The encyclopedia · R&D & Science · Operational decision · 2020–2023
Hartford HealthCare's daily ML forecasts cut stays 0.67 days, adding $55-72M
Daily discharge, ICU and mortality forecasts across seven hospitals let 200+ clinicians plan earlier, cutting length of stay and adding $55-72M a year.
Hartford HealthCare
the move
Hartford HealthCare, the largest hospital network in Connecticut, partnered with MIT and Dynamic Ideas to predict eight inpatient outcomes across its seven hospitals. Models trained on electronic medical records forecast 24- and 48-hour discharge, ICU transfer, mortality and discharge disposition with 75.7-92.5% AUC.
The predictions are delivered each morning through software with color-coded alerts and patient-level explanations. Combining 48-hour discharge forecasts with physicians' judgment produced 10-28.7% more discharges and fewer 7- and 30-day readmissions.
Since deployment, more than 200 doctors, nurses and case managers use the tool daily. Hartford HealthCare observed a 0.67-day reduction in average length of stay per patient and projects $55-72 million in annual revenue uplift for the system.
why it works
- Doctors cannot manually weigh hundreds of measurements per patient; the models do it overnight.
- Explaining each score builds the trust that makes clinicians act on a machine recommendation.
- Predicting discharge 48 hours out gives care teams time to arrange transport, rehab and home services.
- An automated pipeline updates every patient's risk each morning with zero extra staff effort.
what transfers
Predictions change operations only when staff see why and when; the tool, not the model, drives adoption.
what came after
The tool is embedded in daily patient-review rounds at all seven hospitals, with predictions refreshed every morning. Hartford HealthCare recorded the length-of-stay reduction and projected revenue gains after controlled adoption, and the same collaboration has produced more than 20 earlier analytics projects at the network.
references
- Patient Outcome Predictions Improve Operations at a Large Hospital Network
- Patient Outcome Predictions Improve Operations at a Large Hospital Network
spotted an error? The archive wants to know.