Randomized controlled trials in de-implementation research: a systematic scoping review

We performed a systematic scoping review, registered the protocol in Open Science Framework (OSF hk4b2) [9], and followed the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) checklist [10] (Additional file 2).

Data sources and searches

We developed a comprehensive search strategy in collaboration with an experienced information specialist (T. L.) (Additional file 1, eMethods 1). We searched MEDLINE and Scopus for individual and cluster RCTs of de-implementation interventions without language limits through May 24, 2021. First, we used terms identified by an earlier scoping review of de-implementation literature [11] (judged useful in earlier de-implementation research [12, 13]). Second, we identified relevant articles from previously mentioned [11] and two other [3, 4] earlier systematic reviews of de-implementation. Using these identified articles, we updated our search strategy with new index terms (Additional file 1, eMethods 1). Third, we performed our search with all identified search terms (step 1 and step 2). Fourth, we identified systematic reviews (found by our search) and searched their reference lists for additional potentially eligible articles. Finally, we followed up protocols and post hoc analyses (identified by our search) of de-implementation RCTs and added their main articles to the selection process.

Eligibility criteria

We included all types of de-implementation interventions across all medical specialties. We included trials comparing a de-implementation intervention to a placebo, another de-implementation intervention, or usual care. We included studies with any target group, including patients with any disease as well as all kinds of healthcare professionals, organizations, and laypeople. In our review, we excluded deprescribing trials as we considered the context of stopping a treatment already in use (deprescribing) to be somewhat different than the context of not starting a certain treatment (de-implementation), for example, stopping use of long-term benzodiazepines for anxiety disorders (deprescribing) vs not starting antibiotics for viral respiratory tract infections (de-implementation) [14]. We also excluded trials only aiming to reduce resource use (e.g., financial resources or clinical visits) and trials where a new medical practice, such as laboratory test, was as an intervention to reduce the use of another practice.

Outcomes and variables

We collected and evaluated the following outcomes/variables: (1) study country, (2) year of publication, (3) unit of randomization allocation (individual vs. cluster), (4) the number of clusters, (5) was an intra-cluster correlation (ICC) used in sample size calculation, (6) duration of follow-up, (7) setting, (8) medical content area, (9) target group for intervention, (10) the number of study participants, (11) mean age of study participants, (12) the proportion of female participants, (13) intervention categories, (14) rationale for de-implementation, (15) goal of the intervention, (16) outcome categories, (17) reported effectiveness of the intervention, (18) conflicts of interest, (19) funding source, (20) risk of bias, (21) implementation theory used, (22) costs of the de-implementation intervention, (23) effects on total healthcare costs, (24) changes between baseline and after the intervention, and (25) tailoring the de-implementation intervention to study context.

Risk of bias and quality indicators

To improve judgements regarding the studies with complex intervention designs and to enhance the interrater agreement [15] in risk-of-bias assessment, through iterative discussion, consensus building, and informed by previous literature [16, 17], we modified the Cochrane risk-of-bias tool for cluster randomized trials [18] (Additional file 1, eMethods 2). Studies were rated based on six criteria: (1) randomization procedure, (2) allocation concealment, (3) blinding of outcome collection, (4) blinding of data analysts, (5) missing outcome data, and (6) imbalance of baseline characteristics. For each criterion, studies were judged to be at either high or low risk of bias. In addition, we collected data on the number of clusters, length of follow-up, intra-cluster correlation, tailoring, theoretical background, level of randomization, and reported differences before and after the baseline, and considered these as quality indicators.

Study selection and data extraction

We developed standardized forms with detailed instructions for screening abstracts and full texts, risk of bias assessment, and data extraction (including outcomes/variables, intervention categorization, and outcome hierarchy). Independently and in duplicate, two methodologically trained reviewers applied the forms to screen study reports for eligibility and extracted data. Reviewers resolved disagreements through discussion and, if necessary, through consultation with a clinician-methodologist adjudicator.

Intervention categorization and outcome hierarchy

To define categories for the rationale of de-implementation, we used a previous definition of low-value care: “care that is unlikely to benefit the patient given the harms, cost, available alternatives, or preferences of the patient” [2].

We modified the Effective Practice and Organisation of Care (EPOC) taxonomy of health systems interventions to better fit the current de-implementation literature [19]. First, we categorized the interventions from eligible studies according to the existing EPOC taxonomy. Second, we discussed the limitations of the EPOC taxonomy with our multidisciplinary team and built consensus on modifications (categories to be modified, excluded, divided, or added). Finally, we repeated the categorization by using our refined taxonomy. Disagreements were solved by discussion and/or by consulting an implementation specialist adjudicator. Full descriptions of intervention categories and the rationale for the modifications are available in the Additional file 1 (eMethods 3 and 4).

To develop outcome categories for effectiveness outcomes (Table 1), we modified Kirkpatrick’s levels for educational outcomes [20]. We identified five categories: health outcomes, low-value care use, appropriate care use, total volume of care, and intention to reduce low-value care. A complete rationale for the hierarchy of outcomes is available in the Additional file 1 (eMethods 5).

Table 1 Outcome categories for de-implementation effectivenessAnalysis

We used summary statistics (i.e., frequencies and proportions, typically with interquartile ranges) to describe study characteristics. We compared quality indicators (see paragraph “Risk of bias and quality indicators”) between studies published in 2010 or before and after 2010 to explore potential changes in trial methodology and execution. Finally, considering the lack of methodological standards in de-implementation literature (also identified by our scoping review), we created recommendations for future de-implementation research. Through discussion and consensus building, we drafted recommendations in several in-person meetings. Subsequently, authors gave feedback on the drafted recommendations by email. Finally, we made final recommendations in in-person meetings.

留言 (0)

沒有登入
gif