som-shahlab / temp_ds_shift_robustness

Model Robustness to Temporal Distribution Shift

This repository contains the code to reproduce "EHR Foundation Models Improve Model Robustness in the Presence of Temporal Distribution Shift"

Repo Structure

• prediction_utils:
  • Library that defines cohort definition, feature extraction, and model evaluation pipelines.
• ehr_ml:
  • CLMBR (see paper and original repo) library used in this study.
• clmbr:
  • This codebase calls prediction_utils and ehr_ml to run the experiments in the paper "EHR Foundation Models Improve Model Robustness in the Presence of Temporal Distribution Shift"
• dg:
  • This codebase calls prediction_utils to run experiments that evaluate domain generalization on improving model robustness under temporal distribution shift in clinical medicine.