search

autonlab / auton-survival

Auton Survival - an open source package for Regression, Counterfactual Estimation, Evaluation and Phenotyping with Censored Time-to-Events
http://autonlab.github.io/auton-survival
MIT License
315 stars 74 forks source link

what is "uncensored time control" and "uncensored time treated" #99

Closed zapaishchykova closed 1 year ago

zapaishchykova commented 1 year ago

Hi there! Thanks for the great repo! I have a q: what are "uncensored time control" and "uncensored time treated" in the CMHE notebook example in the outcomes df and how to derive them? Thanks! https://nbviewer.org/github/autonlab/auton-survival/blob/master/examples/Demo%20of%20CMHE%20on%20Synthetic%20Data.ipynb

chiragnagpal commented 1 year ago

The data is generated synthetically. So we have access to the true outcome as well as the counterfactual outcome.

Uncensored Time Control refers to the True Outcome if there was no censoring and no intervention was performed.

Uncensored Time Treated refers to the True Outcome if there was no censoring and an intervention was performed.

zapaishchykova commented 1 year ago

How about a real-world dataset, is there a way to compute these two variables?

chiragnagpal commented 1 year ago

Good question. In the case of a real world dataset you would only observe the censored outcomes under either treatment assignment or control. For our KDD 22 paper, we aim to resolve this by training separate regression models on the treated and control populations to estimate the factual and counterfactual outcomes. We then use these models to estimate subgroup level treatment effects.

auton-survival has APIs to train counterfactual models in auton-survival.estimators

https://autonlab.org/auton-survival/estimators.html#auton_survival.estimators.CounterfactualSurvivalModel