Noninferiority (NI) phase III trials in oncology: What is the basis for choosing the NI margin?

Abstract

e13070 Background: The choice of the NI margin has become an important aspect of NI cancer trials, but the contemporary practice regarding this choice has not been formally assessed. Methods: We searched PubMed for reports on phase III trials on systemic antineoplastic treatments for advanced breast cancer (BC) or non-small-cell lung cancer (NSCLC) published between 1/1998 and 12/2009 in 11 leading medical journals (Ann Oncol, BCRT, BJC, Cancer, CCR, EJC, JCO, JNCI, Lancet, Lancet Oncol, and NEJM). Results: We retrieved 197 reports (93 on BC, 104 on NSCLC), 23 of which (12%) had a NI primary hypothesis (12 in BC, 11 in NSCLC). There was no trend in the proportion of trials with a NI hypothesis in the two 6-year periods compared. The median (range) sample sizes for NI and superiority trials were 563 (160-1669) and 368 (97-1692) patients, respectively (P=0.011). The primary end point for NI trials was overall survival (OS)/progression-free survival/response in 9/5/9 cases, respectively. The NI margin was reported in all trials, but justified on statistical grounds in only 3 (13%). Among trials with time-dependent end points (N=14), 11 used the hazard ratio to represent the NI margin (3 used 1-year OS), and the lower limit of the confidence interval ranged from 0.75 to 0.86, with mean and median of 0.8. For trials with a response end point, the NI margin ranged from 15% to 25%, with mean of 18% and median of 15%. Conclusions: NI trials represent 13% and 11% of contemporary phase III trials on BC and NSCLC, respectively, with no apparent recent temporal trend in their use. Most trials use similar NI margins but report no justification for their choice.