Derivation and validation of a hospital all-cause 30-day readmission index

Abstract

The study derives and validates a 30-day hospital readmission risk index to predict a patient's likelihood of readmission, utilizing a health systems electronic medical record. A retrospective data extraction and analysis was conducted using data from the electronic medical record to identify risks of 30-day all-cause hospital readmission on adult patients admitted to a large multi-site health system. Univariate and multivariable logistic regression was performed on a derivation cohort of hospital admissions (n=40,668) and analyzed 91 variables associated with 30-day hospital readmission. A10-variable risk prediction equation was generated and validated in a second patient cohort (n=7,820). The prediction index's discriminative ability was determined using the c-statistic, and calibration of the prediction index was assessed with the use of the Hosmer-Lemeshow test. The hospital all-cause thirty-day readmission index (HATRIX) identified 10 variables to be highly associated with 30-day readmission. The discriminative ability of the derived prediction equation was determined using the c-statistic and was calculated to be 0.73 (95% confidence interval [CI] 0.72-0.73) for the derivation cohort. The prediction equation was validated using a second cohort of patients and resulted with an area under the curve (AUC) of 0.72 (95% CI 0.70-0.73), indicating modest discrimination. An original risk prediction index for 30-day hospital readmission was derived and validated using 2 cohorts of patients. Identifying patients who have an increased risk of 30-day hospital readmission with the use of the electronic medical record is an ideal method for targeting interventions and improving transitions-of-care to reduce hospital readmissions.