A Prospective Study of Plasma Lipid Levels and Hypertension in Women

Background  Although dyslipidemia and hypertension occur together more often than can be explained by chance, few studies have carefully explored the nature of the relationship between plasma lipid levels and the risk of developing hypertension.

Methods  We conducted a prospective study of 16 130 middle-aged and older female health professionals in 1992 who provided baseline blood samples and had no history of high cholesterol level (no treatment or diagnosis) or hypertension (no treatment, diagnosis, or elevated blood pressure). Plasma lipid levels were measured, and baseline risk factors were collected. Incident hypertension included a new physician diagnosis, the initiation of antihypertensive treatment, systolic blood pressure of 140 mm Hg or greater, or diastolic blood pressure of 90 mm Hg or greater.

Results  During 10.8 years of follow-up, incident hypertension developed in 4593 women. In multivariate-adjusted models, the relative risks of development of hypertension from the lowest (referent) to the highest quintile of baseline total cholesterol level were 1.00, 0.96, 1.02, 1.09, and 1.12 (P = .002 for trend); for low-density lipoprotein cholesterol level, 1.00, 0.97, 1.00, 1.02, and 1.11 (P = .053 for trend); for high-density lipoprotein cholesterol level, 1.00, 0.93, 0.87, 0.87, and 0.81 (P<.001 for trend); for non–high-density lipoprotein cholesterol level, 1.00, 1.06, 1.11, 1.12, and 1.25 (P<.001 for trend); and for the ratio of total to high-density cholesterol, 1.00, 1.10, 1.14, 1.20, and 1.34 (P<.001 for trend). Similar relative risks were noted for Adult Treatment Panel III clinical cut points and after the exclusion of obese or diabetic women.

Conclusion  In this large prospective cohort, atherogenic dyslipidemias were associated with the subsequent development of hypertension among healthy women.

Hypertension, a major risk factor for the development of cardiovascular disease (CVD),¹ is a progressive condition that emerges as cultures adopt a western lifestyle. Hypertension is often accompanied by other cardiovascular risk factors in populations and among individuals. Dyslipidemia and hypertension occur together more often than can be explained by chance^2,3; however, the precise nature of this relationship remains unclear.

Atherogenic dyslipidemias could lead to hypertension by several mechanisms. First, atherosclerosis can result in structural changes in large conduit arteries, leading to reduced elasticity.⁴ Second, endothelial dysfunction due to lipid abnormalities,^5-7 resulting in reduced nitric oxide production, release, and activity and abnormal vasomotor activity, could manifest as hypertension.⁸ Endothelium-dependent vasodilation is impaired by elevated total cholesterol (TC) levels.⁹ Third, lipid-mediated damage to the renal microvasculature could manifest as hypertension, illustrated by an association between lipid abnormalities and early renal dysfunction.¹⁰ Finally, dyslipidemia and hypertension represent 2 of several components of the metabolic syndrome that may share common mechanistic pathways.^11,12

To date, few studies have examined whether lipid levels are prospectively associated with the risk of developing hypertension.^13-17 Using data from a large prospective study of middle-aged and older women, we examined the association between various plasma lipid levels and the risk of developing hypertension.

The Women’s Health Study is a recently completed trial of low-dose aspirin and vitamin E in the primary prevention of CVD and cancer.^18,19 The Women’s Health Study randomized 39 876 female US health professionals 45 years or older starting in 1992 who were free of prior myocardial infarction, stroke, transient ischemic attack, and cancer (except nonmelanoma skin cancer). All participants provided written informed consent, and the institutional review board at Brigham & Women’s Hospital, Boston, Mass, approved the study protocol.

Before randomization, baseline blood samples were collected from 28 345 participants and stored in liquid nitrogen until analysis. Samples were then transferred to a core laboratory facility, where they were assayed for lipid levels. Total cholesterol and high-density lipoprotein cholesterol (HDL-C) were measured enzymatically using a Hitachi 911 autoanalyzer (Roche Diagnostics, Basel, Switzerland), and low-density lipoprotein cholesterol (LDL-C) was directly measured (Genzyme, Cambridge, Mass). We calculated non–HDL-C level as the difference between TC and HDL-C levels and computed the TC/HDL-C ratio. Of the samples received, 27 939 were assayed for lipid and C-reactive protein levels using a validated, high-sensitivity assay (Denka Seiken, Tokyo, Japan).

On the baseline questionnaire, women provided self-reports of age (in years), weight and height (converted to body mass index [BMI], calculated as weight in kilograms divided by the square of height in meters), smoking status (never, former, or current), alcohol use (rarely/never, 1-3 drinks/mo, 1-6 drinks/wk, or ≥1 drink/d), frequency of exercise (rarely/never or <1, 1-3, or ≥4 times/wk), parental history of myocardial infarction at younger than 60 years (no or yes), history of high cholesterol (no or yes), current treatment for high cholesterol levels (no or yes), history of diabetes mellitus (no or yes), and use of postmenopausal hormone therapy (never, former, or current). Blood pressure (BP) was ascertained from self-reports of 9 ordinal systolic BP (SBP) categories ranging from less than 110 mm Hg to 180 mm Hg or greater and 7 ordinal diastolic BP (DBP) categories ranging from less than 65 mm Hg to 105 mm Hg or greater. A single measurement of self-reported BP in health professionals is highly correlated with measured SBP (r = 0.72) and DBP (r = 0.60).²⁰

We then excluded women at baseline with high cholesterol levels (current treatment or physician diagnosis) or hypertension (current treatment, physician diagnosis, SBP of ≥140 mm Hg, or DBP of ≥90 mm Hg). Women with hyperlipidemia based only on baseline cholesterol levels were not excluded to isolate the effect of lipid levels on hypertension in the absence of treatment. As a result, a baseline population of 16 130 women initially free of hypertension, treatment or diagnosis of high cholesterol levels, and CVD remained for analysis.

Follow-up consisted of annual questionnaires updating information on compliance, adverse effects of the study agents, health outcomes, and risk factors. The most recent follow-up rates for morbidity and mortality were 97.2% and 99.4%, respectively. To become an incident case of hypertension, women must have provided self-reports of at least 1 of the following 4 criteria: (1) new physician diagnosis of hypertension on follow-up questionnaires at years 1 and 3, and on all annual questionnaires thereafter; (2) initiation of antihypertensive treatment at years 1, 3, or 4; (3) SBP of 140 mm Hg or greater; or (4) DBP of 90 mm Hg or greater. New reports of physician-diagnosed hypertension included the month and year of diagnosis. A missing date for a physician diagnosis of hypertension defined by another criterion was assigned a date of incident hypertension by randomly selecting a date between the current and previous annual questionnaires. Those subjects who developed a major cardiovascular end point (myocardial infarction, stroke, or coronary revascularization) with management that may have an impact on BP, after baseline but before the development of hypertension (n = 37), were censored at their cardiovascular event date and not considered incident cases of hypertension. Based on this definition, 4593 cases of incident hypertension developed during a median follow-up of 9.7 years (maximum, 10.8 years).

The distributions of quintiles for each lipid level were first determined. Women were compared for their coronary risk factors according to quintiles of TC level by using mean values or proportions of baseline risk factors to assess potential confounding. Using Cox proportional hazards analyses, we first calculated the hazard ratios (subsequently referred to as relative risks [RRs]) and 95% confidence intervals (CIs) of incident hypertension for increasing quintiles of each lipid variable, with the lowest quintile as the referent. Models were first adjusted for age and randomized treatment assignment, then further adjusted for exercise, smoking status, alcohol intake, use of postmenopausal hormone therapy, parental history of myocardial infarction before age 60 years, BMI, and diabetes for each lipid level. Linear trend tests for each lipid level were performed using the median level for each quintile as an ordinal variable. The assumption of proportional hazards was satisfied for each of the lipid variables (P>.05 for all). Finally, given the previously published association between C-reactive protein level and hypertension in this cohort,²¹ we assessed whether the addition of C-reactive protein level to models of each lipid variable attenuated the RRs of development of hypertension.

A second set of analyses considered the following clinical cut points of each lipid variable set forth by the National Cholesterol Education Program Adult Treatment Panel III²²: for TC level, less than 200, 200 to 239, and 240 or more mg/dL (<5.18, 5.18-6.19, and ≥6.20 mmol/L); for LDL-C level, less than 130, 130 to 159, and 160 or more mg/dL (<3.37, 3.37-4.13, and ≥4.14 mmol/L); for HDL-C level, less than 40, 40 to 59, and 60 or more mg/dL (<1.04, 1.04-1.54, and ≥1.55 mmol/L); for non–HDL-C level, less than 160, 160 to 189, and 190 or more mg/dL (<4.14, 4.14-4.91, and ≥4.92 mmol/L); and for the TC/HDL-C ratio, less than 4.0, 4.0 to less than 6.0, and 6.0 or more. Models paralleled that for the quintiles described in the previous paragraph.

Additional models considered whether baseline SBP or DBP mediated the association between each lipid level and hypertension. Stratified analyses according to the baseline level of SBP (<120 and ≥120 mm Hg) and DBP (<75 and ≥75 mm Hg) sought to classify women as prehypertensive according to the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC 7) (SBP of 120-139 mm Hg or DBP of 80-89 mm Hg).²³ Because the ordinal categories of DBP included 75 to 84 and 85 to 89 mm Hg, we could not replicate JNC 7 categories perfectly. Tests for interaction considered continuous variables for each lipid and BP level.

Because we excluded women with CVD or hypertension and those who were treated for or who had received a physician diagnosis of high cholesterol level, many but not all women had favorable baseline lipid levels within the guidelines set forth by the National Cholesterol Education Program Adult Treatment Panel III.²² The mean ± SD levels of TC, LDL-C, HDL-C, and non–HDL-C and TC/HDL-C ratio were 201.6 ± 37.8 mg/dL (5.22 ± 0.98 mmol/L), 116.2 ± 30.8 mg/dL (3.01 ± 0.80 mmol/L), 55.1 ± 15.0 mg/dL (1.43 ± 0.39 mmol/L), 146.5 ± 36.5 mg/dL (3.79 ± 0.95 mmol/L), and 3.88 ± 1.16, respectively. Baseline characteristics of the study population according to quintiles of TC levels are presented in Table 1. Age and BMI were greater with increasing quintiles of TC levels. In addition, women with higher TC levels were more likely to be current smokers, have diabetes, exercise less, and use postmenopausal hormone treatment. Women in higher quintiles of plasma TC levels had increases in SBP and DBP, which were of greater magnitude for SBP.

During a median follow-up of 9.7 years (maximum, 10.8 years) for the 16 130 women constituting the baseline population, hypertension developed in 4593 women on the basis of newly initiated treatment or a diagnosis (51.5% of women), SBP of 140 mm Hg or greater or DBP of 90 mm Hg or greater (20.2%), or multiple indications (28.3%). According to the data presented in Table 2, there was a significant association between increasing quintiles of lipid levels and an increased risk of developing hypertension. Each lipid variable was associated with the risk of hypertension in age-adjusted models (P<.001 for trend for all). On multivariate adjustment, the associations were attenuated but remained significant except for LDL-C level (P = .053 for trend). The HDL-C level was a particularly powerful predictor of hypertension, with progressively greater decreases in the risk of developing hypertension from the second (RR = 0.93) through fifth (RR = 0.81) quintiles (P<.001 for trend). Women in the highest quintile of non–HDL-C level and TC/HDL-C ratio also had significant RRs of 1.25 and 1.34, respectively, for the risk of hypertension.

We then considered several sensitivity analyses of the observed association between lipid levels and hypertension. Models that excluded early cases of hypertension during the first 6 months of follow-up did not change the RRs. When we considered lipid variables as continuous variables, the interpretation of the results paralleled those for quintiles or clinical cut points.

Adjustment for BMI accounted for most of the attenuation in the RRs for each lipid variable, but there was no evidence of effect modification by BMI for the association between lipid levels and hypertension. We also considered adjustment for waist circumference and waist-hip ratio, which was provided on the 6-year follow-up questionnaire. Contrary to BMI, controlling for either measure of central adiposity had a minimal impact on the RRs. We then limited analyses to nonobese, nondiabetic women (n = 13 841; 3576 cases of hypertension) and found that lipid levels remained predictive of hypertension in the absence of these 2 strong risk factors, paralleling the results in Table 2. In these 13 841 nonobese, nondiabetic women, the multivariate RRs (95% CIs) of hypertension for increasing quintiles of TC level were 1.00 (referent), 0.97 (0.87-1.09), 1.00 (0.90-1.12), 1.08 (0.97-1.21), and 1.09 (0.98-1.22) (P = .03 for trend); for increasing quintiles of LDL-C level, 1.00 (referent), 0.95 (0.86-1.04), 1.01 (0.92-1.12), 0.95 (0.85-1.06), and 1.08 (0.96-1.22) (P = .35 for trend); for increasing quintiles of HDL-C level, 1.00 (referent), 0.92 (0.83-1.02), 0.83 (0.75-0.93), 0.86 (0.77-0.96), and 0.84 (0.75-0.94) (P<.001 for trend); for increasing quintiles of non–HDL-C level, 1.00 (referent), 1.00 (0.89-1.11), 1.09 (0.98-1.22), 1.09 (0.98-1.22), and 1.19 (1.06-1.33) (P<.001 for trend); and for increasing quintiles of the TC/HDL-C ratio, 1.00 (referent), 1.08 (0.97-1.21), 1.08 (0.97-1.21), 1.14 (1.02-1.28), and 1.30 (1.16-1.46) (P<.001 for trend). Conversely, had we added 4355 women currently receiving treatment for or diagnosed as having a high cholesterol level into our analyses, it had a minute impact on the reported RRs and did not change the overall interpretation of the results.

Another important consideration was adjustment for C-reactive protein level, previously shown to be predictive of hypertension in this population of women.²¹ The median level of C-reactive protein among all 16 130 women was 1.59 mg/L, and increased with increasing levels of each plasma lipid. When plasma C-reactive protein was added to the multivariate models for each lipid variable, the RRs remained virtually identical. Finally, in addition to alcohol consumption, we examined the impact of confounding by various dietary factors and found no evidence of any impact on the RRs.

We next examined lipid level cut points following clinical guidelines²² and the risk of developing hypertension, overall and stratified by SBP or DBP. Similar patterns of confounding were evident as observed in analyses of quintiles. As shown in Table 3, levels of HDL-C and non–HDL-C and the TC/HDL-C ratio were the strongest multivariate predictors of hypertension using well-established clinical guidelines. Stratification by normal (<120 mm Hg) and prehypertension (120-139 mm Hg) SBP categories according to JNC 7 guidelines²³ revealed significant interactions for levels of HDL-C and non–HDL-C and the TC/HDL-C ratio (P<.05 for interaction for all). Stratification by DBP according to categories of less than 75 mm Hg and 75 to 89 mm Hg revealed significant interactions with every lipid variable except HDL-C level for the risk of developing hypertension (P<.05 for interaction). Not all significant interactions reflected meaningful differences in RRs by SBP and DBP categories. However, women with normal levels of baseline SBP and DBP had consistently higher RRs of hypertension for all lipid variables except HDL-C levels. Alternatively, we adjusted for baseline SBP and DBP in our overall multivariate models from Table 2 for each plasma lipid level to evaluate whether plasma lipid levels have an impact on the risk of hypertension through a mechanism separate from BP. Adjustment for SBP and DBP greatly attenuated the effects of each lipid level on the risk of hypertension, but significant associations (P<.05 for all for trends) remained for levels of HDL-C and non–HDL-C and the TC/HDL-C ratio.

This study provides evidence that baseline levels of lipids, particularly HDL-C and non–HDL-C and the TC/ HDL-C ratio, are associated with an increased risk of incident hypertension, even among those with low initial SBP and DBP as defined by JNC 7 guidelines.²³ Adverse lipid levels therefore appear to occur well before the development of hypertension; however, the precise biological mechanism by which lipids may give rise to elevations in BP remain unclear. This finding appeared to extend to women free of baseline diabetes and obesity. The finding of an association between lipid levels and hypertension also persisted after various sensitivity analyses.

The risk stratification of lipid levels according to National Cholesterol Education Program²² guidelines appeared to be clinically relevant for identifying women at greatest risk for development of hypertension. Subsequent stratification by JNC 7²³ criteria for those women who are healthy or prehypertensive suggested that even women with low baseline BP levels but elevated lipid levels may be at risk for hypertension, as the absolute risks of hypertension remained relatively high.

Genetic²⁴ and cross-sectional^25,26 studies support a link between dyslipidemia and hypertension. Hypertensive individuals have a higher prevalence of dyslipidemia,^11,27 and 12% of subjects with early-onset hypertension have an increased frequency of lipid disorders.^28,29 Another study of 1999 healthy men aged 40 to 59 years found an association between resting BP and triglyceride levels.³⁰

Limited, smaller prospective studies have provided evidence of an association between plasma lipid levels and the risk of hypertension. In 1 study, 1482 adult men and women were followed up for 7 years, with 40 cases of hypertension developing. Increases of 1 SD in triglyceride (110 mg/dL [1.24 mmol/L]) and HDL-C levels (11 mg/dL [0.28 mmol/L]) had age-adjusted RRs of 1.42 (95% CI, 1.06-1.89) and 0.82 (95% CI, 0.59-1.15), respectively.¹⁵ In another 10-year study of 2322 middle-aged men (168 incident cases of hypertension), baseline triglyceride and TC levels significantly increased with increasing levels of DBP.¹⁶ In 2 separate studies from the San Antonio Heart Study, subjects with higher baseline triglyceride and lower HDL-C levels had a significantly greater risk of developing hypertension, whereas higher TC and LDL-C levels were associated with nonsignificantly increased risk.¹³ In a second analysis of 1039 nonhypertensive, nondiabetic men and women observed for 7 years (93 incident cases of hypertension), higher triglyceride levels were associated with an increased risk of hypertension, with a multivariate RR of 2.40 (95% CI, 1.28-4.49).¹⁴ Although trial data are limited, 1 small study among 22 patients with isolated systolic hypertension demonstrated that therapy to lower lipid levels with atorvastatin calcium improved systemic arterial compliance and significantly lowered SBP and DBP.³¹ Additional prospective data on the relationship between lipid levels and hypertension will help confirm or refute these associations.

Oparil et al⁴ suggest several mechanisms by which dyslipidemias may result in hypertension over time. First, arteriosclerosis in the large conduit arteries resulting from smooth muscle cell hypertrophy and collagen deposition leads to arterial stiffness. This may translate to elevated SBP seen with advancing age. In addition, a healthy endothelium responds to intravascular cues whether to dilate or constrict. Hypertensive patients clearly have increased vascular reactivity.⁴ Dyslipidemias lead to endothelial dysfunction and improper vasoregulation,³² as nitric oxide production, release, and subsequent activity are reduced among those with high TC and low HDL-C levels. Furthermore, endothelin-1, a potent vasoconstrictor mediated by endothelin-A and -B receptors, increases resistance vessel tone in hypertensive subjects more than in subjects with hypercholesterolemia.³³ Dyslipidemia has been associated with increased circulating levels of endothelin-1,³⁴ which in turn has been linked with hypertension in some^35,36 but not all³⁷ studies. Furthermore, endothelium-dependent vasodilation is inversely correlated with TC levels.⁹ In addition, dyslipidemia may cause damage to the renal microvasculature with the downstream effect of hypertension. A recent report has linked dyslipidemia to renal decline.¹⁰

Finally, dyslipidemia and hypertension are both parts of the metabolic syndrome. Some suggest that hypertension represents a late-stage manifestation of a metabolic syndrome.^13,38,39 It remains unclear why hypertension develops in patients with the metabolic syndrome. Dyslipidemia leading to large conduit vessel stiffness, endothelial dysfunction, and renal microvascular disease⁴ is perhaps one mechanism. Furthermore, lipid levels remained predictive of hypertension in our study, even after we excluded women with obesity and diabetes. However, residual confounding by factors influencing lipid and BP levels may still be present. Although we comprehensively controlled for many common risk factors, our study did not consider confounders that represented insulin resistance, endothelial function, and other relevant biological pathways that may affect lipid and BP levels. In addition, any measurement errors for self-reported BMI may be correlated with the errors in reporting incident hypertension.

Potential limitations of the present study warrant discussion. First, lipid levels remained predictive of incident hypertension in analyses restricted to women free of baseline hypertension, diabetes, treatment for high cholesterol levels, and obesity. Second, we relied on a single baseline measurement of plasma lipid levels that is uncorrected for regression dilution, therefore potentially underestimating the strength of the true long-term association with incident hypertension.⁴⁰ Because this biological variability is likely to be random with respect to hypertension, we likely underestimated the true association between lipid levels and risk of hypertension. Third, the stability of our measurements of plasma lipids stored at −140°C in liquid nitrogen–chilled freezers since 1993 has not been directly assessed. However, studies suggest that a single measurement of lipid levels is reasonably stable over long periods of follow-up.⁴¹ Next, incident hypertension was based on self-reported BP, treatment, and/or physician diagnosis. Because of the potential for random measurement error in lipids, BP, hypertension, and selected confounders, we have likely biased our RRs toward the null and underestimated the true effect of lipid levels on developing hypertension in our study. To address this potential limitation, we performed a separate study in which an 86% validation rate for self-reported hypertension was observed, consistent with other studies.^42,43 Sensitivity analyses considering various definitions of hypertension also yielded consistent RRs for increasing levels of each lipid variable. With 4593 cases of hypertension during follow-up, power was excellent to detect modest RRs and trends across quintiles.

Despite the fact that up to 50 million Americans have hypertension,²³ with its prevalence still on the rise,⁴⁴ surprisingly little is known about the primary prevention of hypertension beyond conventional lifestyle and dietary factors. These data from the Women’s Health Study thus represent potentially important clinical information for the use of screening lipid levels to identify dyslipidemic middle-aged and older women at risk for development of hypertension. The relationship of the individual lipid levels was similar in magnitude and direction to that for CVD. Further studies in other populations susceptible to development of hypertension are needed to clarify the potential role of lipid levels in the development of hypertension in a primary prevention setting and to better understand whether this association is independent of or influenced by other unmeasured clinical factors.

Correspondence: Howard D. Sesso, ScD, MPH, Division of Preventive Medicine, Brigham & Women’s Hospital, 900 Commonwealth Ave E, Boston, MA 02215-1204 (hsesso@hsph.harvard.edu)

Accepted for Publication: June 6, 2005.

Financial Disclosure: None.

Funding/Support: This study was supported by research grants CA-47988 and HL-43851 from the National Institutes of Health, Bethesda, Md, and grants from the Doris Duke Charitable Foundation, New York, NY, and the Donald W. Reynolds Foundation, Las Vegas, Nev.

Acknowledgment: We thank the entire staff of the Women’s Health Study, under the leadership of David Gordon, for their crucial contributions, and the 39 876 dedicated and committed participants of the Women’s Health Study.