Associate Director AstraZeneca NORTH WALTHAM, Massachusetts, United States
Background: An important assumption in the analysis of survival data is that censoring is independent to the likelihood of developing the event of interest. However, this assumption does not hold if the censored patients are at a different risk for treatment failure than those who remain on study leading to biased survival estimation [1]. We studied the cause and the direction of this bias in publicly available datasets and a survival data simulator. Methods: Using NCCTG Lung Cancer Data [2], the cox PH analysis was performed to evaluate the effect of covariates on the overall survival probability. Then the censoring pattern was studied in subgroups of patients with higher and lower risk. In addition, time-to-event data with censoring rate, hazard function and type of censoring distribution as inputs was simulated using a code developed in R. Results: In this analysis, sex and ECOG performance score have significant effect on survival probability. However, females and ECOG=0 participants, with lower hazard ratio, are more likely to be censored with overall survival as endpoint (censored patients were of 59% females and 41% had ECOG=0 vs. 9.5% had ECOG=2). In this case, the Kaplan-Meier method may underestimate the overall survival. Our analysis shows that if the hazard is constant through the study time and distributed uniformly through the study period, the median survival is not affected by censoring. Conclusion: Survival rates is not biased if censoring in survival analysis is truly non-informative [3]. Failure to test the accuracy of this assumption, could lead to biased estimation of the endpoint of interest.
1) Prinja, S. et al. Censoring in Clinical Trials Review of Survival Analysis Techniques, IJCM. 35,217-221 (2010) 2) Loprinzi, C. L. et al. Prospective evaluation of prognostic variables from patient-completed questionnaires. North Central Cancer Treatment Group. Journal of Clinical Oncology. 12, 601-7 (1994). 3) Templeton et al. Informative censoring — a neglected cause of bias in oncology trials, Nat Rev Clin Oncol; 17, pages327–328 (2020)
