The trajectory analysis identified 3 homogeneous groups as the best model based on the BIC (). The procedure calculates the probability of each subject belonging to each trajectory and identifies a subject as belonging to one trajectory based on the largest probability. shows both actual trajectory (solid line; using exact probabilities of belonging to each trajectory for each subject) and estimated trajectory (dotted line; using the model-assigned group identification for each subject) including all covariates except domain-specific cognitive scores. Although the BIC varied slightly depending on the specific cognitive measure included in the model, the 3-trajectory model always provided the best fit to the data, with trajectories virtually identical to those in . Based on the shapes of IADL trajectories, we named them No Decline (group 1: stable at disability state), Moderate Decline (group 2: numbers of IADL disabilities increased somewhat over time), and Sharp Decline (group 3: number of IADL disabilities increased sharply followed by death).
Trajectories of total numbers of IADL disabilities over time
In this community-dwelling cohort of 953 adults free of dementia, the majority entered the study with either no IADL disabilities (78.2%) or only one disability (14.1%). shows the baseline characteristics of the overall cohort and of the 3 latent trajectory-defined groups described above.
shows the association of the baseline cognitive score and other covariates with the three trajectories, using the Sharp Decline trajectory as a reference group. Each column of represents the result of a model with all covariates including MMSE. The basic model (Model 1) includes no specific cognitive domains; Models 2 through 8 each includes one of seven cognitive domains. As an example, a one standard deviation increase in the Learning composite at baseline score is associated with a 95% increase in odds of being in the No Decline group (i.e., OR=1.95) and a 71% increase in odds of being in the Moderate Decline group (i.e., OR=1.71), compared with the Sharp Decline group.
Results of Trajectory Model, adjusting for age, sex, education, recruitment status and numbers of IADLdisabilities at baseline.
Compared with the Sharp Decline group, the No Decline group was associated with higher baseline scores on all cognitive domains as well as fewer total prescription medications, and higher frequency of social engagement, but not with depression. MMSE was not significant in any models.
Compared with the Sharp Decline group, the Moderate Decline group was associated with higher baseline scores on all cognitive domains except psychomotor speed and naming. Total prescription medications, and higher frequency of social engagement were also significantly associated. Neither depression nor MMSE was significant.
We also examined the difference between the Moderate and the No Decline groups by making the reference group the No Decline Group. None of the cognitive functions distinguished the two trajectory groups, but number of prescription drugs was significantly associated with the Moderate Decline group.
As a reference for interested clinician readers, shows the actual mean (SD) baseline cognitive test scores for each of 3 trajectory groups and the overall sample.
Raw cognitive test scores at baseline for each trajectory group
The effect of death on the association between trajectories and covariates
shows the proportion of those who died during the follow-up among each of the three trajectories and follow-up waves during which the death occurred. Only one subject died among the No Decline group. Death occurred during later waves (3rd and 4th follow-up) among the Moderate Decline group, while it occurred during earlier waves (1st through 3rd follow-up) among the Sharp Decline Group.
Proportion of those who died during follow-up among each trajectory group and the waves where death occurred
shows the IADL trajectories after excluding those who died during follow-up from the sample. The three-trajectory model was still found to be the best model. However, none of the cognitive functions distinguished the trajectory groups; the total numbers of prescription medications and social engagement remained significant as previously found.
Trajectories of total numbers of IADL disabilities over time excluding those who died during follow-up
In post hoc analyses we repeated the analyses limiting them to participants free from any disabilities at baseline. The results remained the same comparing the No Decline and Sharp Decline groups, except social engagement was no longer significant. In the comparison of the Moderate Decline and the Sharp Decline groups, visuo-spatial composite, psychomotor speed, executive function, and naming were all nonsignificant. The total number of prescription medications remained significant, but, as with the No Decline group, social engagement no longer distinguished the Moderate Decline group from the Sharp Decline group.
In post hoc cross sectional analysis, executive function, indicated by Trails B, was the only significant variable (OR=0.28, 95%CI: 0.10-0.79) in the model where the outcome was disability in 3 or more IADL tasks. However, in the model where the outcome was disability in 2 or more IADL tasks, none of the cognitive domains was significant.