1

Subjects are monitored until one of two possible events occur -- say, after a time, they succeed at a task or fail at some point. I have the time to event, the type of event, and some covariate information. There is no censoring. Observation ends when one of the two events occurs.

Should this data be analysed as a competing risk model? Or should I treat it more like a classification problem?

Or to put this question differently, I know that Cox Regression and survival analysis are used when there is censoring in a longitudinal study. But in the absence of censoring, is survival analysis still the best way to approach competing outcomes.

kjetil b halvorsen
  • 77,844
Placidia
  • 14,361
  • 1
    When follow up time is variable, then cox regression is appropriate. If, on the other hand, you are assessing something like risk of the outcome at 2 years then I think a logistic regression may be sufficient (but someone correct me if I am wrong on that). The interpretation of the results also changes between the two. – Demetri Pananos Feb 11 '20 at 17:18
  • Logistic regression is not a good method for estimating Pr(outcome by 2y) because (1) it requires there to be no censoring before 2y and (2) it is very inefficient because so much information was discarded in its construction. It ignores close calls, treating failure at 1.99y and failure at 2.01y as drastically different. And there is nothing wrong with doing survival analysis with no censoring. – Frank Harrell Sep 15 '23 at 12:03

0 Answers0