A breakthrough clinical trial demonstrates that integrating machine learning algorithms with patient-reported surveys can significantly improve the rate of early dementia diagnosis. This innovative combination offers a scalable solution for identifying previously unrecognized cases of Alzheimer’s disease and related dementias (ADRD) within busy primary care settings, where screening often lags.
An estimated six million Americans currently live with dementia, and experts agree that early diagnosis remains critically important for better patient management and treatment planning. However, primary care clinicians, frequently managing heavy workloads, often lack the time for extensive cognitive screening. This new study, conducted by researchers at the Indiana University School of Medicine, sought to streamline the process using technology.
The researchers conducted a randomized trial involving over 5,300 adults aged 65 and older across nine federally qualified health centers. None of the participants had a prior diagnosis of mild cognitive impairment or dementia. The clinics were split into three arms to test different diagnostic approaches. The control group received standard, usual care without routine screening. The second group used an algorithm that silently scanned electronic health records (EHR) data, flagging high-risk patients to clinicians. The third, and most successful, group used the EHR-scanning algorithm plus a short, 10-question patient-reported survey. This survey asked about memory, daily functions, mood, and behavior.
The results clearly favored the combined approach. Clinics utilizing the algorithm and the patient survey reported a 31% higher odds of new ADRD diagnoses compared to those in the usual care group. Furthermore, this combined method dramatically increased the utilization of necessary follow-up care. After 12 months, 36.7% of patients in the AI-plus-survey group underwent a dementia-related diagnostic test. In contrast, clinics using the algorithm alone saw only a 27.8% testing rate, and the usual care group showed 29%.
This study confirms that technological support alone is not the only answer; incorporating the patient’s own perspective is vital. The algorithm, which researchers validated with close to 80% accuracy in pre-trial testing, efficiently identifies quantitative risk factors hidden in a patient’s medical history. Adding the 10-question survey then captures crucial subjective and behavioral details that doctors might miss during brief appointments.
According to Dr. Malaz A. Boustani, who led the study, transforming primary care to meet the extensive needs of people with unrecognized cognitive impairment requires a low-cost, sustainable, and time-efficient approach. The combination tool achieves this goal by removing significant barriers for both patients and clinicians.
The findings suggest that the most effective dementia screening strategy leverages the speed and power of machine learning to sort through massive amounts of data, while still grounding the diagnostic process in efficient patient input. Health systems can readily scale this dual approach, moving closer to a future where hidden dementia cases become routine, early diagnoses.
