Cognitive reserve is associated with education, social determinants, and cognitive outcomes among older American Indians in the Strong Heart Study

Cognitive resilience may be conceptualized using three different facets: as cognitive reserve, or resilience in cognition despite advancing pathological status; as brain reserve, or resilience in brain functionality despite advancing pathological status; or as brain maintenance, or resilience to change in neuropathologic status despite substantive risk factors such as advancing age¹. Cognitive reserve is challenging to study—first because direct measures are not available, so operationalization using proxy and latent variables is still needed. Furthermore, in vivo measures of neuronal pathology are not widely available, further distancing available measures from reflecting true status. When neither the injury nor the mediating construct can be directly measured, conceptual frameworks can be difficult to construct and interpret.

Nevertheless, cognitive reserve, may be conceptualized as the ability to overcome the influence of accumulating brain pathology, in order to maintain cognition and functioning^2,3. Thus, those with more cognitive reserve may be better equipped—perhaps via enhanced cortical thickness and/or connectivity—in compensating for accumulating disease burden, and thus preventing or delaying the emergence of clinical symptoms^4,5,6,7. Such an abstract concept may be quantified in multiple ways: one option is to declare intact cognitive status, despite substantive accumulated pathology or advanced age. However, dichotomizing cognitive state as intact versus impaired may be overly reductionistic, arbitrary, and may introduce bias and measurement error in contexts where neuropsychological tests are not validated and normative data do not yet exist^8,9.

A more ideal approach utilizes continuous estimation, thus also enabling adjustment for confounding by premorbid function¹⁰. In such an approach, cognitive reserve is represented as a continuous measure of association between selected measures of cognition and neuropathology, and could theoretically reflect those with abnormally positive aging, negative aging, or both⁹. Continuous estimation of cognitive reserve by regressing cognitive test scores against measures of imaged brain volumes, adjusting for confounders such as age and sex, and predicting the expected residual values from each regression provides an intraindividual metric of observed-over-expected performance¹⁰. As an estimate of continuous, comparative cognitive reserve, regressed residual metrics have unknown specificity to the chosen cognitive domain and pathologic marker⁹.

The selection of specific measures to define the metric of cognitive reserve can influence the patterns of association for that metric. For example, a metric using a measure of processing speed over general brain volume might be demonstrative of global pathologies^1,10, whereas a metric of phonemic fluency over white matter hyperintensities (WMH) might be more sensitive to vascular pathologies¹¹, and a metric of memory over hippocampal loss might be more reflective of neurodegenerative pathologies¹².

Despite increasing interest in resilient aging and cognitive reserve, research is still needed to examine both risk and protective features in American Indians. Note that we use the term “American Indian” per specific direction from our partner Tribal communities and Community Advisory Board. American Indians convey many features associated with strength-based aging¹³, including psychological resilience^14,15,16, community, and social support¹⁷. However, this population is also at particularly high risk for vascular disease, Alzheimer’s disease (AD), and related dementias^18,19, in part due to the high burden of modifiable, clinical risk factors—although not all conventional AD risk factors appear generalizable in effect²⁰. Thus, estimation of cognitive reserve in this unique population may help elucidate some of the observed, potentially modifiable effects from factors such as diabetes and depression^21,22,23, as well as some of the observed variance in mediating factors such as education. In previous research, we detected the strongest associations of brain, WMH, and hippocampal volumes with cognitive test performance²⁴, and thus structured candidate metrics using these imaging features, in combination with corresponding test scores for processing speed, fluency and executive function, and memory, respectively.

This study aimed to develop understanding of these metrics in this population facing unique challenges, with the ultimate goal to identify social determinants across the life-course that have the potential to promote healthier aging trajectories for American Indian older adults and their communities.

Setting

The Strong Heart Study (SHS) is a longitudinal study of American Indians from the Northern Plains, Southern Plains, and Southwest, recruited for multiple examinations between 1989–1999²⁵. Using the SHS as a sampling frame, the ancillary Cerebrovascular Disease and its Consequences in American Indians study (CDCAI) conducted two neurological examinations in 818 consenting study survivors in 2010–2013 (Visit 1) and in 403 of the 818 in 2017–2019 (Visit 2), as previously reported²⁶.

Ethics and inclusion

Institutional, Indian Health Service, and Tribal research review boards approved all procedures. All participants provided written, informed consent, and were compensated for their participation in this study. This study analysis was pre-proposed to the Strong Heart Study Publication & Presentations committee (Proposal 846, Approved 2024-06-13); a list of approved publications is available at https://strongheartstudy.org/Research/Papers-and-Abstracts/Approved-Papers-Dissertation. There was no additional preregistration for this study. All partnering Tribes and communities approved all written dissemination materials. Of note, the term “American Indian” is used herein, per specific direction from our study partner Tribal communities and Community Advisory Board, who have the final say in the representation of their identity. Terminology preferences vary substantively by Tribe, community, and individual, but this term is generally considered acceptable in the United States; is commonly the preferred term when individual Tribe or Tribal nations cannot be directly named; and is also the preferred term based on guidance from nationally-representative, self-governing, Indigenous-led organizations such as the National Congress of American Indians.

Cognitive testing

The Weschler Adult Intelligence Scale 4th edition digit symbol coding subtest (WAIS-DSST, processing speed or visuomotor speed)²⁷ Controlled Oral Word Association FAS (COWA FAS, verbal fluency, executive function)²⁸, California Verbal Learning Test II Short Form (CVLT II-SF, verbal learning and memory)²⁹, were administered at both CDCAI Visit 1 and Visit 2. Modified Mini-Mental Status Examination (3MSE, general multidomain cognitive screening)³⁰, was also administered at both Visit 1 and Visit 2; other tests added at CDCAI Visit 2 included the National Alzheimer’s Coordinating Center Uniform Data Set version 3.0 form C2 battery³¹, with Montreal Cognitive Assessment (MoCA, multidomain general screening)³², Number Span Test forward and backward (attention and working memory), Benson Complex Figure copy and recall (visuospatial skills and memory)^32,33, animal and vegetable naming tests (semantic fluency), Craft Story immediate and delayed recall (contextual verbal memory)³⁴, Trail Making Test A and B (processing speed, simple and divided attention)³⁵, and Multilingual Naming Test (MINT, semantic memory and naming)³⁶. Of note, lower cognitive scores correspond with poorer cognition for all measures except the Trail Making timed tests wherein more time (seconds) needed to complete the activity corresponds to poorer performance. Internal consistency-reliability for cognitive tests in this population appear to follow a pattern of excellent fit and performance for overall or summary scores (omega coefficients >0.9), but poor metrics of fit and performance for individual domain, subcomponent, or tasks (omega <0.8;)^37,38,39. Composite, multidomain cognitive test performance was also defined as the average of Z-scored cognitive tests.

MRI

At both CDCAI Visit 1 and Visit 2, participants underwent 1.5T MRI scans with six sequences including sagittal T1-weighted localizer, co-registered 5 mm axial-T1, 5 mm axial-T2, 5 mm axial-T2* susceptibility-weighted images in the anterior commissure/posterior commissure plane, 3 mm axial fluid-attenuated inversion recovery (FLAIR) images, and 1.5 mm sagittal T1-weighted volumetric gradient echo. Additional details regarding repetition time, echo time, inversion time, slice thickness, slice spacing, acquisition matrix, number excitations, echo train, flip angle, and field of view have been provided in prior reports. MRI were processed for structural volumetrics, including total gray and white tissue for the whole brain, white matter hyperintensities (on FLAIR) or hypointensities (on T1), and hippocampus volume using version 5.3 of the FreeSurfer image analysis suite²³, with motion correction and visual checks at intermediate points in the process for gross misregistrations and other processing failures. Such problems were corrected manually. Only images that exceeded the field of view were censored, although these did not usually pass quality control. Skull stripping using cortical reconstruction by voxel-based parcellation of T1 images, skull stripping using cortical reconstruction, removal of non-brain tissue using a hybrid watershed-surface deformation procedure, automated Talairach transformation, segmentation of subcortical white matter and deep gray matter volumetric structures, intensity normalization, tessellation of the gray matter-white matter boundary, automated topology correction, and surface deformation following intensity gradients to place optimally the gray matter-white matter and gray matter-CSF borders at the location where the greatest shift in intensity defines the transition to the other tissue class. Overall, the results of these procedures include brain and cranial volumetric estimates that include cerebellum but not ventricles, CSF, and dura. All features were estimated in millimeters cubed, and then standardized to intracranial volume to account for interindividual variations in head size.

Other data

Participants self-reported age (years), sex (male; female), years of formal education, annual household income, and typical use of alcohol (drinks per day). Participants underwent clinical and anthropometric examinations, including basic blood and urine laboratory testing, medical history, and transcription of medications. Diabetes mellitus was defined as fasting blood glucose ≥126 mg/dL, prior physician diagnosis, or use of antihyperglycemic medications or insulin. Body mass index (BMI) was defined as weight in kilograms divided by height in meters squared, and categorized using standard criteria (normal BMI <25 kg/m²; overweight BMI 25 to 29.9; obese BMI 30+). Hypertension was defined as averaged, seated systolic blood pressure ≥130 mmHg, diastolic ≥90, or the use of antihypertensive medications. Chronic kidney disease was defined as an estimated glomerular filtration rate <60 mL/min. Participants self-reported emotional and mental health using the 20-item Centers for Epidemiologic Studies Depression (CESD) symptoms scale, and depression was dichotomized using a cutoff ≥16. Plasma biomarkers related to Alzheimer’s disease were measured (using samples from Visit 2) on the Quanterix platform including phosphorylated tau (pTau181), amyloid beta 42 and 40 (AB42/40 ratio), glial fibrillary acidic protein (GFAP), and neurofilament light chain (NfL). Apolipoprotein E epsilon 4 carrier status was assessed using standard genotyping.

Cognitive reserve metrics

Cognitive reserve was operationalized as observed-over-expected cognitive performance, by regression of cognitive test score (WAIS-DSST score, COWA FAS overall score, or CVLT long delay free recall score) over a continuous estimate of brain pathology (total brain volume, WMH volume, or hippocampus volume, respectively; all standardized to IC volume); adjusted for age and sex, resulting in three metrics: WAIS-Brain, COWA-WMH, and CVLT-Hipp. Predicted values for each individual participant according to the regression line were subtracted from their observed values, producing a “residual”, or estimated degree of positive or negative cognition-beyond-pathology. In this paradigm, these residuals represent a continuous estimate for the latent variable of cognitive reserve, with positive values indicating higher reserve and negative values lower reserve. Residuals were then analyzed continuously, as well as in tertiles of low, middle, and high-reserve groups. This method was similar to those used by other groups, and the specific pairings (WAIS-Brain, COWA-WMH, CVLT-Hipp) were defined on the basis of theoretical associations of changes in cognition with associated affected region or pathology¹⁰.

Statistical analyses

Analyses were conducted as complete analysis, with exclusion only on the basis of missing data; for transparency, numbers included in each analysis figure and table were reported separately. Descriptive analyses estimated mean and standard deviation or count and percent, by examination visit, for participant characteristics, including the three metrics of cognitive reserve (WAIS-Brain, COWA-WMH, CVLT-Hipp). Graphical representations included histogram, scatterplot, linear fit, boxplot with whiskers, and heatmap. Heatmap colorations were based on pairwise correlation coefficients (Pearson) and P values. Statistical tests for association included one-way ANOVA (F-test) for between-group differences, T-tests for equivalence in distributions, or regression. Causal mediation analysis estimated direct and indirect effects as well as the proportion of effect mediated for models with education, depression, or diabetes as exposure, composite cognitive test score as outcome, and metric of cognitive reserve as mediator. Evaluation of assumptions included assessment of normality in residuals, residual association with age and sex after adjustment, and heteroscedasticity. All analyses were done using Stata v17-18 (College Station, Texas) or R v4 (Vienna, Austria).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

At Visit 1 (2010–13), this population-based cohort of American Indians had a mean age of 73, was majority female, and had about half with at least some college (Table 1). Depression symptom (CESD) scores were high at both Visit 1 and Visit 2. Diabetes was detected in about half of the cohort at both visits. As noted in prior reports, income was relatively low, overweight and obesity, hypertension, and kidney disease were relatively common, and alcohol use was generally uncommon.

Metrics of cognitive reserve, defined as residuals of cognitive test scores over brain imaging features as % IC volume and adjusted for age and sex, included WAIS DSST over brain volume (WAIS-Brain), COWA fas over WMH volume (COWA-WMH), and CVLT LF over hippocampus volume (CVLT-Hipp). WAIS-Brain residuals were not significantly different comparing Visit 1 to Visit 2, over age, and were normally distributed (Fig. 1). COWA-WMH residuals were somewhat lower from Visit 1 to Visit 2 (mean residual difference 2.6 (95% CI: 1.5, 3.7; t-test P < 0.0001), significantly associated with age (beta -0.6; 95% CI: -0.5, -0.8; P < 0.0001), and were normally distributed. CVLT-Hipp residuals were not significantly different comparing Visit 1 and Visit 2, or by age, and were somewhat normally distributed but with a mild left skew.

Graphical representation of three metrics of cognitive reserve, with (top) residuals of Weschler Adult Intelligence Scale Digit Symbol Coding Test (WAIS DSST) over brain volume as percent intracranial volume (WAIS-Brain), (middle) residuals of Controlled Oral Word Association f/a/s test (COWA) over white matter hypointensities volume (MRI T1 sequence) as percent intracranial volume (COWA-WMH), and (bottom) residuals of California Verbal Learning Test version II short form long delay free recall test (CVLT LF) over hippocampus volume as percent intracranial volume (CVLT-Hipp). Metrics of cognitive reserve defined by regression of cognitive test score over brain imaging feature, adjusted for age and sex. Left: scatterplots of cognitive test scores (Y-axis) by brain imaging features (X-axis). The coloration of scatter points with fit lines by tertile of metric for cognitive reserve, with the highest reserve in green, middle in blue, and lowest in red. Note: COWA-WMH scatterplot shown with X-axis on a log scale, and CVLT-Hipp plot with point-jitter for visibility purposes. Left-middle: Boxplots (mean, 25th, and 75th percentiles), with scatter overlay, and whisker bars (10th and 90th percentiles) for tertiles of (A, B, C) metrics of cognitive reserve. Right-middle: histograms for distribution of each metric with normal tracing. Right: scatterplots of metrics over age, with polynomial fit line and 95% confidence interval. Numbers included in the analysis: WAIS-Brain Visit 1 n = 765, COWA-WMH Visit 1 n = 765, CVLT-Hipp Visit 1 n = 744.

Full size image

Separating all three metrics of cognitive reserve into tertiles enabled comparison of low, middle, and high categories of cognitive reserve, independent of age and sex (Fig. 1). Graphically represented in boxplots by tertile or in scatterplots of each metric’s cognitive test score (WAIS DSST, COWA fas, or CVLT LF, respectively) over brain imaged volumetric feature (overall volume, WMH volume, or hippocampus volume as % intracranial volumes, respectively), with low tertile colored red, middle in blue, and high in green, tertiles had clear separation and consistent linear associations for the WAIS-Brain metric, less clear separation but consistent linear associations for the CVLT-Hipp metric, and somewhat clear separation and inconsistent linear associations for the COWA-WMH metric.

Comparing in a heatmap these three metrics of cognitive reserve with sociodemographic, clinical, and psychological features (Fig. 2), better cognitive reserve (higher residual values) was significantly associated with more years of formal education (WAIS-Brain ρ = 0.45; 95% CI: 0.40, 0.51; P < 0.001and COWA-WMH ρ = 0.48; 95% CI: 0.42, 0.53; P < 0.001), and not significant for CVLT-Hipp (ρ = 0.08; 95% CI: 0.01, 0.15; P = 0.06). Higher income was also significantly associated with the three metrics of cognitive reserve (WAIS-Brain ρ = 0.3, P < 0.001, COWA-WMH ρ = 0.3, P < 0.001, CVLT-Hipp ρ = 0.1, P < 0.001). Diabetes (WAIS-Brain ρ = −0.1 P = 0.001, COWA-WMH ρ = −0.1, P = 0.001) and depression (WAIS-Brain ρ = −0.1, P < 0.001, COWA-WMH ρ = −0.1, P < 0.001, CVLT-Hipp ρ = −0.1, P = 0.02) were significantly associated with lower cognitive reserve, although diabetes was not associated with CVLT-Hipp (ρ = 0.0, P = 0.5). APOEe4 allele status, field center, overweight/obesity, hypertension, chronic kidney disease (CKD), and alcohol use were not statistically associated with the three metrics of cognitive reserve. Boxplots with scatter overlay (Fig. 3) display the distribution of each of the three metrics of cognitive reserve, over four categories of education graphically suggested dose-dependent associations for WAIS-Brain and COWA-WMH metrics of cognitive reserve.

Heatmap with coloration from red (strongest negative association) to green (strongest positive association), with yellow as intermediary values), and overlaid rho Pearson correlation coefficients and P values for associations of three metrics for cognitive reserve: residuals of Weschler Adult Intelligence Scale Digit Symbol Coding Test (WAIS DSST) over brain volume as percent intracranial volume (WAIS-Brain), residuals of Controlled Oral Word Association f/a/s test (COWA) over white matter hypointensities volume (MRI T1 sequence) as percent intracranial volume (COWA-WMH), and residuals of California Verbal Learning Test version II short form long delay free recall test (CVLT LF) over hippocampus volume as percent intracranial volume (CVLT-Hipp) with each other and with sociodemographic & clinical features. All metrics of cognitive reserve defined by regression of cognitive test score over brain imaging feature, adjusted for age and sex. Numbers included in the analysis: WAIS-Brain Visit 1 n = 765, COWA-WMH Visit 1 n = 765, CVLT-Hipp Visit 1 n = 744.

Full size image

Boxplots (mean, 25th and 75th percentiles), with scatter overlay, and whisker bars (10th and 90th percentiles) for three metrics of cognitive reserve, over categories of education (less than high school, <HS; high school graduate, HS; any college; college graduate degree and beyond, BA/BS). Cognitive reserve metrics include (top) residuals of Weschler Adult Intelligence Scale Digit Symbol Coding Test (WAIS DSST) over brain volume as percent intracranial volume (WAIS-Brain), (middle) residuals of Controlled Oral Word Association f/a/s test (COWA) over white matter hypointensities volume (MRI T1 sequence) as percent intracranial volume (COWA-WMH), and (bottom) residuals of California Verbal Learning Test version II short form long delay free recall test (CVLT LF) over hippocampus volume as percent intracranial volume (CVLT-Hipp). All metrics of cognitive reserve defined by regression of cognitive test score over brain imaging feature, adjusted for age and sex. Numbers included in the analysis: WAIS-Brain Visit 1 n = 765, COWA-WMH Visit 1 n = 765, CVLT-Hipp Visit 1 n = 744.

Full size image

Using ANOVA to examine associations of tertiles of the three metrics of cognitive reserve—defined at Visit 1, with plasma biomarkers for Alzheimer’s disease and related dementias—defined at Visit 2 (Table 2), only NfL had a significant association with COWA-WMH (tertile 1 to tertile 3 residual mean difference 10.2, P = 0.006). Of note, several of the comparisons showed significant departure from equal variance (i.e., significant Bartlett chi² test) across tertiles (categories) of cognitive reserve: WAIS-Brain metric with all four of the plasma biomarkers, COWA-WMH and CVLT-Hipp metrics with pTau and amyloid beta, and CVLT-Hipp with NfL.

Longitudinal regression analyses for tertiles of the 3 metrics of cognitive reserve—defined at Visit 1, with cognitive test scores—collected at Visit 2 (Table 3) detected generally consistent, significant associations of earlier cognitive reserve with later cognitive test performance. General cognition (3MSE; MoCA), processing speed (WAIS DSST) verbal fluency (COWA fas), semantic fluency (animal naming, vegetable naming, MINT), contextual memory (Craft story immediate and delayed recall), and executive function or attention (number span forward and backward, Trail Making Test A and B) were all significantly associated with cognitive reserve measured a mean of 6.7 years earlier (range: 4–9 years). Analysis of visual processing and memory (Benson figure copy and recall) and verbal learning and memory (CVLT LF) scores were statistically associated with COWA-WMH and CVLT-Hipp metrics, but did not have statistical evidence of association with the WAIS-Brain metric.

Given positive associations of cognitive reserve with education, depression, and diabetes, as well as with cognitive test scores, causal mediation analysis examined the proportion of mediated effect for models with education, depression, or diabetes as exposure, metric of cognitive reserve—WAIS-Brain, COWA-WMH, or CVLT-Hipp—as mediator, and composite cognitive test score performance (Z-scale) as outcome, with adjustment for age and sex and robust standard errors estimation (Table 4). Of the three exposures considered, only education had a significant mediation (indirect) effect: accounting for an estimated 24% (95% CI: 5-44%, P = 0.014) of the WAIS-Brain metric association with later overall cognitive test performance, and accounting for estimated 15% (95% CI: 0–29%, P = 0.045) of the COWA-WMH metric association with later overall cognitive test performance.

Overall, these findings suggest that cognitive reserve—conceptualized as observed-over-expected cognitive performance and defined using three separate metrics—is positively associated with sociodemographic, clinical, psychological, and cognitive features of aging in American Indian older adults. This examination of cognitive reserve in American Indians compared cognitive reserve using 3 separate metrics: WAIS-Brain, a general metric reflecting processing speed and overall volumetric change; COWA-WMH, a vascular-sensitive metric reflecting changes in phonemic fluency with small vessel disease; and CVLT-Hipp, a memory-sensitive metric attuned to changes in a region targeted preferentially by Alzheimer’s disease.

Income and depression were strongly and consistently associated with all three metrics of cognitive reserve (WAIS-Brain; COWA-WMH; CVLT-Hipp); education and diabetes are consistently associated with 2 of the metrics (WAIS-Brain, COWA-WMH). Kidney disease was associated only with COWA-WMH metric, as was the NfL plasma biomarker. In longitudinal analyses, all three metrics (WAIS-Brain; COWA-WMH; CVLT-Hipp) were also strongly related to cognitive test performance over time, in all cognitive domains except memory. Finally, causal analysis of education with cognitive test performance showed significant mediation (direct effect estimation) by WAIS-Brain (24%) and COWA-WMH (15%) metrics of cognitive reserve. Overall, these findings suggest that some metrics of cognitive reserve may be more informative than others, with patterns of association specific to their definition, but overall such metrics are generally consistent in associations with key social and clinical determinants, thus representing potentially modifiable protective processes in aging.

Our findings of strong and consistent associations for observed-over-expected performance with education, independent of age, are consistent with theoretical models that cognitive reserve is built over a lifetime, developed in early childhood through educational development in cognitive plasticity^40,41. Such a model suggests that cognitive reserve, once built, may also be deconstructed through exposure to especial challenges such as socioeconomic deprivation⁴² or chronic stress⁴³, also consistent with our findings of negative associations with depression. Aside from the processing speed-overall volume (WAIS-Brain) associations, the strongest associations were for executive function and phonemic fluency and small vessel injury (COWA-WMH), especially with income and depression; associations for the metric related to memory (CVLT-Hipp) were not as strong or consistent as for the other two metrics.

APOEe4, ADRD plasma biomarkers, hypertension, body mass, field center (region), or alcohol use did not have detectable statistical evidence of association with metrics of cognitive reserve in these analyses. Other ways to conceptualize cognitive reserve may yield different results. Also, some of these results may reflect from lack of variance in the measures, such as for alcohol; with few users and low use among users, associations due to variability in alcohol use may be difficult to detect. However, education and income, depression, and diabetes—and to a lesser extent kidney disease—have consistently had strong associations with cognitive function, brain imaging and neuropathological findings, and other measures of ADRD in this population^{20,23,26,44,45}, which is consistent with these findings. It is possible that these are key variables to risk and resilience in this population. Future research should examine whether these association patterns are consistent for other Tribes and regions, other populations, and whether addressing the much higher prevalence of vascular disease and dementia may influence the balance of these associations for future generations^18,19.

Prior, cross-sectional research in non-Hispanic white Americans identified associations for a similarly-defined measure of cognitive reserve with cognitive test scores, with non-Alzheimer’s disease pathology, and with APOEe4 genotype¹⁰. We found associations with education, depression, diabetes, and with other socioeconomic and clinical conditions. In longitudinal analyses, later cognition was associated with earlier cognitive reserve, independent of age, sex, baseline cognition, and time between examinations. Future research should examine associations with other neurological conditions.

Limitations

Although causal sequence and temporality is not directly observable in cross-sectional associations, formative and environmental features such as education are likely to represent precursors to cognitive reserve, both directly and indirectly. On the other hand, differential cognitive reserve may be either a contributor or an effect of disease comorbidities such as diabetes and depression. In our longitudinal analyses, cognitive reserve was a significant mediator of only the associations for education. However, two timepoints may not be adequate for answering causal longitudinal questions, and future research may benefit from evaluation of younger populations, more timepoints, and additional measures of cognitive reserve.

Our population included American Indian participants from multiple Tribes and communities in the United States Southwest, Southern Plains, and Northern Plains, but these findings may not generalize to other Tribal groups, regions, or generational cohorts. Cognitive tests may have differential validity across cultural groups, which can affect the generalizability of findings, although preliminary psychometrics in our study population (with more forthcoming) suggest similar test score interpretability as in other populations^37,38, with the exception that scores must be interpreted according to education and bilingual experience. Discriminant properties of cognitive test score cutoffs, in particular, require adjustment and validation for appropriate diagnostics—both for impaired, as well as for evaluation of “super-agers”, or those who maintain healthy cognition despite advanced age and irrespective of pathology status. Also, due to the selection structure of this longitudinal cohort, our findings may be subject to an increased likelihood of Type II error if the outcome (e.g., cognitive resilience) is unduly influenced by the likelihood of survival to participation over Visits, although the metrics evaluated appear similar between the two visits. Finally, other types of resilience, such as psychological or community resilience have yet to be explored in this context. Ongoing research in these communities aims to provide insights into this area in the future.

In summary, this description of cognitive reserve in American Indians suggests that cognitive reserve is strongly associated with education, other socioeconomic factors, depression, diabetes and other clinical factors, and later cognitive changes. This work also suggests that some metrics of cognitive reserve (e.g., residuals of processing speed over total brain volume or residuals of verbal fluency over white matter lesion burden) may be better markers or determinants of overall cognitive reserve than others (e.g., residuals of memory over hippocampal volume), although each metric has unique and specific association patterns. This work has the potential to inform understanding of risk and prevention for Alzheimer’s disease and related dementias.