Background: The dynamic effects of immune checkpoint inhibitors (ICIs) are a challenge when designing and analysing data in non-proportional hazards (PH) scenarios. Herein, we present the risk of making type II errors, affecting pharmacotherapeutic development when methods that assume constant effects are applied.
Patients and methods: Individual patient data from six clinical trials (KEYNOTE-062/061, IMvigor211, CA184-143 y CheckMate-057/037) were extracted. The most relevant time-varying effects were examined using the Royston-Parmar spline model (RPSM), time-driven analyses and weighted log-rank and Renyi tests.
Results: The RPSM yields an appropriate fit in non-PH contexts, enabling dynamic descriptions of the hazard rate, and time-varying differences of overall survival (OS)/progression-free survival. In the KEYNOTE-061, CheckMate-057 and 037 trials, 12-, 18-, and 24-month OS rates were higher with immunotherapy (differences of some 10%) (P-value <0.05). In KEYNOTE-062, CA184-043 and IMvigor-211 trials, OS rate differences were significant for past 20 months. Flemming-Harrington and Renyi tests with late weighting (e.g. with ¿-value = 0 and ¿-value = 1) captured the existence of significant differences on all curves. The Cox models and log-rank tests were inefficient at detecting the effect.
Conclusion: This analysis highlights the risk of declaring studies with ICIs negative, despite associating substantial OS benefits. Effort and consensus are needed with respect to methodology to design and evaluate trials with ICIs in non-PH settings.