Client: Regional hospital network (name withheld)
Industry: Healthcare
Challenge: High 30-day readmission rates straining capacity and reimbursement.
The challenge
Care teams lacked a reliable way to prioritize discharge planning. Risk was assessed manually with inconsistent criteria across wards.
Our approach
We trained an XGBoost classifier on structured EHR fields, comorbidities, and prior utilization patterns. The model output risk tiers consumed by care coordinators at discharge.
Results
Readmissions fell 24% within two quarters. Nurses reported the scores matched clinical intuition while surfacing cases they would have missed.
