Identifying Potentially Avoidable Readmissions: A Medication-Based 15-Day Readmission Risk Stratification Algorithm

BACKGROUND Stratifying patients according to 15-day readmission risk would be useful in identifying those who may benefit from targeted interventions during and/or following hospital discharge that are designed to reduce the likelihood of readmission. METHODS A prediction model was derived via a case-control analysis of patients discharged from a tertiary hospital in Singapore using multivariate logistic regression. The model was validated in two independent external cohorts separated temporally and geographically. Model discrimination was assessed using the C-statistic, while calibration was assessed using the Hosmer-Lemeshow chi(2) and the Brier score statistics. RESULTS A total of 1291 patients were included with 670, 101, and 520 patients in the derivation, temporal, and geographical validation cohorts, respectively. Age (odds ratio [OR] 1.02, 95% confidence interval [CI] 1.01-1.03, p=0.008), anemia (OR 2.08, 95% CI 1.15-8.05, p=0.015), malignancy (OR 3.37, 95% CI 1.16-9.80, p=0.026), peptic ulcer disease (OR 3.05, 95% CI 1.12-8.26, p=0.029), chronic obstructive pulmonary disease (OR 3.16, 95% CI 1.24-8.05, p=0.016), number of discharge medications (OR 1.06, 95% CI 1.01-1.12, p=0.026), discharge to nursing homes (OR 3.57, 95% CI 1.57-8.34, p=0.003), and premature discharge against medical advice (OR 5.05, 95% CI 1.20-21.23, p=0.027) were independent predictors of 15-day readmission risk. The model demonstrated reasonable discrimination on the temporal and geographical validation cohorts with a C-statistic of 0.65 and 0.64, respectively. Model miscalibration was observed in both validation cohorts. CONCLUSION A 15-day readmission risk prediction model is proposed and externally validated. The model facilitates the targeting of interventions for patients who are at high risk of an early readmission.