UMass Lowell Professor Hong Yu (Photo: Edward Brennen for UMass Lowell)

LOWELL – UMass Lowell researchers have uncovered an innovative way to spot Alzheimer’s disease long before an official diagnosis. By using artificial intelligence to analyze clinical notes in electronic health records, a team led by Computer Science Professor Hong Yu has demonstrated that the risk for the disease can be identified up to 15 years earlier than current standards.

About 7.2 million Americans over the age of 65 were living with Alzheimer’s dementia in 2025, according to a recent report by the Alzheimer’s Association. An additional 200,000 Americans have younger-onset dementia. The number of Americans with Alzheimer’s disease is expected to increase to 13.8 million in 2060.

While there is no cure for Alzheimer’s disease, early detection can be a game changer. Early diagnosis allows for behavioral interventions and medications that slow progression in mild cases. It also carries a massive economic impact: In one study, the Alzheimer’s Association estimated savings of $7 trillion in health and long-term care costs thanks to early diagnosis of the disease.

While diagnosing dementia in general is based on outward symptoms, pinning the cause to Alzheimer’s disease requires, for the average patient, either a spinal tap or extensive imaging studies. “The existing diagnostic techniques are invasive or expensive,” said Yu, who teaches in UMass Lowell’s Miner School of Computer and Information Sciences.

Yu was the lead researcher on the study, which received $6 million in funding from the National Institutes of Health. Results from the research were published online in January in “Communications Medicine,” part of the Nature portfolio of journals. The work is being pursued through UMass Lowell’s Center of Biomedical and Health Research in Data Sciences (CHORDS), which brings together experts in diverse fields to improve health through innovative AI and data science approaches. Yu founded the center in 2019 and is its director.

For the study, the researchers accessed the records of 61,537 patients diagnosed with Alzheimer’s disease and more than 234,000 similar patients without an Alzheimer’s diagnosis. That access was made possible with the cooperation of the U.S. Department of Veterans Affairs’ Veterans Health Administration.

Some of the keywords and phrases found in the clinical notes were medical terms such as “visuospatial,” “dysphagia” and “agnosia.” Others were conversational, such as “mood,” “fluency,” “pain,” “wandering,” “hearing,” “delusion” and “getting lost.”

When the keywords found in the medical records of the patients diagnosed with Alzheimer’s disease were compared with those in the records of the control group, the differences were noticeable almost immediately, Yu said, allowing the analysis to uncover the increased risk up to 15 years before the official diagnosis was made.

“We believe this is the most comprehensive electronic health record data set in the United States,” Yu said. “It includes patients from all 50 states, so it’s truly national data.” The high quality of care at the VA means that patients remain in the VA system for years, making data sets of 20 years’ length available to researchers, she added.

Yu’s team has applied fundamental AI principles to data to examine social issues such as suicide risk, food pantries and the risk of drug overdose, in addition to the current look at Alzheimer’s disease risk.

“Our goal is to develop innovative AI technologies to help understand human diseases and to change people’s behavior for better health,” she said.

An important part of her research, she said, is examining social and behavioral determinants of health such as socio-economic status, education, habits like smoking, access to health care, and social isolation.

When such information is added to existing AI models, it makes the predictive model more accurate, she said. She believes understanding these factors can help researchers design solutions to improve health outcomes.

She plans to continue working on the Alzheimer’s disease model by adding social and behavioral determinants of health data.

Even though the relationship between these factors and the risk for Alzheimer’s disease may be complex, she is hopeful.

“We can help people,” she said.