Predicting 30-Day Readmissions With Preadmission Electronic Health Record Data

Abstract

BACKGROUND: Readmission prevention should begin as early as possible during the index admission. Early identification may help target patients for within-hospital readmission prevention interventions. OBJECTIVES: To develop and validate a 30-day readmission prediction model using data from electronic health records available before the index admission. RESEARCH DESIGN: Retrospective cohort study of admissions between January 1 and March 31, 2010. SUBJECTS: Adult enrollees of Clalit Health Services, an integrated delivery system, admitted to an internal medicine ward in any hospital in Israel. MEASURES: All-cause 30-day emergency readmissions. A prediction score based on before admission electronic health record and administrative data (the Preadmission Readmission Detection Model-PREADM) was developed using a preprocessing variable selection step with decision trees and neural network algorithms. Admissions with a recent prior hospitalization were excluded and automatically flagged as "high-risk." Selected variables were entered into multivariable logistic regression, with a derivation (two-thirds) and a validation cohort (one-third). RESULTS: The derivation dataset comprised 17,334 admissions, of which 2913 (16.8%) resulted in a 30-day readmission. The PREADM includes 11 variables: chronic conditions, prior health services use, body mass index, and geographical location. The c-statistic was 0.70 in the derivation set and of 0.69 in the validation set. Adding length of stay did not change the discriminatory power of the model. CONCLUSIONS: The PREADM is designed for use by health plans for early high-risk case identification, presenting discriminatory power better than or similar to that of previously reported models, most of which include data available only upon discharge.