Researchers at Los Angeles-based UCLA Health have developed an AI model that uses EHRs to identify patients with undiagnosed Alzheimer’s disease.
The model was trained on records from more than 97,000 patients and achieved sensitivity rates between 77% and 81% across non-Hispanic white, non-Hispanic African American, Hispanic/Latino and East Asian groups, according to a Dec. 10 health system news release. In comparison, conventional supervised models produced sensitivity rates between 39% and 53%.
The tool employs a semi-supervised machine learning technique called positive unlabeled learning, which allows it to analyze records without needing confirmed diagnoses for all patients. Researchers incorporated fairness measures to address long-standing disparities in Alzheimer’s diagnosis.
The UCLA Health team said the model could help identify high-risk individuals who may benefit from further evaluation or screening. Researchers plan to validate the tool in other health systems to assess its generalizability, the release said.
The study’s results were published Nov. 27 in NPJ Digital Medicine.
