A novel, more reliable approach to use of progression-free survival as a predictor of gain in overall survival: The Ottawa PFS Predictive Model

• Published in: Critical reviews in oncology/hematology Vol. 148; p. 102896
• Main Authors: Stewart, David J., Bossé, Dominick, Goss, Glenwood, Hilton, John F., Jonker, Derek, Fung-Kee-Fung, Michael
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01-04-2020
• Subjects: Antineoplastic Agents - therapeutic use; Disease Progression; Disease-Free Survival; Half-lives; Hazard ratios; Humans; medians; Medical Oncology; Neoplasm Recurrence, Local - mortality; Neoplasms - drug therapy; Neoplasms - mortality; Neoplasms - pathology; Population kinetics; Prediction of overall survival; Progression-Free Survival; Proportional Hazards Models; Randomized Controlled Trials as Topic; Survival Analysis; Half-lives; Hazard ratios; Population kinetics; medians; Progression-free survival; Prediction of overall survival
• ISSN: 1040-8428; 1879-0461
• DOI: 10.1016/j.critrevonc.2020.102896

Progression-free survival (PFS) hazard ratios and gain in median PFS are suggested predictors of overall survival (OS) gain (with gain defined as experimental arm minus control arm values). We assessed use of half-lives (time to progression/death of half remaining patients). We reviewed randomized trials from Journal of Clinical Oncology and New England Journal of Medicine, 01/2012-06/12/2017 (discovery series) and 01/01/2007-12/31/2011 (first validation series). If PFS or OS gains were significant, we used PFS/OS curve nonlinear regression analysis to estimate half-lives and defined “half-life gain” as experimental minus control arm half-life. With low crossover and significant PFS differences, PFS half-life gains ≥1.5 months had positive-predictive-values for OS gains ≥2 months of 79 % and 86 % and PFS half-life gains <1.5 months had negative-predictive-values for OS gains <2 months of 95 % and 75 %, in discovery and validation series, respectively. PFS half-life gains more reliably predicted OS gains than PFS hazard ratios or gains in median PFS. Findings were confirmed in a second validation series.