Univariable vs. multivariable regression in Cox’s proportional hazard model

[giemmecci]

Hi,
Thank you for developing and sharing this package.

I have a basic question (I'm new to survival analysis).

If we run the Cox's proportional hazard model in lifelines, and we don't specify a formula, are the results referring to multiple univariable regression models run on each of the variables, or a multivariable regression model?

In other words, how can we run univariable and multivariable regressions in lifelines?

My understanding is that formulas allow you to select the variable to use (among other things), so in that case, we should interpret the results as a multivariable regression model. Is this correct?

Thanks!

[CamDavidsonPilon]

Hi @giemmecci, good question

The formula is there to specify more precise transformation you may wish to do (think interaction terms, or quadratic terms). If no formula is specified, all variables in the dataframe are used in a single regression (so multivariable regression model).

[giemmecci]

Thanks so much for the quick reply!

So to wrap up, if I run this:

cph.fit(rossi, duration_col='week', event_col='arrest', formula="fin")

Will the result correspond to a univariable regression model run on the "fin" variable?

Thanks!

[CamDavidsonPilon]

Yup, that's exactly right.

[giemmecci]

Thank you!