Fig. 1: Pictorial overview for uncovering potential confounders.
From: Uncovering interpretable potential confounders in electronic medical records
a shows the data processing done for each patient. We preprocess and concatenate the structured and unstructured covariates before applying our method. For the data sources, we compile data from the Stanford Cancer Institute Research Database (SCIRDB), the California Cancer Registry (CCR), and the Epic System. We present the timeline for patient i with both structured (\({X}_{i}^{(s)}\)) and unstructured (\({X}_{i}^{(e)}\)) features Xi. b shows the workflow for identifying how potential confounders affect survival analysis for each treatment group. We uncover covariates that are predictive of both the treatment and outcome as potential confounders. We then perform survival analysis on different combinations of the selected covariates.
