No abstract

No abstract

Background To decrease health care costs and burden, reducing preventable readmissions is essential. A 2017 Singapore‐developed prediction model for readmission 15 days postdischarge was developed and validated. However, it has yet to be validated independently, and its predictability of 30‐day readmission is unknown. The model utilizes variables including age, anemia, malignancy, peptic ulcer disease, chronic obstructive pulmonary disease, number of discharge medications, discharge destination, and premature discharge against medical advice to calculate readmission risk. This study aimed to evaluate the model's performance in predicting readmissions 15 and 30 days postdischarge. Methods This single‐center, independent, prospective cohort study involved Singapore's Alexandra Hospital inpatients from September to November 2017. Readmission risk was calculated utilizing the developed model and information from electronic medical records. Primary and secondary outcomes were the model's performance for 15‐ and 30‐day readmission rates, respectively. Results One hundred thirteen and 112 patients were analyzed for primary and secondary outcomes, respectively. The model performed reproducibly against the original study derivation and validation cohorts. C‐statistics were 0.64 and 0.65, and Hosmer‐Lemeshow χ2’s were 7.28 and 6.85, respectively, for 15‐ and 30‐day readmissions. The Brier score was 0.26 for both end points. The low positive predictive value of 14% to 18% and correctly classified prediction rate of 53% to 54% could possibly be overcome by making the model more stringent in predicting readmissions. This may enhance efficiency in allocation of resources to assist patients with a higher likelihood of readmission. Conclusion This model's performance was reproducible in comparison to the original study's derivation and validation cohorts. Given the multifaceted nature of readmissions, it is challenging to accurately predict readmission. Along with clinical judgment, this model can potentially be a useful tool to health care providers for discharge planning purposes. Further research into optimizing the cut‐off probability may be useful to determine model utility in actual clinical practice.