New AI-powered screening method helps primary care doctors quickly identify dementia

Primary care clinics carry a tough burden. You walk in with concerns about blood pressure, pain or medication questions. There is rarely time to talk about memory slips or moments when everyday tasks feel harder than they should. That pressure on both sides often means Alzheimer disease and related dementias stay hidden until months or even years after early signs appear. More than half of older adults never receive a timely diagnosis, even though early changes in daily habits or thinking can show up long before the condition becomes severe.

Researchers across several institutions have spent years trying to close this gap. They know that lifestyle changes can lower the risk of Alzheimer disease by as much as 45 percent. They also know new therapies can slow the disease in its first stage. Medicare has even launched a plan to support families through a more coordinated system of dementia care. Still, these advances help only if the condition is actually recognized.

Traditional screening tools take time and attention, two things in short supply during a busy clinic day. Paper-based memory tests do not scale well. New blood biomarkers may help detect early Alzheimer disease, but similar tests for other dementias are not yet available. Many people also hesitate to bring up memory concerns because of stigma or fear. This often leaves clinicians guessing or relying on brief impressions during a short visit.

Because of these barriers, scientists began testing digital tools that can work quietly in the background. One of these is the Quick Dementia Rating System, or QDRS. It is a short, 10-question survey that asks people to reflect on changes in problem-solving, communication, behavior and daily function. Most people finish it in under three minutes. Earlier research found that the QDRS correctly identified dementia about 85 percent of the time at common scoring thresholds.

The second tool takes a very different path. Known as a passive digital marker, or PDM, it uses machine learning to analyze years of information stored in electronic health records. It looks for signals such as notes about forgetfulness, vascular risk, missed medications or repeated questions during visits. It then produces a probability score that reflects dementia risk. Early studies showed accuracy near 80 to 85 percent.

To see whether combining these two tools would improve early detection, researchers ran a randomized clinical trial across nine Eskenazi Health clinics in Indianapolis. These federally qualified centers serve diverse and low-income neighborhoods, often with strong interpreter support. Clinics, not patients, were randomized to one of three paths: usual care, PDM alone or QDRS plus PDM.

More than 5,300 adults age 65 and older took part. The average participant was 71. Women made up more than 60 percent of the group. Over half were Black or African American. A smaller share identified as Hispanic or Latino. No one in the study had a prior diagnosis of dementia or mild cognitive impairment, and all had at least three years of electronic health data.

Both digital tools were built directly into the Epic electronic health record system. When patients in the combined-tool clinics checked in, the portal invited them to complete the QDRS. At the same time, the passive digital marker ran silently in the background and flagged anyone with a risk score above 59 percent. Any positive result sent a clear alert to the clinician with the information needed to guide decisions. Suggested next steps included referral to a specialized memory care program that offered structured interviews, neurologic exams, brain scans and lab testing.

Only about one in five people in the combined clinics actually completed the QDRS. Even with this low rate, the blend of patient-reported insight and machine learning still changed the pattern of care.

After 12 months, the clinics using both tools saw a 15.4 percent rate of new dementia diagnoses. Usual care reached 12.4 percent. Clinics using only the passive digital marker reached just 10.3 percent. When researchers adjusted for age, sex, race and ethnicity, the combined approach raised the odds of a new diagnosis by 31 percent compared with usual care.

The timing of diagnosis also shifted. People in the combined clinics were diagnosed sooner, with a hazard ratio of 1.37. Once someone completed any diagnostic service such as a brain scan or lab panel, the chance of receiving a diagnosis increased even more.

The dual approach also led to more diagnostic assessments. Nearly 37 percent of people in the combined clinics received at least one form of follow-up testing within a year. That compared with 29 percent in usual care. Clinics using only the passive digital marker did not differ from usual care in either detection or follow-up.

These results held across multiple analytic methods. Clinic-level differences were minimal. Time-to-event models showed the same pattern. Researchers believe the QDRS helped clinicians trust the automated risk signal, especially in a patient population facing many social and health challenges.

During interviews, lead investigator Malaz Boustani, M.D., MPH, from the Regenstrief Institute, emphasized that the passive digital marker is open source and free to use. “Any healthcare system with an electronic health record and the right personnel can implement it,” he said. “It is zero cost and requires no clinician time.”

Other members of the research team echoed that message. James E. Galvin, M.D., MPH, from the University of Miami Miller School of Medicine, who helped create the QDRS, noted that the survey empowers families to share early concerns. Zina Ben Miled, PhD, from Lamar University, said the digital tools help reach people who might otherwise be overlooked because of limited resources or restricted access to specialty care.

For primary care teams stretched thin by the demands of chronic disease management, this work offers a model that fits within the flow of a typical visit. The QDRS asks patients to reflect for a few minutes at home or in the waiting room. The passive digital marker updates itself as medical data accumulate. Clinicians receive concise alerts only when needed.

The trial showed that this dual system can lift the rate of dementia detection by nearly one third while also speeding up the steps that follow. That can shorten the long and often stressful path toward clarity for families who sense that something is not right.

Research findings are available online in the journal JAMA Network Open.

Like these kind of feel good stories? Get The Brighter Side of News' newsletter.