Pitfalls in Time-to-Event Analysis of Registry Data: A Tutorial based on Simulated and Real Cases

Abstract

ABSTRACTSurvival analysis (also referred to as time-to-event analysis) is the study of the time elapsed from a starting date to some event of interest. In practice, these analyses can be challenging and, if methodological errors are to be avoided, require the application of appropriate techniques. By using simulations and real-life data based on the French national registry of patients with primary immunodeficiencies (CEREDIH), we sought to highlight the basic elements that need to be handled correctly when performing the initial steps in a survival analysis. We focused on right censoring, left truncation, competing risks, and recurrent events. Our simulations show that ignoring these aspects induces a bias in the results; we then explain how to analyze the data correctly in these situations. Rare disease registries are extremely valuable in medical research. We discuss the application of appropriate methods for the analysis of time-to-event from the CEREDIH registry. The objective of this tutorial article is to provide clinicians and healthcare professionals with better knowledge of the issues facing them when analyzing time-to-event data.Key messages boxWhen comparing naïve approaches and the proper methodology, we show that:Not considering right censoring leads to underestimation of survivalNot considering left truncation leads to overestimation of survivalTreating competing risks as right-censoring leads to overestimation of survivalAppropriate recurrent event methods allow to study all events for each patient and not only account for the first event.