Acute Care for Elders Risk Score: A Practical Machine Learning–Based Tool for Screening High-Risk Older Inpatients

To develop and validate a machine learning-based risk score-the Acute Care for Elders (ACE) Risk Score-that integrates the Clinical Frailty Scale (CFS) with readily available clinical and laboratory data to predict adverse outcomes in older hospitalized patients. A retrospective cohort study using a machine learning framework (AutoScore) to develop an interpretable clinical risk score. This study included 21,757 hospital admissions of patients aged ≥65 years at a tertiary academic hospital between May 2021 and November 2023. The patients were randomly allocated to training (50%), test (45%), and validation (5%) cohorts. Eighteen candidate variables were extracted from electronic medical records (EMRs) on the first day of admission. The composite outcome included in-hospital delirium, pressure ulcers, falls, and mortality. Using the AutoScore framework, a parsimonious model with 5 variables (CFS, albumin, C-reactive protein, hemoglobin, and number of pre-admission medications) was selected. Predictive performance was evaluated using receiver operating characteristic analysis and compared with CFS and age. The ACE Risk Score showed superior predictive performance for the composite outcome (area under the curve [AUC] 0.837) compared with CFS alone (AUC 0.798) and age (AUC 0.630; P < .001 for both). Higher ACE scores were also associated with greater risks of safety events, 30-day readmissions and emergency visits after discharge, longer hospital stays, and rapid response team activations. The ACE Risk Score is a practical, interpretable, and scalable tool for early identification of high-risk older inpatients, using data available on the first day of admission. It supports timely, targeted interventions by geriatric teams. The score may facilitate broader implementation of risk-guided care strategies for older adults in hospital settings.