Can Patient-Reported Outcome Measures Help Predict Unplanned Hospital Readmission?

Abstract

Background: Administrative data used to predict unplanned hospital readmissions often lack patient-reported symptoms and functional status. Integrating patient-reported outcome measures (PROMs) may improve risk prediction.

Objectives: To assess the incremental value of PROMs in predicting unplanned readmissions to inform postdischarge monitoring and ongoing care management.

Methods: This population-based retrospective cohort study used linked administrative and PROMs data from British Columbia, Canada. Adults discharged from acute care who provided response to the EQ-5D-5L and Veterans RAND 12-Item Health Survey (VR-12) within 60 days were included. Aggregated Cox proportional hazards models were fitted to estimate unplanned readmission risk across 30-, 180-, and 360-day horizons. The primary prediction horizons were 30 and 180 days. The 360-day horizon was a secondary focus. Model performance was assessed using the concordance statistics and calibration, with subgroup analysis for Ambulatory Care Sensitive Conditions (ACSC).

Results: Among 11,177 individuals, observed unplanned readmission rates within 30, 180, and 360 days of discharge were 5.6%, 18.4%, and 25.0%, respectively. Conditional on surviving to weekly landmarks (23-60 days postdischarge), PROMs modestly improved discrimination. For the 180-day horizon following landmarks, the C-index was 0.762 (95% CI, 0.761-0.763) using predictors from administrative data alone, increasing to 0.774 (95% CI, 0.773-0.774) with EQ-5D-5L and 0.782 (95% CI, 0.781-0.783) with VR-12. Similar gains in discrimination were observed at 30-day and 360-day horizons. All models showed adequate calibration. Among patients with ACSCs, including PROMs improved discrimination by 2.4%-3.0%.

Conclusions: PROMs added predictive value for unplanned hospital readmissions, particularly among patients with ACSCs.