Objective: To develop a predictive model of 30-day readmission using clinical and administrative data.* Design: Retrospective cohort study. After dividing data into developmental and validation sets, multivariable logistic regression was performed.* Participants: Adults with data in Health Facts, a database composed of participating hospitals' electronic medical records. The index hospitalization was a patient's first qualifying hospital admission between 1 October 2008 and 31 August 2010. We excluded observation stays, admissions with length of stay of 0 days, obstetric stays, and patients whose predominant care setting was a psychiatric or rehabilitation unit.* Measurements: Readmission within 30 days of live discharge from the index hospitalization.* Results: There were 463,351 index admissions to 91 hospitals, with 45,098 (9.7%) patients readmitted. In multivariable modeling, factors associated with readmission included prior hospital admission, low hemoglobin, longer stays, and increasing Charlson index; arthroplasty procedures were associated with lower risk of readmission. Model discrimination was modest in developmental data (c-statistic = 0.668) and slightly lower (0.657) in validation data.* Conclusions: Increased comorbidity and prior hospital exposure are associated with unplanned readmission. Despite the availability of many potentially relevant clinical variables, model performance was modest and few clinical variables were associated with readmission in a multivariable model. Focusing on specific conditions with a narrower set of relevant variables may facilitate identifying patients at particularly high risk for readmission.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below