DEVELOPMENT AND VALIDATION OF THE ELECTRONIC HEALTH RECORD RISK OF ALZHEIMER’S AND DEMENTIA ASSESSMENT RULE (ERADAR)

Abstract

Clinical guidelines emphasize the importance of early recognition of dementia for improving care, yet nearly half of patients with dementia are undiagnosed. Our objective was to develop and validate the electronic health record (EHR) Risk of Alzheimer's and Dementia Assessment Rule (eRADAR), an automated tool to detect patients with unrecognized dementia. Study participants were 4,330 older adults enrolled in the Adult Changes in Thought (ACT) study, a prospective cohort study of dementia set within the Kaiser Permanente Washington (KPWA) integrated healthcare delivery system. Participants were assessed for dementia every two years through ACT using standard clinical criteria. Dementia detected in ACT before being recognized in the KPWA EHR (based on dementia/memory loss diagnosis codes or medication fills) was considered unrecognized. We extracted KPWA EHR data for the two years prior to each ACT visit and considered 56 potential predictors representing demographic characteristics, medical diagnoses, vital signs, healthcare utilization and medication fills. Logistic regression with LASSO (least absolute shrinkage and selection operator) was used to develop a prediction model for unrecognized dementia. Analyses were performed at the visit level, with visits divided into 70% development and 30% validation samples. 1,015 of 16,665 (6%) ACT visits resulted in a study diagnosis of dementia, of which 498 (49%) were unrecognized. Our final prediction model for unrecognized dementia included 31 EHR variables: demographics (age, sex); diagnoses (e.g., congestive heart failure, cerebrovascular disease, diabetes, psychoses); vital signs (body mass index, blood pressure); healthcare utilization (e.g., emergency department visits); and medication fills (e.g., anti-depressants). Discrimination was good in the development sample (c statistic, 0.78; 95% CI: 0.76, 0.81) and validation sample (0.81; 0.78, 0.84). Calibration was good based on plots of predicted vs. observed risk. If scores above the 95th percentile were to trigger dementia screening, we estimate that 1 in 6 (15%) visits would result in a dementia diagnosis. eRADAR uses information readily available in most EHR systems to detect patients with unrecognized dementia with good accuracy. Additional studies are needed to examine the impact of implementing eRADAR in healthcare delivery systems on dementia detection rates and patient outcomes.