Overall, the present study used latent growth modeling to examine baseline and longitudinal change in five cognitive factor scores derived from the ADNI neuropsychological battery in healthy controls and MCI patients. At baseline, compared with controls, MCI patients demonstrated lower performance on all of the five cognitive factors (memory, executive function/processing speed, language, attention and visuospatial). Both controls and MCI patients declined on memory over 36 months; however, the MCI patients declined at a significantly faster rate than controls on memory. The MCI patients also declined over 36 months on the remaining four cognitive factors (i.e., executive function/processing speed, language, attention and visuospatial). In contrast, the controls did not exhibit change on any of these non-memory cognitive factors over 36 months, which differed significantly from the MCI patients. Within the MCI group, executive function declined faster than memory, while the other factor scores changed at the same rate as memory over time. The results also suggest that executive function/processing speed declines faster than other cognitive factors, including memory, in patients with MCI. Thus, these findings suggest different patterns of cognitive decline in healthy older adults and MCI patients both at baseline and over time (e.g. diagnostic group by cognitive domain interactions).
As expected, MCI patients had lower performance on tests of memory at baseline in the ADNI study. As discussed above, the inclusion criteria for the ADNI study require that MCI patients have impaired memory and controls have intact memory performance. It is also not surprising that the MCI patients had lower baseline factors scores than controls on the other cognitive domains, as several studies suggest that MCI patients have deficits in non-memory domains relative to healthy controls (Hodges et al. 2006
; Kramer et al. 2006
; Matsuda and Saito 2009
). An international work group outlined criteria for both single- and multiple-domain amnestic MCI (Winblad et al. 2004
), and it is likely that the MCI patients in the ADNI study are classified as multiple-domain amnestic MCI rather than single-domain MCI according to this classification. Indeed, the memory cut-off on the Logical Memory subtest used for inclusion in ADNI is conservative, which likely identified MCI patients who are more impaired than the average MCI patient and have deficits in multiple cognitive domains. Thus, the baseline results in the current study are consistent with other studies suggesting that MCI patients often have deficits in multiple cognitive domains.
When considering change in cognitive performance over 36 months, the results from the latent growth models suggest that memory declined in both MCI patients and healthy controls. However, the rate of memory decline was different, with the MCI patients declining faster than controls. These results confirm other studies that document age-related memory decline in healthy older adults (Albert et al. 1995
; Hayden et al. 2011
; Wilson et al. 2002
) and also other studies that document a steeper slope of memory decline in MCI patients compared with healthy controls (Bennett et al. 2002
; Mungas et al. 2010
). One recent study using random effects regression analysis reported that a majority (65 %) of healthy older adults declined at approximately 0.04 SD annually (equivalent to 0.4 factor units in the present study) on a global cognitive composite score (Hayden et al. 2011
). Thus, the current findings about memory decline in both MCI and controls derived from the latent growth models are consistent with most studies using different methods and approaches.
The novel finding of the current study is that, while memory declined in both MCI and healthy controls, only MCI patients declined significantly on the other four non-memory cognitive factors (i.e., executive function/processing speed, language, visuospatial, and attention). This finding suggests that decline in non-memory domains may be an important feature for distinguishing healthy older adults and persons with MCI. That is, monitoring decline on non-memory domains may be more predictive of clinical progression than a decline in memory. A recent paper by Gomar and colleagues (Gomar et al. 2011
), also using ADNI data, found that the change on the Trailmaking test B was a better indicator of conversion to Alzheimer's disease than change in memory. Although Carter and colleagues (Carter et al. 2011
) argue that deficits in semantic cognition appear before executive dysfunction in MCI patients, several other authors have proposed that executive dysfunction is the second cognitive domain to be affected in MCI patients who progress clinically (for review see (Perry and Hodges 1999
)). Other studies also suggest that the risk of converting to dementia is increased when multiple domains are impaired, including executive function (Albert et al. 2001
; Nordlund et al. 2011
). If only memory is assessed over time, it would be difficult to observe differences between healthy controls and MCI patients. Thus, examining non-memory domains may be a powerful tool for differentiating normal, age-related decline from cognitive decline due to underlying neuropathology.
Numerous studies suggest that healthy aging is associated with brain changes in both gray and white matter and also declines in several cognitive domains, including working memory and attention (Buckner 2004
; Hedden and Gabrieli 2004
; Raz et al. 1997
). Although progress has been made in the understanding of the relationship between changes in the brain and patterns of cognitive aging, the neuroanatomical basis of these age-related changes in cognition continues to be a topic of debate (Raz and Kennedy 2009
; Salthouse 2011
). It is possible that the lack of decline on non-memory cognitive factors over 3 years by the controls in the current study reflects the profile of healthier “successful” agers compared with more “typical” agers (Hsu and Jones 2012
; Negash et al. 2011
). Additional longitudinal studies using comprehensive neuropsychological batteries are needed to help identify which cognitive features are the best predictors of conversion to dementia versus healthy aging.
In addition, when comparing the rates of change of different cognitive factors within the MCI group, the latent growth models suggest that executive function/processing speed changed at a faster rate than memory in MCI patients. In contrast, there was no decline in executive function/processing speed in the healthy controls. This diagnostic group by domain interaction shows a significant difference in the magnitude of change between memory and executive function between MCI and healthy controls. This suggests that executive function may be a potentially more sensitive measure of cognitive decline due to underlying neuropathology and may be a useful tool to distinguish healthy aging and MCI. Future studies are needed to further test this hypothesis. In contrast, the trajectory of decline in language, visuospatial, and attention in MCI was similar to the decline in memory, while there was no change over time on visuospatial and attention factors relative to memory in healthy controls. These domains appear less helpful for distinguishing MCI patients and healthy controls. Finally, because memory declined in both groups, it may also be useful to control for the rate of memory change when examining decline in executive function in future studies.
The second-order latent growth models used in the current study were robust and addressed the theoretical question at hand. Other multivariate alternatives are also available (McArdle 1988
; Salthouse and Ferrer-Caja 2003
). For example, to bolster causal inferences, it is possible to estimate two-stage piecewise longitudinal growth models in which earlier growth in one process predicts future change in another process or developmental stage (Chou et al. 2004
). In the present study, we sought to compare trajectories among cognitive factors but do not make any causal attributions. It is also feasible to expect that mean levels and rates of decline in each cognitive domain do not describe trajectories of all participants equally well, and that there are subgroups of participants who decline faster or slower, depending on where they are in the pathological cascade of Alzheimer disease; such groups might be teased apart through growth mixture modeling (Leoutsakos et al. 2012
). In yet another variation of longitudinal structural equation modeling, a bivariate dual change score model might be used to explicitly explore whether change in one cognitive domain is associated with change in another domain (McArdle and Prindle 2008
); such models require equally spaced visits, which we do not have in ADNI unless we exclude the 6-month visit. These and other research questions should be pursued in future research to extend our knowledge of domain-specific cognitive decline in older adults.
Several limitations of the present study are important to mention. First, because of the way the indicators in the current study are scaled, the factor scores are not generalizable to other samples. We strictly compared relative change in various domains between healthy controls and MCI patients in the ADNI sample. Second, the latent growth models in the present study accommodated linear change in cognitive factors, and the model fit statistics in this study suggested this assumption was adequate. However, it is possible that certain tests declined faster or slower than others within a domain, or even that change in some tests demonstrated quadratic change. We did not explore indicator-specific trajectories because our goal was to make inferences at the level of cognitive domains and not individual tests. It is also important to keep in mind that we did not exclude participants in either group that had a change in diagnosis (e.g., converted to dementia or reverted to normal). The ADNI study also focuses on prodromal stage of Alzheimer's disease, so the results do not generalize to the MCI stages of non-Alzheimer's disease dementias. In addition, the homogeneous racial composition, high educational level, and age composition of the participants in ADNI limit the generalizability to other community samples. As discussed above, the controls in the study may represent a group of healthier “successful” agers compared with more “typical” agers (Hsu and Jones 2012
; Negash et al. 2011
), which also limits the generalizability to other community samples.
In conclusion, latent growth modeling appears to be a useful tool for investigating longitudinal change in neuropsychological performance in both healthy older adults and persons with MCI. The results of the latent growth models suggest that cognitive decline differs between healthy aging and MCI. The latent growth models also suggest that executive function, in particular, declines at a faster rate than memory in MCI patients. The findings also underscore the importance of examining non-memory cognitive decline for potentially differentiating healthy aging and MCI. It is important to replicate these results in other large studies, particularly those with more ethnic diversity, and to determine how the latent factors might also help predict conversion to dementia or functional decline.