Predicting on right-censored data

[jlim13]

When predicting on right-censored data, is there an option to condition on prior duration like in the lifelines package? Or does the predict function do this conditional for us. https://lifelines.readthedocs.io/en/latest/Survival%20Regression.html#prediction-on-censored-subjects

[sebp]

There's no built-in function to do that, but as the lifelines documentation states, all you need to do is divide by S(s). If you have an individual survival function as StepFunction, which predict_survival_function can give you, you just need to evaluate it at s. Once you have S(s), compute S(t) as usual and divide by S(s) from before. Hope that helps.

[jlim13]

so lets say I am predicting an individual's survival function conditioned on them being alive for s days. I would:

1. call predict_survival_function which gives me S()
2. get the value at S(s)
3. divide all of S() by S(s)?

[sebp]

Yes, predict_survival_function will give you S (functions) for each sample, from which you can compute S(t) and S(s). You can have a look at the introduction notebook to see how the code would look like.