32EMF Validation of the Admission for Geriatric patients in the Emergency Department (AGED) Algorithm

Abstract

Listen

Translate article

Identifying high risk older adults in the emergency department (ED) is essential for resource allocation and targeted interventions. The goal of this study was to develop and validate a geriatric risk score predicting emergency department admission using data available on initial assessment. The risk score was developed using an observational cohort of patients age 65 years and older who visited an urban academic ED between 9/1/19 and 2/28/20. The score was then validated using visits between 6/1/20 and 12/31/20. Patients were excluded if they left without being seen, against medical advice, were seen by the geriatric specialty service, or had been seen in the ED within the past 9 days. Forty-four key variables available upon triage were identified using the EMR including Estimated Severity Index (ESI), Clinical Frailty Scale (CFS), triage vital signs, demographics, comorbidities, and chief complaint. A Random Forest model with all key variables was performed to predict admission. The fifteen most important variables were included in a logistic regression model. These were compared to logistic regressions using CFS alone, ESI alone, and CFS, ESI, sex, and age. Secondary outcomes included ED return visit within 9 days and subsequent admission within 30 days of ED visit. The model was then validated using the second dataset for all 3 outcomes. Results/Findings: Of 6863 visits of patients age 65 and older, 5606 (81.7%) met inclusion criteria for this study. Mean age was 74.5 years, 45.6% male. The Random Forest model with all predictors had an AUC of 0.800 [95% Confidence interval (CI): 0.789, 0.812], sensitivity of 76.0% and specificity of 69.3% for admission compared to AUC of 0.661 (CI: 0.647, 0.675) for CFS alone and AUC of 0.681 (CI: 0.666, 0.695) for ESI alone. The logistic regression with the top 15 predictors for admission had an AUC of 0.786 (CI: 0.775, 0.798), sensitivity of 66.3% and specificity of 76.3%. The Random Forest model for ED return visit had an AUC of 0.540 (CI: 0.507, 0.573) and for subsequent admission, AUC was 0.632 (CI: 0.610-0.655). A risk prediction algorithm, the AGED algorithm, that incorporates clinical characteristics known about older adults at triage in the ED including CFS can predict hospital admission with moderate accuracy. The AGED algorithm has poor predictive performance for ED return visit within 9 days and subsequent admission within 30 days of ED visit. Next steps include incorporating the AGED algorithm into clinical practice to see if it has an impact on patient-centered and administrative outcomes.