We identified 75,851 Veterans without ADRD who completed at least three exercise tolerance tests (ETTs), two of which were at least one year apart. We conducted latent class growth analyses to cluster the trajectories of changes in age-adjusted CRF using all ETT data up to one year before first ADRD diagnosis. We selected the optimal number of classes using the Bayesian Information Criterion (BIC), the estimated average posterior probabilities of class membership in each trajectory class, with values >0.7 preferable, and interpretability of the solution. 

Unadjusted and adjusted Kaplan-Meier curves were plotted to evaluate the association between CRF trajectory cluster and survival probability of free-of-ADRD for different age groups.

Average number of ETTs was 18.7 (42.3), with the mean gap between first and last tests of 6.5 (3.8) years (range, 1.1 to 19.2) years. Five clusters of trajectories were identified: four stable groups differing in baseline fitness (groups 1-4), and one group (n=6,691) with substantial decline (>1.0 MET) from relatively high to low CRF level (group 5). Compared to group 4 (non-declining trajectory)  with similar baseline CRF as Group 5, the probability of ADRD-free survival was significantly lower for those in group 5. However,  Group 5 continued to perform better than the groups (groups 1 and 2) that had lower baseline and equivalent endpoint CRF.