Introduction: Automated models that predict cognitive risk in older adults can aid decisions about which patients to screen in busy primary care settings.
Methods: In this retrospective prediction model development study, we conducted formal cognitive testing on 337 older primary care patients to establish cognitive status. We used up to 5 years of prior discrete-field electronic health record (EHR) data to develop a multivariable prediction model that differentiates patients with impaired versus intact cognition.
Results: The final model included seven easily extractable variables with known associations to cognitive decline: age, race, pulse, systolic blood pressure, non-steroidal anti-inflammatory use, history of mood disorder, and family history of neurological disease. The model demonstrated good discrimination of cognitive status (concordance statistic = 0.72).
Discussion: The cognitive risk model may be useful clinically to prompt for objective cognitive screening in high-risk patients. The use of common, discrete variables ensures relative ease of implementation in EHRs.
Highlights: 337 older primary care patients completed full neuropsychological assessment.Risk modeling used data available in a typical primary care record.The model successfully differentiated patients with/without cognitive impairment.This EHR model offers a passive workflow to identify patients at cognitive risk.
Keywords: decision support; dementia; early diagnosis; electronic health record; machine learning; mild cognitive impairment; prediction model.
© 2025 The Author(s). Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring published by Wiley Periodicals, LLC on behalf of Alzheimer's Association.