Tutorials¶
Step-by-step Jupyter notebooks covering the full OneEHR workflow.
| # | Tutorial | Description |
|---|---|---|
| 1 | Quickstart | End-to-end mortality prediction on the TJH COVID-19 dataset |
| 2 | Custom Dataset | Bring your own data into OneEHR's three-table format + medical code mapping |
| 3 | Model Comparison | Compare ML vs DL models with bootstrap CI and DeLong/McNemar statistical tests |
| 4 | Fairness & Explainability | Bias detection across demographic groups + feature importance visualization |
| 5 | Survival Analysis | DeepSurv, DeepHit, concordance index, and Kaplan-Meier curves |
Running Tutorials¶
Local¶
pip install oneehr jupyter
jupyter notebook tutorials/Google Colab¶
Upload any notebook to Google Colab and add this cell at the top:
!pip install oneehrWhat You'll Learn¶
Tutorial 1: Quickstart¶
- Define an experiment with a TOML config file
- Preprocess EHR events into binned features
- Train XGBoost (ML) and GRU (DL) models
- Evaluate on a held-out test set
- Run cross-system comparison
Tutorial 2: Custom Dataset¶
- Prepare dynamic, static, and label CSV files
- Use built-in dataset converters (MIMIC-III/IV, eICU)
- Apply medical code ontologies (ICD-9/10, CCS, ATC)
- Aggregate codes by ontology group for dimensionality reduction
Tutorial 3: Model Comparison¶
- Configure multiple models in one experiment
- Bootstrap confidence intervals for AUROC, AUPRC, F1
- Pairwise statistical tests (DeLong, McNemar)
- Include LLM agents in the comparison
Tutorial 4: Fairness & Explainability¶
- Detect bias across age, sex, ethnicity groups
- Compute demographic parity, equalized odds, predictive parity
- Feature importance: permutation, SHAP, attention weights
- Visualize attributions with waterfall and heatmap plots
Tutorial 5: Survival Analysis¶
- Build DeepSurv and DeepHit models
- Train with Cox partial likelihood loss
- Evaluate with concordance index
- Plot stratified Kaplan-Meier survival curves