Randomized, placebo-controlled trials of antidepressants for acute major depression: thirty-year meta-analytic review

Abstract

Antidepressant-placebo response-differences (RDs) in controlled trials have been declining, potentially confounding comparisons among older and newer drugs. For clinically employed antidepressants, we carried out a meta-analytic review of placebo-controlled trials in acute, unipolar, major depressive episodes reported over the past three decades to compare efficacy (drug-placebo RDs) of individual antidepressants and classes, and to consider factors associated with year-of-reporting by bivariate and multivariate regression modeling. Observed drug-placebo differences were moderate and generally similar among specific drugs, but larger among older antidepressants, notably tricyclics, than most newer agents. This outcome parallels selective increases in placebo-associated responses as trial-size has increased in recent years. Study findings generally support moderate efficacy of clinically employed antidepressants for acute major depression, but underscore limitations of meta-analyses of controlled trials for ranking drugs by efficacy. We suggest that efficiency and drug-placebo differences may be improved with fewer sites and subjects, and better quality-control of diagnostic and clinical assessments.

Figure 1

Summary of meta-analytically computed relative rates (RR) of response after randomization to drug vs placebo) with 95% confidence intervals (CI, horizontal bars when n⩾2 trials per drug) for controlled trials of each of 19 antidepressants (with numbers of trials on the left axis, and numerical values on the right). Drugs are listed by descending apparent efficacy, with symbol-size approximately proportional to weighting by trials per drug. The vertical solid line=null (1.0); vertical dotted line and solid diamond (width=CI)=pooled RR for all agents tested (^*p<0.05; ^**p⩽0.01; ^***p⩽0.001). Overall pooled RR=1.42 (CI: 1.38–1.48), indicating an average of 42% superiority of antidepressants over placebos. Note that phenelzine, clomipramine, tranylcypromine and trazodone (n=1 trial each) appear to be outliers.

Figure 2

Correlations with Spearman nonparametric correlation coefficients (r_s): (a) Subjects per trial vs year of trial reporting (p<0.0001); (b) Collaborating sites per trial vs year (p<0.0001); (c) Meta-analytic responder rate (% of subjects) for antidepressants vs year (p=0.58); (d) Responder rate (%) for placebo vs year (p<0.0001); (e) Responder rate ratio (RR: drug–placebo) vs year (p<0.0001); (f) RR vs subjects per trial (p=0.002). In addition RR decreased significantly with more sites per trial (r_s=−0.302, p=0.004). Note that response after randomization to placebo but not antidepressant drugs selectively increased over years, as subject and site counts per trial increased, with corresponding decreases in drug–placebo relative response rate-ratio (RR).