Midlife depressive symptoms, smoking status, hypertension, diabetes, dyslipidemia and obesity modify the association of long-term vascular risk factor variability from childhood with midlife cognitive function: The Bogalusa Heart Study

Abstract

Background

Variability of vascular risk factors (VRFs) is linked to increased dementia risk. Whether midlife depression, smoking, hypertension, diabetes, obesity, and dyslipidemia modify the association between VRF variability from childhood and midlife cognitive function (CF) is unclear.

Methods

Bogalusa Heart Study participants with ≥ 3 repeated measurements of systolic blood pressure (SBP), glucose, non-HDL, and BMI across the life course and one midlife CF assessment were included. CF was evaluated using the global cognitive score (GCS), derived from averaging eight standardized neuropsychological test scores. VRF variability was assessed using coefficient of variation. Associations between VRF variability and GCS were examined using multivariable linear regression models stratified by depressive symptoms (10- item Center for Epidemiologic Studies Depression scale ≥10), smoking status, hypertension, T2DM, dyslipidemia, and obesity.

Results

A total of 1010 participants were evaluated (34% Black, 61% women; mean age 48 years; follow-up 39 years). Higher glucose variability was associated with poorer GCS among individuals with hypertension (highest vs. lowest tertile, β = −0.12, 95% CI -0.22, −0.02), and among those without depression, obesity, dyslipidemia, diabetes, and among never smokers. Higher SBP variability was associated with poorer GCS only among former smokers (highest vs. lowest tertile, β = −0.16, 95% CI -0.30, −0.02). Middle-tertile BMI variability was associated with better GCS among individuals without depression, without diabetes, and among never smokers. Middle-tertile non-HDL variability was associated with poorer GCS only among individuals without depression.

Conclusions

Tracking VRF variability from early life even in low-risk populations could facilitate optimal cognitive aging.

Graphical abstract

Highlights

• •High glucose variability from childhood was linked to poorer midlife cognition among those with hypertension.
• •High systolic blood pressure variability from childhood was linked to poorer midlife cognition in former smokers only.
• •Moderate BMI variability from childhood was linked to better midlife cognition among those without depression, diabetes, and never smokers.
• •Moderate non-HDL variability from childhood was linked to poorer midlife cognition in those without depression.
• •Residual risk of poorer midlife cognition may persist in individuals of low-risk populationswith higher vascular risk factor variability.

Abbreviations

BHS	Bogalusa Heart Study
BMI	body mass index
CF	cognitive function
CV	coefficient of variation
GCS	global cognitive score
Non-HDL	Non-high density lipoprotein cholesterol
NP	neuropsychological
SD	Standard deviation
T2DM	Type 2 diabetes mellitus
VRF	Vascular risk factor

1. Introduction

Midlife is when cognitive trajectories diverge among individuals [1]. Since aging is a major risk for dementia, examining midlife cognitive function (CF) will help differentiate age-related and pathologic changes in cognition [2]. Dementia pathology may precede symptoms by many years, and a life-course approach targeting modifiable risk factors is advocated to prevent dementia [3]. Examining how modifiable risk factors for dementia affect heterogeneity of midlife CF will help identify strategies to delay the dementia onset [1].

Exposure to hypertension, type 2 diabetes (T2DM), dyslipidemia, and obesity may result in ischemia and atherosclerosis, common pathways of cardiovascular disease and dementia [4]. Importantly, levels of vascular risk factors (VRFs) such as systolic blood pressure (SBP), glucose, non-high density lipoprotein cholesterol (non-HDL), and body mass index (BMI) values fluctuate throughout the lifespan. This fluctuation is a result of the interaction between the body's homeostasis mechanism and the external environment [5]. Higher variability of VRFs from adulthood have been associated with a higher dementia risk [6]. However, the influence of long-term variability of VRFs from childhood on midlife CF are unclear.

Depressive symptoms and smoking status are common midlife risk factors for dementia and may interact with vascular processes across the life course. According to the vascular depression hypothesis, depressive symptoms and VRFs may have a bidirectional relationship [7,8]. Life-long smoking patterns are typically established from early to midlife period, and a dose-response relationship was observed between early to midlife smoking and midlife cognition [9]. These pathways support the role of depressive symptoms and smoking status as potential modifiers of the association between long-term VRF variability and midlife CF.

Midlife hypertension, diabetes, obesity, and dyslipidemia may also modify the influence on CF by VRF variability. As mean VRF levels rise, variability often increases, making it difficult to disentangle whether risk of cognitive dysfunction is driven by elevated mean levels, greater variability, or both [10]. Also, midlife VRFs may modify the relationship between VRFs and CF through additive or multiplicative interactions [[11], [12], [13]]. Few studies have evaluated how VRF variability interacts with midlife VRFs to impact CF.

Depression, smoking, hypertension, diabetes, obesity, and dyslipidemia are classified as midlife dementia risk factors, often originate earlier in life and frequently coexist.

[3,14] Long-term VRF variability across the life course may impact CF by vascular injury and subsequent amyloid beta and tau accumulation [15]. A bidirectional relationship may exist between long-term VRF variability and depression, smoking, hypertension, diabetes, obesity, and dyslipidemia, and may exacerbate one another across the life course. These midlife modifiers may act as a subsequent pathway of brain injury contributing to further cognitive dysfunction [16]. However, how these midlife modifiers influence the relationship between long-term VRF variability from childhood and midlife CF is unclear, and this represents the knowledge gap.

The aim of this study was to test the hypothesis of whether midlife depressive symptoms, smoking, hypertension, diabetes, obesity, and dyslipidemia modify the association between long-term VRF variability from childhood and midlife CF. It was hypothesized that higher VRF variability would be associated with poorer midlife CF in those having the midlife effect modifiers with greater effect sizes compared to the counterparts.

2. Methods

The BHS, started in 1973 and continuing, is a community-based cohort study that examines VRFs from early life [17]. Nine cross-sectional surveys among children 4 to 17 years of age, and eleven cross-sectional surveys among adults previously examined in childhood were administered between 1973 and 2016. The study sample included 1010 midlife participants of the BHS with ≥ three VRF measurements across the life course, who underwent a comprehensive neuropsychological (NP) examination from 2013 to 2016 and covariates examination. At each examination, all study participants provided written informed consent, and the Institutional Review Board of the Tulane University Health Sciences Center approved the study protocols.

Beginning in 1973, trained examiners have used validated questionnaires to collect information on age, race, sex, medication status, smoking, and education [17]. Before examination, participants were asked to fast for 12 -14 h [18]. A participant's BP was examined by two trained examiners from the right arm while relaxed in a sitting position, with each examiner measuring three BP measures. The six measurements were averaged for BP analysis [17]. Standardized protocols were used to measure participants' glucose and cholesterol levels [17]. Non-HDL (total cholesterol – HDL) is reported to reliably predict stroke and cardiovascular events [19]. Weight and height at each examination were measured twice to the nearest 0.1 kg and 0.1 cm, and were averaged before analysis [17]. Weight (kg) was divided by height (m²) to calculate the body mass index (BMI).

2.1. Long-term vascular risk factor variability from childhood

The coefficient of variation (CV) of repeated measures of SBP, glucose, BMI, and non-HDL across the life course was calculated to represent long-term VRF variability [20]. The following formulas were used (using SBP as an example) [20].

$$\text{SD}=\sqrt{\frac{\sum {\left(SBPi-SBPmean\right)}^2}{n-1}},\text{CV}=\frac{SD}{SBPmean}$$

(SBP i is a measurement of SBP of a person at time i, SBP mean is the mean of all measurements of SBP of the person's life course, and n is the number of measurements of SBP of the person's life course.) The CV is unitless and measures the relative variability against the mean VRF level, facilitating direct comparisons of variability among different VRFs. CV of a VRF was ranked into tertiles for analysis. The lowest and highest tertiles of CV represented the lowest and the highest variability.

2.2. Midlife cognitive outcomes

Midlife CF was assessed using a global cognitive score (GCS), which included the performance of three cognitive domains: executive function, episodic memory, and attention and processing speed [21]. The GCS was calculated by averaging the scores of eight neuropsychological (NP) tests which represented the three cognitive domains [22]: 1) executive function (Trail- Making Test part B, and Digit Symbol Coding (WAIS-IV) [23]); 2) episodic memory (Logical Memory I, Logical Memory II, (Wechsler Adult Intelligence Scale IV (WAIS-IV)) and Delayed Recognition (Wechsler memory scale-IV) [24]), and 3) attention and processing speed (Digit Span Task Forwards (WAIS-IV) and Digit Span Task Backwards (WAIS-IV) and Trail-Making Test part A) [20]. Higher scores represented better performance for all NP tests except for the Trail-Making Tests; for consistent interpretation, Trail-Making Test scores were reversed by subtracting raw completion times from the maximum scores. Cultural constructs may impact performance in NP tests, leading to disproportionate distribution of demographics in the higher score ranges [25]. Therefore, NP test scores were first standardized by age in years, race and sex, then averaged to form the GCS [25].

2.3. Covariates

Age at cognitive assessment, race, and sex were excluded from the covariates, since they were used to standardize the NP test scores forming the GCS [26]. Covariates were chosen considering their association with cognition and VRF variability [5].

Clinical and behavioral covariates were hypertension (SBP/DBP > 130/80 mmHg, or on antihypertensive medication), T2DM (glucose ≥126 mg/dL or on diabetes medication), dyslipidemia (non-HDL ≥130 mg/dL or on cholesterol medication), obesity (BMI ≥ 30 kg/m²), smoking status, and mean VRF levels over the life course. VRF levels measured repeatedly over the life course were averaged, and the mean was included as a covariate. Smoking status was included as current smoker/former smoker/never smoker, with never smoker as reference. Hypertension, T2DM, dyslipidemia, and obesity were dichotomized (yes/no).

Socioeconomic covariates were employment status (currently employed/not employed), depressive symptoms (yes/no), and education, measured as the achieved education parameter, described in detail below. The 10- item scale of Center for Epidemiologic Studies Depression (CESD-10) were used to evaluate depressive symptoms. A score of ≥10 on the CESD-10 was considered as having depressive symptoms, with a sensitivity of 91.9%, and a specificity of 92.8% [27]. Genetic covariate was the APOE ε4 carrier status, which was dichotomized (0: having no ε4 allele, 1: having 1 or more ε4 allele). Two single nucleotide polymorphisms (SNPs), rs429358 and rs7412, were imputed from a reference panel, then were used to infer the APOE ε4 status. (Each SNP has a unique identifier (rs#) that facilitates cross-referencing and tracking of SNPs.) [28,29] High correlations between imputed and directly genotyped techniques are reported for the two SNPs (rs 429358, Cohen's kappa = 0.94; rs7412, Cohen's kappa = 0.93) [30]. Follow-up duration was entered as a covariate to adjust for participants' unequal follow-up.

• 1)Achieved education parameter

Years of education by itself may not represent the quality of education optimally [31]. The NP test performance is significantly influenced by an individual's reading level [31]. Therefore, the word and letter reading (Wide Range Achievement Test 4 (WRAT-4)) and vocabulary (WAIS-IV vocabulary subtest) test scores were standardized by age, race, and sex and averaged to calculate the achieved education parameter [26]. In this study, this parameter served as a covariate to account for the NP test performance difference across individuals who had the same level of education in years [31].

2.4. Statistical analysis

Participants were stratified by depressive symptoms, smoking status, obesity, dyslipidemia, hypertension, and T2DM. Participant characteristics were presented as mean (SD) or number (%). Demographic and clinical characteristics between subgroups were compared using independent-samples t-tests for continuous variables and Pearson's Chi-square tests for categorical variables.

Within each effect modifier subgroup, each VRF variability was entered into a multivariable-adjusted linear regression model as tertiles of CV, with the lowest tertile as the reference. First, models were unadjusted, then sequentially adjusted. For SBP variability, Model 1 was adjusted for follow-up duration, smoking, T2DM, dyslipidemia, and obesity; Model 2 was additionally adjusted for mean SBP over the life course; Model 3 was additionally adjusted for employment, achieved education parameter, and depressive symptoms; and Model 4 was additionally adjusted for APOE ε4 status. For glucose, BMI, and non-HDL variability, the models were adjusted for the same covariates as those of SBP variability, except that Model 1 was adjusted for hypertension instead of T2DM/obesity/dyslipidemia; Model 2 was adjusted for the corresponding mean glucose/BMI/non-HDL levels across the life course instead of SBP. In each subgroup analysis, the stratifying variable was excluded from covariates. For example, when stratified by depressive symptoms, depressive symptoms was excluded from the covariates. Interaction terms between each VRF variability and its corresponding effect modifier were tested in the full sample with all covariates included (Model 5). Statistical analysis was performed using SAS and R version 4.4.1. A two-tailed p-value < 0.05 was used as the significance threshold. For brevity, the results are presented as beta coefficients and 95% confidence intervals from Model 4 of each VRF variability. Multiplicity correction was not conducted. The results should be interpreted as exploratory and confidence intervals may not be used for hypothesis testing.

2.5. Data availability

Data supporting the findings of this study are available from the corresponding author upon reasonable request.

3. Results

3.1. Participant characteristics

Table 1 shows the characteristics of 1010 study participants by midlife depressive symptom status, smoking status, hypertension, T2DM, obesity, and dyslipidemia status. Age at cognitive assessment and follow-up duration were 48 ± 5 years and 39 ± 3 years. The study population comprised of 34% Black participants and 61% women. There were more current smokers and more on antihypertensive medications in the depressive symptoms group (p < 0.001, p = 0.001). Mean levels of glucose (p = 0.018), non-HDL (p < 0.001), and BMI levels (p = 0.005) across the life course were higher in the depressive symptoms group. Mean BMI levels across the life course were higher in never smokers compared to current smokers (p = 0.004).

Table 1.

Clinical characteristics of study population by depression, smoking status, and vascular risk factors at time of cognitive assessment.

	Depression		p-value	Smoking status			p-value	Obesity		p-value
	yes	no		current	former	never		yes	no
N	327	683		197	158	655		532	478
Follow up	38.9 (3.0)	38.7 (3.2)	0.238	38.2 (3.6)	38.9 (3.2)	38.9 (2.9)	0.026	38.5 (3.3)	39.1 (2.9)	0.006
Smoking status			<0.001				N/A			0.004
never	192 (59%)	463(68%)		0 (0%)	0 (0%)	655(100%)		364 (68%)	291 (61%)
current	87 (27%)	110(16%)		197 (100%)	0 (0%)	0 (0%)		83 (16%)	114 (24%)
former	48 (15%)	110(16%)		0 (0%)	158 (100%)	0 (0%)		85 (16%)	73 (15%)
Life course mean
SBP	111 (9)	110 (8)	0.093	110 (9)	110 (9)	110 (8)	0.994	112 (8.3)	109 (8.7)	<0.001
Glucose	90.3 (14.7)	88.1(11.2)	0.018	88.8 (12.2)	90.0 (13.4)	88.5 (12.3)	0.437	91.0 (14.0)	86.4 (9.9)	<0.001
Non-HDL	123.4 (28.3)	116.5 (24.5)	<0.001	118.4 (25.5)	119.5 (24.5)	118.6 (26.5)	0.851	121 (27)	116 (24)	0.001
BMI	25.2 (5.7)	24.1 (5.1)	0.005	23.4 (5.0)	24.3 (5.1)	24.8 (5.4)	0.004	27.4 (5.3)	21.1 (2.7)	<0.001
Medication
HTN	138 (42%)	218 (32%)	0.001	54 (27%)	52 (33%)	250 (38%)	0.017	250 (47%)	106 (22%)	<0.001
T2DM	49 (15%)	75 (11%)	0.07	15 (8%)	22 (14%)	87 (13%)	0.083	94 (18%)	30 (6%)	<0.001
Dyslipidemia	39 (12%)	97 (14%)	0.322	20 (10%)	19 (12%)	97 (15%)	0.207	88 (17%)	48 (10%)	0.003
Achieved education parametera	−0.1(0.6)'	0.1 (0.6)	<0.001	−0.1 (0.6)	0.0 (0.5)	0.0 (0.6)	0.001	0.0 (0.6)	0.0 (0.6)	0.282
Currently employed	149 (46%)	496 (73%)	<0.001	90 (46%)	108 (68%)	447 (68%)	<0.001	344 (65%)	301 (63%)	0.577
Depressive symptoms	327 (100%)	0 (0%)	N/A	87 (44%)	48 (30%)	192 (29%)	<0.001	183 (34%)	144 (30%)	0.147
APOE ε4 status	111 (34%)	219 (32%)	0.551	63 (32%)	51 (32%)	216 (33%)	0.96	182 (34%)	148 (31%)	0.272

	HTN		p-value	T2DM		p-value	Dyslipidemia		p-value
	yes	no		yes	no		yes	no
N	637	373		178	832		649	361
Follow up	39.1 (3.0)	38.2 (3.3)	<0.001	39.5 (2.7)	38.6 (3.2)	<0.001	39.0 (3.0)	38.3 (3.3)	0.001
Smoking status			0.793			0.096			0.898
never	417 (66%)	238 (64%)		120 (67%)	535 (64%)		422 (65%)	233 (65%)
current	124 (19%)	73 (19%)		25 (14%)	172 (21%)		124 (19%)	73 (20%)
former	96 (15%)	62 (17%)		33 (19%)	125 (15%)		103 (16%)	55 (15%)
Life course mean
SBP	114 (8)	105 (7)	<0.001	114 (9)	110 (8)	<0.001	111 (8)	109 (9)	0.017
Glucose	90.6 (13.5)	85.9 (9.9)	<0.001	105 (19.8)	85.3 (5.8)	<0.001	90.0 (13.0)	86.8 (11.0)	<0.001
Non-HDL	120.2 (26.8)	116.2 (24.4)	0.017	125.5 (29.3)	117.3 (25.0)	0.002	129.8 (23.6)	98.7 (16.4)	<0.001
BMI	25.4 (5.5)	22.7 (4.5)	<0.001	27.9 (6.1)	23.7 (4.8)	<0.001	24.8 (4.9)	23.8 (6.0)	<0.001
Medication
HTN	356 (56%)	0 (0%)	N/A	115 (65%)	241 (29%)	<0.001	249 (38%)	107 (30%)	0.005
T2DM	99 (16%)	25 (7%)	<0.001	124 (70%)	0 (0%)	N/A	93 (14%)	31 (9%)	0.008
Dyslipidemia	109 (17%)	27 (7%)	<0.001	53 (30%)	83 (10%)	<0.001	136 (21%)	0 (0%)	N/A
Achieved education parametera	0.0 (0.6)	0.0 (0.6)	0.831	0.0(0.6)	0.0(0.6)	0.851	0.0 (0.6)	0.0(0.6)	0.837
Currently employed	393 (62%)	252 (68%)	0.061	99 (56%)	546 (66%)	0.012	418 (64%)	227 (63%)	0.629
Depressive symptoms	227 (36%)	100 (27%)	0.004	74 (42%)	253 (30%)	0.004	221 (34%)	106 (29%)	0.127
APOE ε4 status	215 (34%)	115 (31%)	0.34	66 (37%)	264 (32%)	0.167	233 (36%)	97 (27%)	0.003

Values are presented as mean (SD) for continuous variables or n (%) for categorical variables. P values were obtained from student t-test, ANOVA, or Chi-square test.

Abbreviations: SBP, systolic blood pressure; non-HDL, non-high density lipoprotein cholesterol; HTN, hypertension; T2DM, type 2 diabetes mellitus.

^aAchieved education parameter was derived by standardizing the word, letter, and vocabulary neuropsychological test scores and calculating their average.

Associations between VRF variability and GCS were overall stronger in individuals without the effect modifiers. Principal effect estimates are shown in Fig. 1. All effect estimates are presented in Supplemental Figs. 1–3. Being in the highest tertile of glucose variability was associated with poorer GCS in individuals without depression (β = −0.14, 95%CI -0.23,-0.04), never smokers (β = −0.12, 95%CI -0.22,-0.03), without obesity (β = −0.13,95% CI -0.23,-0.02), without dyslipidemia (β = −0.18, 95% CI -0.30,-0.06), and without diabetes (β = −0.10, 95% CI -0.18,-0.02). Of note, being in the highest tertile of glucose variability was associated with poorer GCS only in individuals with hypertension (β = −0.12, 95%CI -0.22, −0.02). Associations between SBP variability and GCS was modified by smoking status only. Being in the highest tertile of SBP variability was associated with poorer GCS in former smokers only (β = −0.16, 95% CI -0.30, −0.02) In contrast to SBP and glucose variability, being in the middle tertile of BMI variability was associated with better GCS only in individuals without depression (β = 0.11 (95% CI 0.02, 0.20), in never smokers (β = 0.09, 95% CI 0.00, 0.17), and in individuals without T2DM (β = 0.10, 95% CI 0.02, 0.18). Being in the middle tertile of non-HDL variability was associated with poorer GCS only in individuals without depression (β = −0.11, 95% CI -0.20, −0.03) When interactions between VRF variability and effect modifiers were evaluated, only non-HDL variability showed interaction with depression status in the association with GCS (middle vs. lowest tertile, β = 0.16, 95% CI = 0.01, 0.32).

Fig. 1.

Association of vascular risk factor variability with global cognitive score stratified by midlife effect modifiers Panels: A, glucose variability stratified by depressive symptoms, smoking status, obesity, dyslipidemia, hypertension and diabetes status; B, SBP, BMI, and non-HDL variability stratified by the same effect modifiers as glucose variability Number of participants in subgroups:Depression yes: n = 327, depresson no: n = 683; current smoker: n = 197 former smoker: n = 158 never smoker: n = 655; obesity yes: n = 532, obesity no: n = 478; hypertension yes: n = 637, hypertension no: n = 373; T2DM yes: n = 178, T2DM no: n = 832; dyslipidemia yes: n = 649, dyslipidemia no: n = 361. For all effect modifiers, beta coefficients and 95% CIs are shown for models adjusted with the same covariates as follows, except effect modifiers were removed as a covariate due to stratification. (For example, when stratified by smoking status, smoking status was removed as a covariate due to stratification.) Only the principal subgroup effect estimates are shown. For SBP and glucose variability, highest tertile effect estimates are shown. For BMI and non-HDL variability, middle tertile effect estimates are shown. SBP variability: adjusted for smoking status, T2DM, obesity, dyslipidemia, follow-up duration, mean SBP over the life course, achieved education parameter*, employment, and APOE ε4 status. Glucose variability: adjusted for smoking status, hypertension, obesity, dyslipidemia, follow-up duration, mean glucose over the life course, achieved education parameter*, employment, and APOE ε4 status. BMI variability: adjusted for smoking status, hypertension, T2DM, dyslipidemia, follow-up duration, mean BMI over the life course, achieved education parameter*, employment, and APOE ε4 status. Non-HDL variability: adjusted for smoking status, hypertension, T2DM, obesity, follow-up duration, mean non-HDL over the life course, achieved education parameter*, employment, and APOE ε4 status. Abbreviations: SBP, systolic blood pressure variability; Glucose, glucose variability; BMI, body mass index variability; Non-HDL, non-high density cholesterol variability. *Achieved education parameter was derived by standardizing the word, letter, and vocabulary neuropsychological test scores and averaging them.

4. Discussion

In this community-based cohort, depression, smoking status, obesity, dyslipidemia, hypertension, and T2DM modified the associations between long-term VRF variability from childhood and midlife cognitive function. Patterns differed by VRF and by variability level. Glucose variability showed the broadest modification, with most associations observed in subgroups without the modifying factor. High-level SBP variability was associated with poorer GCS only among former smokers. In contrast, middle-level BMI variability was associated with better GCS among individuals without depression, without diabetes, and among never smokers, while middle-level non-HDL variability was associated with poorer GCS only among individuals without depression. These findings suggest that residual risk of poorer midlife cognitive function may be present among individuals with higher long-term VRF variability from childhood, even in low-risk subgroups.

Our study adds novel insights on the association between VRF variability and CF by extending the exposure window to childhood, compared to previous studies which focused on older adults or young adulthood [32,33]. Increased VRF variability from early life may reflect impaired cardiometabolic homeostasis, and contribute to concomitant elevation of mean VRF levels [34]. Therefore, our findings support the clinical relevance of tracking both mean and variability of VRF levels across the life course to inform personalized prevention strategies for dementia. In addition, even though CF assessment was conducted as a single time measurement in midlife, evaluating effect modification by midlife dementia risk factors such as depressive symptoms, smoking status, hypertension, T2DM, dyslipidemia, and obesity influencing the association between VRF variability from childhood and midlife cognition may help identify which VRF variability to focus for monitoring and intervention to preserve brain health [2].

Glucose variability showed the broadest pattern of associations with cognition across effect modifier subgroups, consistent with evidence linking glucose dysregulation to global cognition through metabolic pathways [35]. Also, associations between glucose variability and poorer cognition even in low-risk groups (no depression, obesity, dyslipidemia, or T2DM) align with previous studies in that residual risk of cognitive dysfunction may persist even when mean glucose levels are below clinical thresholds [6,33]. Our findings align with the results of a meta-analysis that found a multiplicative interaction of diabetes and hypertension and AD risk (diabetes OR 2.6, hypertension OR 3.3 and interaction term, 4.89) [12].

In our study, associations between higher glucose, SBP, and non-HDL variability and poorer cognition were observed mainly in subgroups without effect modifiers. However, for smoking status, our findings align with those of the Ohasama Study, when compared to never smokers, day-to-day SBP variability were associated with cerebral infarction in smokers only [36]. This pattern may reflect long-standing hypertension among individuals who quit smoking to control hypertension, in whom elevated SBP variability persists despite cessation. Our findings of association between higher glucose variability and poorer cognition in individuals without obesity and in individuals without dyslipidemia may reflect underlying central obesity or metabolic dysfunction not captured by BMI or lipid categories [37]. Our findings align with a previous study in adults showing no effect modification by obesity when comparing the number of vascular parameters in high variability and dementia risk [6].

Multiple mechanisms may explain how VRF variability influence cognition. Long-term VRF variability may influence midlife CF through shared pathways which involve inflammation and oxidative stress [15]. In addition, SBP variability can contribute to cognitive dysfunction through shear stress, recurrent hypoperfusion, and cerebral vessel remodeling [38]. Hyperglycemia from glucose variability may lead to reactive hyperinsulinemia and peripheral and central insulin resistance [39]. Associations between non-HDL variability and CF may be an epiphenomenon of worsening of other VRFs [40]. Alternatively, non-HDL variability may play a role in increasing the risk of cerebrovascular events by increasing the instability of atherosclerotic plaques [41]. BMI variability, or weight cycling, may redistribute body fat from the thighs to the abdomen. The resulting visceral fat may be the driver of insulin resistance, compromising insulin transport across the BBB [42]. Also, the levels of BP, glucose, and lipid levels may rise and fall with BMI variability, and the resulting VRF variability may adversely affect brain health (“repeated overshoot” hypothesis. [43]).

Several findings from our study diverged from the original hypothesis. In our study, no effect modification by hypertension of the association between SBP variability and cognition were found. In contrast, in a post-hoc analysis of the ASPirin in Reducing Events in the Elderly (ASPREE) trial, BP variability among men on antihypertensive medication was associated with the highest incident dementia risk [32]. Differences in hypertension severity and duration, as well as medication adherence may explain the discrepancies between the two studies [32] Effect modification by diabetes status of the association between glucose variability and cognition yielded mixed findings, with some reporting no association after excluding incident T2DM and others showing stronger associations in individuals without T2DM [33,44]. Differences in exposure duration of VRF variability, or differences in the severity and duration of T2DM may contribute to these discrepancies [44].

Contrary to the original hypothesis, middle-level BMI variability was associated with better CF among individuals without depression, never smokers, and those without T2DM. Multiple factors may explain these discrepancies. BMI may represent both fat and lean mass, therefore middle-level BMI variability may represent both intentional weight loss or an increase in muscle mass [37]. [45] Also, BMI variability may differentially influence cognition according to the pattern of weight change. In older adults, greater BMI changes and higher BMI variability were associated with faster cognitive decline, while in another study, weight loss and higher BMI variability had a negative impact on older adults’ cognition [46,47].

In our study, middle-level non-HDL variability was associated with poorer midlife cognition. This may be explained by a threshold inflection point for non-HDL and its action on cognition. Below the threshold non-HDL level, non-HDL variability may have an adverse effect on cognition only below a threshold non-HDL level; however, above this threshold, non-HDL variability may not influence cognition [48].

Strength of this study is the observation that residual risk of poorer midlife CF may be present among individuals with higher long-term VRF variability from childhood, even in low-risk subgroups. This suggests that monitoring VRF variability and implementing targeted interventions may benefit even in low-risk groups with subclinical VRF levels. Another strength is the longitudinal examination of variability in SBP, glucose, non-HDL, and BMI from early life.

Limitations of this study are the following. Clinical diagnosis and duration of depression or smoking were unavailable. Midlife CF was assessed once. Alternative measures of adiposity such as weight circumference were not assessed [49]. Unmeasured confounding is possible due to unavailable data on physical activity, childhood dietary status, and neuroimaging studies. Information regarding antihypertensive, lipid-lowering, or glucose-lowering medication use during follow-up was unavailable; therefore the influence of these medication use during follow-up could not be evaluated. Also, in low-risk individuals (for example, participants without hypertension, diabetes, or obesity), unmeasured subclinical cardiometabolic abnormalities may partly account for the associations between higher VRF variability and poorer CF, thereby warranting cautious interpretation. Also, due to the large number of subgroup analysis, findings may not be significant due to lack of power as opposed to an absence of a true effect.

Race and sex pooled analysis was conducted, as adding race and sex stratification to the subgroup analysis would have introduced two-way interactions and substantially reduced statistical power and complicated interpretability. Depression, smoking status, and VRFs measured earlier in life may modify the association between VRF variability and cognition differently across the life course. Future studies such as trajectory analysis would be needed to evaluate how these modifiers shape these associations over time.

In this study, CV was chosen over Variability Independent of the Mean (VIM) since it is computationally simpler. VIM is calculated using the regression exponent β from fitting log (SD) on log (mean) across the study population. Therefore, VIM would be affected by the fit of the regression model [20,50]. Also, CV was chosen over average real variability, since in this study, the number of VRF measurements and the time interval between measurements were not equal across participants.

5. Conclusion

Residual risk of poorer midlife CF may be present among individuals with higher long-term glucose or SBP variability from childhood, even in subgroups considered low risk. Higher glucose variability was associated with poorer midlife CF among individuals without depression, obesity, dyslipidemia, or T2DM, and among those with hypertension. Higher SBP variability was similarly associated with poorer GCS among former smokers. These findings underscore the importance of monitoring VRF variability from early life and implementing targeted strategies to reduce VRF variability as a potential approach to support long-term brain health.

CRediT authorship contribution statement

Soo Jung Kang: Writing – review & editing, Writing – original draft, Visualization, Validation, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Jeanette Gustat: Writing – review & editing, Methodology. Wan Tang: Writing – review & editing, Methodology. Phillip H. Hwang: Writing – review & editing, Methodology. Owen T. Carmichael: Writing – review & editing, Methodology, Funding acquisition. Ileana De Anda-Duran: Writing – review & editing, Methodology. Lydia A. Bazzano: Writing – review & editing, Supervision, Resources, Methodology, Funding acquisition, Conceptualization.

Ethical review statement

At each examination, all study participants provided written informed consent, and the Institutional Review Board of the Tulane University Health Sciences Center approved the study protocols.

Sources of funding

This work was supported by an American Heart Association Predoctoral Fellowship[grant DOI: AHA.24PRE1022654.pc.gr.190680" title = "https://doi.org/10.58275/AHA.24PRE1022654.pc.gr.190680">https://doi.org/10.58275/AHA.24PRE1022654.pc.gr.190680 ] and by the National Institutes of Health grant R01 [R01AG041200]. The findings of this study do not necessarily reflect the views of the American Heart Association or National Institutes of Health.

Declaration of competing interest

The authors have nothing to disclose.

Appendix A. Supplementary data

The following is the Supplementary data to this article: