Risk prediction models for hospital readmission: a systematic review

Area of Science:

• Health Services Research
• Medical Informatics
• Clinical Epidemiology

Background:

• Predicting hospital readmission risk is crucial for targeted care transition interventions.
• Accurate risk adjustment is essential for fair hospital performance comparisons.

Purpose of the Study:

• To systematically review validated readmission risk prediction models.
• To evaluate model performance and assess their suitability for clinical and administrative applications.

Main Methods:

• Searched MEDLINE, CINAHL, Cochrane Library, and EMBASE databases.
• Included studies published in English with derivation and validation cohorts for medical patients.
• Extracted data on patient populations, settings, model performance (discrimination, calibration), and data collection methods.

Main Results:

• 30 studies of 26 unique models were identified, primarily focusing on 30-day readmissions.
• Models for hospital comparison (n=14) showed poor discrimination (c-statistic 0.55-0.65).
• Models for identifying high-risk patients at discharge (n=5) demonstrated better discrimination (c-statistic 0.68-0.83).
• Functional and social variables improved model discrimination, but were underutilized.

Conclusions:

• Current readmission risk prediction models generally exhibit poor performance for both comparative and clinical purposes.
• While some models may offer utility in specific contexts, significant improvements are necessary for widespread adoption.
• Future model development should incorporate a broader range of variables, including social determinants of health.