Development of a Fall Risk Score for Older Adults Incorporating Electronic Health Record and Emergency Department...

Older adults have high rates of falls after Emergency Department (ED) discharge; yet existing screening tools either underperform or are too difficult to deploy. This study aimed to evaluate a parsimonious predictive model for falls within 6 months post-ED discharge, utilizing both typical electronic health record (EHR) data and brief ED-based screenings.

In a prospective cohort study from September 2023 to May 2024, 412 community-dwelling adults aged ≥ 60 years who ambulate without assistance were enrolled during ED visits. Baseline data included EHR-derived variables (e.g., comorbidities, medication use) and ED screens (e.g., living situation, fall history). Participants were followed for 6 months to document fall occurrences. Multivariable LASSO logistic regression models to predict any fall were then constructed: Model 1 (EHR), Model 2 (ED screens), and Model 3 (combined). Model performance was evaluated using discrimination and calibration metrics, including area under the receiver operating characteristic (AUC) curves.

Of the 356 participants with complete follow-up, 104 (29.2%) experienced at least one fall. Model 3 demonstrated superior predictive performance (AUC = 0.75) compared to Model 1 (AUC = 0.67) and Model 2 (AUC = 0.71). Significant predictors in the combined model included anemia (OR = 3.19), use of oral hypoglycemics (OR = 2.26), living with less than two other people (OR = 3.79), infrequently leaving home (OR = 1.97), and a history of ≥ 3 falls in the prior 6 months (OR = 12.11). A risk score made up of 9 items (6 EHR; 3 ED screen) categorizing participants as high risk (score 6-25) or low risk (score 0-5) resulted in sensitivity = 64%, specificity = 75%, positive likelihood ratio = 2.54, and negative likelihood ratio = 0.49.

Integrating EHR data with brief ED-based screenings enhances the prediction of fall risk among older adults post-ED discharge. The developed risk score effectively stratifies patients into low versus high risk, facilitating targeted prevention interventions. Further validation in independent cohorts is needed.