Analysis code for Digital Biomarkers of Patient-Reported Symptom Burden in Pancreatic Surgery Patients

Installation instructions are in RAPIDS beta installation

Descriptions of features are in RAPIDS beta features

Check the end of the config.yaml file (ANALYSIS section) for all analysis parameters.

Check the rules/models.smk for all the rules and scripts involved in the analysis.

Check the tools/create_figure1_for_paper.py script for Figure 1.png of the paper. The script contains two steps: 1) rename top 20 important features with the readable names; 2) create density scatter plot which shows SHapley Additive exPlanation (SHAP) values for each feature, reflecting how much impact each feature has on model output.

Reproducing results

1. Extract features and train models

Run snakemake -j6. Results are saved in the data/processed/output_population_model/20hours_10bins/0.3|0.3_5_True folder.

2. Create Figure 1.png of the paper

Run python ./tools/create_figure1_for_paper.py

Notes

By default, the code is for next-day total symptom burden prediction (binary classification). You can change the TARGET_COLS parameter in config.yaml to predict next-day diarrhea or fatigue or pain symptom class.