EHR Risk Chronic disease & population health
Diabetes, heart failure, and related conditions shape U.S. health system demand. Horizon-based risk research needs index-time integrity so published metrics reflect real predictive signal—not leakage.
Open source · MIT
Leakage-aware
Calibrated · SHAP
Docker
EHR Risk Framework
An open-source framework for leakage-safe, calibrated, and explainable EHR risk prediction.
Clinical machine learning workbench for students, labs, and health-informatics research: temporal splits, leakage audits, Brier/ECE calibration, and SHAP — on longitudinal EHR-style data. Research and education only. Outputs are not clinical recommendations and are not intended for patient care. We are working toward broader general-purpose use in the future.
Free demo server — it may be slow. Check it with a small amount of data. For larger workloads or freer experimentation, run locally or on your own server. Open live demo
Why this matters for U.S. healthcare research
Chronic disease remains a major driver of morbidity, mortality, and cost in the United States. Progress in health AI depends less on another opaque notebook and more on methods that universities, academic medical centers, and public-health partners can audit, teach, and reproduce.
Trustworthy health AI methods
Leakage audits, calibrated probabilities (Brier / ECE), and SHAP explanations support the U.S. research community’s push toward reliable, transparent clinical machine learning.
Health-informatics education
A one-command Docker workbench helps U.S. students and trainees practice a full train → audit → explain → serve loop used in serious labs—not only a single AUROC printout.
Open, shareable research infrastructure
MIT-licensed code, task YAML, schema-driven APIs, and downloadable result packs lower the barrier for academic and nonprofit teams—without shipping PHI.
- Encourages equity-aware subgroup reporting when group columns are available.
- Documents limits clearly so methods papers stay scientifically honest.
- Complements—not replaces—credentialed MIMIC and institutional data workflows.
Read more: Why it matters · Limits & model card
IntegrityLeakage audit
ProbabilitiesBrier · ECE
ExplainSHAP UI + API
RunDocker · ZIP
What you get
Configurable index time and horizon. Chronic-risk demos by default; other binary outcomes (for example 30-day readmission) via task YAML.
Temporal integrity + leakage audit
Features use only data at or before index time. An audit job flags common future-leak mistakes.
Calibration
Optional isotonic calibration with Brier and ECE on hold-out evaluation.
SHAP in workbench and API
Generate SHAP from Results; show drivers on Predict with schema-aligned forms.
Docker researcher loop
docker compose up → Datasets → Train → Analytics → Predict. Named runs and a results ZIP.