Blood Biomarker Signatures for Slow Gait Speed in Older Adults: An Explainable Machine Learning Approach

Abstract

Maintaining physical function is crucial for independent living in older adults, with gait speed being a key predictor of health outcomes. Blood biomarkers may potentially monitor older adults’ mobility, yet their association with slow gait speed still needs to be explored. This study aimed to investigate the relationship between blood biomarkers and gait speed using the Midlife in the United States (MIDUS) study biomarker dataset. A cross-sectional design was employed for analysis, involving 405 individuals aged 60 years and over. We used a machine learning framework, specifically the XGBoost algorithm, feature selection methods, and the Shapley Additive Explanations, to develop an explainable prediction model for slow gait speed. Our model demonstrated the highest cross-validation score with the six most important features among 35 variables, as elevated interleukin-6, C-reactive protein, glycosylated hemoglobin, interleukin-8, older age, and female sex were significantly associated with reduced gait speed (area under the curve = 0.75). Our findings suggest that blood biomarkers can play a critical role in integrated models to assess and monitor slow gait speed in older adults. Identifying key blood biomarkers provides valuable insights into the underlying physiological mechanisms of mobility decline and offers promising avenues for early intervention to preserve mobility in the aging population.