Sex differences in the association between major cardiovascular risk factors in midlife and dementia: a cohort study using data from the UK Biobank

We examined the associations with incident all-cause dementia for a range of major cardiovascular risk factors: blood pressure (systolic blood pressure (SBP), diastolic blood pressure (DBP)), smoking status and intensity, diabetes mellitus (type 1 and type 2), adiposity (body mass index (BMI), waist circumference (WC), waist-to-hip ratio (WHR) and waist-to-height ratio (WHTR)), prior stroke, socioeconomic status (SES) and lipids (total cholesterol, high-density lipoprotein (HDL) cholesterol and low-density lipoprotein (LDL) cholesterol).

Blood pressure was taken at baseline using the Omron HEM-7015IT digital blood pressure monitor by taking the mean of two sitting measures. Based on the American Heart Association (AHA) 2017 guidelines [11], we categorised blood pressure into four groups (normal: SBP < 120 mmHg and DBP < 80 mmHg; elevated: SBP 120–129 mmHg and DBP < 80 mmHg; stage 1 hypertension: SBP 130–139 mmHg or DBP 80–89 mmHg; stage 2 hypertension: SBP ≥ 140 mmHg or DBP ≥ 90 mmHg). Smoking status was self-reported and categorised as never, former, or current smokers. Daily consumption of cigarettes was collected among current smokers. Self-reported diabetes was recorded: if the age at diagnosis was less than 30, and the participant was using insulin, it was classified as having type 1 diabetes, otherwise as type 2 diabetes. BMI was calculated as the weight of the individual in kilogrammes, measured using the Tanita BC-418 MA body composition analyser, divided by the square of the individual’s standing height in metres. BMI was used as a continuous measure, as well as categorised into four groups (underweight: < 18.5 kg/m²; healthy weight: 18.5–24.9 kg/m²; overweight: 25.0–29.9 kg/m²; obese: 30.0 kg/m² and above). Waist and hip circumference were measured using the Wessex non-stretchable sprung tape measure. History of stroke was self-reported, based on a touchscreen questionnaire, and a nurse-led interview was conducted to confirm the medical history. SES was determined using the Townsend Deprivation Index and categorised into three levels using the national cut-off points (high ≤ − 2.08; middle − 2.08 – 1.40; low ≥ 1.40). The Townsend Deprivation Index is a measure of area deprivation, derived from the national census data about unemployment, car ownership, household overcrowding and owner occupation, with higher scores indicate higher levels of deprivation [12]. Blood lipid levels were measured using the Beckman Coulter AU580, and elevated cholesterol was defined as total cholesterol ≥ 6.2 mmol/L.

The primary study endpoint was incident fatal or non-fatal all-cause dementia. The International Classification of Diseases ICD-10 codes (A81.0, F00, F01, F02, F03, F05, G30, G31.0, G31.1, G31.8, and I67.3) were used to identify participants with dementia if one or more of these codes were recorded as a primary or secondary diagnosis in the health records or recorded as the underlying or contributory cause of death in the death registers. Outcome adjudication for incident dementia was conducted by the UK Biobank Outcome Adjudication Group, and ICD-10 codes were used to determine Alzheimer’s disease (AD) (F00, G30) and vascular dementia (F01, I67.3).

Baseline characteristics for women and men were presented as number (percentage) for categorical variables and as mean (standard deviation) for continuous variables. The crude incidence of all-cause dementia was modelled using Poisson regression, with sex as a covariate and a log offset for person-years. Cox proportional hazards regression models were used to estimate the sex-specific hazard ratios (HR) and 95% confidence intervals (CI) for each risk factor for dementia, with interaction terms fitted between each exposure variable and sex [13].

Analyses for all risk factors were adjusted for age, with different sets of covariates, determined a priori, for each risk factor according to perceived probable causal relationships. All models were adjusted for age. In addition, SBP, diabetes, socioeconomic status and total cholesterol were adjusted for each other, and adjusted for smoking status, BMI, lipid-lowering drugs and antihypertensive drugs. For DBP and AHA hypertension, the same adjustments were made as SBP. Stroke and smoking variables were adjusted for socioeconomic status. Body adiposity was adjusted for smoking status and socioeconomic status. HDL and LDL cholesterol were adjusted the same way as total cholesterol. The interaction term between each risk factor and sex was used to obtain the women-to-men ratio of hazard ratios (RHR) for each risk factor [13]. The effects of the risk factors were evaluated separately for vascular dementia and AD using Cox models, with the same adjustments used in the main analyses.

We conducted subgroup analyses investigating whether sex differences in risk factors for dementia varied by age group (< 60 years and ≥ 60 years). Similarly, subgroup analyses were undertaken by SES groups, categorising individuals as above or below the national median Townsend score in the UK (− 0.56).

In addition to making relative comparisons, we evaluated sex differences on the absolute scale, by estimating sex-disaggregated unadjusted and multiple-adjusted rates per 10,000 person-years, and the women-to-men difference of the rate differences, using Poisson regression models. Covariate adjustments were the same as those made in the Cox models.

Post-hoc exploratory analyses were conducted after observing unexpected results from the initial analyses for blood pressure and BMI. We assessed the shape of the associations for continuous SBP, DBP and BMI, with the risk of dementia for women and men using penalised smoothing splines, adjusted for the same set of covariates as outlined for the Cox models. Extreme values in the upper and lower 0.1% of the blood pressure and BMI distributions, which are associated with unreliable estimates, were omitted. The reference was 120.0 mmHg for SBP, 80.0 mmHg for DBP and 22.5 kg/m² for BMI. Sex differences in blood pressure were further explored by disaggregating the results by baseline antihypertensive use.

Kaplan-Meier survival curves were additionally constructed to assess the survival probability of death and dementia by sex.

Sensitivity analysis was conducted, by fitting the proportional sub-distribution hazards regression model described by Fine and Grey [14], in which death was considered as a competing risk.

All analyses were performed on complete case data using R Studio Version 4.0.2 (R Core Team, 2020) and Stata 16.0 (StataCorp, 2019).

Sex was a significant risk factor for dementia, with women at a lower risk of dementia than men (HR, 0.83 [0.77–0.89]) after multiple adjustments of systolic blood pressure, baseline age, smoking status, lipids medications, total cholesterol, antihypertensive medications, body mass index, diabetes and Townsend deprivation index.

The age-adjusted HRs (see Additional file 1) are broadly similar to the multiple adjusted HRs (Fig. 1), although the associations between lipids and dementia were no longer significant after multiple adjustments.

The results for vascular dementia and AD were broadly similar to all-cause dementia (see Additional file 4). There was some evidence indicating that diabetes, smoking and stroke may be more strongly associated with vascular dementia than with AD. The sex difference in the effect of raised BP was consistent for vascular dementia and AD.

There was no evidence for effect modification of the sex differences by age group (≥ 60 vs < 60 years), except for among those who smoked ≥ 20 cigarettes per day; it appeared that the risk of dementia was higher in men among the younger age group, while the risk of dementia was higher in women among the older age group (p = 0.03) (see Additional file 5).

There was some evidence of heterogeneity by SES in sex differences between blood pressure (systolic and diastolic) and dementia (p = 0.03 for SBP and p = 0.01 for DBP; see Additional file 6), driven by the comparatively stronger association of blood pressure with dementia among women of higher SES.

The crude incidence rates for dementia was 5.88 [5.62–6.16] in women and 8.42 [8.07–8.78] in men per 10,000 person-years. For all risk factors considered, and for all the categories in these risk factors, the rates of dementia were higher in men than women (Table 2). Unadjusted rates were broadly similar to the multiple-adjusted rates (see Additional file 7).

In addition, the Kaplan-Meier survival curves indicated that the survival probabilities of both death and dementia were lower for men in comparison with women throughout follow-up (see Additional file 8).

The competing risk of death was considered in the analyses, and the results are presented in the supplementary material (see Additional file 9). There were minor differences between the competing risk analysis and the main analysis presented in Fig. 1. Most notably, the ratios of sub-distribution hazard ratios for BMI categories were no longer significant compared with our main analysis.

High blood pressure has been consistently linked to dementia and impairment of some cognitive domains [15, 16], but few studies have evaluated potential sex differences, with mixed empirical results [17‐21]. A study of included members of the Kaiser Permanente Northern California (KPNC) found that midlife hypertension was associated with a higher risk of dementia in women only [17]; with two other studies also reporting that hypertension was associated with vascular dementia [18] and mild cognitive impairment [19] in women but not in men. Our study not only used guideline-based definitions to categorise hypertension, blood pressures were also included as continuous measures, offering insights to more precise blood pressure targets in improving cognitive health.

Several mechanisms may provide some explanations for the sex differences between blood pressure and dementia observed in our study. First, women may have better cerebral autoregulation than men [22], and reduced cerebral perfusion in men may be associated with a higher incidence of orthostatic hypotension hospitalisation [22], which has been associated with a greater risk of dementia [23]. Second, sex differences in medical treatment are possible. Previous studies reported that treatment adherence was generally lower in women than men, which was in part explained by greater polypharmacy and more side effects in women [24].

Cardiovascular pathology is thought to underlie the increased risk of cognitive impairment and dementia among smokers [3], although smokers also have a higher risk of premature death, precluding them from reaching the age at which dementia might develop; as such, potential uncertainty exists in determining this relationship [3]. Similar to our study, a meta-analysis examined the effects of cigarette smoking on dementia showed that current smokers had a greater risk of all-cause dementia, vascular dementia and AD, but not for former smoker when compared with never smokers [25]. Studies that accounted for the competing risk of death found that the effect of smoking attenuated [26]. Our competing risk analysis indicated that in comparison to never smokers, current smokers and smoking more than 20 cigarettes per day remain to be strong risk factors for all-cause dementia in both women and men. Further clarification is necessary for smoking as a risk factor for dementia.

A previous meta-analysis found that type 2 diabetes was associated with a 60% increased risk of any dementia in both sexes [27]. Similarly, the present study showed that compared to no diabetes, type 1 and type 2 diabetes were associated with an increased risk of dementia in both women and men. The link between diabetes and dementia is alarming, with the number of people with diabetes projected to increase to 68,000 by 2027 in the UK Biobank [10], as it is increasing globally [28]. While routine screening for cognitive dysfunctions in diabetes is necessary [29], novel treatment options and prevention strategies are crucial for reducing the burden of dementia in this at-risk population.

The relationship between body adiposity and dementia is complex and appear to be non-linear depending on dementia subtypes [30]. Our recent meta-analysis, with the inclusion of data from the UK Biobank, showed comparable findings where underweight (BMI < 18.5 kg/m²) was strongly associated with the risk of both vascular and non-vascular dementia [30]. Although no sex difference was reported in the meta-analysis, these effects were only significant in women [30]. Underweight is an important subclinical consequence of prodromal dementia, such that the possibility of reverse causality cannot be ruled out.

A previous meta-analysis showed that female sex was associated with a 30% greater risk for post-stroke dementia, although this may have been influenced by women’s older age [31]. While there was no evidence for a sex difference in the effects of stroke in the present study, a history of stroke was associated with more than double the risk of dementia in both sexes. Our results complement the joint call for action by the World Stroke Organisation in prevention for stroke and dementia [32], as it was estimated that more than a third of dementia could be prevented by preventing stroke [32].

The Three-City Study found that women living in deprived neighbourhoods have a greater risk of dementia, but not men [33]. Another study concluded that lower educational attainment was independently associated with a greater risk of dementia death in women, after considering other comorbidities and behaviours [34]. Given deprivation was associated with the risk of dementia to a similar extent in women and men, reducing socioeconomic inequities may thus have significant effects on dementia prevention, as well as many other conditions on a population level.

The evidence for lipids as risk factors for dementia has been mixed, with a meta-analysis suggesting that high total cholesterol in midlife is associated with a greater risk of developing AD [35]. A recent meta-meta-analysis, which explored the effect of cholesterol fractions concluded that LDL cholesterol increased the risk of AD, while HDL and total cholesterol showed a non-significant association with AD risk [36]. A Mendelian Randomisation study confirmed that high circulating total cholesterol and a reduced level of HDL cholesterol might be associated with an increased risk of AD [37]. Our results do not support such associations.

To our knowledge, this study is the first to systematically evaluate sex differences in a range of cardiovascular risk factors for dementia in a large general population, using standardised methodology. This study was further strengthened by its prospective cohort design and large sample size. Sex differences were compared on both relative and absolute scales, providing insights to inform public health and clinical practice in risk reduction for dementia. This study also has limitations. The UK Biobank cohort is a relatively healthy and affluent population, predominantly of Caucasian ancestry, which may limit the generalisability of the results. Self-reported smoking, diabetes and history of stroke may be subject to reporting bias. Multiple testing of interactions can raise the concern of false positives. With the threshold of 5%, one significant test would be expected in every 20 tests performed, even if there were no real effects. Thus, our results should be interpreted with caution. Lastly, while we were able to distinguish the effects of blood pressure by baseline antihypertensive use, we had no further information on medication use duration or dose, hence limiting our ability to interpret the results.