A systematic review of 30 studies of 26 unique readmission risk prediction models for hospitalized patients found that most models have poor predictive ability. The most common outcome was 30-day readmission, with only one model specifically addressing preventable readmissions. Fourteen models based on retrospective administrative data could be used for hospital comparison, but nine had poor discriminative ability (c-statistics 0.55–0.65). Seven models could identify high-risk patients for early intervention (c-statistics 0.56–0.72), and five could be used at discharge (c-statistics 0.68–0.83). Models incorporating real-time data showed better performance, with one model (c-statistic 0.72) identifying high-risk heart failure patients. Nine models using primary data (e.g., surveys, chart reviews) had moderate performance, with one model (c-statistic 0.77) combining administrative and functional data to predict readmissions and nursing home transfers. Most models lacked variables related to overall health, function, illness severity, and social determinants of health. Only one model explicitly addressed potentially preventable readmissions, finding about 25% of readmissions were clearly preventable. Despite some models showing better performance, most had poor predictive ability, and few incorporated variables related to social and functional factors. The review highlights the need for improved models to accurately predict readmission risk, especially for clinical use. While some models may be useful in specific settings, their overall performance is limited, and efforts to improve their accuracy are needed. The findings suggest that readmission risk prediction remains a complex and underdeveloped area, with current models lacking the ability to effectively standardize risk or guide clinical interventions. The review also notes that readmission rates may not fully reflect the quality of care, as preventability of readmissions remains unclear. Overall, the study concludes that most current models are inadequate for clinical or administrative use and that further research is needed to develop more accurate and applicable risk prediction models.A systematic review of 30 studies of 26 unique readmission risk prediction models for hospitalized patients found that most models have poor predictive ability. The most common outcome was 30-day readmission, with only one model specifically addressing preventable readmissions. Fourteen models based on retrospective administrative data could be used for hospital comparison, but nine had poor discriminative ability (c-statistics 0.55–0.65). Seven models could identify high-risk patients for early intervention (c-statistics 0.56–0.72), and five could be used at discharge (c-statistics 0.68–0.83). Models incorporating real-time data showed better performance, with one model (c-statistic 0.72) identifying high-risk heart failure patients. Nine models using primary data (e.g., surveys, chart reviews) had moderate performance, with one model (c-statistic 0.77) combining administrative and functional data to predict readmissions and nursing home transfers. Most models lacked variables related to overall health, function, illness severity, and social determinants of health. Only one model explicitly addressed potentially preventable readmissions, finding about 25% of readmissions were clearly preventable. Despite some models showing better performance, most had poor predictive ability, and few incorporated variables related to social and functional factors. The review highlights the need for improved models to accurately predict readmission risk, especially for clinical use. While some models may be useful in specific settings, their overall performance is limited, and efforts to improve their accuracy are needed. The findings suggest that readmission risk prediction remains a complex and underdeveloped area, with current models lacking the ability to effectively standardize risk or guide clinical interventions. The review also notes that readmission rates may not fully reflect the quality of care, as preventability of readmissions remains unclear. Overall, the study concludes that most current models are inadequate for clinical or administrative use and that further research is needed to develop more accurate and applicable risk prediction models.