Interestingly, seems that lifelines only draws the 95% errorbars and confidence intervals, did you also find no way to change this @aGuyLearning...?
If not, its something larger, and for now until someone requesting this, can keep like that

in our coxph wrapper, we don't let user set the confidence intervalls, so this is the default. can be checked after fitting in the .confidence_intervals_ dataframe of the coxphfitter object.

Could we actually redesign the interface and have this function take an AnnData object instead? We'd have to store the coxph results in the AnnData object. This would be much more in line with the rest of the ehrapy experience.

Also ping @eroell

Yes, good point. Thanks. We should do this indeed - storing the results used for the plot in adata.uns I suppose

Might also need a key_added argument then or something to ensure that we don't always overwrite results. But yeah this behavior must be consistent across all of the surv analysis methods

the kmf plot takes a list of KaplanMeierFitter Objects, so i went with that. in the survival analysis plots, ols is the only one taking an anndata.

We should be consistent about it and I think we should consider modifying all of them to take AnnData. It's what the rest of the API does

the kmf plot takes a list of KaplanMeierFitter Objects, so i went with that. in the survival analysis plots, ols is the only one taking an anndata.

Yes, noticed where the idea here came from. Having .tl writing to AnnData, and .pl to read from it should be our go to. It has not been done like that everywhere yet unfortunately, and we strive for this consistency now :)

one way to go about this is to store the required table for plotting of the lifeline's fitter object in adata.uns[key_added] upon the .tl call;

.uns seems to me the only slot suitable for this table for the univariate functions such as kaplan_meier.

For cox_ph here, as the coefficients are on a per-variable level, I think adding them into the .var column would be kinda clean; basically adding into .var the columns {key_added}_{"coef"}, {key_added}_{"coef lower 95%"}, {key_added}_{"coef upper 95%"}, .... There could also be a boolean {key_added}_{cox_ph_variable} column, which the plotting function uses to pick only the variables used in the cox ph model

Can your test not work with

@pytest.fixture
def mimic_2_10():
    mimic_2_10 = ep.dt.mimic_2()[:10]

    return mimic_2_10

?

using the smaller set we get the follwoing error:

ConvergenceError: Convergence halted due to matrix inversion problems. Suspicion is high collinearity. Please see the following tips in the lifelines documentation: https://lifelines.readthe

so i took the whole mimic_2 set

I see, makes sense. I think it's okay if we use the smaller subset and filter the warning here https://github.com/theislab/ehrapy/blob/main/pyproject.toml#L132