FHIR-native ML model predicting 30-day readmission risk at discharge
by HealthData LabsOur Hospital Readmission Risk Predictor uses an ensemble of gradient boosting and transformer models trained on 18.4 million inpatient discharges across 340 US hospitals. The model ingests FHIR R4 resources including Encounter, Condition, MedicationRequest, Procedure, Observation, and Patient demographics to generate a 0–100 risk score at point of discharge. It outperforms LACE+ by 18% AUROC on external validation. The model includes fairness evaluations across race, ethnicity, and insurance status and is deployed as a CDS Hooks service compatible with Epic, Cerner, and athenahealth.
0.84
Model AUROC
+18%
vs LACE+
18.4M
Training Discharges
< 200ms
Prediction Latency
< 2%
Bias Disparity (race)
340
Hospitals Validated
HIPAA BAA included with all enterprise plans