Objective: To identify potential bias in non-inferiority design of published cancer trials, and to provide suggestions for future practice.
Study design and setting: We systematically searched MEDLINE, Embase and CENTRAL databases (until April 17, 2020) to obtain non-inferiority phase III cancer trials and protocols. Distribution of essential characteristics and study design parameters was compared between trials with and without concluding non-inferiority using multivariable logistic regression.
Results: A total of 291 eligible trials were included. We observed that increased odds of concluding non-inferiority was significantly associated with more lenient non-inferiority margins (OR = 1•94, 95% CI 1•02-3•69) and higher hypothesized event rate (OR = 1•24, 95% CI 1•06-1•47). Trials that established non-inferiority adopted margins that were more dispersedly distributed (dispersion OR = 2•90, 95% CI 1•88-4.48).
Conclusion: Although limited by the exploratory nature, our study demonstrated existence of possible distorted non-inferiority design which could incur excess non-inferiority in cancer clinical trials. Pre-registration and transparent reporting of detailed non-inferiority design is imperative for future research.
Keywords: Bias; Event rate; Margin; Non-inferiority; Oncology; Randomized clinical trial.
Copyright © 2021. Published by Elsevier Inc.