Benchmarking Interpretable Machine Learning for Frailty Screening in Older Adults Using Routine Clinical Variables...

Area of Science:

• Gerontology
• Biostatistics
• Machine Learning

Background:

• Frailty is prevalent in Latin America and the Caribbean, necessitating accessible screening tools.
• Existing screening methods often require extensive clinical data, limiting their use in resource-limited settings.
• Validated, low-cost screening tools for frailty are scarce in the region.

Purpose of the Study:

• To develop and validate interpretable machine learning models for frailty screening in older adults.
• To assess the added value of the SARC-F questionnaire beyond routine hemodynamic and anthropometric measures.
• To identify effective and cost-efficient frailty screening methods for geriatric care.

Main Methods:

• A cross-sectional pilot study involving 100 older adults classified using the FRAIL scale.
• Development and evaluation of three logistic regression models with 5-fold cross-validation.
• Comparison of models including basic clinical data (Model A, B) versus those incorporating SARC-F (Model C).

Main Results:

• The SARC-F questionnaire was the sole significant predictor of frailty (p < 0.001).
• Models using only hemodynamic and anthropometric data showed poor discrimination (AUC = 0.546).
• Model C, including SARC-F, achieved high discrimination (AUC = 0.921), with SARC-F alone yielding an AUC of 0.942.

Conclusions:

• Routine hemodynamic and anthropometric data are insufficient for accurate frailty screening.
• The SARC-F questionnaire is a dominant and effective predictor of frailty.
• SARC-F shows promise for cost-effective frailty screening in resource-limited geriatric environments, pending external validation.