Abstract
Background/Aims Identifying and preventing 30-day readmissions has become a national focus. There have been attempts at developing all-cause predictive models to identify ‘high risk’ readmission patients such as LACE. Historically, in-hospital severity predictive models have been developed for various services and disease states. Some of the most widely used severity measures are the Modified Early Warning Score (MEWS), Glasgow Coma Score (GCS), Acute Physiology, and the American Society of Anesthesiologists (ASA) Score. We wish to compare in-hospital severity scores to predict 30 day readmission.
Methods A retrospective a cohort of patients was assembled who were admitted to two Geisinger hospitals during calendar year 2012. Admissions related to a child’s birth, unknown acuity, and in-hospital deaths. Data was obtained during the year preceding the admission and outcomes were assessed up to 30 days post-discharge. Additional data including initial laboratory results, comorbidities, length of stay, and discharge disposition was obtained. For severity scores measured longitudinally during the admission, we selected the initial, last, and worse. A base model of known readmission predictors was built using logistic regression with a flexible specification of continuous measures, and the severity scores were considered individually by forcing them into the model. The net reclassification index (NRI) was used to quantify improvement in prediction.
Results A total of 41,413 admissions met cohort entry criteria (mean age 58.5 y, 45% (18,628) male). Readmissions within 30 days occurred 13.3% (5,502) during 2012 with another 2.9% (1,217) that died without a readmission. The base model consisting of demographics, comorbidities, discharge disposition, and utilization in past year resulted in an area under the ROC curve (AUC) of 0.710. LACE significantly contributed to the model (AUC = 0.715, NRI = 2.0%, OR = 1.12 [1.10, 1.13], P <0.001). The last reported GCS was significantly associated with 30-day readmission (OR = 1.05 [1.02, 1.09]), however there was no increase in the AUC and the NRI was non-significant. The MEWS was not significantly associated with readmission.
Conclusions The usefulness of in-hospital severity scores was not consistent in their association with readmission. Additional research is needed to understand performance in subgroups and to utilize serial measurements when available.
