Abstract
Background:
Methods that can identify subgroups with different trajectories of cognitive decline are crucial for isolating the biologic mechanisms which underlie these groupings.
Objective:
This study grouped older adults based on their baseline cognitive profiles using a latent variable approach and tested the hypothesis that these groups would differ in their subsequent trajectories of cognitive change.
Methods:
In this study we applied time-varying effects models (TVEMs) to examine the longitudinal trajectories of cognitive decline across different subgroups of older adults in the Rush Memory and Aging Project.
Results:
A total of 1,662 individuals (mean age = 79.6 years, SD = 7.4, 75.4%female) participated in the study; these were categorized into five previously identified classes of older adults differing in their baseline cognitive profiles: Superior Cognition (n = 328, 19.7%), Average Cognition (n = 767, 46.1%), Mixed-Domains Impairment (n = 71, 4.3%), Memory-Specific Impairment (n = 274, 16.5%), and Frontal Impairment (n = 222, 13.4%). Differences in the trajectories of cognition for these five classes persisted during 8 years of follow-up. Compared with the Average Cognition class, The Mixed-Domains and Memory-Specific Impairment classes showed steeper rates of decline, while other classes showed moderate declines.
Conclusion:
Baseline cognitive classes of older adults derived through the use of latent variable methods were associated with distinct longitudinal trajectories of cognitive decline that did not converge during an average of 8 years of follow-up.
INTRODUCTION
Emerging evidence demonstrates both clinical and biological heterogeneity in cognitive aging [1–6]: While some individuals remain relatively stable, many show steady or step-wise declines in various, or all, domains of cognitive performance [7]. Latent variable methodology has repeatedly shown that specific cognitive patterns of impairment exist across different subgroups of individuals [6, 8–10]. The application of this methodology has resulted in significant improvements in diagnostic and prognostic rigor [6, 11].
In independent studies of samples with clinically-diagnosed mild cognitive impairment (MCI), patterns representing amnestic, dysexecutive, and mixed impairment subtypes have been identified in cross-sectional studies [5, 6]. In our prior work on two independent samples of community-dwelling older adults [3, 11], we identified five classes with distinct cognitive profiles based on their baseline evaluation. We identified two classes without impairment: Superior Cognition and Average Cognition, and three classes with impairment: Memory-Specific Impairment, Frontal Impairment, and Mixed-Domains Impairment. Individuals in the Mixed-Domains Impairment class performed poorly on all neuropsychological measures, while individuals in the Memory-Specific Impairment class performed poorly on episodic memory measures; the third class consisted of individuals who performed poorly on just tests of frontal executive function (the Frontal Impairment class), while the remaining two classes consisted of individuals who scored average relative to the rest (the Average Class), and of individuals who showed superior performance (the Superior Cognition class). The largest class in both samples was the Average class (46.1%in MAP), which was expected because the samples consist of community-dwelling participants. Despite superficial similarities to pre-established criteria of MCI, our classes were not assigned based on cut-scores on individual tests and thus, provided more nuanced characteristics, and were shown to accurately and reliably predict conversion rates to dementia [9]. We also showed that these groups had different rates of progression to dementia [3, 11]. Lastly, in studies by ourselves and others, profiles also showed distinct brain-behavior associations. For example, in one study by a different group, the dysexecutive subtype was associated with white matter pathology [12], while the amnestic subtype was associated with cortical thinning [6]. We also found differential associations between latent class profiles and postmortem indices of brain pathology [13]. Deceased and autopsied participants who, upon enrollment at approximately 7 years prior to death, had belonged to the Mixed-Domains Impairment and Memory-Specific Impairment classes, were more likely to have higher amyloid- β load (Aβ) and tau density, while the Frontal-Impairment class was associated with arteriolosclerosis [13].
Thus, despite existing heterogeneity in manifestations of cognitive impairment in older adults, distinct groups have been identified based on their initial cognitive profiles. Prior work finds that specific patterns of cognitive impairment are associated with different combinations of pathology in the brain. Indeed, mounting evidence suggests that cognitive impairment may be detectable via neuropsychological testing at preclinical phases of clinical diagnoses [14]. Therefore, if these groups are clinically relevant, their baseline profiles should predict distinct trajectories of cognitive decline. Although there is some prior work that has identified longitudinal trajectories of differential cognitive aging [7, 16], and validated the resultant trajectories using structural brain indices and clinical diagnoses, such work was longitudinal from the outset, i.e., prior work did not analyze baseline assessment, and only included one domain of cognitive function [15, 16], or only one global measure of cognition [7].
The aim of this study was to test the hypothesis that the latent cognitive profiles identified at baseline in our prior work continue to be associated with different patterns of cognitive decline longitudinally. We also examined which cognitive abilities are driving decline within each subgroup. Our hypotheses were that the Mixed-Domains Impairment class and the Memory-Specific Impairment class would have the steepest rates of decline driven mainly by the episodic memory and semantic memory domains, that the Frontal Impairment class would have the highest rates of decline in the domains of perceptual speed, perceptual orientation, and working memory, and the Average and Superior Cognitive classes would show minimal or no decline. These data are important for demonstrating the potential clinical utility of these groupings.
MATERIALS AND METHODS
Participants
This study included participants from the Rush Memory and Aging Project (MAP), which involves older individuals recruited from the greater Chicago area [17, 18]. Older persons without known dementia consented to annual clinical evaluation and signed an informed consent and an Anatomical Gift Act for organ donation at the time of death, and a repository consent that permitted data to be repurposed. The study was approved by Institutional Review Boards of Rush University Medical Center and the Albert Einstein College of Medicine. MAP data can be requested at http://www.radc.rush.edu. To be consistent with our prior studies [3, 13], we included participants: 1) without dementia at the baseline assessment, and 2) who had at least one follow-up visit with valid cognitive assessment. We also selected neuropsychological tests that were analogous to our prior studies [4].
At the time of these analyses, data on 1,924 MAP participants were available. Of these, 79 participants were excluded due to diagnosis of dementia at baseline, and 183 were excluded because they did not have a follow-up visit due to enrollment within the prior year, death, or withdrawal. A total of 1,662 individuals were included in these analyses.
Neuropsychological battery
Up to 21 waves of longitudinal neuropsychological data were available for the Rush MAP cohort. Each participant underwent annual structured neuropsychological evaluations that included 21 tests, which assess a broad range of cognitive processes that are affected by aging and dementia. The standardized scores within the cohort were computed by converting raw scores on each of the cognitive measures into z-scores, using the mean and standard deviation from the baseline evaluation, and then averaged to yield a summary score of the global cognition, with a mean of 0 and a standard deviation of 1, with higher scores indicating better cognitive performance. Specifically, composite scores for the following domains of cognitive function were formed: episodic memory (Word List Memory, Word List Recall, Word List Recognition, Immediate Story Recall, Delayed Story Recall, Logical Memory I, Logical Memory II); semantic memory (Boston Naming Test, Verbal Fluency, Reading Test), working memory (Digit Span Forward, Digit Span Backward, Digit Ordering), perceptual speed (Symbol Digits Modalities Test, Number Comparison, Stroop color naming, Stroop word reading), and Perceptual Orientation (Judgment Line Orientation, Standard Progressive Matrices. More detailed information can be found in our prior work [19–21].
Statistical analyses
Latent class analysis
We previously fit latent class analysis (LCA) models in 1,662 MAP participants on baseline performance of 10 neuropsychological indicators [11, 13] (episodic memory: Logical Memory [22] and Word list Recall [23]; semantic memory: Boston Naming Test [24] and Category fluency [25]; working memory: Digits [26] and Digit Ordering [27]; perceptual orientation: Progressive Matrices [28] and Line Orientation [29]; perceptual speed: Symbol Digits Modalities Test [30] and Number Comparison). Baseline age, sex, and education were included in the LCA models. A five-class model was deemed optimal based on fit indices, including the Bayesian information criterion (BIC) and entropy. The five profiles reflected impaired cognition (Mixed-Domains Impairment, Memory-Specific Impairment, Frontal Impairment) and intact cognition (Superior Cognition and Average Cognition) (see [3, 11], and Supplementary Figure 1), suggesting that applying categorical models of cognitive function (as opposed to continuous models) could be a useful tool that distinguishes between normative and pathological aging on the basis of characterization rather than severity. After identification of the classes, participants were assigned to their most likely class based on posterior probabilities. We then characterized and validated our model using pre-existing characteristics to determine if the classes are distinguishable on core neuropsychological characteristics and external validators including vascular risk and postmortem neuropathology [13]. Consistent with previous findings, we observed that participants belonging to classes with a poorer cognitive profile (Mixed-Domains Impairment and Memory-Specific Impairment) also had a higher prevalence of vascular disease, including diabetes and hypertension [11]. More details on our previous findings are provided in prior publications [3, 11].
Time-varying effects models
Time-varying effects models (TVEM) were applied to the data to explore dynamic effects of the latent classes at baseline (predictors) on their longitudinal cognitive trajectories (outcomes). TVEM is a novel, flexible statistical method that does not assume any functional forms (such as linear or quadratic) for the effects of predictors or for the longitudinal outcomes [31–34], and has been previously applied in behavioral and psychological studies, including in cognitive aging research [35]. Since prior studies have shown that cognitive aging is a non-linear process [36], this model allows us to capture complex and constantly changing cognitive trajectories, including their coefficients and confidence bands for the association between each latent class at enrollment and its trajectory of decline over annual assessments. TVEM is well-equipped in studying how time effects unfold over days, months, and years, hence focusing on age-varying associations. Although we have up to 21 years of follow-up for some participants, the sample size for some classes (e.g., the Mixed-Domains Impairment class) were smaller at the outset (baseline N = 71, 4.3%), and due to loss to follow-up, selective attrition, and death, and the number of participants in each class decreased over time. Because power is reduced with 8 years of follow-up (see [11] for more information), we chose to study up to 8 years of follow-up to retain power; however, for comprehensiveness, we also present results using the entire follow-up period in the Supplementary Material. In this study, “time” is the number of years from first evaluation. Our primary outcome was longitudinal cognition over the 8 years of follow-up, and the primary predictor was class-assignment at enrollment. The Average Cognition class was treated as the reference class since it contained the most individuals (n = 767, 46.1%). In other words, the TVEM analyses in this study showed the time-varying difference in cognition of each individual class compared to the Average Cognition class. The time-varying effect model can be expressed as:
where y ij is the cognition of individual i at time j, β 0 (t ij ) is the mean cognition over time for female subjects with average baseline age and average education, β 1 (t ij ), β 2 (t ij ), and β 3 (t ij ) represents the coefficient function of time for age, sex, and education, respectively. COG-PRO it represents the cognitive class assignment for individual i at time (year) t, and β 4 (t ij ),. . . 7 (t ij ) is the coefficient function for the Mixed Domains Impairment class, the Memory Specific Impairment class, the Frontal Impairment class, and the Superior class, respectively. Baseline age and education were centered in the analysis to facilitate interpretation of the intercept function. All variables were allowed to have time-varying effects. Although TVEM is also flexible in treating missing observations, we had complete data on age at baseline, gender, education, and latent class assignment at baseline.
All models were run using the P-spline method for model estimation which can automatically choose the optimal number of knots. Standard errors of the estimates were calculated using a ‘sandwich’ formula. Analyses were programed in SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and R version 4.0.0 (https://www.r-project.org/).
RESULTS
Characteristics of the study participants
Participants (n = 1,662) included 1,253 (75.4%) females, had a mean baseline age of 79.6 years (standard deviation (SD) = 7.4), and a mean education of 14.8 years (SD = 3.2). Mean follow-up was 4.5 (SD = 3.9) years. Descriptive data stratified by latent class-assignment at baseline can be found in Table 1. Briefly, there were 71 participants (4.3%) in the Mixed Domains Impairment class, 274 participants (16.5%) in the Memory-Impairment class, 767 participants (46.1%) in the Average Cognition class, 222 participants (13.4%) in the Frontal Impairment class, and 328 participants (19.7%) in the Superior Cognition class.
Baseline demographic and neuropsychological characteristics of the participants from the Memory and Aging Project (n = 1,662)
SD, standard deviation.
Figure 1 illustrates the mean level of cognition (black dotted line), and trajectories over time for each domain estimated by the model for a female belonging to the Average Cognition class with an average age at baseline of 79.6 years and average education of 14.8 years.
The mean trajectory of global cognition and the five domains, including episodic memory, semantic memory, working memory, perceptual speed, and perceptual orientation. Results were estimated using TVEM for a female in the Average Cognition class, with a baseline age of 79.6 years and 14.8 years of education (average values for this group).
Associations of baseline class-assignment on cognitive trajectories
Mean cognition
The mean trajectory of cognition in each of the cognitive domains for each of the five classes over the eight years of follow-up can be seen in Fig. 2. The trajectories show that over time, differences in each of the cognitive domains across the classes are sustained through the eight years of follow-up. Specifically, the Superior Cognition and Average Cognition classes continue to perform at superior or average levels relative to the other classes, while the Frontal Impairment, Memory-Specific Impairment, and Mixed-Domains Impairment classes continue to perform poorly over time. For global cognition, there is a slow decline across all the classes with minor differences in the rate of decline (e.g., the Mixed domains Impairment class shows steeper declines). For episodic memory, the Memory-Specific and Mixed-Domains Impairment classes show similar trends of low and declining performance. For semantic memory, there are steeper declines for the Memory-Specific class. For working memory, the Memory-Specific and the Frontal Impairment classes show indistinguishable patterns of decline, which may suggest that these classes have similar performance on working memory. All of the classes show declines of perceptual speed, while Mixed-Domains Impairment show steeper decline of perceptual orientation.
Marginal raw means of global cognition, and each of the five cognitive domains for each class over 8 years of follow-up. Baseline Ns for each of the classes are as follows: Mixed-Domains Impairment = 71; Memory-Specific Impairment = 274; Frontal Impairment = 222; Average Cognition = 767; Superior Cognition = 328.
Time-varying effects of class assignment
To directly compare the cognitive trajectories for each group defined at baseline, we applied TVEM to compare the trends of cognition in the other 4 classes, compared with the Average Cognition class (i.e., the biggest class) (Fig. 3). We describe below the results for up to 8-years’ of follow up. Complete TVEM results can be found in Supplementary Figure 2.
The relative rate of mean change for cognition (the panels) for all class trajectories relative to the rate of change in the Average Cognition class. The red line and the shaded blue are the pointwise estimates and 95%confidence bands. The horizontal black line is the reference line at zero. All models have been adjusted for age at baseline, sex, and education.
Global cognition
As shown in Fig. 3, compared to the Average Cognition class, all the classes displayed significantly different effects on global cognition. Specifically, for the Mixed Domains Impairment class, the effect on global cognition became more prominent throughout the first five years of follow-up, with the largest effect observed at five years from baseline (est. = –1.56, 95%CI = –1.90, –1.23, SE = 0.17) (Fig. 3, panel 1, “Effect on global cognition”). Differences between the Average Cognition class and the Memory-Specific Impairment class were evident beginning at about 1 year of follow-up (est. at 1 year = –0.75, 95%CI = –0.81, –0.70, SE = 0.03). While significant, the gaps between the Average Cognition class and the Superior Cognition class, and the Average Cognition class and the Frontal Impairment class, remained stable throughout the duration of the follow-up.
Episodic memory
Each of the classes had significantly different effects on episodic memory during follow-up when compared to the Average Cognition class, with some time-points showing larger effects. Specifically, for the Mixed-Domains Impairment class, differences from the Average Cognition class became larger over time until about four years from baseline (est. at four years = –1.39, 95%CI = –1.75, –1.02, SE = 0.18) (Fig. 3, panel 2, “Effect on episodic memory”). Relative to the Average Cognition class, the effect on episodic memory was most prominent at about four years from baseline for both the Memory-Specific Impairment class (est. = –1.15, CI = –1.29– –1.01, SE = 0.07) and the Frontal Impairment class (est. = –0.28, CI = –0.41, –0.15, SE = 0.07), but remained relatively stable thereafter. The gap between the Average Cognition class and the Superior Cognition class remained stable for the duration of the follow-up period.
Semantic memory
Similarly, the effects on semantic memory for all the classes relative to the Average Cognition class were significantly different throughout follow-up. Specifically, for the Mixed-Domains Impairment class, large differences were seen until about four years from baseline (est. = –2.0, 95%CI = –2.23, –1.58, SE = 0.17). By contrast, for the Memory-Specific Impairment class, relative to the Average Cognition class, differences increased during the whole 8 years of follow-up (est. at 8 years = –1.00, 95%CI = –1.26, –0.75, SE = 0.13) (Fig. 3, panel 3, “Effect on semantic memory”). The Frontal Impairment class and the Superior Cognition class showed relatively stable differences from the Average Cognition class during the 8 years of follow-up.
Working memory
For working memory, performance at all follow-ups were significantly different between the Average Cognition class and the rest of the classes, with the Mixed-Domains Impairment and the Frontal Impairment Classes showing differences throughout follow-up (Mixed-Domains Impairment: est. = –1.00 at 1 year of follow-up, 95%CI = –1.18, –0.83, SE = 0.09; est. = –1.03 at 8 years of follow-up, 95%CI = –1.40, –0.65, SE = 0.19); Frontal Impairment: est. = –0.53, 95%CI = –0.62, –0.44, SE = 0.05, est. = –0.63, 95%CI = –0.78, –0.47, SE = –0.47 for the corresponding time point, respectively) (Fig. 3, panel 4, “Effect on working memory”). The Memory-Specific Impairment class showed slightly larger effects compared to the Average Cognition class from 4 years of follow-up onwards (est. = –0.64, 95%CI = –0.75, –0.53, SE = 0.05 at 4 years of follow up; est. = –0.71, 95%CI = –0.89, –0.53, SE = 0.09 at 8 years of follow-up). The Superior Cognition class had stable and significantly better performance compared to the Average Cognition class at all time-points.
Perceptual speed
Relative to the Average Cognition class, all classes had significantly different effects on perceptual speed throughout follow-up. The Mixed-Domains Impairment class showed the most prominent differences at about 5 years of follow-up (est. = –1.70, 95%CI = –1.93, –1.47, SE = 0.12) (Fig. 3, panel 5, “Effect on perceptual speed”). The Memory-Specific Impairment class showed gradual differences from the Average Cognition class (est. = –0.65, 95%CI = –0.69, –0.52, SE = 0.04 at 1 year of follow-up; est. = –0.76, 95%CI = –0.95, –0.58, SE = 0.09 at 8 years of follow-up). An opposite trend was seen for the effects of the Frontal-Impairment Class, which showed gradual reduction in the differences relative to the Average Cognition class (est. = –0.81, 95%CI = –0.90, –0.72, SE = 0.05 at 1 year of follow- up; est. = –0.73, 95%CI = –0.85, –0.60, SE = 0.06 at 8 years of follow-up).
Perceptual orientation
While all the classes showed significant differences in performance at all follow-up time-points when compared to the Average Cognition class, participants in the Mixed-Domains Impairment class showed the largest effects, with the most prominent differences from the Average Cognition class taking place at 7 years of follow-up (est. = –1.52 at 7 years of follow-up, 95%CI = –1.92, –1.12, SE = 0.20) (Fig. 3, panel 6, “Effect on perceptual orientation”). Relative to the Average Cognition class, the Memory-Specific Impairment class also showed differences in perceptual orientation (range: est. = –0.37 at 1 year of follow-up, 95%CI = –0.46, –0.28, SE = 0.05; est. = –0.57 at 8 years of follow-up, 95%CI = –0.77, –0.37, SE = 0.10). Relative to the Average Cognition class, the Frontal Impairment class differed with a slight improvement at 3 years of follow-up (est. = –0.68 at 3 years of follow-up, 95%CI = –0.78, –0.58, SE = 0.05), and maintained this trend until the end of follow-up.
DISCUSSION
In this study we characterized the time-varying dynamic effects of baseline class-assignment on cognitive aging trajectories over 8 years of follow-up using time-varying effects models. This work extended previous cross-sectional work by incorporating a longitudinal design to explore the trajectories of the baseline classes over-time. Our results further reinforce the applicability of empirical identification of specific types of cognitive impairment.
Our results showed that: 1) Baseline cognitive groupings of older adults derived through the use of latent variable methods are associated with distinct longitudinal trajectories of cognitive decline that do not converge during an average of 8 years of follow-up for our primary analysis, 2) these differences are maintained for up to 21 years of follow-up, even as our data becomes sparse, and 3) the differences in patterns of change in cognition are subtle when compared to the large differences at study baseline. When comparing the classes relative to the effects of the Average Cognition latent class, the Mixed-Domains, and Memory-Specific Impairment classes had larger effects on all cognitive domains, with the largest effects taking place at around five years from baseline for both classes (Fig. 3), while the Frontal Impairment class only showed stable effects on Working Memory from about six years into follow-up. The Superior Cognition class remained stable relative to the Average Cognition latent class for the duration of follow-up. These findings have important translational implications and underscore the importance of further work elucidating the mechanisms and factors that may account for differences among subgroups and subsequent patterns of cognitive decline.
First, our results extend prior work using an independent outcome. We previously showed that latent class membership based on baseline cognitive data predicted time to incident dementia and AD [11]; we now show that these same cognitive subgroups exhibited distinct trajectories of cognitive decline. This suggests that while these classes were constructed based on cognitive data from a single point in time, the factors which contributed to these initial groupings reflect enduring aspects of health and disease that have a sustained influence on cognitive course. While the Frontal and Superior Cognition classes reflected a course of stable cognitive function or slow decline relative to the Average Cognition class, participants in the Memory-Specific and Mixed-Domains Impairment classes showed more rapid decline affecting all cognitive domains. These results are also consistent with literature showing that multidomain, as well as memory impairment early on, is associated with dynamic dedifferentiation [37], more rapid decline [16, 38] and worse outcomes [11, 39], including terminal decline [40].
Second, the longitudinal patterns of cognitive decline based on class-assignment at enrollment continued to differ over time. This raises the possibility that there may be distinct underlying biological mechanisms which initiate or maintain these ongoing differences [11]. There is a considerable body of literature documenting a pattern of selective preservation and dissociable impairment for episodic memory and executive function associated with AD pathology and vascular pathology up to five years prior to death [41–43]. Our group and others have shown that class-assignment based on neuropsychological assessment at enrollment are associated with different imaging patterns of brain structure [44], white matter lesions [12], and an ordered degenerative and cerebrovascular neuropathological pattern that coincides with the profiles of increasing cognitive impairment up to seven years prior to autopsy [1, 39]. The longitudinal patterns of results in the current study further support cross-sectional findings in that the classes did not converge but rather continued to display differential patterns of decline.
Third, different groupings of cognitive impairment and associated trajectories of decline suggest the utility of more granular cognitive testing, as accurate groupings have the potential to aid clinicians and caregivers for future decision-making in allocating resources for medical care and support. These results may have translational consequences for aging research by suggesting that administering a comprehensive neuropsychological assessment may result in better characterization of cognitive phenotypes and improve stratification of older adults at risk of different patterns of cognitive decline. Risk stratification offers potential to improve eligibility criteria for clinical trials designed to develop targeted interventions because the subgroups are stable and specific. Complex statistical modelling has become more interpretable, whereby estimation of risk scores prior to clinical onset of disease is becoming more widespread [45–49]. An approach that integrates neuropsychological assessment with latent variable modelling, places individuals on a risk spectrum, defining the level and profile of impairment, and potentially offering a personalized medicine approach to treatment. Our results identified points in time in which different subgroups of individuals showed the steepest rates of decline, suggesting that the optimal window for clinical interventions may vary in the different subgroups. Our prior work suggested that when compared to individuals in the Average Cognition class, individuals in the Mixed- and Memory-Specific Impairment classes were at a higher risk of converting to dementia within the first five years of follow-up [3]. Results in the current study supported this finding by showing that the steepest declines for the Mixed-Domains and Memory-Specific Impairment took place in domains of episodic memory, semantic memory, and perceptual speed in the first five years of follow-up. The application of TVEM can help identify critical periods of risk to inform on the optimal timing of targeted interventions for specific groups of individuals [32].
Strengths are noted. Participants in the study underwent annual cognitive assessments allowing us to reliably characterize their trajectories over time. The Rush MAP cohort has very high rates of follow-up. We applied a novel technique, TVEM, which provided us with a unique opportunity to flexibly evaluate the effects of baseline class assignment on the trajectories of cognitive decline. One of the advantages of TVEM is that this modeling approach can accommodate non-linear change in cognition, with opportunities to identify critical time-windows for potential intervention strategies. Our study also has some limitations. First, our findings were based on a particular cognitive battery, which is broader than what may be available in clinical care. In addition, the cognitive battery did not contain some tests that are widely used to assess executive function (see [50, 51]; e.g., the Trail-Making Test B [52], and Wisconsin Card Sorting Test [53]). However, our working memory, perceptual speed, and perceptual orientation measures tap frontal executive functions. As there is no universally agreed upon cognitive battery for use in older adults, this limitation seems inevitable. We note that the basic latent class structure has been identified in the Einstein Aging Study, which measured the same domains but used different instruments than the ones used in the Memory and Aging Study [3, 11]. This suggests that our work is generalizable to different samples using distinct batteries. Cohort-differences may have impacted the results in terms of initial class-assignment and their trajectories [54]. Specifically, a large proportion (36%) of individuals in the Superior Cognitive class were born between 1936 and 1945, while a large proportion of individuals in the Mixed-Domains Impairment class were born in different birth cohorts (21.1%were born between 1905 and 1914, and 21.1%were born between 1925 and 1930). Further, 24.8%of participants in the Memory-Specific Impairment class were born between 1915 and 1919, and another 24.8%were born between 1920 and 1924. Effects of birth cohort on late life cognition and health warrant further investigation.
Further, we demonstrated in a previous study that some participants move across classes over time [9], and although the majority remain in their baseline class assignment, the trajectories of individuals who move may affect the results. We used continuous cognitive measures to identify and assign individuals to categorical latent variables; however, given previous findings implicating that specific patterns of cognitive impairment are driven by types and combinations of pathology in certain brain regions, it is reasonable, in this context, to consider cognitive impairment and decline as a categorical latent variable. Lastly, we did not take death or terminal decline into account in this study.
To the best of our knowledge, this is the first study to show that the trajectories of cognitive decline differ in community-dwelling older adults with different initial cognitive profiles that were derived using a latent variable approach. Despite the varied person-specific differences and heterogeneity of late-life cognitive impairment, our approach offers the potential of identifying more homogeneous groups of older adults with similar initial cognitive profiles and patterns of cognitive decline. Further validation of neuropsychological classes using biological markers is particularly important for future research in attempts to bridge links between cognitive phenotypes and their underlying etiologies and is especially crucial for developing targeted treatments. Much of late-life cognitive impairment is unexplained; genetic and epigenetic characteristics may influence levels of cortical proteins which provide resilience but may not manifest a known pathologic footprint [55]. Further work is needed to identify these factors which may account for these class-specific trajectories of cognitive decline observed in this study.
Footnotes
ACKNOWLEDGMENTS
The authors thank the many participants in the Rush Memory and Aging Project; Traci Colvin, MPH, for coordination of the clinical data collection; Karen Skish, MS, for coordination of the pathologic data collection; and John Gibbons, MS, and Greg Klein, MS, for data management.
This work was supported by the Memory and Aging Project (R01AG17917, R01AG343749, and R01AG42210) from the National Institute on Aging, the Einstein Aging Study (P01 AG03949) from the National Institutes on Aging program. Andrea R. Zammit was funded by the National Institute on Aging of the National Institutes of Health under Award Number K01AG054700, and by the Sylvia and Leonard Foundation. Aron S. Buchman was supported by the National Institutes on Aging under award number R01AG056352; Sue E. Leurgans was supported by the National Institutes of Aging under award number P30AG010161 (PI: DAB). The contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.
