|
Abstract:
|
In personalized treatment regimen, recommendation of tailored preventative interventions to patients in clinical practice requires to estimate the heterogeneous treatment effect. The growing data in modern electronic health records (EHRs) provide the chance to go beyond the information guided by the average treatment effect estimation. A rich source of observational data provided by EHRs offers the plausibility for the unconfoundedness assumption in causal inference, but it also comes along with the data complexity including the curse of dimensionality, high order of interaction, heterogeneity as well as outliers, which makes the least square type estimation problematic. Therefore, we propose robust machine learning estimation procedures of heterogeneous treatment effect. According to our previous work of general learner for heterogeneous treatment effect estimation that including inverse propensity score weighting, doubly robust, R-learning, modified covariates methods etc., we show how to do estimation using ML algorithm instead of regression. This could improve the flexibility and speed of regression based methods.
|
