Stratified Predictors of Gait Speed Decline in Aging Adults: An Interpretable Machine Learning Approach using data from the CLSA

Gait speed, known as the ‘sixth vital sign’ is an indicator of functional health and healthy aging. Identifying drivers of midlife gait decline is critical for timely intervention, yet most evidence focuses on older adults and overlooks non-linear, heterogeneous effects. We applied machine learning with quantile regression to Canadian Longitudinal Study on Aging (n=23,419) to identify predictors of gait speed over 3 years. While higher BMI and comorbidity burden were consistently associated with slower gait across quantiles, distinct predictors emerged at distributional extremes. Among slower walkers, higher HbA1c, depressive symptoms, and prolonged QTc predicted further decline, while emotional support and outdoor activity were protective. Among faster walkers, fruit and vegetable intake, lower white blood cell count, and quicker reaction time predicted faster gait. These findings highlight the importance of distribution-sensitive modeling to identify modifiable, stage-specific predictors for tailored interventions to improve health span and quality of life in aging populations.