Degree and centrality-based approaches in network-based variable selection: Insights from the Singapore Longitudinal...

Area of Science:

• Gerontology
• Network Science
• Biomedical Informatics

Background:

• The Singapore Longitudinal Aging Study (SLAS-2) collected extensive clinical data.
• Predicting successful aging (SAGE) requires analyzing numerous variables.
• Reducing data dimensionality while preserving predictive power is crucial for efficient analysis.

Purpose of the Study:

• To develop and evaluate a network-based method for selecting a representative subset of clinical variables from the SLAS-2 dataset.
• To compare the efficacy of degree-based versus centrality-based variable selection strategies.
• To assess the predictive performance of reduced variable sets for the SAGE index.

Main Methods:

• A network-based approach was used to identify variable subsets from SLAS-2 clinical data.
• Four subsetting strategies were implemented: two degree-based and two centrality-based.
• Machine learning models predicted the SAGE index using these variable subsets.

Main Results:

• All models identified important predictors across physical, cardiovascular, cognitive, and immunological domains.
• A centrality-based approach yielded the smallest variable subset with high predictive accuracy (AUC).
• Adding more central variables minimally improved prediction but significantly increased subset size.

Conclusions:

• Network-based centrality analysis offers an efficient method for data reduction in aging studies.
• This approach balances predictive performance and variable set size, creating representative data subsets.
• The identified variable subsets can effectively represent the comprehensive SLAS-2 clinical dataset.