Choosing statistical tests for survival analysis

Abstract

In survival analysis, researchers are not interested in a disease per se, its symptoms, diagnostics, treatment or outcomes are not their main concern either. The time, however, the time lapsed to the outcome of a disease, is the main focus of the survival analysis studies. Nevertheless, not for all subjects researchers might observe the event due to various reasons. The subjects might be censored from the study at different time periods: at the end of the course if the event was not observed at all; within the course if the subject was lost to follow-up, or enrolled erroneously. The censoring makes the survival data unfeasible to be analysed with standard non-parametric tests. Kaplan-Meier estimate handles the censored data well, providing in addition to the test results, the survival probabilities and survival curves. On the other hand, Kaplan-Meier estimate does not give us the information on the significance of the difference in the survival of two groups but a few statistical tests specifically used in survival analysis do. The choice of a test is always challenging since there is a fine line between the tests, and the one should have enough expertise and knowledge of the data in hand to be able to identify the assumptions of what test are addressed by the survival data more.