search

CamDavidsonPilon / lifelines

Survival analysis in Python
lifelines.readthedocs.org
MIT License
2.37k stars 560 forks source link

IPTW adjusted KM plotting #1147

Open JUNGYO opened 4 years ago

JUNGYO commented 4 years ago

Combined with Zepid package, I want to plot IPTW adjusted KM curve and calculate adjusted log-rank test. Does anyone have examples of this?

pzivich commented 4 years ago

I currently don't have the weighted-KM supported within zepid (but it looks like I should move up on my list). You can do it using the underlying weight estimation and inputting it into lifelines' KaplanMeierFitter.

This all assumes that treatment A is assigned as baseline (and it not a time-varying treatment). The outcome indicator is Y with confounder L and time of event T.

For the zepid side:

from zepid.causal import IPTW

ipw = IPTW(data, exposure="A", outcome="Y")
ipw.treatment_model("L + L_sq")  # estimating IPTW
data['iptw'] = ipw.iptw  # extracting IPTW

On the lifelines side:

from lifelines import KaplanMeierFitter

kmf = KaplanMeierFitter(label="ip_weighted_km")
kmf.fit(data['T'], data['Y'], weights=data['iptw'])

For confidence intervals you would need to use a bootstrapping procedure. I haven't thought about how a IPW-log-rank test would work, so I will let others weigh in