PROGNOSTIC MODEL FOR PREDICTING PATIENTS AT HIGH RISK OF EMERGENT HOSPITAL READMISSION WITHIN 30 DAYS AFTER DISCHARGE USING ADMINISTRATIVE DATA

Purpose: Frequent hospital readmissions contribute significantly to rising healthcare burden. Improving the management of frequently admitted patients has been increasingly one of the important strategies for improving healthcare outcomes and reducing healthcare expenditure.  Early and accurate prediction of patients’ risk of readmission within 30 days after discharge is therefore critical. This study aimed to develop an easy-to-use prognostic model to predict patients’ risk of emergent hospital readmission within 30 days upon discharge.

Method: This was predictive modeling using administrative data.  Patients from selected inpatient disciplines discharged from a tertiary general hospital in Singapore in 2008 were used.  Information on demographics, previous hospital admission, previous visit to emergency department (ED), the clinical diagnoses and utilization of the index hospital admission were extracted from the hospital administrative database. The primary outcome of the study was emergent hospital readmission in 30 days after discharge.  Two models were developed for medical and surgical patients separately. The models were developed using randomly chosen 60% samples and validated using the remaining 40% samples. Random effect logistic regression was applied to identify important predictors and estimating the associated regression coefficients. Significance level of 0.05 and the Schwarz’s Bayesian Infor­ma­tion Cri­te­rion (SBIC) were used to select the best-fit model.  C-statistics of the receiver operation characteristic (ROC) plot was applied to assess the discrimination power of the model.  Hosmer-Lemeshow test was applied to test the goodness-of-fit of the prognostic model. All statistical analyses were conducted using SAS 9.2 (SAS Institute, Cary, NC).

Result: Five factors were identified as significant predictors of emergent hospital readmission within 30 days after discharge;  i.e. previous hospital admission in preceding year, age, hospital length of stay, number of secondary diagnoses, and admission type (emergent or elective). The c-statistics of the ROC plot of the prognostic model for medical patients was 0.83 (95%CI: 0.82-0.83), while it was 0.81 (95%CI: 0.80-0.82) for surgical patients. The goodness-of-fit tests for the two models were not significant.

Conclusion: A simple prognostic model for predicting patients at high risk of emergent hospital readmission in 30 days has been developed and internally validated, which has good discrimination power and goodness of fit.