I see a lot of kaplan-meier plots for survival analyses, and many distinguish between groups (usually male or female). I was wondering if it was (statistically) possible to distinguish between more groups (for example: male, female, child, adult), or if that violated some assumption?
I have seen Aalen-Johansen estimates / plots that seem to include more groups, but my understanding is that is for when there are multiple (competing) outcomes rather than different starting groups under consideration.
Best Answer
Kaplan-Meier curves certainly can accommodate more that 2 groups, as can the log-rank and other non-parametric tests of differences among groups. See, for example, Section 7.3 of Klein and Moeschberger.
Once you start with multiple groups and combinations of groups, the plots can become unwieldy. Also, Kaplan-Meier curves don't correct for other covariates that might affect survival. If the data can meet the requirements of a Cox proportional hazards model, then you can use hazard ratios to distinguish among groups more compactly while correcting for covariates, and display illustrative modeled survival curves.