Cognitive Activity Is Associated with Cognitive Function over Time in a Diverse Group of Older Adults, Independent of Baseline Biomarkers

The longitudinal effect of cognitive activity [1], or cognitive engagement [2], on cognitive function is well established. In longitudinal studies of older adults with no cognitive diagnosis at baseline, cognitive activity has been associated with slower cognitive decline, reduced risk of mild cognitive impairment, and reduced risk of Alzheimer’s disease (AD) and related dementias [3‒5]. Nonetheless, debate remains as to whether more frequent cognitive activity confers a benefit to cognitive health or if less frequent cognitive activity represents a consequence of cognitive decline, referred to as reverse causation. A systematic review and bias analysis acknowledged that late-life cognitive activity may lower the risk of AD and all-cause dementia but noted that more data were needed to confirm the finding [6]. The authors concluded that reverse causation was a potential source of bias, given the fact that it is not possible to exclude participants with nonobvious or preclinical phases of cognitive impairment in most studies [6]. Conversely, in epidemiological studies, cognitive activity has long been recognized as an important correlate of the cognitive reserve concept, which has established that there is not a 1:1 correlation between brain pathology and clinical symptoms of cognitive impairment [7]. Despite an incomplete understanding of the mechanism of influence, there is evidence that cognitively active persons maintain higher levels of cognitive functioning, compared to less cognitively active persons, despite increasing neuropathological changes. In particular, Wilson et al. [3] and colleagues found that the relationship between cognitive activity and cognitive function was unrelated to postmortem markers of neurodegenerative and cerebral vascular pathologies believed to be responsible for cognitive decline [3], thereby suggesting that cognitive activity may benefit cognitive function.

The advent of blood biomarkers in cognitive aging research and their potential to predict and stage AD and related dementias [8‒13] provide a new environment in which to examine the relationship between cognitive activity and cognitive functioning. We chose to study this relationship in the contexts of three biomarkers, total tau (tau), neurofilament light (NfL), and glial fibrillary acidic protein (GFAP), due to their association with cognitive decline [9]. Plasma t-tau has been established as a general sign of neurodegeneration with elevated concentrations found in a number of conditions, including Alzheimer’s dementia [14]. Higher concentrations of plasma t-tau have been associated with cognitive decline and risk of mild cognitive impairment [9]. There is accumulating evidence that elevated GFAP levels are found early in the Alzheimer’s continuum and increase in tandem with disease progression, alongside increasing t-tau pathology markers [15]. NfL levels in blood have been associated with neurodegeneration [16], and there is now evidence that NfL may be able to discriminate between neurodegenerative dementias and healthy controls [17]. Similarly, GFAP concentrations aid in distinguishing AD from frontotemporal dementia, and both AD and frontotemporal dementia from healthy controls [17]. The exact threshold of brain neuropathology necessary for presenting cognitive symptoms is unknown, given the fact that neuropathologies do not explain all of the variation in late-life cognitive decline [18]. Our objective was to better understand the influence of cognitive activity in older adults with variable cognitive functioning and a range of underlying disease status. Since biomarkers provide a proxy indicator for neuropathology [19], we examined the association of cognitive activity and cognitive function, in the context of three biomarkers.

Participants were recruited door-to-door as part of the Chicago Health and Aging Project (CHAP); they are described in Table 1. All participants lived in the catchment area and agreed to participate in the study, which had and continues to have approval by the IRB at the Rush University Medical Center. Entry into the study required participants to be at least 65 years old, with no additional exclusion criteria. Data collection for this population-based cohort study occurred in 3-year cycles from 1993 to 2012. Participants agreed to in-home interviews and provided written informed consent. A stratified random sample of roughly one-third of participants took part in clinical evaluations in which blood samples were collected. The baseline for this study is taken as the time of the first in-home evaluation in which the first blood draw and cognitive activity were measured.

Blood samples were collected from 5,696 persons in CHAP, with 11,600 blood draws from 1993 to 2012. Given budgetary restraints, we could only obtain immunoassays in 3,000 samples. Still, we optimized our power to examine the associations of biomarkers with cognitive decline and activity by selecting participants who provided multiple blood draws and underwent a clinical evaluation. Of the 5,696 participants with blood draws, 1,534 underwent clinical evaluations. We selected participants with a baseline blood sample and at least two global cognitive function outcome measurements, resulting in 1,168 participants for analyses. Study design and sample selection for blood biomarker assignment have been described and illustrated in a previous publication [9].

The goal of this cognitive activity measure, as described previously [20], was to tap common activities of older adults that involved processing of information, while minimizing social, physical, and income demands. In other words, the scale was designed to understand what older adults of different backgrounds, income levels, and resources do on a regular basis. The 7 activities in the scale are positively correlated with one another, and Cronbach’s coefficient alpha is 0.58, indicating an acceptable level of internal consistency. In prior research, the total score on this measure has been associated with lower risk of developing dementia [5] and slower rate of cognitive decline [3].

At baseline, participants reported the frequency with which they engaged in late-life cognitive activities [4]. They reported how often they (1) viewed television, (2) listened to the radio, (3) went to a museum, (4) played games such as cards, checkers, crosswords, or other puzzles or games, (5) read books, (6) read magazines, and (7) read newspapers. Participants were presented with five choices to indicate frequency: 5 = every day or almost every day, 4 = several times a week, 3 = several times a month, 2 = several times a year, or 1 = once a year or less. We averaged the seven items to create a composite cognitive activity score. This self-report measure has been described previously and validated in several other studies [1, 3‒5].

We chose to measure biomarkers via our blood samples for two reasons. First, there is less participant burden in obtaining a blood sample, compared to collecting cerebrospinal fluid. Second, technological advances have made this method comparable to cerebrospinal fluid collection, in terms of analytical sensitivity [21]. Study personnel collected blood during in-home visits, put the samples on dry ice, and transported them to the Rush Biorepository freezer. The samples collected between 1994 and 2012 were stored at -80°C. The samples remained frozen until 2019 when they were sent to Quanterix Corporation in Billerica, MA, USA, where personnel assayed the three neuronal biomarkers. All three biomarkers underwent ultrasensitive immunoassays performed in duplicates through the single molecular assay bead-based HD platform and the Neurology 4-Plex A kit [9]. We calculated the mean concentration based on the average of duplicate measurements. The coefficients of variation between the replicates were as follows: (1) Tau – 7.3%; (2) NfL – 3.0%; and (3) GFAP – 3.0%. All biomarkers were analyzed as continuous variables. We created our figures based on the median distribution of biomarker concentrations specific to each biomarker. Total tau was categorized into low with less than or equal to 0.40 pg/mL or high with greater than 0.40 pg/mL. NfL low concentration was defined as less than or equal to 25.5 pg/mL and high as greater than 25.2 pg/mL. A low concentration of GFAP was defined as less than or equal to 232 pg/mL and high as greater than 232 pg/mL. We had previously established in this population that these values provided meaningful cutoffs in predicting cognitive decline [9].

We measured cognitive function with four tests, including tests of immediate recall, delayed recall (East Boston Test; [22, 23]), perceptual speed (Adapted Digit Symbol Test; [24]), and the Mini-Mental State Examination (MMSE) [25]. Since the four tests are scored with different scales, we standardized them to a Z score by using the baseline means and standard deviation (SD) of the CHAP population. Therefore, the average score is 0 with a SD of 1. We then averaged the scores into a global cognitive function score [4]. A composite global test score is used, which reduces ceiling and floor effects, and allows us to accommodate a wider range of abilities. This facilitates the goal of our longitudinal study, making it possible to capture cognitive changes over time [26]. There is substantial evidence from previous studies that the four tests load on a single factor and account for 75% of the variance of each test score [4]. We also used the immediate and delayed recall scores of the East Boston Test to create a memory score and the adapted digit symbol test to create a speed score. Cognitive function was measured at every cycle, which occurred on average every 3 years.

We conducted descriptive analyses of participant characteristics at baseline. We ran a correlation matrix to characterize the co-occurrence of biomarker concentrations. Additionally, we demonstrate the correlations among cognitive activity levels and cognition over time (global cognition, memory score, and speed score). We treated the first blood draw as the baseline for cognitive decline analyses. Because the concentrations of the biomarkers (tau, NfL, and GFAP) were positively skewed, they were log10 transformed for regression analyses as described previously [9].

We conducted separate linear mixed-effects models to evaluate the following relationships at baseline and over time, controlling for age, sex, race, and education: (1) cognitive activity with global cognitive function, (2) biomarker with global cognitive function, (3) cognitive activity and biomarker with global cognitive function, and (4) cognitive activity and biomarker with global cognitive function, with a term added for the interaction of biomarker and cognitive activity. The main effect of cognitive activity tests for the baseline effect and a cognitive activity × time interaction term is used to test for change over time. To help interpret relative differences in annual decline, we calculated the percent difference in yearly decline of the cognitive score. Plots were generated using R, version 4.0.3 (R Group for Statistical Computing, Vienna).

Participants were 1,168 older adults, with 63% women and mostly non-Hispanic Blacks (60%), at least 65 years old, and had an average of 12.6 (SD = 3.5) years of education at baseline (Table 1). The average cognitive activity score for all participants was 3.18 (SD = 0.64; range = 1–5).

Participants had an average global cognition score of 0.21 (SD = 0.67), based on a standardized Z score, as described previously [2]. The geometric means of total tau concentration was 0.37 pg/mL (95% CI: 0.35–0.39). For NfL, the geometric mean was 26.1 pg/mL (95% CI: 25–27), and for GFAP, it was 227 pg/mL (95% CI: 220–234). Biomarkers were positively correlated with each other. Tau and NfL had a correlation of 0.23; tau and GFAP had a correlation of 0.23; and NfL and GFAP had a correlation of 0.53.

We found that cognitive activity was associated with global cognitive function at baseline (Table 2; models 1, 5, 9) and continued to be associated when each biomarker was added to the model (Table 2; models 3, 7, and 11), with comparable effect sizes.

We did not find an association between tau and cognitive function at baseline in the model with only tau (Table 2; model 2) or when we added cognitive activity (Table 2; model 3), in contrast to NfL and GFAP. NfL was associated with cognitive functioning in the model with only NfL (Table 2; model 6) and after adding cognitive activity to the model (Table 2; model 7). GFAP was associated with cognitive functioning in the model with only GFAP (Table 2; model 10) and after adding cognitive activity to the model (Table 2; model 11).

We found no effect on baseline cognition for any of the interactions between cognitive activity and biomarker, as demonstrated in Table 2, models 4, 8, and 12.

Cognitive activity was associated with global cognitive function over time in the basic model, as shown in Table 3, models 1, 5, 9. In models with both cognitive activity and a biomarker (Table 3; models 3, 7, and 11), cognitive activity continued to be associated with rate of change in global cognitive function for all three biomarkers but models with NfL (model 7) and GFAP (model 11) were weakened just enough to not reach significance (Table 3).

All three biomarkers were associated with decline in cognitive functioning in the model with only the biomarker (Table 3; models 2, 6, 10) and after adding cognitive activity to the model (Table 3; models 3, 7, 11).

We found no interactions between cognitive activity and biomarker for any of the biomarkers over time. See Table 3, models 4, 8, and 12.

Figure 1 shows the longitudinal changes in cognitive function by biomarker concentrations (high vs. low) and cognitive activity level (high activity vs. low activity). The solid lines represent person who participated in the 90th % of cognitive activity (average cognitive activity score of 4), and the dotted lines represent persons who participated in the lowest 10th % of cognitive activity (average cognitive activity score of 2). Persons who report very high cognitive activity continue to function much better than those with low levels of cognitive activity even at high levels of biomarkers.

We examined the relationship between self-reported cognitive activity and cognitive function at baseline and over time in the contexts of three different biomarkers (tau, NfL, and GFAP) measured at baseline. Baseline cognitive activity robustly predicted cognitive function, independent of biomarker. At baseline, NfL and GFAP were associated with lower cognitive function, independent of cognitive activity, but tau was not. Over time, cognitive activity was associated with better cognitive functioning in each of the models with the biomarker added, despite some attenuation in the effect in the models with NfL and GFAP. We found no interaction between cognitive activity and biomarker on its effect on cognitive function at baseline or over time.

The main finding of our study is that more frequent cognitive activity is associated with better cognitive function in old age, despite high concentrations of biomarker. Since the presence of biomarkers is an indication of underlying disease presence, our finding has important implications for the role of cognitive activity in increasing cognitive reserve. Cognitive activity may contribute to cognitive reserve and exert its influence in at least two ways. First, by engaging in a life with enriching activities, a person arrives to old age with a higher level of cognitive function and the onset of clinical features of cognitive impairment will be delayed. Second, cognitive activity may contribute to a slower rate of cognitive decline, even in the presence of variable disease accumulations as measured by biomarkers. The finding that cognitive decline is slower in persons who engage in more frequent cognitive activity has been established in this population previously [4, 27] and in other studies [3, 6]. Our results follow the same trajectory, despite the attenuation of the association in the models with NfL and GFAP.

The finding that tau was not associated with cognitive function at baseline may be due to the nonlinear nature of tau as established previously [9]. Additionally, we only included participants with two or more cognitive observations. Since our sample included participants who were available for at least one data collection cycle (N = 3 years) past baseline, we likely excluded participants who were less healthy, had lower cognitive function scores, and had higher tau scores, reducing our power to find an association. Over time, the association between tau and cognitive function was apparent. We currently do not know why this would be the case for only tau, as we do not have a complete understanding of how each of the biomarkers functions cross-sectionally and over time. However, our findings are overall consistent with the growing body of literature in which higher concentrations of biomarkers are associated with lower cognitive function.

The finding that none of the biomarkers modified the relationship between cognitive activity and cognitive function at baseline and over time has several implications. An association at baseline between cognitive activity and cognitive function suggests that interventions may be most effective if implemented sometime before the 6th decade of life – the age at which our baseline data were collected. The fact that there was also no interaction between biomarker and cognitive activity over time provides evidence that older persons with a range of underlying disease states may continue to benefit from enrichment activities across the lifespan. These findings complement results from Wilson et al. [27] and colleagues, who previously established that more frequent cognitive activity was associated with slower cognitive decline in those who did not have any cognitive disorder, despite accumulating neuropathology. Together, these studies provide evidence for the idea that cognitive activity benefits cognitive function consistent with the idea of cognitive reserve. While these findings are inconsistent with the concept of reverse causation, the mechanism underlying the relationship between cognitive activity and cognitive function is not fully understood. Changes in brain structures identified on brain imaging [28‒32] provide support for the role of extensive learning, behavioral activation, or practice in improved neural structures that likely offset the effect of neuropathological changes in the brain. It is also possible that cognitive activity is associated with personality characteristics, such as openness to experience [2], that are responsible for enhanced cognitive functioning. More research that simultaneously looks at engagement in enriching cognitive activities throughout the lifespan and other psychosocial factors is needed to better understand the mechanism.

This study has several limitations. Our measure of cognitive activity or cognitive engagement may fail to fully capture the entire range of cognitive activities in which older adults from diverse backgrounds engage. Our cognitive activity measure is based solely on self-report and therefore vulnerable to error. Our cognitive battery is limited to four tests and may fail to accurately capture the strength of the cognitive activity-cognitive decline association that is related to each of the biomarkers. Our study sample of 1,168 has an overall cognitive score that is slightly higher than the average of the original cohort baseline of 6,157, and therefore, this study may result in an underestimation of the effect of cognitive activity on cognitive function. It is also possible that sex differences exist in these relationships that we were unable to examine due to lack of power. Finally, the last data collection cycle was 11 years before the statistical analysis for this study, which may limit the applicability of the findings due to cohort-specific changes (e.g., access to technology).

The main strength of our study is that we report on the associations of biomarkers with cognitive activity, an important modifiable lifestyle factor, and thereby extend the evidence for the relevance of biomarkers. Additionally, our participants in this population-based study represent a diverse group of older adults, making our findings relevant for a large number of persons. More research is needed on how modifiable risk factors, such as cognitive activity, may influence or even interrupt biological processes – indicated by biomarkers – that contribute to cognitive decline.

We thank the many participants who contributed to our study and all members of our study team.

Our study was conducted in accordance with the World Medical Association Declaration of Helsinki and was approved by the Rush University Medical Center Institutional Review Board, approval number ORA: 20090204. Study personnel obtained written informed consent from each participant.

The authors have no conflicts of interest to declare.

This study was supported by grants RO1AGO3154, RO1AGO51635, RF1AGO57532, and RO1AGO58679 from the National Institutes of Health.

Drs. Kristin R. Krueger and Kumar B. Rajan contributed to conception and design of the study, leading the writing of the manuscript. Drs. Kristin R. Krueger and Kumar B. Rajan, along with Todd Beck, were responsible for the acquisition and analysis of data. All authors contributed to the interpretation of the data and oversight of the final version.

External researchers can use our public data portal to request data at https://www.riha.rush.edu/dataportal.html.