Serial gait speed measurements over time and dynamic survival prediction in older adults

Abstract

Background: A one-time gait speed measurement predicts mortality risk. A framework for updating a clinician’s mortality risk perception with new information from each clinic visit is needed. We used joint modeling of longitudinal and survival data for dynamic prediction of mortality risk. Methods: We fit sex-stratified joint models to 20-meter (bi)annual longitudinal gait speed measured every 6 months and 14-year survival data from the Health, Aging and Body Composition Study allowing for non-linear fluctuations of gait speed and controlling for important covariates such as age, recent hospitalization, blood pressure, obesity, and comorbidities. Results: Participants (N = 3048) were 74 years old with gait speed 1.18 m/s. They were 42% Black, and 52% died over 14 years. Higher gait speed of 0.1 m/s was associated with 23% (95% confidence interval or CI = 20–25%) and 25% (CI = 21–28%) reductions in mortality risk in men and women; and a 0.05 m/s annualized slowing (slope) with 31% (CI = 13–51%) increase in men (all p < 0.05), with findings persisting after covariate adjustment. Distant gait speed history over a year prior contributed little for mortality risk prediction with mean change of only 1–2% in 5-year risk. Conclusion: The two most recent gait speeds appear sufficient to consider for mortality risk in the present initial analysis. More frequent gait speeds need to be considered in mortality risk prediction before definitive conclusions supporting real-world application.

Keywords

• Gait speed
• Survival
• Mortality
• Joint modeling
• Older adults