Survival time

#1
Hey all,

I'm trying to use different ML models for Survival analysis of surgical patients after surgery.
I have 7500 patients with different times of followups (from 1 year to 10 years) in days.
I want to focus on a certain period of time - let's say 1 year from procedure.
I have a status at 1 year variable (0/1).
Should i use the time variable as is (0-5000 days), or create a new time variable which will be up to 365 days only, and whoever's over 365, to change it 365.
Thus, use the dead at 1 year as status, and the new time365 variabale for time.

What's your advice?

thank you
 

staassis

Active Member
#2
Both approaches are fine but keeping the data as they are (not artificially truncated) may present you with further insights regarding the goodness of fit of various modeling frameworks. Oftentimes, when you are utilizing a particular (semi-)parametric method of survival analysis (like Cox proportional hazard rate model), if the model does a good job about the right tail it does a good job in the middle.
 
Last edited: