Related Experiment Video
Updated: Apr 12, 2026
07:31
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
8.3K
Development and implementation of a real-time 30-day readmission predictive model
Patrick R Cronin1, Jeffrey L Greenwald2, Gwen C Crevensten2
1Laboratory of Computer Science, Massachusetts General Hospital, Boston, MA.
Summary
Hospitals can now predict patient readmissions within 30 days using a new model. This tool helps identify at-risk individuals for targeted interventions, improving patient care and reducing hospital readmissions.
Area of Science:
- Healthcare Operations
- Clinical Informatics
- Predictive Analytics
Background:
- Hospitals face significant pressure to reduce patient readmissions.
- Accurate prediction of patients at high risk for rehospitalization is crucial for implementing timely interventions.
- A functional predictive model is needed to support real-time clinical operations.
Purpose of the Study:
- To develop and validate a predictive model for 30-day hospital readmissions.
- To assess the model's performance in a real-time clinical setting.
- To determine if a functional predictive model can be implemented in a large academic hospital.
Main Methods:
- A predictive model was developed using retrospective data from 45,924 hospital admissions.
- The model incorporated factors available by the day after admission.
- Prospective validation was conducted in real-time for 3,074 admissions.
Main Results:
- The retrospective model achieved an Area Under the Curve (AUC) of 0.705 with good calibration.
- The real-time implementation showed an AUC of 0.671.
- The real-time model tended to overestimate readmission risk.
Conclusions:
- A moderately discriminative predictive model for 30-day readmissions can be developed.
- This model can be implemented in a large academic hospital setting.
- Real-time implementation requires calibration adjustments to mitigate overestimation of risk.
