Effect Size Deflation With Increasing Sample Size in Adjunctive Trials for Negative Symptoms in Schizophrenia: A Meta-Analysis and Meta-Regression Across Mechanisms of Action
Joshua T. Kantrowitz, Megan Mayer, Tse‐Hwei Choo, Kevin Liu, Viraj Govani, Shyanne Francis, Daniel C. Javitt
Abstract
OBJECTIVE: Meta-analytic evidence supports the effectiveness of several mechanisms of action adjunctive to antipsychotics for the treatment of negative symptoms in schizophrenia. Nevertheless, the results have not been replicated consistently in multicenter randomized clinical trials. The authors' primary objective in this study was to assess whether differential scaling of response in the placebo arm by sample size could account for the discrepant results between smaller- and larger-scale trials. METHODS: The authors conducted a meta-analysis and meta-regression of the potential contribution of differential scaling by sample size, site number, and other potential moderators in randomized clinical trials of adjunctive mechanisms of action in schizophrenia that included negative symptom outcomes. RESULTS: Eight mechanisms of action were identified across 159 active-placebo comparisons with 13,020 unique participants (59.2% male [SD=26.8]; 57.2% multicenter studies). Between-group differences in negative symptoms were observed for four mechanisms of action. The size of the between-group treatment effects decreased significantly with increasing sample size (b=-0.60) and site number (b=-0.33). This was attributable primarily to significantly greater improvement in the placebo arm (b=-0.45, SE=0.21), but not the active treatment arm (b=0.14, SE=0.21). Consequently, significant between-group effects were observed preferentially in studies with ≤150 participants and ≤10 sites, along with a mean of >20 participants per site ("recruitment density") (b=-0.52). CONCLUSIONS: This differential placebo-active scaling with increased sample size is unlikely to be of physiological origin and may indicate that site management issues contribute to negative clinical results in large-scale trials for the negative symptoms of schizophrenia. Current power analytic approaches do not adequately consider the consequences of sample size-dependent effect size deflation, leading to excessive risk of type II error with increasing sample size or site numbers and decreasing recruitment density.