A Machine Learning Approach to Modify the Neurocognitive Frailty Index for the Prediction of Cognitive Status in the...

Area of Science:

• Gerontology
• Neuroscience
• Computational Biology

Background:

• Frailty is a key predictor of health outcomes in older adults.
• The Neurocognitive Frailty Index (NFI) assesses physical and cognitive elements for age-related health decline.
• High dimensionality of the NFI may limit its clinical adoption.

Purpose of the Study:

• To reduce the dimensionality of the NFI using machine learning.
• To maintain the predictive power of the NFI for cognitive decline.
• To develop a more accessible frailty assessment tool.

Main Methods:

• Employed machine learning techniques including Network Analysis, neural networks, LASSO, Random Forest, and XGBoost.
• Calibrated models using data from the Canadian Study of Health and Aging.
• Identified and removed variables with minimal impact on outcome prediction.

Main Results:

• Reduced the NFI from 42 to 36 elements by removing six non-impactful variables.
• The modified NFI scale maintained predictive performance for cognitive change, comparable to the original NFI.
• Demonstrated the feasibility of using machine learning for predictive model modification.

Conclusions:

• Machine learning can effectively reduce the dimensionality of complex health indices like the NFI.
• A 36-element NFI offers a feasible and accurate alternative for assessing frailty and predicting cognitive decline.
• This approach has broader implications for refining predictive models in neurodegenerative diseases and aging research.