Background: There is concern that non-inferiority trials might be deliberately designed to conceal that a new treatment is less effective than a standard treatment. In order to test this hypothesis we performed a meta-analysis of non-inferiority trials to assess the average effect of experimental treatments compared with standard treatments.
Methods: One hundred and seventy non-inferiority treatment trials published in 121 core clinical journals were included. The trials were identified through a search of PubMed (1991 to 20 February 2009). Combined relative risk (RR) from meta-analysis comparing experimental with standard treatments was the main outcome measure.
Results: The 170 trials contributed a total of 175 independent comparisons of experimental with standard treatments. The combined RR for all 175 comparisons was 0.994 [95% confidence interval (CI) 0.978-1.010] using a random-effects model and 1.002 (95% CI 0.996-1.008) using a fixed-effects model. Of the 175 comparisons, experimental treatment was considered to be non-inferior in 130 (74%). The combined RR for these 130 comparisons was 0.995 (95% CI 0.983-1.006) and the point estimate favoured the experimental treatment in 58% (n = 76) and standard treatment in 42% (n = 54). The median non-inferiority margin (RR) pre-specified by trialists was 1.31 [inter-quartile range (IQR) 1.18-1.59].
Conclusion: In this meta-analysis of non-inferiority trials the average RR comparing experimental with standard treatments was close to 1. The experimental treatments that gain a verdict of non-inferiority in published trials do not appear to be systematically less effective than the standard treatments. Importantly, publication bias and bias in the design and reporting of the studies cannot be ruled out and may have skewed the study results in favour of the experimental treatments. Further studies are required to examine the importance of such bias.