy 2009; Scott 2013; Schouten 2007).
Fifteen studies compared tailored interventions to a non‐tailored intervention (Baker 2001; Beeckman 2013; Cheater 2006; Coenen 2004; Engers 2005; Evans 1997; Foy 2004; Fretheim 2006; Karuza 1995; Lakshminarayan 2010; Langham 2002; Leviton 1999; Simon 2005; Soumerai 1998; van Bruggen 2008) (see Table 1). Eight of these were included in a meta‐regression; the pooled odds ratio was 1.79 (95% CI 1.06 to 3.01, P value = 0.033) (Baker 2001; Beeckman 2013; Cheater 2006; Coenen 2004; Evans 1997; Fretheim 2006; Leviton 1999; Simon 2005). In all but one of these trials, the non‐tailored intervention consisted of the dissemination of written educational materials or guidelines. Beeckman 2013 issued a standard protocol and group lecture to the control group. Removing this study from the meta‐regression gave a pooled odds ratio of 1.48 (95% CI 1.24 to 1.75, P value = 0.002) (plots for these two comparisons are shown in Figure 5 and Figure 6).
Meta‐regression plot for the eight studies that had a non‐tailored control
Meta‐regression plot for the seven studies with a control of no intervention
We carried out analyses to investigate possible sources of heterogeneity between trial results. Study attributes assessed were risk of bias, explicit utilisation of a theory when developing the intervention, adjustment to local factors, and number of domains addressed by the determinants identified. Separate models were fitted dependent on the intervention delivered to the control group (none or non‐tailored), but none were found to be associated with the reported effectiveness of the tailored interventions. We carried out sensitivity analyses assuming a larger clustering effect than had been accounted for in the standard analyses, by using higher intra‐class correlation coefficient (ICC) estimates taken from Campbell (Campbell 2005). For the main analysis with all 15 studies, the pooled odds ratio was 1.54 (95% CI 1.27 to 1.86, P value < 0.001). For the eight studies where the control group was a non‐tailored intervention, the OR was 1.63 (95% CI 1.11 to 2.40, P value = 0.020), and for the eight studies where the control group received no intervention the OR was 1.30 (95% CI 0.78 to 2.15, P value = 0.243).
Since the studies were of interventions designed to implement appropriate clinical practice, adverse effects may have been unlikely, although it is possible that the implementation interventions could have unintended effects. Clear reports of such effects were not included in the studies.
Discussion
Summary of main results
In this update of our review of randomised trials of tailored implementation interventions, we identified an additional nine studies, seven of which had been published since the previous version of the review (Baker 2010). With 23 studies included from the previous version of the review, the finding of nine further studies suggests that the number of research studies into the effectiveness and mechanisms of tailored implementation is increasing. Despite the increase in the number of studies, however, our overall finding is similar to that of the previous review. Tailored implementation can be effective, but the effect is variable and tends to be small to moderate.
In the subsidiary comparisons, the effect of tailoring appeared to be less in comparison with no intervention than with non‐tailored interventions. This unexpected finding may be due to imprecision, chance, the small number of studies, and other unexplained factors. The best estimate of the effectiveness of tailored interventions is most likely to be that of the overall analysis that included all 15 studies.
Overall completeness and applicability of evidence
The completeness and applicability of the evidence are limited by the current level of development of the methods of tailoring. In the included studies, tailoring was undertaken in different ways and agreement on which methods should be used appeared to be absent.
The methods used to identify determinants and to select interventions to address them, including the rationale underpinning the approach to intervention selection, varied between studies. Determinants may be investigated by various methods and, if several methods are used together, a large number of determinants may be identified (Krause 2014a). Determinants may be classified in different ways (Légaré 2009). We used a recently developed classification, which employs descriptive categories derived from a review of studies of determinants (Flottorp 2013). Once determinants have been identified and those to be addressed have been chosen, strategies to address them have to be selected, but this process may be undertaken in different ways (Wensing 2014). This process was not, however, described in detail in the included studies and they did not suggest that a generally accepted method had emerged. The adoption of a standard approach to reporting interventions, such as TIDieR, might help to overcome this problem (Hoffmann 2014).
The studies in our review also did not investigate whether identified determinants had been overcome by the chosen interventions other than through assessment of changes in professional behaviour or health outcomes. In future, researchers should consider investigating whether determinants have indeed been addressed, by undertaking process evaluations or investigations of programme theory (Rogers 2008) alongside trials, perhaps incorporating some of the methods initially used to identify the determinants, with investigation taking place in both the intervention and control arms of trials. Studies to compare different ways of selecting interventions are also required, for example studies that compare the use of different theories, or the use of an explicit theory with no explicit theory.
Furthermore, it is not clear which element of the tailored strategy approach explained effectiveness. The studies employed various interventions to improve professional practice and it is possible that use of such interventions (for example, audit with feedback, educational outreach) would have improved professional practice whether or not tailoring had been undertaken. Eight of the trials in the meta‐regression included a control group that received a non‐tailored intervention, but in all but one study the control intervention was limited to the dissemination of educational materials or guidelines. Therefore, our review shows that tailored strategies can be effective, but is unable to determine whether this approach is more effective than selecting other interventions. Evidence on the applicability of the method to low‐income countries and with disadvantaged groups is also limited.
It should be pointed out that the studies included in this review do not enable any assessment of the costs of tailored strategies. Since the identification of determinants and tailoring of strategies involve additional steps beyond the application of a particular strategy, such as education alone, the economic costs of tailoring may be higher than several other interventions. Conversely, they may be lower through enabling the more expensive elements of interventions to be reserved for situations when they are likely to be effective. Consequently, evidence of the cost‐effectiveness of tailoring in comparison with other implementation methods is required from well‐designed evaluation studies. There are, therefore, several important questions to be addressed in future research into the effectiveness of tailored strategies.
Certainty of the evidence
It was possible to include 32 trials in this update, whereas 26 were included in the previous version (Baker 2010). We excluded three studies from this update that had been included in the previous version of the review because we assessed them as no longer meeting our criteria for tailoring. Fifteen studies could be included in the meta‐regression analysis in this update. Therefore, the amount of evidence has improved. Nevertheless, applying the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) system (GRADE Working Group 2004), the certainty of evidence is still moderate (summary of findings Table for the main comparison). The reasons for this remain the variable risk of bias of the included studies and the inconsistent results.
A number of questions remain about the design of tailored strategies and their impact on identified determinants, as described above. It is possible to have reasonable confidence that tailored implementation strategies are more likely to lead to improved performance than the dissemination of guidelines alone, but further well‐planned studies are required to determine how the tailored strategies approach should be designed to maximise effectiveness, and how the approach compares to other more intensive implementation strategies.
Agreements and disagreements with other studies and reviews
The only reviews that have directly investigated the effects of tailored interventions are the previous versions of this review (Baker 2010; Shaw 2005). In their review of 32 studies (randomised controlled trials, controlled trials, controlled before and after studies and interrupted time series) of implementation of clinical guidelines in obstetric care, Chaillet 2006 reported that the proportion of strategies that were effective was higher among studies that included a prospective identification of determinants compared with standardised interventions. Bosch 2007 undertook qualitative analysis of 20 purposefully selected quality improvement studies that reported investigations of determinants of practice. They found that attention to determinants did not always mean that the chosen intervention was based on determinants identified, although determinants were often used to adjust interventions, and concluded that the design of quality improvement interventions was in its infancy, the translation of identified determinants into implementation interventions still being a black box. Our findings concur with these reviews in showing that tailored strategies can be effective, but that the methods of tailoring are not yet well developed and are not described in detail in published studies.
Potential biases in the review process
The review was limited to randomised controlled trials (RCTs) and whilst the randomised trial design is considered to be less susceptible to bias in comparison with other study designs, it is possible that good quality interrupted time series or controlled before‐after studies could provide further insight into the effectiveness of tailored implementation strategies.
A potential limitation of electronic handsearching is that this approach, in contrast to handsearching print journals, risks overlooking otherwise unpublished studies reported in (non‐indexed) conference abstracts and journal supplements (Hopewell 2002). However, this is more likely to be a source of bias for reviews in which interrupted time series and controlled before‐after studies are included, since in comparison with these types of studies, randomised trials are more likely to be identified through electronic database searches. Using a complex search, including a sensitive RCT filter, in the key electronic databases should have identified the majority of relevant, published trials.
Of the 32 trials reviewed, only 15 could be included in the meta‐regression analysis. In the meta‐regression analysis, the outcomes included were either those reported as the primary outcome or, when this was not possible, we selected the most clinically relevant measure and therefore the introduction of bias is unlikely. We pooled a relatively wide variety of outcomes in the meta‐regression analysis, although in all studies the study outcomes related to processes of care, and the studies all addressed the same question about the effectiveness of tailored interventions. The small number of studies, however, limited the power to detect study attributes that could explain the variation in intervention effectiveness.
'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.
'Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Meta‐regression plot for the 15 studies in the analysis
Meta‐regression plot for the eight studies that had a non‐tailored control
Meta‐regression plot for the seven studies with a control of no intervention
| Interventions tailored to address identified determinants of practice compared with no intervention for implementing appropriate clinical practice | ||||||
| Patient or population: healthcare professionals Settings: mostly primary care in the USA and Europe Intervention: tailored interventions to implement practice guidelines Comparison: no intervention or dissemination of guidelines alone | ||||||
| Outcomes | Absolute effect | Relative effect | No of participants | Certainty of the evidence | Comments | |
| Without tailored intervention | With tailored intervention | |||||
| Difference (95% CI) | ||||||
| Implementation of recommended practice, e.g. clinical practice guideline recommendations | Moderate adherence* 60 per 100 patients | 70 per 100 patients | OR 1.56 (95% CI 1.27 to 1.93, P value < 0.001) | 15 studies with at least 7990 health professionals (numbers unclear in 5 studies) | ⊕⊕⊕⊝ | 17 other studies could not be included in the meta‐regression. The effect of tailored interventions in these studies varied from no effect to moderate effect between studies and between outcomes within studies, a finding consistent with the meta‐regression |
| Difference: 10 more patients receiving recommended practice per 100 patient encounters (Margin of error: 6 to 14 more patients) | ||||||
| Low adherence* 20 per 100 patients | 28 per 100 patients | |||||
| Difference: 8 more patients receiving recommended practice per 100 patient encounters (Margin of error: 4 to 13 more patients) | ||||||
| Healthcare outcomes | ‐ | ‐ | ‐ | No studies | ‐ | The studies did not include sufficient evidence to enable an assessment of effect on healthcare outcomes to be made |
| Costs | ‐ | ‐ | ‐ | No studies | ‐ | The studies did not include sufficient evidence to enable an assessment of effect on healthcare outcomes to be made |
| Adverse effects | ‐ | ‐ | ‐ | No studies | ‐ | The studies did not include sufficient evidence to enable an assessment of effect on healthcare outcomes to be made |
| Margin of error = Confidence Interval (95% CI) OR: Odds Ratio *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). † The OR and confidence intervals shown are taken from a meta‐regression. The results of 14 studies not included in the meta‐regression indicated that, on average, tailored interventions improve professional practice. However, the effects were mixed. CI: confidence interval; OR: odds ratio | ||||||
| GRADE Working Group grades of evidence High = This research provides a very good indication of the likely effect. The likelihood that the effect will be substantially different† is low. Moderate = This research provides a good indication of the likely effect. The likelihood that the effect will be substantially different† is moderate. Low = This research provides some indication of the likely effect. However, the likelihood that it will be substantially different† is high. Very low = This research does not provide a reliable indication of the likely effect. The likelihood that the effect will be substantially different† is very high. †Substantially different = a large enough difference that it might affect a decision | ||||||
| 1The assumed risks without a tailored intervention were selected to help interpret the overall odds ratios in situations in which there is a high risk of undesirable professional practice without intervening (20% desired practice) and a medium risk (60% desired practice). | ||||||
| Risk of bias | Study ID | Primary outcome(s) | Effect size | Authors' conclusions |
| Tailored intervention compared to no intervention | ||||
| Unclear | Avorn 1983 | 1. Prescribing of targeted drugs (amount and costs) No suitable dichotomous outcome reported | Costs reduced in intervention arm versus control by 14% (P value = 0.0001) | Academic‐based 'detailing' was a useful and cost‐effective way to improve the quality of drug therapy decisions and reduce unnecessary expenditures |
| Unclear | Avorn 1992 | 1. Residents not on psychoactive drugs | 1. Decrease of 27% in intervention arm and 8% in control arm (P value = 0.02) | An educational programme targeted to physicians, nurses, and aides can reduce the use of psychoactive drugs in nursing homes without adversely affecting the overall behaviour and level of functioning of the patient |
| High | Callahan 1994 | 1. Frequency of recording a depression diagnosis 2. Stopping medications associated with depression 3. Initiating antidepressant medication 4. Psychiatry referral | 1. 12% control and 32% intervention arm (P value < 0.01) 2. 22% control and 23% intervention arm 3. 8% control and 26% intervention arm (P value < 0.01) 4. 14% control and 12% intervention arm | Intensive screening and feedback of patient‐specific treatment recommendations increased the recognition and treatment of late life depression by GPs |
| Low | Fairall 2012 | 1. Time to death 2. Proportion with undetectable viral loads | 1. Time to death did not differ (hazard ratio 0.94, 95% CI 0.76 to 1.15) 2. Viral load suppression was similar in each group, 72% in the intervention and 70% in the control groups; risk difference 1.1% (95% CI ‐2.4 to 4.6) | Expansion of primary care nurses' role to include ART initiation and represcription can be done safely, but might not reduce time to ART or mortality |
| Low | Figueiras 2006 | 1.Number of reported adverse drugs reactions (ADRs) 2. Number of serious ADRs 3. Number of high causality ADRs 4. Number of unexpected ADRs 5. Number of new‐drug related ADRs Results not in a suitable format | 1. RR 10.23 (95% CI 3.81 to 27.51) 2. RR 6.32 (95% CI 2.09 to 19.16) 3. RR 8.75 (95% CI 3.05 to 25.07) 4. RR 30.21 (95% CI 4.54 to 200.84) 5. RR 8.04 (95% CI 2.10 to 30.83) | The intervention increased reporting of ADRs, with effect maximal at 4 months, but no longer from 13 months after intervention |
| Low | Flottorp 2002 | 1. Rate of antibiotic use | 1. 3% less likely to receive antibiotics after intervention in sore throat arm (P value = 0.032), no change in UTI arm | Passively delivered, complex interventions targeted at identified barriers to change had little effect in changing practice |
| Unclear | Goodwin 2001 | 1. Rate of up‐to‐date preventative services Results reported as percentages, numbers of patients not given | 1. Intervention: 31% to 42%, control: 35% to 37% (P value = 0.015) | An approach to increasing preventive service delivery that is individualised to meet particular practice needs can increase global preventive service delivery rates |
| High | Hux 1999 | 1. Median antibiotic cost Results reported as percentages, numbers of patients not given | 1. Change of 0.05% intervention versus 3.37% control, P value < 0.002 | A simple programme of confidential feedback and educational materials blunted cost increases, increased the use of first‐line antibiotics, and was highly acceptable to Ontario primary care physicians |
| Unclear | Looijmans 2007 | 1. The proportion of healthcare workers vaccinated against influenza | 1. Uptake was 9% higher than in the control group (RR 1.59, 95% CI 1.08 to 2.34) | The intervention resulted in higher, though moderate, influenza vaccine uptake among healthcare workers in nursing homes |
| High | Matchar 2002 | 1. % time in target range No suitable dichotomous outcome reported | 1. Difference (intervention minus control) adjusted for minor baseline differences was 5% (95% CI ‐5% to 14%), P value = 0.32 | A properly administered anticoagulation service can successfully manage the anticoagulation of most patients with atrial fibrillation; however, these services did not improve anticoagulation compared to usual care |
| Unclear | Murphy 2009 | The proportion of patients at 18‐month follow‐up above target levels for (1) blood pressure, (2) cholesterol, and (3) hospital admissions | 1. intervention (systolic) 27.2% versus 32.8 % in controls, OR 1.52 (0.99 to 2.30); 2. 15.2% versus 16.4%, OR 1.13 (0.63 to 2.03); 325.8% versus 34.0%, OR 1.56 (1.53 to 2.60) | There was a reduction in hospital admissions, but no other clinical benefits, possibly because of a ceiling effect |
| Unclear | Ross‐Degnan 1996 | 1. Sales of oral rehydration salts Results reported as percentages, numbers of patients not given | 1. Increased by 21% in intervention arms compared to controls (P value < 0.05) | Face‐to‐face training of pharmacy attendants, which targets deficits in knowledge and specific problem behaviours, can result in short‐term improvements in product sales and communication with customers |
| High | Santoso 1996 | 1. Prescribing of oral rehydration solution Results reported as percentages, numbers of patients not given | 1. Increase after intervention, but not after both interventions | The small group face‐to‐face intervention did not appear to offer greater impacts over large seminars in improving the appropriate use of drugs in acute diarrhoea |
| Unclear | Schouten 2007 | 1. Guideline‐adherent antibiotic prescription 2. Adjustment of antibiotic to renal function 3. Switches in therapy 4. Streamlining of therapy 5. Gram staining and culture of sputum samples (No primary outcome specified) | 1. Difference between intervention and control hospitals OR 2.63 (95% CI 1.57 to 4.42) 2. OR 12.9 (95% CI 3.64 to 45.8) 3. OR 1.20 (95% CI 0.02 to 76.51) 4. OR 1.94 (95% CI 0.34 to 11.03) 5. OR 1.13 (95% CI 0.64 to 2.00) Baseline: 1.24 (0.43 to 3.56) Follow‐up: 2.21 (0.79 to 6.17) | For some indicators, the intervention led to improvements. Secular trends may have had an effect on indicators that did not improve to a greater extent in the intervention group |
| Low | Scott 2013 | 1. Administration of alteplase in patients with stroke in emergency departments | 1. Increase from 1.25% to 2.79% in intervention hospitals, 1.25% to 2.10% in controls (RR 1.37, 95% CI 0.96 to 1.93) | The increase in use of alteplase was smaller than the effect to which the study was powered |
| Unclear | Van Gaal 2011 | 1. The incidence of adverse events per patient week (the sum of the incidents of pressure ulcers, urinary tract infections, and falls divided by the total number of weeks) | At follow‐up, the rate was 0.06 in the intervention group, 0.09 in the control group (RR 0.57, 95% CI 0.34 to 0.95) | It is possible to implement multiple guidelines simultaneously |
| Unclear | Zwarenstein 2007 | 1. Asthma symptom score | The decline in score on 1‐year follow‐up was 4.08 in the intervention group and 3.24 in the control group (adjusted for baseline, OR 1.48, 95% CI 1.00 to 2.20) | Educational outreach was effective in reducing children's asthma symptoms |
| Tailored intervention compared to non‐tailored intervention | ||||
| Unclear | Baker 2001 | 1. 3 or more symptoms recorded at diagnosis 2. Suicide risk assessed at diagnosis 3. Treated with antidepressant or cognitive therapy 4. Therapeutic dose of antidepressant 5. Reviewed after 3 weeks 6. Suicide risk reassessed 7. 2 or more follow‐up consultations 8. Treated for 4 months or more | 1. OR 1.9 (97% CI 0.9 to 3.8) 2. OR 5.6 (95% CI 2.8 to 11.3) 3. OR 2.5 (95% CI 0.7 to 9.2) 4. OR 1.3 (95% CI 0.6 to 3.2) 5. OR 1.1 (95% CI 0.5 to 2.4) 6. OR 0.7 (95% CI 0.2 to 3.0) 7. OR 2.0 (95% CI 0.9 to 4.0) 8. OR 1.2 (95% CI 0.6 to 2.4) (ORs adjusted for baseline) Baseline: 1.10 (0.74 to 1.64) Follow‐up: 1.57 (0.98 to 2.51) | The findings suggest that this approach to implementation may be effective and should be further investigated |
| Unclear | Beeckman 2013 | 1. Residents receiving fully adequate pressure ulcer prevention in bed 2. Residents receiving fully adequate prevention in a chair | 1. 4.6% in intervention group, 1.5% in control group 2. 60.0% in intervention group, 13.2% in control group, P value = 0.003 | Positive effects were observed when residents were in a chair, but not when in bed |
| Low | Cheater 2006 | 1. Nurse performance assessed by examining patients' nursing records against a list of review criteria (primary outcome) | Mean improvement in aggregate compliance scores in percentage points: 1. ‐2.3 (95% CI ‐1.63 to 1.7) for audit and feedback compared to control 2. 0.9 (‐3.3 to 5.1) for educational outreach compared to control | In comparison with educational materials alone, the implementation methods did not improve care at 6 months follow‐up |
| Unclear | Coenen 2004 | 1. Antibiotic prescribing rate by GPs for adult patients with acute cough | 1. OR 0.56 (95% CI 0.36 to 0.87) Risk of prescribing antibiotics for intervention group versus controls, adjusted for relevant clinical symptoms | Implementing a guideline for acute cough is successful in optimising antibiotic prescribing |
| Unclear | Engers 2005 | 1. Referrals to a therapist 2. Prescription of pain medication on a time‐contingent basis 3. Prescription of paracetamol versus NSAIDs No baseline data reported | Intervention compared to control: 1. OR 0.8 (95% CI 0.5 to 1.4) 2. OR 1.0 (95% CI 0.3 to 3.0) 3. OR 2.0 (95% CI 0.8 to 5.5) | The intervention modestly improved implementation of the Dutch low back pain guideline by GPs |
| High | Evans 1997 | 1. Rate of diagnosis of asthma | 1. 40/1000 versus 16/1000, P value < 0.01 | The intervention substantially increased child health staff's ability to identify children with asthma, involve them in continuing care, and provide them with state‐of‐the‐art care for asthma |
| Unclear | Foy 2004 | 1. Assessment appointment within 5 days 2. Ascertainment of cervical cytology history 3. Screening or antibiotic prophylaxis for genital tract infection 4. Misoprostol used for cervical priming and early and mid‐trimester abortion 5. Supply of contraception at discharge Results reported as percentages | Difference between intervention and control groups 1. OR 0.89 (95% CI 0.50 to 1.58) 2. OR 0.93 (95% CI 0.36 to 2.40) 3. OR 1.70 (95% CI 0.71 to 5.99) 4. OR 1.00 (95% CI 0.27 to 1.77) 5. OR 1.11 (95% CI 0.48 to 2.53) | The intervention was ineffective, possibly because of high pre‐intervention compliance and limited impact of the intervention on barriers outside the control of clinical staff |
| Low | Fretheim 2006 | 1. Proportion of patients prescribed a thiazide among patients prescribed an antihypertensive for the first time 2. Proportion of those started on antihypertensive or cholesterol‐lowering treatment having a cardiovascular risk assessment 3. Proportion satisfying treatment goals for BP or cholesterol | 1. Prescribing thiazides relative risk intervention versus control 1.94 (1.49 to 2.49) 2. Risk assessment done relative risk intervention versus control 1.04 (0.60 to 1.71) 3. Treatment goal achieved, intervention versus control relative risk 0.98 (0.93 to 1.02) | The intervention had an impact on prescribing patterns, but not on other outcomes |
| Unclear | Karuza 1995 | 1. Physician vaccination rates for influenza Results reported as percentages, numbers of patients not given | 1. The intervention arm had a higher adjusted vaccination rate (62.39%) compared to controls (46.46%), P value < 0.001 | Interventions using small groups can be useful in facilitating adoption of guidelines by physicians |
| Unclear | Langham 2002 | 1. Adequate recording of 3 risk factors n/N not reported | 1. Difference of 10.5% (95% CI ‐3.9 to 24.9) between information and no information and 6.6% (95% CI ‐8.9 to 22.0) between evidence and no evidence | Adequate risk factor recording did not differ between the information (versus not information) or the evidence (versus not evidence) intervention groups |
| Unclear | Lakshminarayan 2010 | Adherence to indicators for stoke care for (1) acute care; (2) in‐hospital care; (3) discharge care | 1. OR 1.8 (95% CI 0.44 to 7.6); 2. OR 1.05 (0.83 to 1.3); 3. OR 1.04 (0.64 to 1.7) | No intervention effect was demonstrated, although there was a secular trend |
| Unclear | Leviton 1999 | 1. Use of corticosteroids | 1. Use increased by 108% in active dissemination hospitals and by 75% in usual dissemination hospitals (P value < 0.01) | An active, focused dissemination effort increased the effectiveness of usual dissemination methods when combined with key principles to change physician practices |
| Unclear | Simon 2005 | 1. Proportion of patients with hypertension receiving a diuretic or beta‐blocker | Difference between control and group detailing OR 1.40 (95% CI 1.11 to 1.76) Difference between control and individual detailing OR 1.30 (95% CI 0.95 to 1.79) Difference between group and individual detailing OR 1.10 (95% CI 0.86 to 1.42) | Both detailing interventions resulted in an approximately 13% absolute increase in guideline‐recommended drugs |
| Unclear | Soumerai 1998 | 1. Appropriateness of the prescribing of selected drugs (aspirin in eligible elderly patients) Data reported as percentages, numbers not given | 1. Median change +0.13 in intervention and ‐0.03 in controls, P value = 0.04 | Working with opinion leaders and providing performance feedback can accelerate adoption of some beneficial acute myocardial infarction therapies |
| Unclear | Van Bruggen 2008 | 1. % of patients with poor control achieving HbA1c of < 8% | 70% in the intervention and 58% in the control groups achieved adequate control (not after controlling for baseline value, potential confounders and clustering) | The process of diabetes care did improve, but intermediate outcomes hardly changed |
| ADR: adverse drugs reaction | ||||
| Study ID | Outcome | Baseline odds ratios (95% CI) | Follow‐up odds ratios (95% CI) |
| Avorn 1992 | Residents not on antipsychotic drugs | 0.90 (0.42 to 1.90) | 1.08 (0.49 to 2.34) |
| Baker 2001 | Antidepressants in therapeutic dose | 1.10 (0.74 to 1.64) | 1.57 (0.98 to 2.51) |
| Beeckman 2013 | Fully adequate prevention of ulcers | 1.02 (0.35 to 2.94) | 10.59 (3.56 to 31.45) |
| Callahan 1994 | Depression diagnosis | 1.23 (0.57 to 2.63) | 2.65 (1.40 to 5.03) |
| Cheater 2006 | Recording of management criteria | 1.37 (0.85 to 2.22) | 1.65 (0.99 to 2.71) |
| Coenen 2004 | Antibiotics not prescribed | 0.80 (0.49 to 1.32) | 1.07 (0.59 to 1.92) |
| Evans 1997 | Returning asthma patients from previous year | 0.94 (0.48 to 1.83) | 2.88 (1.28 to 6.46) |
| Flottorp 2002 | Antibiotics not prescribed | 1.12 (0.94 to 1.31) | 1.26 (1.06 to 1.50) |
| Fretheim 2006 | Thiazides prescribed for hypertension | 0.63 (0.42 to 0.95) | 1.68 (1.20 to 2.35) |
| Leviton 1999 | Use of antenatal corticosteroids | 1.00 (0.65 to 1.51) | 1.59 (0.88 to 2.83) |
| Looijmans 2010 | Uptake of flu vaccine | 0.98 (0.63 to 1.52) | 1.71 (1.10 to 2.65) |
| Murphy 2009 | Numbers below the target level for BP | 0.99 (0.68 to 1.41) | 1.31 (0.87 to 1.96) |
| Schouten 2007 | % where key quality indicators performed | 1.24 (0.43 to 3.56) | 2.21 (0.79 to 6.17) |
| Scott 2013 | Use of alteplase for stroke | 1.00 (0.14 to 7.05) | 1.34 (0.36 to 4.92) |
| Simon 2005 | Beta‐blockers or diuretics prescribed for hypertension | 1.03 (0.88 to 1.21) | 1.40 (1.18 to 1.65) |
| Note: Odds ratio = odds of outcome in treatment group/odds of outcome in control group, calculated at baseline and follow‐up and adjusted for clustering BP: blood pressure | |||
