J Korean Geriatr Psychiatry.  2024 Oct;28(2):33-40. 10.47825/jkgp.2024.28.2.33.

Machine Learning-Based Multi-Modal Prediction of Cognitive Decline in Community-Dwelling Older Adults

• Kim J
• Kim N
• Park B
• Roh HW
• Hong CH
• Son SJ
• Research Committee of Korean Association for Geriatric Psychiatry

• KMID: 2561291
• DOI: http://doi.org/10.47825/jkgp.2024.28.2.33

Abstract

Objective
This study aimed to develop a machine learning model to predict cognitive decline in community-dwelling older adults. By integrating multimodal data, including demographic, psychosocial, and neuroimaging information, we sought to en-hance early detection of cognitive decline.
Methods
Data were obtained from 159 participants in the Biobank Innovations for Chronic Cerebrovascular Disease with Alzheimer’s Disease Study. Participants underwent clinical assessments, neuropsychological testing, and magnetic resonance im-aging scans. Cognitive decline was defined as an increase in the Clinical Dementia Rating-Sum of Boxes of greater than 2.05 points per year at follow-up. Models were developed using the logistic classification, combining demographic, psychosocial as-sessments, and neuroimaging data. Model performance was evaluated using area under the curve (AUC), accuracy, and F1 score, while Shapley additive explanation values were used to assess feature importance.
Results
The model that incorporated all data types achieved the highest performance, with an AUC of 0.834. The top predictor of cognitive decline was years of education, underscoring the importance of non-invasive, easily accessible data for prediction.
Conclusion
This machine learning model demonstrates significant potential for early cognitive decline prediction, offering a scalable tool for improving dementia screening and timely intervention, especially in resource-limited settings.

Keyword

Cognitive dysfunction; Dementia; Machine learning; Geriatrics