Censored patients in Kaplan–Meier plots of cancer drugs: An empirical analysis of data sharing

• DOI: 10.1016/j.ejca.2020.09.031
• Publication Date: 2020-11-05T13:44:07Z
• Authors: Kate Rosen; Vinay Prasad; Emerson Y. Chen
• Keywords: Randomised trial; Oncology and Carcinogenesis; 610; Antineoplastic Agents; Kaplan-Meier Estimate; Informative censoring; 03 medical and health sciences; 0302 clinical medicine; Neoplasms; Humans; Oncology & Carcinogenesis; Cancer; Randomized Controlled Trials as Topic; Retrospective Studies; Biomedical and Clinical Sciences; Information Dissemination; Prevention; Oncology and carcinogenesis; Survival analysis; Oncology trial; 3. Good health; Good Health and Well Being; Kaplan-Meier curve; Public Health and Health Services; Kaplan–Meier curve

IntroductionKaplan–Meier survival analysis, the cornerstone of evaluating efficacy oncology drugs in randomised controlled trials (RCTs), assumes censored patients are neither healthier nor sicker than those followed. We sought to examine whether censoring patterns differ between control and experimental arms one journal that mandates reporting number censored.MethodsIn this retrospective review, proportion study design data were gathered from RCTs published The Lancet Oncology reported Kaplan–Meier curves May 2018 August 2019. Differential rates analysed at 1st, 3rd, 6th, overall time points each study. Analysis was stratified by progression-free (PFS) or (OS) end-points.ResultsOf 160 articles reviewed, 29 studies with 51 eligible. In both OS (N = 25) PFS 26), absolute weighted difference initially positive, indicating more arm (first point OS: 0.32%; PFS: 2.00%). then became negative, as progressed (end-of-study −7.54%; −9.09%).ConclusionDifferences cancer suggest there could be systematic bias present various may influence key results. Further investigation is needed, possible reasons include assignment disappointment, inappropriate follow-up length, lack efficacy, intolerable toxicity, predominant specific after randomisation.