Error in using CoxTimeVaryingFitter due to NaN in base-df

[Asieh-A-Mofrad]

Discussed in https://github.com/CamDavidsonPilon/lifelines/discussions/1435

^{Originally posted by Asieh-A-Mofrad July 8, 2022} Hi,

I'm new to survival analysis and the lifelines library. I'm using Cox’s time varying proportional hazard model and trying to use the CoxTimeVaryingFitter. My data has several time-varying covariates and is in the event-based type (i.e. NaN represents if an event is not happening).

When I create the dataset based on the lifelines guideline using lifelines.utils.covariates_from_event_matrix(), The final dataset has NaN in the first period of each event. It is in line with the example in the guideline:

   start   E1  var1  stop  id  event
0    0.0  NaN   0.1   1.0   1  False
1    1.0  1.0   0.1  10.0   1   True
2    0.0  NaN   0.5  12.0   2   True

But when I use this dataset as the input to the fit, I got the error: NaNs were detected in the dataset. Try using pd.isnull to find the problematic values.

PS. If I remove the rows with NaNs, the fitter works, but in this case, I miss some data, similar to the data for id=2 in the example, and none of the time periods starts at time 0.

Thanks in advance, Asieh