Researchers at UCLA have created a new artificial intelligence (AI) tool that may help find people with undiagnosed Alzheimer’s disease by looking at their electronic health records.
This could be a big step forward in fixing a major problem in Alzheimer’s care—many people who have the disease never get diagnosed, especially in communities that are often left out of healthcare research.
The study was published in the journal npj Digital Medicine. It explains how this new AI tool could help doctors catch Alzheimer’s earlier and more fairly.
Alzheimer’s is one of the top causes of death in the United States. It affects about one in nine Americans over age 65. But not everyone is equally likely to be diagnosed.
For example, African Americans are nearly twice as likely to have Alzheimer’s as white Americans, but they are only slightly more likely to be diagnosed. Hispanic and Latino people also face a similar gap between having the disease and being diagnosed.
Dr. Timothy Chang from UCLA’s Department of Neurology led the study. He said the difference between who has Alzheimer’s and who gets diagnosed is huge, especially for people from underrepresented groups.
In the past, researchers have tried using machine learning to find people with Alzheimer’s from health records. But many of these tools were built using older methods that might carry bias.
The new tool from UCLA uses a more advanced method called “semi-supervised positive unlabeled learning.” This method doesn’t rely only on confirmed cases. Instead, it learns from both people who have a diagnosis and those who may have it but haven’t been diagnosed yet. This helps the AI make better predictions and treat all groups more fairly.
The UCLA team studied health records from over 97,000 patients. Their AI tool looked at patterns in things like age, diagnoses, and other medical issues. It noticed not only the usual signs of Alzheimer’s, like memory loss, but also less expected signs like skin sores and irregular heartbeats, which might point to hidden cases.
The AI tool did a much better job than older models. It had success rates between 77% and 81% across all the groups studied, including white, Black, Hispanic, and East Asian patients. In contrast, older models only had 39% to 53% success.
To make sure the tool was working properly, the researchers also checked genetic data. People the tool predicted to have Alzheimer’s had more known Alzheimer’s genes, like the APOE ε4 gene. This extra step gave more confidence that the tool was finding real, hidden cases.
Dr. Chang said the AI tool could help doctors flag high-risk patients earlier so they can be tested or treated sooner. That’s important because early treatment and lifestyle changes can slow the disease. The team now plans to test the tool in other health systems to make sure it works in different places and for different people.
This AI tool has the potential to close the gap in Alzheimer’s diagnosis, especially for people who often don’t get the care they need. By making sure the tool is fair and accurate for everyone, it could help give more people a chance to get help earlier.
If you care about Alzheimer’s, please read studies about the likely cause of Alzheimer’s disease , and new non-drug treatment that could help prevent Alzheimer’s.
For more health information, please see recent studies about diet that may help prevent Alzheimer’s, and results showing some dementia cases could be prevented by changing these 12 things.
Copyright © 2025 Knowridge Science Report. All rights reserved.
