Efficient and Explainable Risk Assessments for Imminent Dementia in an Aging Cohort Study

Area of Science:

• Gerontology
• Neurology
• Artificial Intelligence

Background:

• The growing aging population necessitates scalable dementia risk identification methods.
• Current prediction tools are often costly, lack predictive power, or require extensive testing, limiting their applicability.
• There is a need for accessible and accurate methods to predict imminent dementia onset.

Purpose of the Study:

• To develop a machine learning model for accurate and scalable prediction of imminent dementia onset (within three years).
• To reduce the burden of dementia risk assessment by optimizing cognitive testing and visit frequency.
• To provide individualized risk explanations for dementia prediction.

Main Methods:

• Applied a machine learning model to a longitudinal aging cohort study with extensive clinical variables.
• Identified a subset of highly predictive cognitive tests and optimized clinical visit schedules.
• Utilized individualized prediction explanations that retain non-linear feature effects.

Main Results:

• The developed machine learning model accurately identifies individuals at risk for dementia.
• The final model uses only four cognitive tests (under 20 minutes) for assessment.
• Predictive performance is comparable to a 100-minute neuropsychological battery, with personalized risk explanations.

Conclusions:

• The proposed method offers an efficient and scalable tool for dementia risk screening in the aging population.
• Individualized risk explanations enhance patient understanding and engagement.
• This approach addresses the limitations of current dementia prediction strategies.