Identifying Key Variables in AUD Multisite Trials: A Systematic Review

Purpose: Multi-site randomized control trials (MS-RCTs) are critical for producing evidence on pharmacotherapy as a treatment for alcohol use disorder (AUD-RX). However, trial design characteristics and analytic approaches are heterogeneous across studies, creating a challenge for synthesizing evidence and a missed opportunity for informing the design of new studies. This study is a systematic review characterizing study design features and reported design-specific empirical findings from AUD-RX MS-RCTs.

Methods: A series of iterative database searches were deployed to identify MS-RCTs relevant to AUD-RX. In October 2021, we searched PubMed, Scopus, Embase, Cochrane (CENTRAL), and Ovid Medline. English-language main findings papers from between 1975 and 2021 with at least 12 weeks of study treatment were included. Pilot studies were excluded. Study features included sample sizes, number of sites, treatment length, data collection length, intra-class correlation(ICC), endpoint measures, statistical approach and missing data, attrition, and imputation. Frequencies are calculated for categorical features. Simple and meta-analytic statistics are estimated for quantitative features. Meta-regression is used to assess how different study design features and strata are associated with empirical findings.

Results: The initial search identified 566 main findings papers of which 25 met inclusion/exclusion criteria. Naltrexone was the most common AUD-RX (8) followed by disulfiram (3) and acamprosate (2). The majority (16) were two-arm trials. All trials had a placebo group. Average trial treatment length was 18.6 weeks with 12 weeks (12) and mode 16 (n=5). Median sample size was 270 with an interquartile range of 150 to 371. Studies relied on an average of 10.6 (SD = 9.8) sites. ICCs were not reported by any study. Seventeen studies reported using site as a covariate in outcome models. One study reported combing study sites due to small site sample sizes. The median patients/site was 38 (IQR: 21, 85). Number of study sites was negatively correlated with patients/site (-1.5; p=.011). The odds of detecting statistical significance (p<.05) for a primary hypothesis increased with higher patients/site (OR=1.046; p=.045).

Conclusions: Studies varied substantially in many of their design features. Reported information on study site characteristics was minimal. Results suggest that detecting statistically significant effects is sensitive to the tradeoff between overall sample size and number of study sites.