Introduction: Randomized controlled trials (RCTs) designed to demonstrate non-inferiority of an intervention compared with control have become increasingly common in cardiovascular medicine. Such RCTs may be biased toward null findings through low enrollment, post-randomization exclusions, loss to follow-up, or wide inferiority margins. We characterized the features of non-inferiority cardiovascular RCTs published in high-impact journals that could lead to bias.
Methods: We searched PubMed for non-inferiority cardiovascular RCTs published between January 1, 1990 and August 11, 2016 in The New England Journal of Medicine, Lancet, and JAMA. We reviewed methodological characteristics, including sample size, power estimates, selected non-inferiority margin, and success of studies in achieving non-inferiority.
Results: Of 3,689 screened studies, we identified 104 non-inferiority RCTs. Publication increased over time (P<0.001), as more than 50% (n=53) were published since 2010. Of 101 trials with eligible data, 80 (77%) trials claimed non-inferiority (19 of which also demonstrated superiority), whereas 21 (20%) did not (including 7 which showed worse outcomes with the tested intervention, and 14 that had inconclusive results, Figure). Only 1 study had >10% of participants lost to follow-up. Of 75 studies with available data, 14 reported >10% post-randomization exclusions. Of 89 studies with available information, 10 analyzed a cohort >20% smaller than their calculated sample size. Only 55 studies (53%) reported all the randomized patients in the primary endpoint analyses. Only 52 trials (50%) reported analyses from both the intention-to-treat and per-protocol cohorts, of which 2 found a discrepancy in analyses. Treatment adherence was reported in 18 trials (34%). Pre-specified non-inferiority margins ranged widely, with absolute differences between 0.4-14%, hazard ratios between 1.05-2.85, odds ratios between 1.1-2.0, and relative risks between 1.1-2.0. Only 9 studies (8.7%) used a placebo or no-intervention arm.
Conclusion: Non-inferiority designed RCTs in cardiovascular medicine are increasingly published in high-impact journals, commonly conclude non-inferiority of the new intervention, and frequently have design features that might bias the studies toward non-inferiority.