UCLA researchers have developed an artificial intelligence tool that can use electronic medical records to identify undiagnosed Alzheimer's patients, addressing critical gaps in Alzheimer's treatment, particularly significant underdiagnosis in underrepresented communities.
of study Published in npj Digital Medicine magazine.
Disparities in diagnosis of Alzheimer's disease and dementia among certain populations are a long-standing problem. African Americans are almost twice as likely to have a neurodegenerative disease as non-Hispanic whites, but only 1.34 times as likely to be diagnosed. Similarly, Hispanics and Latinos are 1.5 times more likely to have the disease, but only 1.18 times more likely to be diagnosed.
“Alzheimer's disease is the sixth leading cause of death in the United States, affecting one in nine Americans over the age of 65,” said study lead author Dr. Timothy Chan of the Department of Neurology at the UCLA Department of Health. “There is a huge gap between those who actually have this disease and those who are diagnosed, and the gap is even greater in underrepresented communities.”
Previous studies have leveraged machine learning models to attempt to predict Alzheimer's disease using electronic medical records, but they were designed using traditional frameworks that may not account for certain diagnostic biases.
The new model developed by the UCLA team takes a different approach, known as semi-supervised positive unlabeled learning, and is specifically designed to promote fairness while maintaining high accuracy.
Electronic medical records for more than 97,000 patients at UCLA Health. This includes patients with a confirmed diagnosis of Alzheimer's disease and patients with unconfirmed diagnosis.
This model achieved sensitivity rates of 77% to 81% across non-Hispanic white, non-Hispanic African American, Hispanic/Latino, and East Asian groups, compared to sensitivities of 39% to 53% for traditional supervised models.
The UCLA researchers built on previous AI models used to predict various diseases, including Alzheimer's disease, but there were gaps in reducing bias and disparities. The UCLA tool analyzed patterns in health records such as diagnosis, age, and other clinical factors. They also identified key predictive features of Alzheimer's disease, including both neurological indicators such as memory loss and unexpected patterns such as pressure sores and heart palpitations that may signal undiagnosed cases.
Unlike traditional approaches that require a confirmed diagnosis for all training data, the UCLA model learns from both confirmed cases and patients whose Alzheimer's disease status is unknown. The researchers incorporated equity measures throughout the development of the model, using population-specific criteria to reduce diagnostic disparities.
The tool was validated using multiple approaches, including genetic data. Patients predicted to have undiagnosed Alzheimer's disease showed significantly higher polygenic risk scores and genetic markers for the disease, known as APOE ε4 allele counts, compared to patients predicted not to have Alzheimer's disease. Chang said the tool could help clinicians identify high-risk patients who may benefit from further evaluation or screening. Early detection is critical as new Alzheimer's disease treatments become available and lifestyle interventions can slow the progression of the disease.
The research team plans to prospectively validate this model in affiliated health systems to assess its generalizability and clinical utility prior to potential implementation into routine practice.
“Our model helps improve significant underdiagnosis in underrepresented populations by ensuring unbiased predictions across populations,” said Chan. “It could potentially address disparities in Alzheimer's disease diagnosis.”
reference: Tran T, Fu M, Fung J et al. Significant positive label-free learning to predict undiagnosed Alzheimer's disease in diverse electronic medical records. npj digitmed. 2025;8(1):730. doi: 10.1038/s41746-025-02111-1
This article has been reprinted from the following material: Note: Materials may be edited for length and content. Please contact the citation source for details. You can access our press release publishing policy here.
