Comparing the outcomes of noninferiority (NI) and superiority phase III trials.

Abstract

e13055 Background: We compared the outcomes of NI and superiority trials on advanced breast cancer (BC), non-small-cell lung cancer (NSCLC), and colorectal cancer (CRC). Methods: We searched PubMed for phase III trials on systemic antineoplastic treatments for advanced BC, NSCLC and CRC published between 1/1998 and 12/2009 in 11 leading journals. We categorized primary endpoints (PEP) as time-to-event (overall survival or any variant of progression-free survival), response rate, or other (quality of life or toxicity). We used the PEP (defined as the one stated explicitly, used for N calculation, or cited first) to ascertain trial positivity. Results: We retrieved a total of 262 trials (93 on BC, 102 on NSCLC, and 67 on CRC), 36 of which (13.7%) used a NI design (12 in each tumor type). There was no significant trend in the proportion of NI trials in the two 6-year periods compared (1998-2003 vs 2004-9). The median number of patients/arm for NI and superiority trials were 284 and 164, respectively (p<0.001). There was no significant difference in the distribution of the PEP categories between NI and superiority trials. We could ascertain trial positivity in all but six trials: 24 (66.7%, 95% confidence interval [CI], 49.1% to 81.2%) NI trials were positive, compared with 89 (39.4%, 95% CI, 33.0% to 46.1%) superiority trials (p<0.001). NI trial positivity could be determined by finding a CI for the estimated treatment effect that excluded the NI margin in 15 of 27 trials (otherwise, positivity was based on authors’ conclusions, or the experimental therapy was superior to control for the PEP). The overall rates of trial positivity varied across tumor types: 48.4% for BC, 31.4% for NSCLC, and 53.7% for CRC (p=0.002). When adjusted for trial size, NI design (vs. superiority; odds ratio [OR]=4.2; 95% CI, 1.7 to 10.3) and tumor type (BC [OR=2.2; 95% CI, 1.2 to 4.0] and CRC [OR=2.8; 95% CI, 1.4 to 5.5] vs. NSCLC as reference) remained significantly associated with trial positivity. Conclusions: NI trials are more likely than superiority trials to yield positive results. The influence of NI margin width on trial results should be investigated.