IQ moderation of cognitive decline supports cognitive reserve and not brain maintenance

Participants

The participants were drawn from our ongoing studies at Columbia University Irving Medical Center: the Reference Ability Neural Network (RANN) study and the Cognitive Reserve (CR) study (Stern, 2009; Stern, Gazes, Razlighi, Steffener, & Habeck, 2018; Stern et al., 2014). Subjects were recruited primarily by randomized market mailing. An initial telephone screening determined whether participants met basic inclusion criteria (i.e., right-handed, English speaking, no psychiatric or neurological disorders, and normal or corrected-to-normal vision). Potentially eligible participants were further screened in person with structured medical and neuropsychological evaluations to ensure that they had no neurological or psychiatric conditions, cognitive impairment or contraindication for MRI scanning. Global cognitive functioning was assessed with the Mattis Dementia Rating Scale (Lucas et al., 1998), on which a minimum score of 130 was required for retention in the study. In addition, participants who met diagnostic criteria for MCI were excluded. The studies were approved by the Internal Review Board of the College of Physicians and Surgeons of Columbia University. Additional details about procedures can be found in previous reports (Habeck et al., 2016; Stern et al., 2014). This study is currently in the process of completing five-year follow-up on all participants using the same procedures. The current analysis included 254 participants who were assessed at both baseline and follow-up, and had data from at least one of the 23 cognitive tasks that comprise the reference abilities at either time point, in which 209 participants also had pre-post testing on cognitive tasks during fMRI studies. The demographic information for the participants is presented in Table 1. We did not find any systematic difference between the participants who had and had not completed the fMRI procedures. Two participants did not receive any MRI scans and were excluded from cortical thickness analysis.

Cognitive Tasks

To estimate each of the four RAs, three measures from out-of-scanner and three from in-scanner tasks were included in the models to ensure robust estimation of latent abilities.

Statistical Analysis

For descriptive statistics, mean, standard deviation, median, minimum, and maximum were reported for continuous variables and frequency and percent were reported for categorical variables. The correlation among demographic variables were examined using spearman’s correlation and two sample t-test assuming unequal variances between males and females.

Latent change score model

For all tests of the study hypotheses, a multiple indicator latent change score model (LCSM) (Kievit et al., 2018) was used as depicted in Figure 1. The LCSM models changes in the latent scores rather than the observed scores. We modeled the four RAs in the manner of traditional confirmatory factor analysis as described in previous studies (Salthouse et al., 2015; Stern et al., 2014). The factor loadings at baseline and follow-up were constrained to be the same. With LCSM, we tested two primary questions: 1) does performance on RAs change over time and are changes moderated by age; 2) do cognitive reserve proxies (IQ and education) moderate the changes beyond age. Moderation by age, sex, IQ, and education were tested by adding these factors to the LCSM model as shown in Figure 1. We report the parameter estimates of associations as well as the overall goodness of fit measures (CFI, TLI, RMSEA).

• Download figure
• Open in new tab

Figure 1.

Diagram for the latent change score analysis. Acronyms of the figures are listed by reference abilities. Memory: Logical Memory (LM), Word Order recognition (WOR), Paired Associates (PA), Selective Reminding Task - long-term storage sub-score (SRTlts), Selective Reminding Task - continuous long-term retrieval (SRTctrl), and Selective Reminding Task - the number of words recalled on the last trial (SRTlast); Fluid Reasoning: Paper Folding (PF), Matrix Reasoning (MR), Letter Sets (LSets), WAIS III Block design task (BD), WAIS III Letter–Number Sequencing test (LNSD), and WAIS III Matrix Reasoning test (W3MR); Processing Speed: Digit Symbol (DS), Letter Comparison (LC), Pattern Comparison (PC), Digit Symbol subtest from the Wechsler Adult Intelligence Scale-Revised (DSWAIS), Part A of the Trail making test (TMTA), and Color naming component of the Stroop (CNS); Vocabulary: Synonyms (SYN), Antonyms (ANT), Picture Naming (PN), vocabulary subtest from the WAIS III (VOC), the Wechsler Test of Adult Reading (WTAR), and American National Adult Reading Test (NART).

Demographic characteristics

Table 1 provides a summary of participant characteristics at baseline. Participants aged 20 to 80 years at baseline were followed for approximately 5 years on average. IQ was correlated with baseline age (ρ=0.31, p<0.001) and with years of education (ρ=0.51, p<0.001). IQ was also higher in males (Male: 119±7.5; Female: 116±8.7; t(249.94)=-2.58, p=0.01) while no sex difference was found for age and education (p’s>0.1).

Longitudinal Changes in RAs

As expected, even after controlling for their baseline RAs, the three RAs showed decline over time (Reasoning: β =-0.78, p<0.001; Memory: β =-0.15, p=0.025; Speed: β =-0.68, p<0.001), while vocabulary showed increase over time (β=0.26, p=0.007).

Age moderation of changes in RAs

For all reference abilities, we found age moderation of the changes in RAs after adjusting for the respective baseline RA (Table 2 and Figure 2)., As age increased there was accelerated decline in fluid reasoning (β=-0.77, p<0.001), memory (β=-0.28, p<0.001) and speed (β=-0.57, p<0.001). Vocabulary also showed a negative slope with older age, however, as shown in the bottom panel of Figure 2, its trajectory differed from the other RAs in that, while the magnitude of change is decreasing with age, participants continued to show improvement in performance until the peak at 67 years (β=-0.36, p<0.001). The effect remained even after adjusting for IQ, education, and sex for all four RAs.

• Download figure
• Open in new tab

Figure 2.

Top: Association between changes in RAs and baseline age. All models were adjusted for IQ, sex, and the three other baseline reference abilities. Bottom: Peak detection and inflection points for RAs. Decline in reasoning, speed, and memory were accelerated after the respective inflection points of the quadratic b-spline (red lines), while vocabulary showed more gradual improvement after age 30 and reached the peak at age 67 (blue line).

Sex, Education, and IQ moderation of changes in RAs

Moderation effects on changes in RAs were tested by adding each of the three factors, sex, education, and IQ, separately into the LCSM in Figure 1 to see if there is an association between each of these factors and changes in each of the RAs.

Sex and Education did not show significant moderation effect on the changes in any of the RAs (p’s>0.15). When age, IQ, education and sex were simultaneously entered in the same model, these variables were also not associated with changes in the four RAs (p’s>0.07).

For NARTIQ, even after adjusting for age and baseline RAs, higher IQ was associated with reduced decline in fluid reasoning (β=0.35, p<0.001) and memory (β=0.17, p=0.005), but not in speed (β=0.12, p=0.18). Statistical details are listed in Table 3 and illustrated in Figure 3. These effects remained unchanged after adjusting for education. IQ moderation of vocabulary changes was not tested due to strong collinearity between NARTIQ and the individual vocabulary tests.

• Download figure
• Open in new tab

Figure 3.

Association between changes in RAs and IQ. All models were adjusted for age and the other three baseline RAs.

Test of brain maintenance as a source of IQ moderation

To test whether brain maintenance explains the IQ moderation of cognitive changes, effect of the mean cortical thickness was removed from the cognitive variables before being tested in the LCSM from the previous section. Despite the residualization, the IQ moderation effect on cognitive changes remained the same. Changes in reasoning and memory stayed significant, demonstrating that IQ moderation of changes in the two RAs is independent of any possible brain maintenance effect.

Age moderation of cognitive changes

Our results are consistent with other large epidemiological aging studies in which participants received repeated cognitive testing over the course of two to nine years. As was found in our study, significant age-related declines have been observed in processing speed, memory, and fluid reasoning (Cornelis et al., 2019; Hughes et al., 2018; Salthouse, 2016) with the rate of decline increasing with older age (Singh-Manoux et al., 2012; Zaninotto et al., 2018). In particular, processing speed consistently been reported to show a large decline with advancing age whether age was treated as a categorical or continuous variable (Cornelis et al., 2019; Hughes et al., 2018). To further understand the trajectory of cognitive aging, we used change point analysis, and identified age 59 years to be the point after which processing speed shows accelerated decline, and age 58 years for reasoning. Interestingly, memory showed accelerated decline only after 70 years of age, over a decade later than processing speed and reasoning. The later accelerated decline for memory can serve as a clear delineation between normal and pathological aging, for which memory declines sharply much earlier in life.

Even though all three RAs began to decline in the 20’s, the rate of decline was stable until these change points, after which the rate of decline became steeper. The cause for the steeper decline is likely related to the rate of change in brain structural measures. The later inflection point for memory is consistent with previous brain structural findings. Vinke et al. (2018) conducted a large epidemiological study on the age trajectories of a number of brain structural changes. They examined the age at which each structural metric declined by two standard deviations (SD) from the mean at 45 years old. Hippocampal volume, the structure known for its essential role in memory (Squire, 1992), decline by two SD decades later than that for total brain volume and a number of other white matter metrics such as mean diffusivity, normal appearing white matter volume, and the log of white matter lesion volume. This is consistent with our finding that decline in memory accelerates later in life than other cognitive domains. The earlier decline for the white matter metrics could be consistent with the earlier inflection points for reasoning and processing speed since there is evidence that these two cognitive processes are more reliant on a larger network of white matter tracts: in one study larger number of white matter tracts were found to mediate age-related cognitive differences for these two RAs whereas only two major white matter tracts mediated age-related cognitive differences for memory (Gazes et al., 2016).

Of studies that examined vocabulary (Hartshorne & Germine, 2015; Salthouse, 2019; Singh-Manoux et al., 2012), our study demonstrated similar findings: vocabulary improved throughout most of adulthood. In a cross-sectional study, Hartshorne and Germine (2015) reported that abilities relying on accumulated knowledge, such as vocabulary and information subtest of the WAIS III, peak around 50 to 60 years old in a cohort of 48,537 participants ranging in age from 16 to 89 years. Our study observed the peak for vocabulary to be 67 years old, much later than that observed by Hartshorne and Germine, which, being a cross-sectional study, may have confounded cohort effects in their results whereas our study examined longitudinal data. However, since our data only consist of two timepoints, a baseline and the 5-year follow up, it also may have cohort effects imbedded in the model estimation. Further longitudinal observation is needed to pinpoint the peak performance age for vocabulary.

Previous studies have found smaller longitudinal decline in cognitive abilities than that seen in cross-sectional analyses (Singh-Manoux et al., 2012), with one study estimating that cross-sectional differences to be as much as 12 times the changes observed longitudinally (Cornelis et al., 2019). However, there are inconsistencies in the degree of longitudinal change observed across studies, ranging from maintained (Cornelis et al., 2019; MacAulay et al., 2018) to significant decline in cognitive performance (Hughes et al., 2018; Singh-Manoux et al., 2012). Two main factors likely contribute to these inconsistencies. First longitudinal findings could be influenced by differences in time between follow-ups, which range from two (MacAulay et al., 2018) to nine years (Hughes et al., 2018). Second, most of these longitudinal studies examined performance change in individual tasks, which can be overly influenced by task-specific idiosyncrasies, including measurement error, and might be more sensitive to practice effects. Our study adopted a moderate time frame between baseline and follow-up of five years and used six different measures to estimate each RA. Using the latent change score model, we were able to detect significant change in the latent RA, which is a more robust measure than an observed score. Therefore, using a moderate follow-up interval and performing a robust statistical method to examine cognitive changes, our results confirm previous findings that showed age related longitudinal decline in fluid reasoning, processing speed and memory, while vocabulary improves throughout most of adulthood then plateaus in the 60’s.

IQ moderation of cognitive changes

Despite a number of previous positive studies (Glymour et al., 2012; Pool et al., 2016; Zahodne et al., 2015), CR moderation of cognitive change has not been found consistently. The variability in findings may be attributed to a few key factors: differences in the metric used to estimate CR, the age range of the study sample, and the statistical approach for estimating cognitive change. Most of the studies that did not find CR effects used education either as the sole factor (Gottesman et al., 2014; Karlamangla et al., 2009; Proust-Lima et al., 2008; Zahodne et al., 2011) or as part of a composite score (Soldan et al., 2017), likely due to the ease in obtaining this information. However, years of education may not be an accurate proxy for CR since for middle-aged and older adults, year of education stabilizes in young adulthood and does not grow with an individual’s experience. Furthermore, years of education is usually self-reported, and can be inconsistent across individuals and studies, depending on whether the cumulative years of schooling or the equivalent years of education for degrees earned is reported. For life span samples that include young adults, years of education adds to the inaccuracy as many young adults have not yet completed schooling. Thus, it was not surprising that education in our study did not show significant moderation of cognitive decline.

In addition to education, we also tested whether NARTIQ, another commonly used proxy for CR, moderated cognitive changes. We observed that NARTIQ was associated with slower decline in reasoning and memory but not for processing speed. Moderation of changes in vocabulary was not examined due to the large overlap between NART IQ and vocabulary subtests. CR moderation of changes for the two higher order abilities but not for processing speed may be consistent with the fact that processing speed relies on more basic functions such as stimulus processing and motor programming, while reasoning and memory consist of higher order processes. CR moderation is hypothesized to exert protective effects through providing greater processing capacity and/or through compensatory processes that provide alternative networks when the primary network is damaged (Stern, 2012). These CR mechanisms would be able to alleviate cognitive decline for higher order RAs such as reasoning and memory but not for processing speed, which relies on more peripheral processes such as visual input and motor output.

Examining brain maintenance effect

Brain maintenance theory posits that genetics and life experiences can aid in the preservation of a healthy brain which in turn would contribute to less decline in cognition (Nyberg, Lovden, Riklund, Lindenberger, & Backman, 2012). This protective effect can only be distinguished from CR protective effects by accounting for measures of brain structural health in the CR moderation models. In our study, IQ moderation of changes in cognition was confirmed to be independent of any brain maintenance effect by demonstrating that the IQ moderation effect was unchanged even after removing the effects of mean cortical thickness from cognitive changes in the LCSM. Thus, with results showing slower decline for adults with higher IQ regardless of brain structural health, individuals with higher IQ likely possess superior functional processing of tasks, either in the form of greater processing efficiency, higher processing capacity, and/or more flexibility in task processing (Stern, 2012).

Limitations and conclusions

All longitudinal studies are susceptible to practice effects which can confound the observed cognitive changes also in this study. However, despite potential practice effects, robust cognitive decline was observed over five years for the three abilities across the life span and taking into account a practice factor would only strengthen the main effects further. More importantly, while practice effects might have diminished the main effects, it is unlikely that interaction terms would be affected significantly.

Using latent change score modeling on 24 cognitive measures to robustly estimate the changes in four cognitive abilities, the current study contributes to further understanding of cognitive decline in aging by demonstrating that, over five years of follow-up across the whole adult lifespan, cognitive decline accelerates with older age for memory, reasoning, and speed. We further pinpointed an increase in the rate of decline for the three abilities at certain change points: at 58 years for reasoning, 59 years for speed, and 70 years for memory. For vocabulary, however, we found longitudinal improvement up to 67 years, after which the rate of improvement slowed to zero. Furthermore, IQ, a CR factor, was observed to be an alleviating factor in the 5-year cognitive decline for memory and reasoning, even after accounting for potential brain maintenance effect, demonstrating the influence of life experiences in cognitive aging.