Among 93 participants, the distribution of cognitively intact participants and participants with aMCI and naMCI was 54, 8, and 31, respectively. Their characteristics are given in . Participants with aMCI and naMCI differed in in-home walking speed during the baseline week, suggesting that trajectories in walking speed between these 2 types of MCI are potentially different. Therefore, we did not combine the 2 types of MCI in the statistical analysis, but focused on naMCI, as this latter group had sufficient sample size.
Participant demographics and baseline clinical characteristicsa
Participants with naMCI had lower (worse) Tinetti balance scores and slower mean in-home walking speeds during the baseline week compared with cognitively intact participants. As expected, all neuropsychological test scores were different between the cognitively intact participants and the participants with naMCI. CIRS, body mass index, UPDRS, FAQ scores, and height were not different between the 2 groups. By the third annual assessment, 2 participants were lost to follow-up and 4 participants were deceased (7.0%). These participants were included in the analyses using the available information until their drop out in the framework of mixed-effects latent trajectory models.
Trajectory analyses of walking speed.
Trajectory analyses identified 3 distinct groups of walking speed trajectories as the best model based on the BIC (A). The procedure calculates the probability of each participant belonging to each trajectory and identifies a participant as belonging to one trajectory based on the highest probability. A shows the estimated trajectories, using the model-assigned group identification for each participant. Based on the shapes of walking speed trajectories, we named them fast, moderate, and slow walkers, respectively (trajectory lines from top to bottom in A). The figure shows that speed is relatively stable with a slight decline (i.e., getting slower) over time for the first 2 groups, but the slow walker group showed a steeper declining trend. shows the association between baseline characteristics including MCI status and the 3 trajectories with the slow group as the reference, using multinomial logit models estimated jointly with the trajectory model. The table can be read as follows. For example, the odds of belonging to the fast group is reduced by approximately one-tenth (e−2.17 = 0.11) compared with the odds of belonging to the slow group if the participant has naMCI, after controlling for age, gender, education, and Tinneti balance scores (p = 0.01). In other words, the odds of belonging to the slow group is 9 times higher (1/e−2.17) if the participant has naMCI. Likewise, the odds of belonging to the slow group is approximately 5 times higher (1/e−1.69 = 5.4) compared with the odds of belonging to the moderate group if the participant has naMCI (p = 0.01). MCI status was the only significant covariate in the model. By using the model-assigned group identification for each participant based on the results in the multinomial logit model, the proportion of naMCI among the fast, moderate, and slow trajectories was found to be 16.7%, 34.6%, and 66.7%, respectively (Pearson χ2 test, p = 0.01), indicating that more than half of the slow walker trajectory was populated by participants with naMCI.
Trajectories of in-home walking speed and variability based on latent trajectory analyses
Results of multinomial logit model showing the association between walking speed trajectories and baseline characteristics
Walking speed trajectories as depicted in A showed linear trajectories without any overlap. Therefore, as a post hoc analysis, we used mixed-effects models (random intercept) with an indicator variable of cognitive status (naMCI vs cognitively intact) and its interaction with time to estimate the average difference in the amount of decline between the 2 groups. The results are shown in . Participants with naMCI showed additional decline in walking speed (m/second) by 0.0006 point (p < 0.0001) per week, meaning approximately 10% shorter m/second in 181 weeks (approximately 3.5 years) (e(−0.0006×181) = 0.90). For example, a baseline speed of 0.6 m/second means it takes 50 seconds to walk 30 m (30/0.6 = 50). Given that a cognitively intact participant would remain at this walking speed, the counterpart with MCI would take about 56 seconds to walk 30 m (30/0.54 = 55.6), i.e., requiring about 6 seconds longer to walk the same 30-m distance. Controlling for race and height did not change the result.
Mixed-effects models with outcome being log of mean weekly walking speed
Trajectory analysis of COV of walking speed.
The trajectory analysis identified 4 distinct trajectory groups of walking variability (COV) as the best model based on the BIC. B shows the estimated trajectories. Group 1 can be characterized as having the highest COV at baseline followed by a further increase in COV and then sharply declining COV. Groups 2 and 3 were in the middle in COV at baseline, and their COV remained relatively stable with only a slight increase over time. Finally, group 4 started off with the lowest COV and experienced decreasing COV over time. The results of the association between baseline characteristics and the 4 trajectories based on multinomial logit models estimated jointly with the trajectory model are found in table e-1. The results showed that participants with naMCI were less likely to be in the groups with stable COV but were more likely to be in either the group with the highest (group 1) or the lowest COV at baseline (group 4). By using the predicted group assignment for each subject, the proportion of naMCI among each trajectory were found to be 53.3%, 33.3%, 16.7%, and 61.5% for groups 1, 2, 3, and 4, respectively (Pearson χ2 test, p = 0.02).