Abstract
Up to half of people with dementia in the U.S. are undiagnosed. Current guidelines recommend early diagnosis to improve patient outcomes. Our goal was to develop and validate a tool that uses electronic health record (EHR) data to identify patients likely to have current, undiagnosed dementia. We used gold-standard dementia diagnosis data from the Adult Changes in Thought (ACT)--a prospective cohort study that follows adults age ≥65 years to detect incident dementia--linked with EHR data from Kaiser Permanente Washington. Participants at each ACT visit were classified as no dementia, diagnosed dementia (dementia-related medications or diagnosis codes present in the EHR before ACT diagnosis) or undiagnosed dementia (no EHR evidence of dementia recognition before ACT diagnosis). We divided the data into training (70%) and test (30%) sets. Logistic regression with LASSO penalty was used to identify EHR predictors of undiagnosed dementia versus no dementia. Our sample included 16,655 visits in 4,330 patients (498 unrecognized dementia). The final model included 31 predictors in 5 categories: demographics (age, sex); vital signs (BMI, high blood pressure); diagnoses (e.g., diabetes, psychoses); healthcare utilization (e.g., emergency visits) and medication-related variables (e.g., anti-depressant use). Discrimination based on the c-statistic was 0.78 (95% CI: 0.76, 0.81) in the training set and 0.81 (0.78, 0.84) in the test set. These results suggest that EHR data can identify older patients who may have current, undiagnosed dementia. Additional studies are needed to determine whether implementation of eRADAR in clinical settings results in earlier diagnosis and improved patient outcomes.