Background: non-inferiority trials (NITs) evaluate whether an intervention is not substantially worse than a comparison by a predefined non-inferiority margin. NITs have gained popularity in the medical literature and are increasingly included in systematic reviews and meta-analyses as they theoretically require smaller sample size, compare to an active control, and report patient-centred outcomes rather than surrogate outcomes.
Objectives: to empirically evaluate the characteristics of NITs versus superiority trials including their effect size and risk of bias. Knowing these characteristics can help in integrating them in meta-analyses.
Methods: we searched for all meta-analyses of at least five randomized controlled trials (RCTs) published between 2007 and 2015 in the 10 top medical journals. Meta-analyses had to evaluate a medication or device for chronic medical conditions. We selected one binary outcome deemed to be most important to patients (e.g. mortality, stroke, and myocardial infarction) from each meta-analysis. We excluded meta-analyses of diagnostic studies, prognostic studies, behavioral interventions, and continuous outcomes. We calculated differences in terms of log-transformed odds ratio from each trial compared with log-transformed final pooled odds ratio and compared them between NITs and superiority trials using t test and Chi2 test. We also evaluated a priori hypothesized explanatory variables that might affect the validity of trials.
Results: we identified 70 meta-analyses with 930 RCTs for chronic medical conditions, of which 32.86% (23/70) included at least one NIT. The mean percentage of NITs in a meta-analysis was 15.10% (range 2.08% to 41.67%). We found no statistically significant difference in reported effect size between the NITs and the superiority trials (P = 0.94). The NITs had significantly more participants (P < 0.01) and were more likely to be multi-centre trials (P = 0.02). We found no statistical difference on being an early study in evidence chain, study length, funding source, number of trials stopped early, or any 'Risk of bias' items in the Cochrane 'Risk of bias' tool (Table 1). The primary outcome in 31.58% NITs (12/38) were composite outcomes; 89.47% (34) did not justify the non-inferiority margin; and 71.05% (27) used either an intention-to-treat or per-protocol analysis without reporting both.
Conclusions: this study found no empirical evidence to suggest that NITs importantly differ from superiority trials, suggesting it may be appropriate to pool NITs and superiority trials together. The reporting and methods of NITs need to be improved.
Patient or healthcare consumer involvement: none