Digital Tool Improves Early Dementia Detection

CHICAGO — Combining machine learning and patient-reported tools significantly improved early detection of dementia in primary care, according to new research presented at the annual meeting of the American Geriatrics Society (AGS) 2025 Annual Scientific Meeting.

Researchers from Indiana University Indianapolis found that using a machine learning algorithm and a patient-reported screening tool — together known as the Digital Detection of Dementia (D3) approach — led to a 44% higher likelihood of a clinician diagnosing a patient with dementia over a 1-year period compared with usual care.

The researchers tested the D3 model in a randomized trial across nine federally qualified health centers (FQHCs) in Indianapolis.

Of the more than 5300 patients aged 65 years or older who enrolled, 62% were women, and more than half were Black or Hispanic.

The clinics were assigned to provide three types of care, the first being usual care.

Patients in the second group were evaluated for dementia using the passive digital marker, an artificial intelligence (AI) algorithm that analyzed existing electronic health record data to flag potential cases.

Clinicians at the third clinic evaluated patients by analyzing results from the Quick Dementia Rating System, a 10-item questionnaire that takes patients 2-3 minutes to complete, and viewed flagged cases produced by the algorithm.

Providers at the third clinic diagnosed significantly more patients with dementia than those at clinics using usual care, even after accounting for age, sex, race, and ethnicity (odds ratio, 1.44; 95% CI, 1.19-1.75).

In contrast, clinics using only the AI tool showed no statistically significant increase in diagnoses. Researchers hypothesized that even partial use of the questionnaire — completed by just 21% of eligible patients — may have improved clinicians’ trust in AI alerts, nudging them toward making a diagnosis.

“Providers…are very busy and are bombarded with computerized decision support,” said lead study researcher Malaz Boustani, MD, a geriatrician and professor of aging research at Indiana University. But “with the right messenger and time,” trust in digital tools can grow and lead to behavior change.

According to previous research, 62% of older adults seen in FQHCs have mild cognitive impairment, and 12% have dementia. Yet most go undiagnosed. Black patients were more than twice as likely as White patients to have unrecognized cognitive decline.

The tools were embedded directly into the clinics’ electronic health record system, triggering alerts in the chart view. Patients were offered the questionnaire during electronic check-in. A clinical decision support system then guided clinicians with next steps.

“This hybrid model lets clinicians do the right thing without adding to their workload,” Boustani said.

The D3 model is now being evaluated for broader implementation, Boustani said. He and his colleagues are also exploring ways to increase patient completion of the questionnaire and refine clinician prompts.

“We’re waiting on the result of an ongoing trial that is trying to replicate our study using interruptive alerts instead of non-interruptive alerts,” Boustani said. “If that trial shows similar or better results, we will work with existing consulting companies to distribute the digital tools across the country.”

Boustani serves as a chief scientific officer and co-founder of two private companies and has various financial relationships involving equity with other companies. He also serves on various advisory boards for pharmaceutical companies.

Lara Salahi is a health journalist based in Boston.