How does the cox predict work exactly?

[justkrismanohar]

HI guys,

So, from my understanding, the Cox model does not assume any form of the base link hazard function h0. So, I am wording how lifelines give us probabilities for survival. Is it assuming form for h0? I am really interested to understand how this works as I am new to survival analysis.

Thanks guys!

[CamDavidsonPilon]

You can compute the baseline hazard non-parametrically (not unlike how the Kaplan Meier curve is computed non-parametrically, but the two are not the same). Given h0 and the hazard coefficients, you have the full survival curve => predictions.

[justkrismanohar]

Interesting..... i noticed in #543 you mentioned that

In Cox PH Regression model, the baseline hazard function is the hazard function when all covariates are set to zero, right?

So, no. In lifelines, the baseline hazard is the hazard function when all covariates are set to the mean of the covariate.

So that "baseline survival" in the graph is the "average" subject (which often has no real world meaning tbh, you can't be 0.5 male). Thus, it makes sense to have prio=0 above the baseline survival.

Does that make sense?

I assume in that case the baseline will take on a form? so something like e^(Beta*X) ? But I think I am missing something and the h0 should be function of t?

[CamDavidsonPilon]

well hazard is h(t) = h0(t) * exp(something). That conversation is about what something is s.t. exp(something) is equal to 1, which provides an interpretation to h0(t). However h0(t) is still computed non-parametrically.