Accuracy of Prospective Predictions of 30-Day Hospital Readmission

Abstract OBJECTIVE: This study examined how accurately inpatient case managers predicted 30-dayreadmission and whether objective patient characteristics improved prediction accuracy. METHODS: In this prospective study, inpatient case managers at a psychiatric hospital rated their concern (1, not concerned; 5, very concerned) about readmission after discharge of 282 privately insured patients. Sensitivity and specificity of the ratings were calculated. Logistic regression identified whether patient characteristics that could affect 30-day readmissionimproved prediction accuracy. RESULTS: Concern levels ≥3 yielded 86% sensitivity, 37% specificity, and a positive predictive value (PPV) of 13%; levels ≥4 yielded 39% sensitivity, 78% specificity, and a PPV of 17%. Concern level independently predicted readmission; appointments within seven days postdischarge further improved model accuracy (p=.03) (area under the curve=.67, 95% confidence interval=.58-.78). CONCLUSIONS: Although not highly accurate, case manager concern identified some patients at higher risk of 30-day readmission. Appointments within seven days of discharge improved prediction accuracy.