58 Using Machine Learning to Predict Hospital Disposition With Geriatric Emergency Department Innovations Intervention

Abstract

The geriatric emergency department innovations (GEDI) program is a nurse-based geriatric assessment and care coordination program that reduces preventable admissions for older adults and decreases Medicare expenditures. All adults age 65+ in the ED are currently eligible, but only 5% of older adults receive GEDI care due to resource limitations. Currently available geriatric screenings to identify older adults likely to benefit from programs like GEDI perform poorly in the ED. The objective of this study was to predict likelihood of hospitalization accurately and consistently with and without GEDI care using machine learning models to better target the GEDI program. We performed a cross sectional observational study of ED records from (1/2010 to 3/2018). Using propensity score matching, GEDI patients were matched to non-GEDI patients. Logistic regression, random forest, support vector machine, gradient boosting, and neural network techniques were used to predict hospital admission using demographic data, medical history, ED visit chief complaint, vitals, and geographic data. Hospital admission was then predicted with and without GEDI assessment during the ED visit. Visits were classified into one group with a predicted change in hospital disposition with the GEDI assessment and another group with no predicted change in disposition with GEDI assessment. In this second analysis feature importance was performed on the tree-based models. Final model performance was reported as the area under the curve (AUC) using receiver operating characteristic models for the test data. We included 55,056 patients age 65+ who accounted for 134, 361 ED visits. Of these, 3, 860 visits were included in the training set and 10, 142 in the testing set to predict hospital admission. 5, 071 visits were used in the training and testing sets to predict change in disposition with GEDI. The random forest model had the best performance with an AUC of 0.81 (95% CI 0.79-0.83). In the random forest model, 9, 756 (96.2%) ED visits were predicted to have no change in disposition with GEDI assessment, and 386 (3.8%) ED visits were predicted to have a change in disposition with GEDI assessment (Table 1). Of those with a predicted change in disposition from admitted to discharged with GEDI assessment the top 5 most influential variables out of 86 variables with their relative importance are in Table 2. The higher relative importance the more it influenced the model’s outcome, however ranking does not imply the direction of influence. All importance values add up to a total of 1. Our machine learning models were able to predict who is likely to be discharged with GEDI assessment with good accuracy and thus select a cohort appropriate for GEDI care. Future implementation of this machine learning model into the electronic health record may assist in the identification of older adults who should be prioritized for GEDI care.View Large Image Figure ViewerDownload Hi-res image Download (PPT)