Withdrawal Confounding in Randomized Controlled Trials of Antipsychotic, Antidepressant, and Stimulant Drugs, 2000-2017

Background: Results of relapse prevention randomized controlled trials (RCTs) which discontinue psychotropic drug treatment from some participants may be confounded by drug withdrawal symptoms. We test for the confound by calculating whether >= 50% of the difference in relapse risk between drug-discontinued and drug-maintained groups is present at discontinuation time points (DCTs) with "short" and "long" assumptions regarding onset and duration of withdrawal symptoms. Methods: In eligible RCTs of antidepressants, antipsychotics, and stimulants from 2000 to 2017 (n = 30) selected from a systematic review, differences in relapse risk were examined by arithmetic and graphical comparison of mean behavioral scores or survival plots. Results: Only 14 studies (46.6%) with 15 analyses of relapse risk provided sufficient data. Under short and long DCTs, 9 of 13 (69.2%) and 7 of 9 (77.8%) interpretable analyses, respectively, suggested a withdrawal confound. The proportion of endpoint placebo-maintenance group difference present by the DCT averaged 69.1% (range, 58.7-148.0%, n = 13) for short DCT assumptions, and 79.0% (range, 51.5-183.3%, n = 9) under long DCTs. One study (3.33%) controlled for withdrawal effects, and 1 yielded inconclusive results. Conclusions: These results support suggestions that withdrawal symptoms confound the results of relapse prevention RCTs. Accounting for such symptoms in RCTs is an ethical, scientific, and clinical imperative. Justifications for relapse prevention RCTs employing a discontinuation procedure require more scrutiny. (C) 2019 S. Karger AG, Basel