Survival-inferred fragility index of FDA-approved cancer drugs using surrogate endpoints.

person Jonathan Ofer Davidoff Cancer Center, Rabin Medical Center–Beilinson Hospital, Petah Tikvah, Israel info_outline Jonathan Ofer, Shai Gilboa, Yarden Pras, Gal Markel, David Bomze, Tomer Meirson

Authors person Jonathan Ofer Davidoff Cancer Center, Rabin Medical Center–Beilinson Hospital, Petah Tikvah, Israel info_outline Jonathan Ofer, Shai Gilboa, Yarden Pras, Gal Markel, David Bomze, Tomer Meirson Organizations Davidoff Cancer Center, Rabin Medical Center–Beilinson Hospital, Petah Tikvah, Israel, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel, Davidoff Cancer Center, Rabin Medical Center–Beilinson Hospital, Petah Tikva, Israel, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel., Tel Aviv, Israel Abstract Disclosures Research Funding No funding received None. Background: The survival-inferred fragility index (SIFI) quantifies the robustness of positive trials by estimating the number of changes in a survival analysis that would render the results insignificant. Here, we use a variation of SIFI (SIFI-median) in which the median patient is reassigned between study groups until significance is lost. Our goal was to study the robustness of oncology randomized controlled trials (RCTs) leading to drug approval by the FDA based on surrogate time-to-event endpoints. Methods: In this cross-sectional study, we identified phase II-III RCTs between 2010 and 2020. We reconstructed individual patient data from published Kaplan-Meier plots of surrogate time-to-event endpoints and calculated the SIFI-median, i.e., the minimum number of median survivor reassignments from the intervention group to the control group that would cause the loss of statistical significance. Results: A total of 101 studies comprising 60697 patients met the inclusion criteria. The most common surrogate endpoint was PFS 86 (85.1%), followed by DFS 7 (6.9%). The SIFI-median was 23 (interquartile range [IQR], 12 to 37) and the SIFI-median as a proportion of sample size was 5.3% (IQR, 2.7% to 8.1%). Forty-five studies (44.5%) had SIFI-median less than 5% of the total sample size which was often less than the number of censored patients during early follow-up. The SIFI-median was less than 1% of the total sample size in 10 studies (9.9%). Conclusions: Low fragility indices indicates that the statistical conclusions of clinical studies could be reversed with a small number of variations. The lack of statistical robustness in multiple FDA-approved drugs suggests higher uncertainty in their clinical benefit.