Statistical issues and recommendations for noninferiority trials in oncology: A systematic review.

Abstract

e13073 Background: This systematic review aims to provide an overview of previously published noninferiority (NI) trials in cancer with a special emphasis on methodological issues. Methods: We identified citations of randomized trials assessing NI of antineoplastic treatments by a computerized and manual search in October 2010 and reviewed for inclusion criteria. Data were extracted independently by 2 of 3 authors. Results: We identified 72 articles. An increasing trend was observed (4.9 articles per year between 2001 and 2007, 12.7 articles per year between 2008 and 2010). Quality of reporting in recent published articles was moderately improved after 2008 when the CONSORT was extended to NI trials (p<0.01). NI margins were used in 68 trials (94%, 17 trials used the fixed margin method, 5 trials used the effect retention method, 1 trial used both, 45 trials did not specify the method). Overall, variability in the resulting sizes of margins was small and margins were set to be no larger than the assured effect of active control. There was no trial that took uncertainty in historical data into consideration of the NI margin explicitly. Four trials had a plan of switching from superiority to NI. Five trials conducted a post hoc NI analysis. Conclusions: Statistical considerations in NI trials are less well established than in superiority trials and future methodological researches are needed. Practical recommendations were presented in the table. Design An NI margin should be specified prior to sample size calculation. Design The fixed margin and effect retention methods can be chosen according to objective of the trial. The objective would be to prove absence of a clinically relevant difference or that the experimental treatment has an effect greater than zero. Design A best estimate of the effect of active control, ideally from a meta-analysis, is necessary. It is not recommended to conduct an NI trial in cancers with limited evidence. Potential uncertainty in historical evidence should be examined carefully. Reporting Justification for an NI margin should be reported with reference to historical data in addition to the items in the CONSORT checklist. General It is advisable to follow guidelines from regulatory authorities.