Is Midlife Metabolic Syndrome Associated With Cognitive Function Change? The Study of Women’s Health Across the Nation

Abstract

Context

Metabolic syndrome (MetS) affects cognitive function in late life, particularly in women. But longitudinal research is scarce on associations of MetS with cognitive function during midlife.

Objective

To determine associations between MetS exposure and cognitive function trajectories in midlife women.

Design and Setting

This is a 17-year prospective, longitudinal study of multiracial/ethnic women in 7 US communities, with annual/biennial assessments.

Participants

Participants were 2149 US women traversing menopause.

Exposure

Exposure consisted of MetS assessments (median 4 assessments over 4 years).

Main Outcome Measures

Main outcome measures were assessments of cognitive function in 3 domains: perceptual speed (symbol digit modalities test, SDMT), episodic memory (East Boston Memory Test, EBMT), and working memory (Digit Span Backward Test, DSB).

Results

By their first cognitive assessment (age 50.7 ± 2.9 years), 29.5% met the criteria for MetS. Women completed a median (interquartile range [IQR]) of 6 (IQR 4–7) follow-up cognitive assessments over 11.2 (IQR 9.2–11.5) years. Women with MetS, compared with those without, had a larger 10-year decline in SDMT z-score (estimate –0.087, 95% confidence interval, –0.150 to –0.024; P = 0.007), after adjustment for cognitive testing practice effects, sociodemographics, lifestyle, mood, and menopause factors. As such, MetS accelerated the 10-year loss of perceptual speed by 24%. MetS did not differentially affect the rate of decline in either immediate (P = 0.534) or delayed (P = 0.740) episodic memory or in working memory (P = 0.584).

Conclusions

In midlife women MetS exposure was associated with accelerated decline in perceptual speed, but not episodic or working memory.

Metabolic adaptations are fundamental for functional resilience of every living cell, tissue, and organ, including the brain (1). Metabolic syndrome (MetS) represents “de-tuning” of metabolic adaptations and, thus, a decreased metabolic resilience with detrimental health consequences (2). Specifically, MetS leads to a 5-fold higher risk of diabetes and a 2-fold higher risk of cardiovascular and cerebrovascular disease (3–5). Based on the harmonized definition (5, 6), MetS is clinically diagnosed by meeting ≥3 out of 5 criteria defining abdominal obesity, hypertension, dysglycemia, and dyslipidemia. Postulated mechanisms underlying MetS involve insulin resistance, proinflammatory and prothrombotic impairment, defective endocrine and neurotransmitter signaling, and autonomic dysregulation, provoked by nutrient oversupply and inadequate disposal (2, 7–9). For a long time, glucose and lipid metabolism were considered to be primarily affected. More recent metabolomic evidence suggests the involvement of multiple metabolic superpathways, including dysregulation of amino acid and nucleotide metabolism (10, 11). Moreover, MetS phenotypes depend on genetic, age-related, environmental, and behavioral factors (12–14). Although MetS is typically associated with obesity, MetS also affects the nonobese (11, 15). Given apparent unified versatile pathophysiology, it is prudent to approach MetS as a syndrome rather than as clusters of a few diagnostic components (4, 5).

The age-specific prevalence of MetS is greatest at midlife (16), and in women MetS emerges with the menopausal transition (17, 18). In addition, women compared with men have higher age-specific prevalence of late-life dementia (19). As a prevalent condition (20), MetS attracts considerable interest as a preventive target for aging-related functional impairment (21, 22).

The effects of MetS on cognitive decline in people without dementia have been systematically reviewed (23, 24), admittedly with the relative paucity of longitudinal research in midlife individuals. A meta-analysis of cross-sectional studies (25) supports an association between MetS and most forms of executive function, information processing speed, and word generation, particularly in midlife women (26). The only longitudinal study of 45- to 64-year-old women and men found no association between MetS and a 6-year trajectory of cognitive decline (27), possibly due to insufficient length of follow-up and too few time points of cognitive assessments. The same study reported cross-sectional associations of MetS with executive dysfunction and word generation deficits (27). The discrepancies between cross-sectional and longitudinal findings (28) are often related to practice effects related to the gain of experience with repeating cognitive assessments.

Thus, we explored the longitudinal associations between MetS and change in cognitive function in the Study of Women’s Health Across the Nation (SWAN). SWAN, a study of health and aging during the menopausal transition, has shown an increasing prevalence of MetS during midlife (17). We hypothesized that MetS exposure during midlife in these women would be associated with steeper declines of cognitive function, independent from practice effects and sociodemographic, lifestyle, mood, and menopausal factors.

Materials and Methods

Standard protocol approvals and consents

Women from 7 US sites were enrolled in SWAN, a multiethnic/racial, community-based study. Procedures were approved by each site’s institutional review board, and all participants provided written informed consent. The design and analyses for this manuscript were approved by the SWAN Publications and Presentations Committee.

Participants

The SWAN inception cohort consisted of 3302 women who were recruited from 5 ethnic/racial groups: African American (Boston, MA; Chicago, IL; Detroit, MI area; and Pittsburgh, PA); Chinese (Oakland, CA area); Hispanic (Newark, NJ); Japanese (Los Angeles, CA); and non-Hispanic white (all sites), as detailed previously (29). Eligibility criteria included being aged 42 to 52 years at enrollment (1996/1997); having an intact uterus and at least 1 ovary; not being pregnant, breastfeeding, or lactating; not using oral contraceptives or reproductive hormone therapy; and reporting at least 1 menstrual cycle during the prior 3 months. This report includes data collected at the baseline assessment (1996/1997) and annual/biennial follow-up assessments that occurred through 2013.

Cognition was first tested at the fourth annual follow-up, with 2466 women completing cognitive testing according to the study protocol at ≥2 visits from the fourth to thirteenth follow-up visit (median = 11.2 years, interquartile range, [IQR] 9.2–11.5 years). To ensure longitudinal assessment of MetS, we excluded 53 women with fewer than 2 MetS assessments before the cognitive baseline. To reduce the disease-specific effects of stroke and diabetes on cognitive function, we excluded 264 cases of stroke or diabetes reported at the cognitive baseline. The final analytic sample consisted of 2149 women. Table 1 summarizes the comparison of baseline characteristics between the analytic sample and those excluded (n = 1153).

Table 1.

Baseline Comparison of the Analytical Sample and Excluded Cases

	Characteristics		Analytical Sample (n = 2149)	Excluded Cases (n = 1153)	P Value
Sociodemographic factors	Age at study baseline, mean ± SD		46.4 ± 2.7	46.3 ± 2.7	0.659a
	Site–race/ethnicity, n (%)	Boston black	116 (5.4)	83 (7.2)	<0.001
		Boston white	211 (9.8)	42 (3.6)
		Chicago black	134 (6.2)	114 (9.9)
		Chicago white	142 (6.6)	67 (5.8)
		Los Angeles Japanese	237 (11.0)	44 (3.8)
		Los Angeles white	172 (8.0)	43 (3.7)
		Michigan black	198 (9.2)	127 (11.0)
		Michigan white	136 (6.3)	82 (7.1)
		New Jersey Hispanic	80 (3.7)	206 (17.9)
		New Jersey white	47 (2.2)	99 (8.6)
		Oakland Chinese	208 (9.7)	42 (3.6)
		Oakland white	151 (7.0)	58 (5.0)
		Pittsburgh black	103 (4.8)	59 (5.1)
		Pittsburgh white	214 (10.0)	87 (7.5)
	Married, n (%)	Yes	1466 (68.2)	682 (59.2)	<0.001
		No	656 (30.5)	444 (38.5)
		Missing	27 (1.3)	27 (2.3)
	College education, n (%)	Yes	1045 (48.6)	356 (30.9)	<0.001
		No	1088 (50.6)	782 (67.8)
		Missing	16 (0.7)	15 (1.3)
	Financial strain, n (%)	No	1412 (65.7)	555 (48.1)	<0.001
		Yes	725 (33.7)	587 (50.1)
		Missing	12 (0.6)	11 (1.0)
Lifestyle, mood, and menopausal factors	Smoker, n (%)	Never	1849 (86.0)	871 (75.5)	<0.001
		Yes	298 (13.9)	282 (24.5)
		Missing	2 (0.1)	0 (0)
	Alcohol servings/ week, n (%)	None	971 (45.2)	582 (50.5)	0.002
		1–7	763 (35.5)	376 (32.6)
		> 7	314 (14.6)	127 (11.0)
		Missing	101 (4.7)	68 (5.9)
	Physical activity score, mean ± SD		7.8 ± 1.8	7.3 ± 1.8	<0.001a
	Depressive symptom score, mean ± SD		9.8 ± 9.1	12.4 ± 10.6	<0.001a
	Anxiety symptom score, mean ± SD		2.3 ± 2.2	2.8 ± 2.7	<0.001a
	Sleep difficulty, n (%)	No	1526 (71.0)	737 (63.9)	<0.001
		Yes	612 (28.5)	411 (35.6)
		Missing	11 (0.5)	5 (0.4)
	Vasomotor symptom days/2 weeks, n (%)	6–14	1360 (63.3)	634 (55.0)	<0.001
		1–5	579 (26.9)	342 (29.7)
		None	201 (9.4)	165 (14.3)
		Missing	9 (0.4)	12 (1.8)
	Menopause stagesb | hormone therapy, n (%)	Premenopause	1526 (71.0)	737 (63.9)	<0.001
		Early perimenopause	612 (28.5)	411 (35.6)
		Undetermined	11 (0.5)	5 (0.4)

Abbreviations SD, standard deviation.

^a P-values from 2 sample t tests except for depressive symptom and anxiety symptom scores (2-sided Wilcoxon test).

^bPostmenopausal women were excluded at the study baseline by design.

Assessment of cognitive function

Perceptual speed (a subcategory of processing speed) was assessed with the symbol digit modalities test (SDMT) (30) that involves using a model to match as many unfamiliar symbols with digits as possible in 90 seconds. Verbal episodic memory was evaluated by assessing immediate and delayed (by approximately 10 minutes) recall of 12 ideas contained in the East Boston 36-word paragraph (East Boston memory test, EBMT) (31, 32). Working memory was assessed with digit span backward (DSB) (33), which involved listening to increasingly longer strings of digits, ranging from 2 to 7, and repeating them backward without error. Tests were professionally translated into Cantonese, Japanese, and Spanish. Bilingual participants who indicated a preference for a specific language always took tests in the same language (34). The first available cognitive assessment for each woman in the SWAN study (at SWAN visit 4 or later) was considered to be their cognitive baseline.

Assessment of metabolic syndrome

Cumulative prevalence of MetS assessed over 2 or more visits before or at the women’s cognitive baseline (MetS exposure) was selected as a primary predictor. Consistent with the current diagnostic criteria (5), MetS was defined by the presence of at least 3 of the following 5 components: abdominal obesity (waist circumference ≥80 cm for Chinese, Japanese, and Hispanic women and ≥88 cm for other ethnic/racial groups), hypertension (systolic blood pressure ≥130 mm Hg, diastolic blood pressure ≥85 mm Hg, or the taking of blood pressure medication), low high-density lipoprotein cholesterol level (<1.3 mmol/L), hypertriglyceridemia (fasting triglycerides level ≥1.7 mmol/L), and impaired fasting glucose (fasting glucose level ≥5.6 mmol/L).

Waist circumference was measured to the nearest 0.1 cm with a measuring tape placed horizontally around the participant at the narrowest part of the torso. Measurements were taken over light clothing at the end of a normal exhalation. Blood pressure was measured on the right arm in a seated position after at least 5 minutes of rest using a sphygmomanometer. Respondents had not smoked or consumed caffeine within the prior 30 minutes. Appropriate cuff size was determined based on arm circumference. Two sequential blood pressure values were completed at least 2 minutes apart and averaged. Blood samples for glucose, high-density lipoprotein cholesterol, and triglycerides, collected after a 12-hour fast, were frozen and shipped for testing to Medical Research Laboratories, which are certified by the National Heart, Lung, and Blood Institute, Centers for Disease Control and Prevention Part III program (35).

Covariates

Covariates were chosen based on their potential to confound hypothesized associations.

Sociodemographic factors documented at study baseline included study site (Pittsburgh, PA; Chicago, IL; Oakland, CA; Los Angeles, CA; Boston, MA; Newark, NJ: and Detroit-area, MI), ethnicity/race (African American, Chinese, Japanese, Hispanic, and non-Hispanic white), financial strain (defined as difficulty paying for basic food and housing, dichotomized as “not difficult at all,” and “somewhat difficult/very difficult”), education (less than college degree and college degree or higher), and age calculated from birth date to cognitive baseline. The site and race/ethnicity was a composite variable, with each race/ethnic group at each study site as a separate category because these 2 variables were confounded in the SWAN study.

Lifestyle, mood, and menopausal factors were assessed at follow-up visits close to or concurrent with cognitive function assessments and were analyzed as time-varying covariates. These covariates included self-reported smoking status (no or yes), alcohol use (none, 1–7 servings/week, >7 servings per week) (36), and nonoccupational physical activity as the total sum score of the adapted Kaiser Physical Activity Survey (37).

Sleep difficulty included any 1 of the 3 self-reported sleep symptoms occurring ≥3 times a week during the 2 weeks before assessment: trouble falling asleep (sleep initiation difficulty), waking up several times a night (sleep maintenance difficulty), and waking up earlier than planned and inability to fall asleep again (early morning awakening) (38).

A depressive symptoms score was assessed using a Center for Epidemiological Studies Depression Scale (38); anxiety score was calculated as the sum of the 4 symptom ratings (irritability/grouchiness, tension/nervousness, pounding/racing heart, or feeling fearful for no reason) (39). The frequency of vasomotor symptoms (VMS; hot flashes and/or night sweats) during the previous 2 weeks, was analyzed as a 3-level categorical summary variable (no symptoms; VMS occurring on 1–5 days or on 6–14 days) (40).

Menopausal status based on bleeding criteria and hormone therapy categories was described previously (41). Specifically, premenopause was defined as having had no change in the predictability of menses. Experiencing decreased predictability of menses, but having no gaps of 3 or more months, was the criterion for early perimenopause. No menses for 3 to 11 months characterized late perimenopause. Absent menses for 12 or more months (after the final menstrual period) defined natural postmenopause, and bilateral oophorectomy with or without hysterectomy categories were further categorized whether or not hormone therapy was used. Those who underwent a hysterectomy without bilateral oophorectomy (partial hysterectomy) prior to the final menstrual period and those with undetermined menopausal status due to hormone therapy (for indications other than postmenopausal) were assigned to separate categories.

Statistical analysis

Analyses were performed using SAS v9.1.3 (SAS Institute, Inc., Cary, NC). Descriptive statistics compared those with and without MetS using means, standard deviations, t tests for continuous variables (or nonparametric alternatives); counts, percentages, and chi-squared tests for categorical variables.

The distribution of raw cognitive test scores was close to a bell shape (not multimodal). Thus, the raw scores of the cognitive tests were standardized using z-scores, calculated for each cognitive test from the mean and standard deviation (SD) of the cohort at each subject’s first cognitive assessment (at cognitive baseline), rendering the mean z-score of 0 for the cohort at the cognitive baseline.

Mixed-effects models with separate random intercepts and slopes (allowing for within-individual intercepts and slopes) were fitted to the z-score for each test of cognitive function (dependent variable), and MetS (yes/no) assessed by cognitive baseline (primary predictor). The key advantage of this random-effects approach relates to applicability, even when individuals are not measured at the same number of time points (42). An interaction of MetS with time from the cognitive baseline (“MetS over time”) captured change in cognitive function depending on exposure to MetS by the time from cognitive baseline. The effect size of MetS over time reflected the estimate of the cumulative 10-year difference in change of the respective cognitive scores among women with and without MetS.

Model 1 explored the effects of MetS exposure on each variable of cognitive function without adjustment for defined covariates. Cases with stroke (based on self-report) were censored.

Model 2 adjusted for sociodemographic factors: age (at cognitive baseline), study site and race/ethnicity (composite variable, assigning each race/ethnic group at each study site as a separate category), education, and financial strain (assessed at baseline).

Model 3 (final model) also included an adjustment for lifestyle, mood, and menopausal factors (nonoccupational physical activity; smoking and alcohol use; depressive, anxiety, and VMS; sleep difficulty; menopausal status; and hormone therapy) as time-varying covariates.

Repeated cognitive testing may be associated with initial cognitive score improvement due to practice effect, hindering the evaluation of naturalistic longitudinal change of cognitive function (43). Such practice effects were previously reported in the SWAN study cohort (34). To select the most appropriate practice effect indicator, we compared the models without practice effects with the models including 1 of the following indicator variables for practice effects: (a) sequential indicators for the first 2 follow-up cognitive testing occasions, (b) an indicator for each of the practice occasions after the first cognitive assessment, and (c) and the square root of the number of prior testing occasions (as practice effect tends to be largest at first exposure and diminishes with subsequent assessments) (34, 43, 44). The final strategy to adjust for practice effects (strategy c) was selected based on the best-fitting model according to Bayes Information Criterion (44). The estimates for MetS effects on the rate of change in each cognitive score over time were compared in models using each of the 3 approaches to practice effects. As expected (44), alternative specifications led to large differences in the estimated rates of cognitive change scores but minimal differences in the principal estimate of interest, MetS over time (data not shown). Comparisons of Bayes Information Criterion for the models without practice effect indicators and the models individually including each of 3 practice effect indicators revealed the best model fit of models adjusted for practice effect using indicators as a square root of the number of prior testing occasions (data not shown). Thus, the presented final models 1 to 3 were adjusted for a square root of the number of prior testing occasions as practice effect indicator.

Separate sensitivity analyses explored whether the results of models 2 and 3 were affected by interactions of time with education and with site–race/ethnicity (representing the main effects of education and site–race/ethnicity on the rate of cognitive score change). Since the trajectory of cognitive decline is the quantity of interest, the latter sensitivity analysis adjusts for confounders of cognitive decline not just confounders of cognition level.

As a significantly higher proportion of women from the Newark, New Jersey, site had to be excluded from the analytical sample (Table 1) due to insufficient counts of follow-up cognitive or MetS assessments. We conducted a sensitivity analysis excluding the cases from Newark, New Jersey (as reported in the Results section).

Results

Subject characteristics

The women completed on average [SD] 3.7 [0.7] MetS assessments before or at their cognitive baseline, with MetS diagnosed in 635 (29.5%) women. Although the exact duration of MetS is unknown (53.4% of those women already met criteria for MetS at the start of the SWAN study), 392 (61.7%) of the women had persistent MetS (meeting MetS criteria on ≥2 SWAN study assessments (45)). The observations from cases of incident stroke occurring after cognitive baseline were censored (n = 18, 0.8%). Women with and without MetS were equally likely to be censored due to incident stroke.

Table 2 summarizes the characteristics of women at their cognitive baseline, comparing those with and without MetS. The mean age [SD] at their cognitive baseline was 50.7 [2.9] years, and 51.6% of women had bleeding patterns consistent with early perimenopause or premenopause. MetS was more prevalent among women without a college education and those who reported financial strain, sleep difficulty, smoking, and no alcohol use. The frequency of VMS and mean depressive and anxiety symptom scores did not differ by MetS status, whereas mean physical activity scores were lower among those with MetS than for those women without MetS. The prevalence of MetS was lower among women in premenopause and early perimenopause or treated with hormone therapy.

Table 2.

Characteristics of Women With and Without Exposure to Metabolic Syndrome by First Cognitive Assessment

	Characteristics		Total (N = 2149)	No MetS (N = 1514)	MetS (N = 635)	P Value
Sociodemographic factors	Age at cognitive baseline, mean ± SD		50.7 ± 2.9	50.6 ± 2.8	51.1 ± 3.2	<0.001a
	Site–race/ethnicity, n (%)	Boston black	116 (5.4)	79 (5.2)	37 (5.8)	<0.001
		Boston white	211 (9.8)	162 (10.7)	49 (7.7)
		Chicago black	134 (6.2)	83 (5.5)	51 (8.0)
		Chicago white	142 (6.6)	97 (6.4)	45 (7.1)
		Los Angeles Japanese	238 (11.1)	181 (12.0)	57 (9.0)
		Los Angeles white	171 (8.0)	143 (9.4)	28 (4.4)
		Michigan black	198 (9.2)	123 (8.1)	75 (11.8)
		Michigan white	136 (6.3)	79 (5.2)	57 (9.0)
		New Jersey Hispanic	80 (3.7)	51 (3.4)	29 (4.6)
		New Jersey white	47 (2.2)	31 (2.0)	16 (2.5)
		Oakland Chinese	208 (9.7)	153 (10.1)	55 (8.7)
		Oakland white	151 (7.0)	107 (7.1)	44 (6.9)
		Pittsburgh black	103 (4.8)	71 (4.7)	32 (5.0)
		Pittsburgh white	214 (10.0)	154 (10.2)	60 (9.4)
	Married, n (%)	Yes	1465 (68.2)	1055 (69.7)	410 (64.6)	0.027
		No	656 (30.5)	437 (28.9)	219 (34.5)
		Missing	28 (1.3)	22 (1.5)	6 (0.9)
	College education, n (%)	Yes	1045 (48.6)	777 (51.3)	268 (42.2)	0.001
		No	1187 (50.6)	726 (48.0)	361 (56.9)
		Missing	17 (0.8)	11 (0.7)	6 (0.9)
	Financial strain, n (%)	No	1412 (65.7)	1031 (68.1)	381 (60.0)	<0.001
		Yes	725 (33.7)	472 (31.2)	253 (39.8)
		Missing	12 (0.6)	11 (0.7)	1 (0.2)
Lifestyle, mood, and menopausal factors	Smoker, n (%)	No	1839 (85.6)	1315 (86.9)	524 (82.5)	0.016
		Yes	258 (12.0)	162 (10.7)	96 (15.1)
		Missing	52 (2.4)	37 (2.4)	15 (2.4)
	Alcohol servings/week, n (%)	None	1019 (47.4)	649 (42.9)	370 (58.3)	<0.001
		1–7	955 (44.4)	730 (48.3)	225 (35.4)
		> 7	116 (5.4)	89 (5.9)	27 (4.3)
		Missing	59 (2.7)	46 (3.0)	13 (2.0)
	Physical activity score, mean ± SD	n = 2120b	7.8 ± 1.7	7.9 ± 1.7	7.4 ± 1.7	<0.001a
	Depressive symptom score, mean ± SD	n = 2137b	8.6 ± 8.6	8.4 ± 8.5	8.9 ± 8.8	0.222
	Anxiety symptom score, mean ± SD	n = 2096b	2.2 ± 2.2	2.1 ± 2.1	2.3 ± 2.4	0.108
	Sleep difficulty	No	1313 (61.1)	958 (63.3)	355 (55.9)	0.002
		Yes	781 (36.3)	515 (34.0)	266 (41.9)
		Missing	55 (2.6)	41 (2.7)	14 (2.2)
	Vasomotor symptoms days/2 weeks, n (%)	6–14	413 (19.2)	274 (18.1)	139 (21.9)	0.215
		1–5	674 (31.4)	481 (31.8)	193 (30.4)
		None	1010 (47.0)	720 (47.6)	290 (45.7)
		Missing	52 (2.4)	39 (2.6)	13 (2.0)
	Menopausal stages | hormone therapy (HT), n (%)	Premenopause	148 (6.9)	107 (7.1)	41 (6.5)	0.085
		Early perimenopause	960 (44.7)	682 (45.0)	278 (43.8)
		Late perimenopause	211 (9.8)	142 (9.4)	69 (10.9)
		Postmenopause | no HT	355 (16.5)	239 (15.8)	116 (18.3)
		Postmenopause | +HT	126 (5.9)	96 (6.3)	30 (4.7)
		Bilateral oophorectomy | no HT	11 (0.5)	5 (0.3)	6 (0.9)
		Bilateral oophorectomy | +HT	48 (2.2)	39 (2.6)	9 (1.4)
		Partial hysterectomy	42 (2.0)	26 (1.7)	16 (2.5)
		Undetermined	249 (11.5)	178 (11.8)	70(11.1)

Abbreviation: MetS, metabolic syndrome; SD, standard deviation.

^a P-values from 2 sample t-test, except for depressive symptom & anxiety symptom scores (2-sided Wilcoxon test).

^b n different from total n = 2149.

Cognitive assessments

The first cognitive assessment (cognitive baseline) was conducted at SWAN study visit 4 in a majority of cases (n = 1968 or 91.3% of the analytic cohort). Women with and without MetS completed a median [IQR] of 5 (IQR 4–7) and 7 (IQR 5–7) cognitive assessments, respectively (P < 0.001), over the 11.2 years (IQR 9.2–11.5) of follow-up.

MetS exposure and cognitive function scores

Women with MetS had a larger decline of perceptual speed measured as SDMT z-score (estimate ± systemic error [SE] = –0.087 (95% confidence interval, –0.150 to –0.024; P = 0.017) than those without exposure to MetS (Table 3, model 3 adjusted for practice effects). The mean ± SD SDMT score of 56 [11] at the cognitive baseline was used to transform z-score estimates to a 10-year change in SDMT score (time estimate from model 3 multiplied by SD at cognitive baseline). The absolute 10-year decline of SDMT score was –4 in women without MetS (with mean ± SD SDMT score of 56 ± 11 at their cognitive baseline) and –5 in women with MetS (with average SDMT score of 55 ± 12 at their cognitive baseline); the estimated effect on the SDMT score decline attributable to MetS was –1 over 10-years, and equivalent to a 24% larger decline in perceptual speed (thus perceptual speed “aging” accelerated by an additional 2.4 years in a cumulative 10-year period).

Table 3.

Change in Perceptual Speed (Symbol Digit Modalities Test, SDMT (30)) Over Time in Women With and Without Exposure to Metabolic Syndrome Before Their Cognitive Baseline

		Without Practice Effects			With Practice Effects
	Effects	Estimate	(95% CI)	P Value	Estimate	(95% CI)	P Value
Model 1	Intercept	0.132	(0.084; 0.179)	<0.001	0.020	(–0.028; 0.069)	0.411
	Practice	-		-	0.222	(0.194; 0.249)	<0.001
	Time (10-year)	0.023	(–0.009; 0.055)	0.158	–0.383	(–0.441; –0.326)	<0.001
	MetS	–0.141	(–0.228; –0.053)	0.002	–0.131	(–0.218; –0.044)	0.003
	MetS over time	–0.083	(–0.146; –0.020)	0.010	–0.090	(–0.153; –0.027)	0.005
Model 2	Intercept	0.394	(0.259; 0.529)	<0.001	0.297	(0.163; 0.432)	<0.001
	Practice	-		-	0.203	(0.175; 0.231)	<0.001
	Time (10-year)	0.022	(–0.010; 0.054)	0.1476	–0.352	(–0.411; –0.294)	<0.001
	MetS	–0.015	(–0.090; 0.060)	0.701	–0.010	(–0.085; 0.064)	0.786
	MetS over time	–0.083	(–0.146; –0.020)	0.010	–0.090	(–0.153; –0.027)	0.005
Model 3	Intercept	0.455	(0.307; 0.603)	<0.001	0.320	(0.173; 0.468)	<0.001
	Practice	-		-	0.215	(0.186; 0.246)	<0.001
	Time (10-year)	–0.010	(–0.051; 0.031)	0.7629	–0.365	(–0.428; –0.303)	<0.001
	MetS	–0.007	(–0.069; 0.083)	0.852	–0.012	(–0.063; 0.087)	0.751
	MetS over time	–0.078	(–0.141; –0.014)	0.016	–0.087	(–0.150; –0.024)	0.007

Model 1 represents the unadjusted effects of MetS exposure to SDMT z-score.

Model 2 adjusts model 1 for sociodemographic factors (age at cognitive baseline, education, financial strain, and site–race/ethnicity).

Model 3 adjusts model 2 for time-variable lifestyle and mood and menopause factors: smoking and alcohol use, nonoccupational physical activity, sleep difficulty, depressive state, anxiety and vasomotor symptoms, menopausal stages/hormone therapy.

The significant effects are highlighted in bold.

Abbreviation: CI, confidence interval; MetS, metabolic syndrome.

While adjustment for sociodemographic factors (age, education, financial strain, site–race/ethnicity) eliminated the baseline differences attributable to MetS (Table 3, models 2 and 3, compared with model 1), the SDMT z-score decline was still larger in women with MetS compared with those without MetS (Table 3, effect of MetS over time). Moreover, the effect size was very similar after additional adjustment for lifestyle, mood, and menopause factors (nonoccupational physical activity, smoking, alcohol use, sleep difficulty, depression/anxiety symptoms, VMS, and menopausal stage/hormone therapy—model 3).

We detected a significant 10-year decline in working memory DSB z-scores (Table 4 models adjusted for practice effects) and episodic memory EBMT delayed recall z-scores (Table 5 models adjusted for practice effects) but not episodic memory EBMT immediate recall z-scores. However, no significant effects of MetS over time were detected in models of working memory or episodic memory (immediate or delayed).

Table 4.

Change in Working Memory (Digit Span Backward, DSB (33)) Over Time in Women With and Without Exposure to Metabolic Syndrome Before Their Cognitive Baseline

		Without Practice Effects			With Practice Effects
	Effects	Estimate	(95% CI)	P Value	Estimate	(95% CI)	P Value
Model 1	Intercept	0.081	(0.034; 0.127)	<0.001	0.023	(–0.025; 0.072)	0.349
	Practice	-			0.0117	(0.084; 0.150)	<0.001
	Time (10-year)	0.089	(0.052; 0.125)	<0.001	–0.124	(–0.194; –0.053)	0.001
	MetS	–0.089	(–0.175;–-0.003)	0.043	–0.088	(–0.173; –0.002)	0.046
	MetS over time	–0.030	(–0.102; 0.041)	0.404	–0.034	(–0.105; 0.038)	0.353
Model 2	Intercept	0.460	(0.321; 0.599)	<0.001	0.392	(0.252; 0.531)	<0.001
	Practice	-			0.101	(0.068; 0.135)	<0.001
	Time (10-year)	0.088	(0.051; 0.124)	<0.001	–0.097	(–0.167; –0.026)	0.007
	MetS	–0.026	(–0.105; 0.053)	0.517	–0.028	(–0.106; –0.051)	0.491
	MetS over time	–0.025	(–0.097; 0.046)	0.491	–0.028	(–0.100; 0.043)	0.437
Model 3	Intercept	0.451	(0.294; 0.608)	<0.001	0.372	(0.215; 0.529)	<0.001
	Practice	-			0.097	(0.062; 0.133)	<0.001
	Time (10-year)	0.083	(0.036; 0.130)	0.001	–0.075	(–0.150; –0.001)	0.048
	MetS	–0.023	(–0.104; 0.058)	0.584	–0.024	(–0.104; –0.057)	0.564
	MetS over time	–0.017	(–0.089; 0.055)	0.650	–0.020	(–0.092; 0.052)	0.584

Model 1 represents the unadjusted effects of MetS exposure to digit span backward z-score.

Model 2 adjusts model 1 for sociodemographic factors (age at cognitive baseline, education, financial strain, and site–race/ethnicity).

Model 3 adjusts model 2 for time-variable lifestyle, mood and menopause factors: smoking and alcohol use, nonoccupational physical activity, sleep difficulty, depressive state, anxiety and vasomotor symptoms, menopausal stages/hormone therapy.

The significant effects are highlighted in bold.

Abbreviation: CI, confidence interval; MetS, metabolic syndrome.

Table 5.

Change in Episodic Memory (East Boston Memory Test, EBMT (31,32), Immediate and Delayed Recall) Over Time in Women With and Without Exposure to Metabolic Syndrome Before Their Cognitive Baseline

		Without Practice Effects			With Practice Effects
	Effects	Estimate	(95% CI)	P Value	Estimate	(95% CI)	P Value
Immediate recall: East Boston Memory Test (EBMT)
Model 1	Intercept	0.079	(0.038; 0.119)	<0.001	0.008	(–0.038; 0.053)	0.784
	Practice	-		-	0.140	(0.100; 0.181)	<0.001
	Time (10-year)	0.077	(0.032; 0.122)	0.001	–0.174	(–0.261; –0.087)	<0.001
	MetS	–0.106	(–0.181; -0.031)	0.006	–0.101	(–0.176; –0.026)	0.008
	MetS over time	0.047	(–0.041; 0.135)	0.297	0.044	(–0.044; 0.132)	0.330
Model 2	Intercept	–0.137	(–0.244; –0.030)	0.054	–0.186	(–0.295; –0.078)	0.001
	Practice	-		-	0.091	(0.050; 0.133)	<0.001
	Time (10-year)	0.081	(0.036; 0.127)	<0.001	–0.084	(–0.173; 0.004)	0.063
	MetS	–0.027	(–0.198; 0.043)	0.451	–0.025	(–0.096; 0.045)	0.479
	MetS over time	0.043	(–0.045; 0.132)	0.333	0.042	(–0.046; 0.129)	0.355
Model 3	Intercept	–0.060	(–0.186; 0.066)	0.350	–0.125	(–0.253; 0.004)	0.057
	Practice	-		-	0.092	(0.049; 0.136)	<0.001
	Time (10-year)	0.049	(0.036; 0.127)	0.242	–0.089	(–0.183; 0.004)	0.062
	MetS	0.008	(–0.198; 0.043)	0.823	0.008	(–0.063; 0.079)	0.828
	MetS over time	0.036	(–0.045; 0.132)	0.389	0.028	(–0.060; 0.116)	0.534
Delayed Recall: East Boston Memory Test
Model 1	Intercept	0.080	(0.039; 0.121)	<0.001	–0.010	(–0.056; 0.037)	0.684
	Practice	-		-	0.177	(0.135; 0.218)	<0.001
	Time (10-year)	0.094	(0.048; 0.141)	<0.001	–0.221	(–0.309; –0.133)	<0.001
	MetS	–0.074	(–0.150; 0.002)	0.057	–0.068	(–0.143; 0.008)	0.079
	MetS over time	–0.014	(–0.104; 0.077)	0. 767	–0.016	(–0.107; 0.074)	0.722
Model 2	Intercept	–0.094	(–0.203; 0.015)	0.091	–0.170	(–0.281; -0.060)	0.003
	Practice	-		-	0.133	(0.091; 0.175)	<0.001
	Time 10 years	0.099	(0.052; 0.142)	<0.001	–0.143	(–0.232; –0.053)	0.002
	MetS	0.010	(–0.061; 0.082)	0.777	0.013	(–0.058; 0.084)	0.716
	MetS over time	–0.014	(–0.105; 0.077)	0.762	–0.016	(–0.106; 0.074)	0.728
Model 3	Intercept	0.028	(–0.100; 0.156)	0.669	–0.072	(–0.202; 0.058)	0.275
	Practice	-		-	0.138	(0.094; 0.182)	<0.001
	Time (10-year)	0.068	(0.009; 0.126)	0.023	–0.160	(–0.255; 0.066)	0.001
	MetS	0.033	(–0.040; 0.105)	0.374	0.035	(–0.037; 0.107)	0.335
	MetS over time	–0.012	(–0.103; 0.079)	0.791	–0.015	(–0.106; 0.075)	0.740

Model 1 represents the unadjusted effects of MetS exposure to each EBMT z-score.

Model 2 adjusts model 1 for sociodemographic factors (age at cognitive baseline, education, financial strain, and site–race/ethnicity).

Model 3 adjusts model 2 for time-variable lifestyle, mood and menopause factors: smoking and alcohol use, nonoccupational physical activity, sleep difficulty, depressive state, anxiety and vasomotor symptoms, menopausal stages/hormone therapy.

The significant effects are highlighted in bold.

Abbreviations: CI, confidence interval; EBMT, East Boston memory test; MetS, metabolic syndrome.

Practice effects and sensitivity analyses

Prior cognitive testing experiences tend to have a measurable influence on the magnitude of cognitive change due to increased familiarity (test-related and session-related) with the assessment tools (learning bias due to practice effect) (43, 46). Initial improvements in cognitive testing scores due to practice effects have been observed in SWAN women (34).

Comparison of models without and with adjustment for practice effects (Table 3), reveals that the effects of time on SDMT z-scores in models 1 to 3 unadjusted for practice were obscured. Adjustment for practice effects revealed a significant decline in perceptual speed SDMT z-scores, which was larger in women with MetS. As such, failure to adjust for practice effects would have led to false conclusions.

Likewise, practice effects were associated with significantly higher working and episodic memory scores (Tables 4 and 5). Moreover, adjustment for practice effects corrected the significant change in z-scores of working memory (DSB), and episodic memory delayed recall (EBMT) toward negative instead of positive 10-year time effects. Nonetheless, we did not detect significant association of MetS over time on the change in these z-scores, which was the principal outcome of interest.

Separate sensitivity analyses included the interactions of time with education and with site–race/ethnicity in models 2 and 3. The latter sensitivity analysis adjusts for confounders of cognitive decline, not only confounders of cognition level. The results were similar with or without the adjustments for these confounders (data not shown).

Sensitivity analyses excluding Newark, New Jersey, participants (n = 127, 5.9% of the analytical sample) yielded similar results (data not shown).

The findings were similar after excluding/censoring the cases (n = 560) treated with hormonal therapy at any point after enrollment into the SWAN study (data not shown).

Discussion

MetS was associated with a larger decline of perceptual speed (SDMT z-score) in midlife women in over a decade of naturalistic follow-up with multiple repeated cognitive assessments. Our longitudinal findings are congruent with prior cross-sectional research suggesting the association of MetS with lower perceptual speed scores (47–49). The present results extend the prior work in 2 ways. First, we showed longitudinally that MetS was responsible for a larger midlife decline of perceptual speed, even after controlling for the cognitive baseline differences. Second, these differential effects of MetS were observed in association with a decline in perceptual speed scores but not in episodic memory or working memory scores. The basis for this selective association of MetS with loss of perceptual speed is uncertain. One potential explanation may relate to the overall nonuniform patterns of age-related changes (50, 51) and MetS-related changes (52) across various intellectual abilities during midlife. Specifically, the earlier and larger perceptual speed decline (typically observed since early adulthood) may be more susceptible to effects of MetS, whereas the effects of MetS on verbal and episodic memory abilities are harder to detect until late life (28). Moreover, changes in perceptual speed estimates are more likely to be detectable due to the wider range of the SDMT scale (0–110) than the range for EBMT (0–12 words) and DSB (0–7 strings). Lastly, the neurobiological drivers affecting a decline in these cognitive domains may differ. A decline in episodic memory is an early sign of Alzheimer disease (53) and other forms of neurodegeneration (54), whereas a decline in executive function and perceptual speed is common in vascular dementia (55), with the latter hypothesized as related to a cognitive decline due to MetS and diabetes.

The average decline in perceptual speed over time in the midlife women with MetS was slightly greater than the decline related to normal aging. Should this decline in perceptual speed associated with MetS be considered clinically relevant? The difference of perceptual speed decline in women with MetS compared with those without MetS was equivalent to accelerating the 10-year decline in perceptual speed by an additional 2.4 years of aging. Interestingly, the reductions in brain volume observed in cross-sectional studies of diabetes compared with controls average 0.5% to 2.0%, which are equivalent to 2 to 5 years of normal aging (21, 23), comparable in magnitude to the aging-converted effect size observed in our study. Nevertheless, the mean 10-year decline in perceptual speed z-scores of –0.36 in women without MetS and –0.45 in women with MetS are above the threshold for impaired cognition, which is typically defined as a performance below a z-score of –1.65 (equivalent to the lowest fifth percentile of reference values) (21). Still, even subtle decrements of perceptual speed have a potential to cause functional problems, such as struggling to follow fast-paced conversations, missing subtle social cues, getting overwhelmed when faced with too much information (including a plethora of health-related information). Thus, comprehensive testing of executive functions deserves investigation to further define MetS effects.

The cross-sectional and experimental research suggests association of MetS with poorer executive function, whereas associations with performance on memory, attention, and some other domains have been mixed (52). Admittedly, the longitudinal research on cognitive function is scarce, particularly in midlife subjects (21). The longitudinal analysis in a cohort of 45- to 64-year-old subjects of the Atherosclerosis Risk in Communities (ARIC) study (27) found no association of MetS with the 6-year change (over 2 time points) of cognitive function scores (digit symbol substitution test, delayed word recall and word fluency test). This difference from our findings may be related to confounding by practice effects in cognitive testing (43).

Similar to our findings, the Hoorn study revealed the association of MetS with worse perceptual speed, attention, and executive functioning and did not find longitudinal associations in memory domains (56). However, the Hoorn study included only white subjects with an average age [SD] of 68 [5] years, whereas our SWAN cohort was younger and more diverse from the race/ethnic perspective.

The plausible multifactorial mechanisms of the relationship between MetS and the decline of perceptual speed (and overall cognitive function) suggest a direct relationship or potentiation of negative effects related to degenerative brain changes (1, 23, 57). MetS and insulin resistance are associated with changes in white and gray matter microstructure, related to perceptual speed (23, 58, 59). Impaired mitochondrial, lysosomal, and proteasomal function; oxidative stress; chronic inflammation; prothrombotic state; and other MetS-related mechanisms may compromise the blood–brain barrier integrity, neurotransmitter function, synaptic plasticity, adaptive responses of neuronal bioenergetics, and other processes in the background of epigenetic and genetic susceptibility may affect cognition trajectories (1). MetS-related loss of metabolic flexibility may contribute to lower perceptual speed and overall brain function (60). Moreover, a great deal of MetS is attributed to lifestyle factors (2), particularly food, physical activity, sleep, and social environment—all proven to have effects on the brain and be affected by the brain (1, 59). Although clinically MetS is defined using 5 diagnostic markers (6), at least 49 biomarkers from a variety of metabolic superpathways contribute to MetS (10). The approach to MetS from the perspective of broader integrated pathobiology (rather than the sum of 3–5 MetS factors) may lead the way to a better understanding of related perceptual speed declines at midlife.

Strengths

The strengths of our study include its population-based design and the long follow-up period with multiple repeated assessments of cognitive function. Such longitudinal assessments based the estimates on changes within individuals, reducing the likelihood of error arising from natural variation between individuals, which is a source of bias in cross-sectional investigations.

Multiple longitudinal cognitive assessments over a decade of follow-up permitted adjustment for practice effects in repeated cognitive testing (43). The learning bias due to practice effects can lead to nonlinear associations and misinterpretation of results (44). To our knowledge, our report is the first longitudinal analysis of the MetS-related effects on a decline in cognitive function scores in midlife that accounts for the learning bias related to repeated testing.

Our models were also adjusted for sociodemographics, lifestyle, sleep, mood, and menopausal stages/hormone therapy factors—potential confounders in the association between MetS and cognitive function scores.

Limitations

Sensitivity to participant burden constrained the SWAN study to a limited cognitive test battery. We were unable to investigate the reasons why MetS effects were found only in association with perceptual speed but not with other cognitive domains. We were unable to ascertain the overall duration of MetS exposure (as some women met MetS criteria by the time they entered SWAN study), and the number of incident MetS cases was insufficient to test the association of incident MetS with changes in cognitive function. Excluding the standard MetS criteria, SWAN did not assess the other known superpathways affected by MetS (10) (ie, branched-chain amino acids) that may promote cognitive decline. The relationship between MetS and cognitive function may be affected by menopause and hormonal milieu, which may deserve separate detailed investigations. Moreover, we were not able to assess whether the larger decline of perceptual speed would impair integrated cognitive functions such as health information processing or implementation of lifestyle interventions. Lastly, our study did not include men.

Conclusions

Women with MetS during midlife subsequently experienced a larger decline in perceptual speed scores but not in memory scores in the SWAN study. As the prevalence of metabolic dysfunction steeply increases in midlife women, these findings contribute the knowledge about the association of MetS with several domains of cognitive function in women at midlife without dementia.

Glossary

Abbreviations

DSB

digit span backward

EBMT

East Boston memory test

IQR

interquartile range

MetS

metabolic syndrome

SD

standard deviation

SDMT

symbol digit modalities test

SWAN

Study of Women’s Health Across the Nation

VMS

vasomotor symptoms

Additional Information:

Disclosure Summary: Authors declare no conflict of interest that is relevant to the subject matter or materials included in this work.

Data Availability: All data generated or analyzed during this study are included in this published article or in the data repositories listed in References (61). Investigators who require assistance accessing the public use data set or applying for SWAN investigator status may contact the SWAN Coordinating Center at swanaccess@edc.pitt.edu.