Background: Determining the non-inferiority margin is an essential step in the design and interpretation of non-inferiority trials, and this margin should be preferably justified on clinical and statistical grounds.
Methods: After a PubMed search for phase III trials in advanced breast cancer (BC) or non-small cell lung cancer (NSCLC) published between January 1998 and December 2009 in 11 leading journals, non-inferiority trials were selected by manual search of the full papers.
Results: Twenty-four of 195 trials had a primary non-inferiority hypothesis. When the two six-year study periods were compared, there were time trends within BC and NSCLC, with most non-inferiority trials in BC reported in the first six-year period, and vice-versa for NSCLC. The median sample size was larger for non-inferiority than superiority trials (p < 0.01). The choice of a non-inferiority margin was reportedly justified in only five cases. Non-inferiority trials were more likely than superiority trials to yield positive results (p < 0.001), as were trials in breast cancer (p = 0.02).
Conclusions: Non-inferiority margins for cancer trials appear to be chosen mostly on historical grounds. Since nearly three-quarters of non-inferiority trials achieve their primary objective, the extent to which the choice of margins has influence on trial results remains to be determined.