Temporal order of gait speed decline relative to changes in measures affected early in Alzheimer’s disease

Abstract Background Individuals with dual decline in gait and cognition have a greater risk of developing dementia. Understanding when gait speed declines relative to measures affected early in Alzheimer’s disease can improve risk assessment. Method Using data for 761 participants (1,108 cognitively unimpaired visits) from the Baltimore Longitudinal Study of Aging, we estimated the trajectories of gait speed, memory (California Verbal Learning Test [CVLT] immediate recall score), hippocampal volume, and plasma Aβ 42 /Aβ 40 , glial fibrillary acidic protein (GFAP), and p‐tau181. We fitted a progression score (PS) model that aligns individuals based on the similarity of their multivariate longitudinal observations to reveal long‐term biomarker trajectories, accounting for individual differences in baseline level and rate of progression. We included height as a covariate for gait speed, first administration versus not for CVLT (to control for practice effect), intracranial volume for hippocampal volume, and estimated glomerular filtration rate for each plasma biomarker. 74 participants converted to mild cognitive impairment (MCI) or dementia after the last visit used for fitting the PS model and were used to verify that the estimated PS reflects neurodegenerative disease progression using a Cox proportional hazards model. We then examined the temporal ordering of the estimated biomarker trajectories by comparing the timing of peak relative changes. Result Greater PS at last visit was associated with a higher risk of conversion to MCI/dementia (hazard ratio for interquartile range normalized PS = 6.1, P < 2×10 ‐16 ). The PS‐only model had a better concordance index compared to an age‐only model for predicting conversion (0.880 ± 0.016 vs. 0.852 ± 0.018). Estimated trajectories as a function of PS are shown in Figure 1. The peak relative change in hippocampal volume occurred prior to that of gait speed (difference in PS between peaks: ‐0.57 [95% confidence interval ‐1.32, ‐0.05]), whereas p‐tau181 changed later (difference in PS between peaks: 1.95 [1.19, 2.45]) (Figure 2). Conclusion Decline in gait speed occurs after hippocampal volume loss but alongside other biomarkers known to be affected early in Alzheimer’s disease and it can prove useful in individualized prediction. Alternative approaches for examining temporal ordering will help assess robustness of findings.