Sex Differences in the Association Between Metabolic Dysregulation and Cognitive Aging: The Health and Retirement Study

Abstract

Background

Dysregulation of some metabolic factors increases the risk of dementia. It remains unclear if overall metabolic dysregulation, or only certain components, contribute to cognitive aging and if these associations are sex specific.

Methods

Data from the 2006–2016 waves of the Health and Retirement Study (HRS) was used to analyze 7 103 participants aged 65 and older at baseline (58% women). We created a metabolic-dysregulation risk score (MDRS) composed of blood pressure/hypertension status, glycosylated hemoglobin (HbA1c)/diabetes status, total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), and waist circumference, and assessed cognitive trajectories from repeated measures of the HRS-Telephone Interview for Cognitive Status (HRS-TICS) over 10 years of follow-up. Linear mixed-effects models estimated associations between MDRS or individual metabolic factors (biomarkers) with mean and change in HRS-TICS scores and assessed sex-modification of these associations.

Results

Participants with higher MDRSs had lower mean HRS-TICS scores, but there were no statistically significant differences in rate of decline. Sex stratification showed this association was present for women only. MDRS biomarkers revealed heterogeneity in the strength and direction of associations with HRS-TICS. Lower HRS-TICS levels were associated with hypertension, higher HbA1c/diabetes, and lower HDL-C and TC, whereas faster rate of cognitive decline was associated with hypertension, higher HbA1c/diabetes, and higher TC. Participants with higher HbA1c/diabetes presented worse cognitive trajectories. Sex differences indicated that women with higher HbA1c/diabetes to have lower HRS-TICS levels, whereas hypertensive males presented better cognitive trajectory.

Conclusions

Our results demonstrate that metabolic dysregulation is more strongly associated with cognition in women compared with men, though sex differences vary by individual biomarker.

Biomarkers, Brain aging, Diabetes, Epidemiology

Cardiometabolic abnormalities cause immune and metabolic dysregulation and vascular senescence (1,2), in turn contributing to accelerated cognitive aging. Interventions to modify risk factors including obesity, dyslipidemia, and diabetes are proposed to preserve brain integrity and slow cognitive decline with aging (2–5). However, evidence for the effectiveness of these strategies is mixed; for example, randomized trials of lipid lowering drugs and blood pressure medications do not find evidence supporting prevention or delay of cognitive decline (6). Important gaps also exist in the epidemiology relating these risk factors to cognitive aging. In older adults, lower high-density lipoprotein cholesterol (HDL-C) (7), total cholesterol (TC) (8,9), diastolic blood pressure (DBP) (10,11), higher systolic blood pressure (SBP) (3,10), waist circumference (WC) and obesity (5,12,13), and hyperglycemia (14–17) are associated with declines in global and/or specific cognitive functions. But not all studies observe such associations (18,19). Moreover, the extent to which different indicators of metabolic dysregulation predispose to cognitive decline may vary by sex. For example, some studies find lower levels of HDL-C, having hypertension, and hyperglycemia associated with lower cognition only among women (17,20,21), although not all studies find sex differences (13,22–24).

To address these knowledge gaps, we analyzed metabolic dysfunction biomarker associations with cognitive trajectories in a national sample of older adults in the United States. We first tested how a composite index of cardiometabolic risk and then each of its individual components were associated with trajectories of cognitive decline over 10 years of follow-up and how these associations varied between older men and women.

Method

Study Population

Data come from the Health and Retirement Study (HRS), the largest on-going nationally representative longitudinal survey of adults over the age of 50 in the United States. The study began in 1992 with follow-up interviews every 2 years and new cohorts added every 6 years. The HRS has been approved by the University of Michigan institutional review board. All analyses for this study were approved by the University of North Carolina Institutional Review Board.

HRS survey designs and methods have been described previously (25,26). Data collection consisted of face-to-face baseline interviews and primarily telephone interviews for follow-up waves, until 2006, at which point half of the sample was randomly assigned face-to-face interviews to facilitate collection of physical and biological measures. The half receiving face-to-face interviews alternated at each subsequent wave (25). Our analyses used demographic, social, and cognitive data from the 2006–2016 core interviews, and the 2006, 2008, 2010, and 2012 Biomarker Data. Because biological samples were only collected as part of the enhanced face-to-face interview, which is given to half of the HRS study population at a time, baseline for our participants is either 2006, 2008, 2010, or 2012 (Figure 1). Of the 42 053 participants interviewed in HRS from 2006 to 2016, 7 103 met our inclusion criteria of being at least 65 years old at baseline and having complete covariate data, at least 1 valid biomarker data point, and at least 3 cognitive data points.

Figure 1.

Study flow chart with final study sample, n = 7 103.

Cognitive Function

Cognitive function was assessed at baseline and each follow-up visit for those aged 65 and older using the HRS-Telephone Interview for Cognitive Status (HRS-TICS) administered to the participant either over the telephone or face-to-face. The measure has a possible score range of 0–35 and includes immediate recall (0–10), delayed recall (0–10), serial 7s (0–5), backwards count from 20 (0–2), object naming (0–2), President naming (0–1), Vice President naming (0–1), and date (month, day, year, day of week; 0–4) encompassing cognitive domains of verbal memory, orientation, and executive functioning and attention. Further details on the HRS-TICS procedures can be found in Documentation of Cognitive Functioning Measures in the Health and Retirement Study (DR-006) link http://hrsonline.isr.umich.edu/sitedocs/userg/dr-006.pdf. Data from 2006 through 2016 (up to 6 visits and 10 years of follow-up) were included in this analysis. For analysis, the HRS-TICS score was standardized to a mean of 0 and a SD of 1.

Metabolic Dysfunction Risk Score

We measured metabolic dysfunction from blood chemistries assayed from dried blood spots (27), physical exam data, and participant reports about physician-diagnosed hypertension and diabetes. Blood chemistries included TC, HDL-C, and glycosylated hemoglobin (HbA1c). Physical exam parameters included SBP, DBP, and WC. Biospecimens were collected at the same time of the cognitive assessments. We computed the metabolic dysfunction risk score (MDRS) based on cardiometabolic health using diagnostic cutpoints specific for older adults and known to be important markers for the clinical management of chronic metabolic conditions in aging. Because it was not possible to use other established metabolic scores since not all the relevant data were collected in HRS, an in-house metabolic score (MDRS) was utilized. For the MDRS, we based blood pressure cutpoint measures on the 2017 Guidelines for Prevention, Detection, Evaluation, and Management of High Blood Pressure in Adults from the American College of Cardiology (28). HbA1c cutpoints follow glycemic goals established by the American Diabetes Association for diagnosis of diabetes (29). TC levels were considered poor based on the definition from the American Heart Association (AHA) 2021 updated recommendations for cardiovascular health and low HDL-C follows the AHA cutpoints for metabolic syndrome. The MDRS score is constructed as follows: one point for SBP ≥ 130 mmHg, DBP ≥ 80 mmHg, or report of physician-diagnosed hypertension; one point for HbA1c ≥ 6.5% or report of physician-diagnosed diabetes; one point for TC ≥ 240 mg/dL; one point for low HDL, defined as < 40 mg/dL for men and <50 dL for women; and one point for WC > 94 cm in men or >80 cm in women. A higher MDRS reflects higher metabolic dysregulation.

Covariates

Sociodemographic

Data including self-reported educational attainment, sex, age, and race/ethnicity were collected in participants’ respective baseline interviews. Educational attainment was categorized as less than a high school diploma, high school diploma, some college, or a college diploma and higher. Race/ethnicity was categorized as non-Hispanic White, non-Hispanic Black, Hispanic, or other, and sex was self-reported as either male or female. Age was grand mean centered based on the analytic sample’s mean baseline age of 73.2 years.

Cohort effect

To adjust for cohort differences and heterogeneity in trajectories, we included the HRS-created cohort variable to identify the 6-year birth cohort of participants: Cohort 1 (Asset and Health Dynamics Among the Oldest Old [AHEAD]) includes those born before 1924; Cohort 2 (Children of the Depression [CODA]) includes those born between 1924 and 1930; Cohort 3 (HRS-Original Cohort) includes those born between 1931 and 1941; and Cohort 4 (War Babies) includes those born between 1942 and 1947.

Statistical Analyses

Descriptive statistics were calculated to characterize the overall study population as well as the population stratified by sex. We used linear mixed-effects models with a random intercept and slope to test the relationship between metabolic dysregulation and cognitive trajectories using 3 models. Model 1 tested the association between the independent variable, MDRS or individual biomarker, and HRS-TICS. Model 2 tested the interaction between the MDRS/biomarker and sex. If a statistically significant sex interaction was detected in model 2, then we estimated the association separate for men and women (model 3). Metabolic dysregulation cutoff thresholds were used in the MDRS risk score analyses, but continuous measures were used when analyzing individual metabolic factors.

Model 1 specification is the following:

Level 1 Model: Growth trajectory with age within individuals

$${\displaystyle \begin{array}{ll}{\displaystyle \mathrm{H}\mathrm{R}\mathrm{S}-\mathrm{T}\mathrm{I}\mathrm{C}{\mathrm{S}}_{ti}=}& {\beta }_{0i}+{\beta }_{1i}\mathrm{t}\mathrm{i}\mathrm{m}\mathrm{e}+{\beta }_{2i}\mathrm{t}\mathrm{i}\mathrm{m}{\mathrm{e}}^2+{\beta }_{3i}\mathrm{A}\mathrm{g}\mathrm{e}\\ & {\displaystyle +{\beta }_{4i}\mathrm{A}\mathrm{g}{\mathrm{e}}^2+e_{ti}}\end{array}}$$

Level 2 Model: Heterogeneity in these trajectories across individuals for the intercept:

$${\displaystyle \begin{array}{ll}{\displaystyle {\beta }_{0i}=}& Y_{00}+Y_{01}\mathrm{M}\mathrm{D}\mathrm{R}\mathrm{S}/\mathrm{b}\mathrm{i}\mathrm{o}\mathrm{m}\mathrm{a}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}+Y_{02}\mathrm{A}\mathrm{g}\mathrm{e}+Y_{03}\mathrm{S}\mathrm{e}\mathrm{x}\\ & {\displaystyle +Y_{04}\mathrm{R}\mathrm{a}\mathrm{c}\mathrm{e}+Y_{05}\mathrm{E}\mathrm{d}\mathrm{u}\mathrm{c}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}+Y_{06}\mathrm{C}\mathrm{o}\mathrm{h}\mathrm{o}\mathrm{r}\mathrm{t}+u_{0i}}\end{array}}$$

for the linear growth rate:

$${\displaystyle \begin{array}{l}{\displaystyle {\beta }_{1i}=Y_{10}+Y_{11}\mathrm{M}\mathrm{D}\mathrm{R}\mathrm{S}/\mathrm{b}\mathrm{i}\mathrm{o}\mathrm{m}\mathrm{a}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}+Y_{12}\mathrm{A}\mathrm{g}\mathrm{e}+u_{1i}}\end{array}}$$

The Level 1 model represents the intraindividual change in HRS-TICS score (Y_ij) as a function of time. In this model, the MDRS for person i at time t is modeled as a function of intraindividual time-varying covariates, including a quadratic function of age and time. Time is measured as the number of years a person was in the HRS after the baseline interview, and age is mean centered. At Level 2, the model assesses the interindividual differences in change with time and age and determines the associations between the change with these variables and the MDRS/biomarker. The intercept coefficient (or growth rate) is modeled as a function of the MDRS/biomarker, age (mean centered = 73), sex (reference category is females), race (reference category White Non-Hispanic), education (reference is college degree or higher), and cohort (reference is the fourth cohort). The time and age coefficient are modeled as a function of the MDRS/biomarker. The coefficients Y₀₀–Y₀₆ represent the effects on mean level of HRS-TICS score, and the coefficient Y₁₁ represents the effects on the linear rate of change with time in the HRS-TICS. This minimally sufficient adjustment set was determined a priori by DAG analysis (30), and we further tested the model fit by generalized model fit statistics using the Bayesian information criterion (BIC). In addition, we divided the age and time variables by 5 to make the coefficients more visible.

Model 2 specification is as follows:

Level 1 Model: Growth trajectory with age within individuals

$${\displaystyle \begin{array}{ll}{\displaystyle \mathrm{H}\mathrm{R}\mathrm{S}-\mathrm{T}\mathrm{I}\mathrm{C}{\mathrm{S}}_{ti}=}& {\beta }_{0i}+{\beta }_{1i}\mathrm{t}\mathrm{i}\mathrm{m}\mathrm{e}+{\beta }_{2i}\mathrm{t}\mathrm{i}\mathrm{m}{\mathrm{e}}^2+{\beta }_{3i}\mathrm{A}\mathrm{g}\mathrm{e}\\ & {\displaystyle +{\beta }_{4i}\mathrm{A}\mathrm{g}{\mathrm{e}}^2+e_{ti}}\end{array}}$$

Level 2 Model: Heterogeneity in these trajectories across individuals for the intercept:

$${\displaystyle \begin{array}{l}{\displaystyle \begin{array}{ll}{\beta }_{0i}=& Y_{00}+Y_{01}\mathrm{M}\mathrm{D}\mathrm{R}\mathrm{S}/\mathrm{b}\mathrm{i}\mathrm{o}\mathrm{m}\mathrm{a}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}+Y_{02}\mathrm{S}\mathrm{e}\mathrm{x}\\ & {\displaystyle +Y_{03}\mathrm{S}\mathrm{e}\mathrm{x}\ast \mathrm{M}\mathrm{D}\mathrm{R}\mathrm{S}/\mathrm{b}\mathrm{i}\mathrm{o}\mathrm{m}\mathrm{a}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}+Y_{04}\mathrm{R}\mathrm{a}\mathrm{c}\mathrm{e}}\\ & {\displaystyle +Y_{05}\mathrm{E}\mathrm{d}\mathrm{u}\mathrm{c}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}+Y_{06}\mathrm{C}\mathrm{o}\mathrm{h}\mathrm{o}\mathrm{r}\mathrm{t}+u_{0i}}\end{array}}\end{array}}$$

for the linear growth rate:

$${\displaystyle \begin{array}{ll}{\displaystyle {\beta }_{1i}=}& Y_{10}+Y_{11}\mathrm{M}\mathrm{D}\mathrm{R}\mathrm{S}/\mathrm{b}\mathrm{i}\mathrm{o}\mathrm{m}\mathrm{a}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}+Y_{12}\mathrm{S}\mathrm{e}\mathrm{x}\\ & {\displaystyle +Y_{13}\mathrm{S}\mathrm{e}\mathrm{x}\ast \mathrm{M}\mathrm{D}\mathrm{R}\mathrm{S}/\mathrm{b}\mathrm{i}\mathrm{o}\mathrm{m}\mathrm{a}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}+u_{1i}}\end{array}}$$

If model 2 generated statistically significant interactions between the biomarker and sex, we proceeded to measure effect modification by sex stratifying the models. Model 3 specification follows that of model 1 without the sex variables. To ease interpretation, biomarkers in graphics are modeled as z-scores. All statistical tests were 2 tailed, and p-values were considered significant if <.05. All statistical analyses were conducted in SAS 9.4 (SAS Institute, Inc., Cary, NC).

Results

Of the total available sample (N = 9 379), almost a quarter percent of HRS participants did not meet our inclusion criteria (N = 2 276 [24.3%]). Compared with those eligible, ineligible individuals were older (mean age 78 [SD 7.8]), 4.5% more males, and had lower mean cognitive scores (mean HRS-TICS 18.6 [SD 5.8]). They also appear to be sicker as they had more markers of metabolic dysfunction. Of the 7 103 eligible HRS participants in our study, 42% were male, 77% White Non-Hispanic, and 38% had a high school–level education. Participants had an average of 4 follow-up visits (range 3–6), the mean age at baseline was 73.4 years (SD 6.3), and the mean HRS-TICS score at baseline was 22.3 (SD 4.5) points. The HRS-TICS scores were normally distributed. Women had significantly higher baseline level of cognitive functioning than men (women’s mean 22.5, SD 4.7 [range 4–35] median 22.5 and mode 23; men’s mean 22.1, SD 4.3 [range 3–35] median 22.0 and mode 24.0, p < .001). Less than 2% of the sample (n = 122) presented with no cardiometabolic risk factors, or an MDRS of zero; of these, 48% (n = 58) were men and 52% (n = 64) were women. Over half the study population had an MDRS of 3 or more (n = 3 601) and <1% had the maximum score of 5. Full descriptive statistics are given in Table 1.

Table 1.

Baseline Demographics of Eligible Sample in the HRS, N = 7 103

	Ineligible Sample (N = 2 276)	Eligible Sample (N = 7 103)	Male (n = 2 976), 42%	Female (n = 4 127), 58%
Age, mean (SD)	78.1 (7.8)	73.4 (6.3)	73.2 (6.1)	73.5 (4.4)
Cognition (HRS-TICS), mean (SD)	18.6 (5.8)	22.3 (4.5)	22.1 (4.3)	22.5 (4.7)
Birth cohort, n (%)
AHEAD (median age 86)	623 (27.4)	576 (8.1)	200 (6.7)	376 (9.1)
CODA (median age 79)	601 (26.4)	1 436 (20.2)	610 (20.5)	826 (20.0)
War Babies (median age 66)	973 (42.8)	511 (7.2)	216 (7.3)	295 (7.2)
HRS (median age 70)	78 (3.4)	4 580 (64.5)	1 950 (65.5)	2 630 (63.7)
Education, n (%)
<HS	733 (32.2)	1 535 (21.6)	626 (21.0)	909 (22.0)
HS diploma	798 (35.1)	2 704 (38.1)	1 019 (34.2)	1 685 (40.8)
Some college	403 (17.1)	1 467 (20.7)	569 (19.1)	898 (21.8)
College degree/more	340 (14.9)	1 397 (19.7)	762 (25.6)	635 (15.4)
Race, n (%)
Non-Hispanic White	1 735 (76.2)	5 502 (77.5)	2 348 (78.9)	3 154 (76.4)
Non-Hispanic Black	310 (13.6)	890 (12.5)	328 (11.0)	562 (13.6)
Hispanic	190 (8.4)	589 (8.3)	242 (8.1)	347 (8.4)
Other	39 (1.7)	122 (1.7)	58 (2.0)	64 (1.6)
Metabolic Dysfunction Risk Score, n (%)
None	49 (2.2)	122 (1.7)	58 (1.9)	64 (1.6)
One	278 (12.2)	853 (12.0)	411 (13.8)	442 (10.7)
Two	787 (34.6)	2 527 (35.6)	1 110 (37.3)	1 417 (34.3)
Three	791 (34.8)	2 609 (36.7)	1 041 (35.0)	1 568 (38.0)
Four	355 (15.6)	947 (13.3)	350 (11.8)	597 (14.5)
Five	16 (0.7)	45 (0.6)	6 (0.2)	39 (0.9)
Biomarkers
SBP, mean (SD)	138.9 (49.8)	137.0 (44.7)	138.4 (41.5)	135.9 (46.9)
DBP, mean (SD)	76.7 (12.7)	70.5 (78.1)	78.3 (11.0)	78.0 (11.4)
Hypertension diagnosis, n (%)	1 569 (68.9)	4 583 (64.5)	1 869 (62.8)	2 714 (65.9)
HbA1c, mean (SD)	6.0 (1.1)	5.9 (0.9)	5.9 (0.9)	5.9 (0.8)
Diabetes, n (%)	653 (28.7)	1 580 (22.2)	731 (24.6)	849 (20.6)
HDL-C, mean (SD)	52.2 (15.7)	54.0 (15.6)	48.7 (13.5)	57.9 (15.9)
TC, mean (SD)	190.9 (42.2)	196.8 (41.8)	187.5 (38.4)	203.4 (42.8)
WC, n (%)	100.6 (15.2)	101.3 (14.7)	105.7 (12.4)	98.0 (15.4)

Notes: AHEAD = Asset and Health Dynamics Among the Oldest Old; CODA = Children of the Depression; DBP = diastolic blood pressure (mmHg); HbA1c = hemoglobin A1c (percentage); HDL = high-density lipoprotein cholesterol (mg/dL); HRS-TICS = Health and Retirement Study Telephone Interview Cognitive Status; HS = high school; SBP = systolic blood pressure (mmHg); TC = total cholesterol (mmHg); WC = waist circumference (cm). In the ineligible sample, the following number of participants had missing data: 2 education and race. One ineligible individual from the most recent HRS cohort was ineligible and not included in analysis.

MDRS and Cognitive Trajectories

Older adults with higher MDRS had lower mean HRS-TICS scores (per unit of MDRS, intercept b = −0.03 HRS-TICS SD, 95% CI [−0.04, −0.01]). MDRS was not associated with the rate of decline in HRS-TICS (b = −0.01, 95% CI [−0.02, 0.001]). In sex-stratified analysis, the association between MDRS and mean HRS-TICS was larger for women than for men (women, b = −0.05, 95% CI [−0.07, −0.03]; men, b = 0.01, 95% CI [−0.02, 0.03]). This sex difference was statistically significant (p < .001 for test of sex × MDRS interaction). MDRS associations with HRS-TICS decline were not statistically different from zero for both women and men. Effect sizes are graphed in Figure 2 and reported in Supplementary Table S1. Predicted trajectories of cognitive decline are graphed in Figure 3. For illustration purposes and to possibly address the trajectory instability for those with 4 or 5 metabolic dysregulatory factors and age 85-plus, we combined them and present them as a single category in Figure 3.

Figure 2.

Association between biomarkers and cognitive trajectories. This figure demonstrates the relationship between metabolic dysregulation biomarkers (or diagnosed health conditions) included in the MDRS and population mean (standardized) HRS-TICS score (red) units and rate of change in HRS-TICS score over 5 yr (blue). Output are beta estimates and 95% CIs for the respective variable terms in linear mixed-effects models with standardized biomarker values as the exposure. All shown estimates come from final, fully adjusted overall and sex-stratified regression models. DBP = diastolic blood pressure; HbA1c = hemoglobin A1C (percentage); HDL-C = high-density lipoprotein cholesterol (mg/dL); HRS = Health and Retirement Study; HRS-TICS = Telephone Interview Cognitive Screening from Health and Retirement Study; MDRS = Metabolic Dysfunction Risk Score; SBP = systolic blood pressure (mmHg); TC = total cholesterol (mmHg); WC = waist circumference (cm).

Figure 3.

Predicted cognitive trajectories per MDRS point for all and stratified by sex.

Output represents predicted HRS-TICS scores for each participant and observation from fully adjusted model by MDRS points. We calculated the predicted value for HRS-TICS based on the model estimates as follows: Level 1: Y_ij = β _0i time + β _2i time 2; Level 2: For the intercept: β _0i = γ 00 + γ 01 MDRS + γ 02 age+ γ 03 age 2 + γ 04 race + γ 05 education + γ 06 cohort + u 0i, and for linear growth rate: β _1i = γ 10 + γ 11 MDRS + γ 12 age + u 1i, which indicates the predicted HRS-TICS score for each participant’s covariate values. MDRS scores of 4 and 5 were merged for data figure only. X-axis shows SD from the mean cognitive score, and y-axis shows age in years. MDRS = Metabolic Dysfunction Risk Score.

Individual Biomarkers and Cognitive Trajectories

Effect sizes for all individual biomarkers are graphed in Figure 2 for continuous and categorical individual biomarkers, and statistical details are reported in Supplementary Tables S2–S9. Predicted trajectories of cognitive decline are graphed in Figure 4.

Figure 4.

Predicted cognitive trajectories for individual biomarkers for all and stratified by sex for those biomarkers where the sex stratification was significant. This figure demonstrates the relationship between biomarkers (dichotomized at MDRS cut point values) and predicted HRS-TICS score. Outputs are predicted HRS-TICS scores for each participant and observation from fully adjusted overall and sex-stratified linear mixed-effects models. X-axis shows SD from the mean cognitive score, and y-axis shows age in years. DBP = diastolic blood pressure; HbA1c = hemoglobin A1C (percentage); HDL-C = high-density lipoprotein cholesterol (mg/dL); HRS = Health and Retirement Study; HRS-TICS = Telephone Interview Cognitive Screening from Health and Retirement Study; MDRS = Metabolic Dysfunction Risk Score; SBP = systolic blood pressure (mmHg); TC = total cholesterol (mmHg); WC = waist circumference (cm).

Systolic Blood Pressure

SBP was not associated with levels of cognitive function across follow-up or with the rate of decline (level per unit of SBP, b = 0.01, 95% CI [−0.01, 0.03], p = .202; slope b = −0.01, 95% CI [−0.02, 0.002], p = .105). There were no differences in cognitive function by sex for SBP across study follow-up or trajectories (level b = −0.002, 95% CI [−0.04, 0.03], p = .909; slope b = 0.002, 95% CI [−0.003, 0.01], p = .416).

Diastolic Blood Pressure

DBP was not associated with levels of cognitive function across follow-up or with the rate of decline (level per unit of DBP, b = 0.01, 95% CI [−0.01, 0.02], p = .528; slope b = 0.01, 95% CI [−0.005, 0.02], p = .232). There were no differences in cognitive function by sex for DBP across study follow-up or trajectories (level b = −0.01, 95% CI [−0.05, 0.02], p = .383; slope b = −0.003, 95% CI [−0.01, 0.002], p = .253).

Hypertension

A diagnosis of hypertension was associated with lower levels of cognitive function across follow-up when compared with those without hypertension, but did not differ in their rate of cognitive decline (level b = −0.06, 95% CI [−0.09, −0.03], p = .001; slope b = 0.01, 95% CI [−0.01, 0.04], p = .316). Women with hypertension showed lower levels of cognitive functioning across follow-up when compared with those without hypertension, but did not differ in their rate of cognitive decline (level b = −0.06, 95% CI [−0.10, −0.01], p = .012; slope b = −0.01, 95% CI [−0.04, 0.03], p = .644). Men with hypertension also showed differences in average level of cognitive functioning across follow-up when compared with men without hypertension, but the rate of cognitive decline was slower for those with hypertension (level b = −0.06, 95% CI [−0.10, −0.01], p = .027; slope b = 0.05, 95% CI [0.01, 0.08], p = .014). In the interaction model, there were no differences in cognitive function by sex for hypertension across follow-up (level b = 0.01, 95% CI [−0.05, 0.08], p = .668); however, the rate of cognitive decline differed by sex for those with hypertension, but not for those without hypertension (slope hypertension × sex b = 0.01, 95% CI [0.003, 0.02], p = .004; slope no-hypertension × sex b = −0.001, 95% CI [−0.01, 0.01], p = .723).

Glycated Hemoglobin A1c

Older adults with a higher HbA1c showed lower levels of cognitive functioning across follow-up and differ in the rate of cognitive decline compared with those with lower HbA1c (level per unit of HbA1c, b = −0.03, 95% CI [−0.04, −0.01], p = .002; slope b = −0.03, 95% CI [−0.04, −0.01], p < .001). Women with higher HbA1c also showed differences in average level of cognitive functioning across follow-up and differ in the rate of cognitive decline compared with those with lower HbA1c (level per unit of HbA1c, b = −0.04, 95% CI [−0.07, −0.02], p < .001; slope b = −0.03, 95% CI [−0.04, −0.01], p = .006). In contrast, men with higher HbA1c did not show differences in average level of cognitive functioning across follow-up, but did show faster rate of cognitive decline compared with those with lower HbA1c (level per unit of HbA1c, b = −0.004, 95% CI [−0.03, 0.02], p = .706; slope b = −0.03, 95% CI [−0.05, −0.01], p = .001). There were sex differences in level of cognitive function by HbA1c but not in cognitive trajectories (level per unit of HbA1c × sex b = 0.05, 95% CI [0.02, 0.08], p = .003; slope HbA1c × sex b = −0.001, 95% CI [−0.01, 0.004], p = .669).

Diabetes

Having a doctor diagnosis of DM was associated with lower levels of cognitive functioning across follow-up and faster rate of cognitive decline compared with those without diabetes (level b = −0.08, 95% CI [−0.12, −0.05], p < .001; slope b = −0.05, 95% CI [−0.08, −0.02], p = .001). Diabetic women also showed lower average level of cognitive functioning across follow-up and faster rate of cognitive decline compared with those without diabetes (level b = −0.15, 95% CI [−0.21, −0.10], p < .001; slope b = −0.05, 95% CI [−0.09, −0.01], p = .021). In contrast, diabetic men did not show differences in average level of cognitive functioning across follow-up, but did show faster rate of cognitive decline compared with those without diabetes (level b = 0.01, 95% CI [−0.05, 0.06], p = .797; slope b = −0.05, 95% CI [−0.10, −0.01], p = .015). In the interaction model, there were differences in cognitive function by sex for diabetes across follow-up (level b = 0.17, 95% CI [0.10, 0.25], p < .001). But rate of cognitive decline was different by sex. Nondiabetics had slower rate of cognitive decline (slope b = 0.01, 95% CI [0.0002, 0.01], p = .041), but not diabetics (slope b = 0.005, 95% CI [−0.01, 0.02], p = .407).

High-Density Lipoprotein Cholesterol

Older adults with a higher levels of HDL-C showed higher levels of cognitive functioning across the study follow-up but did not differ in the rate of cognitive decline compared with those with lower HDL-C (level b = 0.02, 95% CI [0.004, 0.04], p = .019; slope b = −0.00004, 95% CI [−0.01, 0.01, p = 1.00). No cognitive level or trajectory differences were observed by sex for HDL-C (level b = −0.03, 95% CI [−0.07, 0.01], p = .151; slope b = −0.004, 95% CI [−0.01, 0.002], p = .229).

Total Cholesterol

Higher levels of TC showed higher levels of cognitive functioning across the study follow-up but faster rate of cognitive decline compared with those with lower TC (level, b = 0.02, 95% CI [0.004, 0.04], p = .017; slope b = −0.01, 95% CI [−0.03, −0.001, p = .038). No cognitive level or trajectory differences were observed by sex for TC (level b = −0.01, 95% CI [−0.05, 0.02], p = .538; slope b = 0.001, 95% CI [−0.005, 0.01], p = .820).

Waist Circumference

Larger WC was associated with lower cognitive functioning across the study follow-up but did not differ in the rate of cognitive decline compared with those with lower WC (level, b = −0.02, 95% CI [−0.04, −0.005], p = .011; slope b = 0.001, 95% CI [−0.01, 0.01], p = .909). No cognitive level or trajectory differences were observed by sex for WC (level b = 0.01, 95% CI [−0.02, 0.05], p = .481; slope b = 0.004, 95% CI [−0.002, 0.01], p = .172).

Discussion/Conclusion

In a large, national longitudinal study of older adults in the United States, we found that higher levels of metabolic dysregulation were associated with lower levels of cognitive functioning, and these associations were stronger for women than men. For each MDRS component, women lost 0.24 HRS-TICS points in a 5-year period, which translates into a maximum of 1.2 HRS-TICS points lower in a woman with all 5 MDRS components present. However, there was no association between MDRS and rate of cognitive decline, either in the total population or in sex-specific populations. In analyses of the individual MDRS components, we found that certain biomarkers (eg, HbA1c) contributed more than others to the association with cognitive function.

These results align in general, with most longitudinal studies (13,31–35), but not all (36). However, it is worth pointing out that some of these other studies that obtained similar results to ours used similar definitions of metabolic risk, but had shorter study follow-ups (33,34,36–38), smaller sample sizes (13,31–33,35,36), and most were national representative samples (33,38,39). In addition, all these studies compared the association between metabolic dysregulation and cognition but did not further explore effect modification by sex. Our finding that metabolic dysregulation was not associated with rate of cognitive decline is supported by one study (33) but contrasts with findings from others (31,36,38). These discrepancies are likely due to differences in study design, population, and measurement and operationalization of key variables but also possibly due to the wide variety of component causes of cognitive change and decline (40).

With aging, women tend to have more metabolic syndrome than men (41), overall score higher on cognitive status measures, and are known to live longer. Thus, to minimize spurious differences by sex, we used a multilevel study design that accounted for sex differences and inequality in sample size. In addition, we included only those who had at least 3 or more cognitive assessments, and adjusted for variables known to correlate with cognitive function differently by sex (ie, cohort, education level, race). Nonetheless, our findings showed that higher MDRSs were associated with lower cognitive function in women only which is consistent with earlier studies (31,41). These observed sex differences suggest possible unidentified metabolic differences in the aging process, which may lead to potential differences in therapeutic approaches for men and women.

Individual components of metabolic dysfunction varied in their association with cognitive function. This has been previously reported in a systematic review (40). In our study, a diagnosis of hypertension was associated with worse cognitive levels for both men and women as previously reported by a few (40), but not all (37,41) prior studies. Hyperglycemia (higher HbA1c and being diagnosed diabetic) was associated with poorer levels of cognitive function, which consistently aligns with prior research (15,17,34,37,39,42–44). We also found that having lower HDL-C and lower TC was associated with poorer levels of cognitive function which parallels extant literature (7,13,20,37,45,46). Because HDL-C is presumed to reverse the cholesterol transport pathway, improve cardiovascular function, and is believed to preserve cognitive function (47), it may be possible that there is a moderating effect between cholesterol type and aging that warrants further study. We also found that having larger WC was associated with poorer levels of cognitive function, which supports most reported findings (13,39,48), but not all (49).

Few individual components included in our metabolic dysregulation risk score led to faster rate of cognitive decline. We found that older adults with higher HbA1c or diabetes had faster cognitive decline. In the Sacramento Area Latino Study of Aging, researchers found elevated fasting blood glucose or use of antidiabetic medications associated with lower delayed recall scores over time (34). Others have found an association with slower rate of cognitive decline (36). These discrepancies are likely due to different lengths of follow-up, sample size, or sample age range in these studies. However, hypertensive men had a slower rate of cognitive decline. Because hypertension in midlife, particularly in men, may lead to cardiovascular events, we cannot rule out that this result in men may be a function of residual survival bias. This finding contrasts with other research that has shown faster declines in cognitive function among younger cohorts with higher blood pressure measures regardless of sex have been observed (10).

We also found that higher levels of TC were associated with better overall cognitive function, yet faster cognitive decline. Prior research has demonstrated that high levels of TC in early and midlife are negatively associated with health outcomes; however, high TC has been linked to longevity and higher levels of cognition among older adults (8,9,45,46). Our association between higher TC and faster cognitive decline contradicts some previous findings (8,9).

Some of the metabolic dysregulatory factors we included in this study are recognized to contribute to arterial aging which triggers chronic inflammatory processes that lead to atherosclerosis and subsequent complications such as increased risk for developing Alzheimer’s disease and vascular dementia (5). Our results provide further evidence that these metabolic markers may lead to lower cognitive function in late life, particularly among women. However, it is important to point that there are several cross-over associations between these cardiometabolic factors and cognitive function across the life span, where a biomarker may be detrimental to survival and cognitive function in early midlife but less supported in late life. For instance, hypertension in midlife may predict cardiovascular disease and lower cognitive function, but hypertension relationships with cognitive outcomes in older ages may be attenuated. Further studies should consider adjustment for additional comorbidities.

Our study has several strengths including use of a large, national sample of older adults for up to a 10-year follow-up period. We also evaluated how the associations between metabolic dysregulation and individual biomarkers and cognition varied by sex, demonstrating a stronger overall association for women. However, there are several limitations to this study. Although the HRS sample was designed to be representative of the United States, the eligibility requirements for the subsample used in the present study may have altered its national representativeness. Additionally, HRS-TICS is a brief measure of global cognitive function. Performance on such measures can be affected by multiple factors, not all are solely due to cognition, such as hearing impairment. Given the brevity of the HRS-TICS, it is not possible to investigate in-depth specific cognitive domains, such as memory or executive function. Domains that others have found to be related to metabolic markers (34,36,37). Although we accounted for cohort membership, it must be noted that HRS participants who experience cognitive decline were less likely to return for cognitive assessments over time. Furthermore, older cohort participants present higher attrition rates than younger ones from more recent waves. In addition, we did not account for medication use, which may have affected some of our findings. It is possible that if we accounted for medications we may have captured people who did not self-report a diagnosis and our findings may have differed.

In the United States, metabolic dysregulation appears to be more prevalent among certain underrepresented groups (23,34) who are also at increased risk of dementia. African Americans are twice as likely to have dementia compared with Whites, and Hispanics are 1.5 times more likely than Whites (50). It is estimated that 10.3% White, 13.8% African American, and 12.2% Hispanic Medicare beneficiaries have Alzheimer’s disease or some other type of dementia (50).

Therefore, our results have implications for the prevention and delay of onset of cognitive decline in aging populations. Although many potential risk factors for cognitive decline and dementia are not modifiable, the cardiometabolic factors included in the MDRS are modifiable, even in later life, through medication and/or behavior modification. Our results suggest that maintaining metabolic factors within expected clinical ranges may maintain cognitive health, and some may even slow the rate of cognitive decline. Future research should identify optimal metabolic sex- and age-stratified biomarker ranges that will maintain cognitive function and possibly deter cognitive decline and dementia.

Funding

This work was supported by the National Institutes of Aging (grant no. R01AG057800) (PI: Y.C.Y.).

Acknowledgments

The authors thank Max Reason (University of North Carolina at Chapel Hill, North Carolina) for his administrative assistance.

Author Contributions

M.C.-K. (Duke University, North Carolina) contributed to design and conceptualized study; performed biostatistical review of results; drafted the manuscript; R.C.S. (University of North Carolina at Chapel Hill, North Carolina) contributed to data analysis, drafting, and revision of manuscript; H.K.O. (University of North Carolina at Chapel Hill, North Carolina) contributed to drafting and revision of manuscript; D. W. B. (Columbia University, New York) contributed to revision of statistical analyses and revision of manuscript. Y.C.Y. (Principal Investigator; University of North Carolina at Chapel Hill, North Carolina) led and coordinated communication among sites. Reviewed manuscript for intellectual content. Revision of statistical analyses and revision of manuscript. B.L.P. (Co-investigator; Duke University North Carolina) contributed to revision of manuscript for intellectual content.

Conflict of Interest

None declared.

Data Availability

Data are publicly available from the HRS website (https://hrs.isr.umich.edu/data-products).