Duke Health and Preemptive validate early detection model for hospital readmissions

Hospital readmissions remain one of the most persistent challenges in healthcare. Predicting them accurately before symptoms escalate allows for interventions that keep patients at home.

DURHAM, NC — Preemptive, in collaboration with Duke Health, today announced results from a retrospective validation study demonstrating the efficacy of continuous physiological modeling in predicting unplanned hospital readmissions.

The Study Design

The study analyzed historical data from a cohort of 200 active heart failure patients discharged from Duke University Hospital. The dataset included 28 unplanned hospital readmissions within the observation window.

Preemptive applied its proprietary signal processing models to continuous physiological data streams. The objective was to determine if the model could identify patients at risk of readmission within a 3-day (72-hour) look-ahead window.

Key Findings

• High Predictive Power: The model achieved an Area Under the Curve (AUC) of 84.7%, significantly outperforming traditional risk scores which often hover around 0.60-0.65.
• Optimized Sensitivity: At the selected setpoint, the model correctly identified 75% of at-risk patients (True Positive Rate).
• Practical Specificity: The model maintained a False Positive Rate of only 15%, striking a critical balance for real-world clinical workflow by minimizing alert fatigue.
• Earlier Intervention: Signals were detected 72 hours (3 days) prior to the readmission event, providing ample time for outpatient medication adjustment or stabilization.

"Detecting instability three days before a readmission changes the game. It moves us from 'rescue care' to true management."