Build a 30-Day Readmission Risk Model on De-Identified EHR Data
Overview
What this challenge is about.
You receive a curated MIMIC-style de-identified EHR cohort (about 28,000 admissions, demographics, comorbidities, labs, prior-admission counts) with 30-day readmission labels. Train a gradient-boosting baseline plus a sequence model (a simple Transformer over coded events). Calibrate both. Audit performance across two protected attributes (sex, age band) with FPR/TPR gaps. Build a model card (Mitchell et al. style). Deliver a 4-page memo for clinical leadership that recommends one model and discusses the equity trade-offs.
The Brief
What you'll do, and what you'll demonstrate.
Build and audit a 30-day readmission risk model on de-identified EHR data with calibration and fairness reported alongside discrimination.
Earning criteria — what you'll demonstrate
- Apply ML to a real EHR-derived clinical-risk prediction problem
- Calibrate clinical-grade classifiers and report ECE alongside AUROC
- Audit subgroup performance gaps and reason about clinical equity
- Produce a model card that survives clinical review
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Applied AI Scientist
Clinical-grade risk models with calibration + fairness audits are the applied-AI-scientist's signature work at any payer-facing healthtech startup.
This challenge sharpens
- risk-stratification
- model-calibration
- fairness-metrics
ML Researcher
Comparing a tree ensemble to a sequence model on EHR data with rigorous reporting is the kind of focused study clinical-ML hiring loops grade.
This challenge sharpens
- ehr-modeling
- transformer
- gradient-boosting
AI Safety Researcher
Subgroup fairness audits and model cards on clinical models are exactly the AI-safety-researcher's contribution to any healthtech product.
This challenge sharpens
- fairness-metrics
- model-calibration
- risk-stratification