New feature request: Add smoothed Kaplan-Meier plot

[alonsosilvaallende]

New feature request. It would be nice to add the possibility to have a smoothed Kaplan-Meier plot instead of the steps plot currently used. I'm thinking something similar to this.

[arturomoncadatorres]

I suggested something similar in #670. Doing a proper smoothing (as suggested by O'Keese et al.,) could help confidentialising the KM curves and contribute to protecting the privacy of individual data points (which very often correspond to patients).

[CamDavidsonPilon]

hmm never thought about this (sorry about dropping both these issues). Are ya'll thinking of just smoothing the plot, or also the attributes like survival_function_?

[arturomoncadatorres]

(No problem!)

From my side and at least on the KM case, I was thinking on just smoothing the curve when plotting.

Further measures for confidentialising the curve could include:

• Suppressing the symbols on the KM curve that indicate a censor event (i.e., making sure that kmf.plot's show_censors is False)
• Add a small amount of noise to the end point of the curve to conceal the study end date

These would also be restricted exclusively to the display (i.e., plot). On top of my head, kmf.plot could have an additional boolean parameter confidential that when True, applies these three (or however) options when plotting.

[alonsosilvaallende]

In my case, just smoothing the curve would be enough.

[alonsosilvaallende]

I found a library (PySurvival) which implements Smooth Kaplan-Meier. The code can be found here.