Reported Methodologic Quality and Discrepancies between Large and Small Randomized Trials in Meta-Analyses

Abstract

Purpose: To explore whether reported methodologic quality affects estimated intervention effects in randomized trials and contributes to discrepancies between the results of large randomized trials and small randomized trials in meta-analyses.

Data Sources: Meta-analyses of randomized trials that included at least one large trial (≥ 1000 participants) were included, regardless of the therapeutic area. Eligible meta-analyses were identified through electronic searches and bibliographies of relevant articles.

Study Selection: Full-length randomized trials.

Data Extraction: Methodologic quality was assessed according to reported randomization, double blinding, and follow-up as separate components and by using the Jadad composite scale.

Data Synthesis: Fourteen meta-analyses involving 190 randomized trials from eight therapeutic areas were included. Compared with large trials, intervention effects were exaggerated in small trials with inadequate allocation sequence generation (ratio of odds ratios, 0.46 [95% CI, 0.25 to 0.83]; P = 0.011), inadequate allocation concealment (ratio of odds ratios, 0.49 [CI, 0.27 to 0.86]; P = 0.014), and no double blinding (ratio of odds ratios, 0.52 [CI, 0.28 to 0.96]; P = 0.01). Large trials did not differ significantly from small trials with adequate generation of the allocation sequence, adequate allocation concealment, or adequate double blinding. No association was seen between reported follow-up and intervention effects. The Jadad scale provided no additional information because the scale and the quality components overlapped substantially.

Conclusions: Inadequate generation of the allocation sequence, allocation concealment, and double blinding lead to exaggerated estimates of intervention benefit and may contribute to discrepancies between the results of large randomized trials and small randomized trials in meta-analyses.