Dementia risk method uses machine learning for scalable and affordable care

Reuters Events SMR and Advanced Reactor 2025

Dementia risk method uses machine learning for scalable and affordable care
by Clarence Oxford
Los Angeles CA (SPX) Nov 21, 2024

Researchers from the Regenstrief Institute, Indiana University, and Purdue University have unveiled a cost-effective method for predicting dementia risk. This innovative approach leverages existing patient data and machine learning to provide a "zero-minute assessment" for under a dollar, enabling early identification of individuals at heightened risk for dementia and mild cognitive impairment.

"Dementia risk detection is crucial for managing care and planning effectively," said Dr. Malaz Boustani, MPH, senior author and researcher at Regenstrief Institute and IU School of Medicine. "We addressed the challenge of early identification with a scalable, cost-effective solution by utilizing existing data in patients' medical notes."

The researchers' technique extracts information from electronic health records (EHRs) using machine learning. The system identifies phrases and sentences relevant to dementia risk from free-text narratives in medical notes. These notes are written by healthcare providers, such as doctors and nurses, and contain valuable insights about the patient's health status.

Key data points include clinician observations, patient remarks, mental status reports from family members, medication histories, and longitudinal data like blood pressure or cholesterol levels. This information is analyzed to create individualized dementia risk predictions or evidence of cognitive decline.

Dr. Zina Ben Miled, a Regenstrief affiliate scientist, emphasized the approach's precision: "Our methodology combines supervised and unsupervised machine learning to extract sentences relevant to dementia. This not only enhances predictive accuracy but also allows providers to verify results quickly by reviewing the specific text driving the risk assessment."

Early dementia risk prediction has numerous benefits. It empowers patients and families to access resources such as support groups and the Centers for Medicare and Medicaid GUIDE model, which aids individuals in staying at home longer. It also enables healthcare providers to deprescribe medications harmful to cognitive health, discuss risky over-the-counter drugs, and consider FDA-approved amyloid-lowering therapies that alter Alzheimer's disease progression.

Dr. Paul Dexter, a co-author and researcher at Regenstrief and IU School of Medicine, highlighted the broader implications: "Regenstrief and Indiana University have long demonstrated the utility of EHRs. This study exemplifies how machine learning can extract maximum clinical value from these records, making early dementia identification vital as new treatments emerge."

The tool's practicality is another major advantage. Primary care clinicians, often pressed for time and lacking specialized training, can integrate the zero-minute assessment seamlessly into their workflows.

The researchers are wrapping up a five-year clinical trial of the tool in Indianapolis and Miami. Insights from this trial will refine the framework for broader use in primary care. Future work will focus on integrating medical notes with other EHR data and environmental factors to further enhance predictive capabilities.

Research Report:Dementia risk prediction using decision-focused content selection from medical notes