Longitudinal Hemodynamic Correlates of and Sex Differences in the Evolution of Blood Pressure Across the Adult Lifespan: The Framingham Heart Study

Abstract

Background

Systolic blood pressure increases with age after midlife, particularly in women, and contributes to development of wide pulse pressure hypertension in middle‐aged and older adults. Relative contributions of aortic stiffness and premature wave reflection to increases in pulse pressure remain controversial.

Methods and Results

We evaluated visit‐specific values and change in key correlates of pulse pressure, aortic characteristic impedance, forward and backward wave amplitude, and global reflection coefficient, at 3 sequential examinations of the Framingham Generation 3 (N=4082), Omni‐2 (N=410), and New Offspring Spouse (N=103) cohorts (53% women). Data were analyzed using repeated‐measures linear mixed models adjusted for age, sex, and risk factor exposures. Pulse pressure increased markedly with age after midlife (age and age‐squared terms, P<0.0001), particularly in women (age slope 3.1±0.2 mm Hg/decade higher in women, P<0.0001). In sex‐specific models, change in pulse pressure was closely related (all P<0.0001) to baseline (6.7±0.2 and 7.3±0.2 mm Hg/SD in men and women, respectively) and change (11.8±0.1 and 11.7±0.1 mm Hg/SD) in forward wave amplitude, whereas relations with baseline (2.1±0.15 and 2.0±0.14 mm Hg/SD) and change (4.0±0.13 and 3.4±0.11 mm Hg/SD) in global reflection coefficient were weaker. Global reflection coefficient fell as aortic characteristic impedance increased (P<0.0001), consistent with the hypothesis that impedance matching reduces relative wave reflection in the arterial system.

Conclusions

Proximal aortic stiffening, as assessed by higher aortic characteristic impedance and larger forward wave amplitude, is strongly associated with longitudinal increase in pulse pressure, especially in women, whereas wave reflection has more modest relations.

Nonstandard Abbreviations and Acronyms

CFPWV

carotid‐femoral pulse wave velocity

FWA

forward wave amplitude

GRC

global reflection coefficient

Z_c

characteristic impedance of the aorta

Clinical Perspective.

What Is New?

• Relative contributions of aortic stiffness and premature wave reflection to increases in pulse pressure remain controversial.

• We evaluated visit‐specific values and change in key correlates of pulse pressure, aortic characteristic impedance, forward and backward wave amplitude, and global reflection coefficient at 3 sequential examinations of Framingham Generation 3, Omni‐2, and New Offspring Spouse cohorts.

• Development of wide pulse pressure is strongly associated with proximal aortic stiffening, assessed as higher characteristic impedance of the aorta and larger forward wave amplitude, especially in women, whereas wave reflection has a modest relation.

What Are the Clinical Implications?

• Wide pulse pressure hypertension is highly prevalent in Westernized societies and is often inadequately controlled by conventional treatments.

• Failure to control hypertension may represent failure to address underlying pathophysiology of blood pressure elevation.

• Additional research into interventions that prevent or reverse proximal aortic stiffening is required to address this unmet need.

Systolic blood pressure increases rapidly after midlife in Westernized societies and is associated with increased risk for various adverse health outcomes. ¹ , ² Although pulse pressure, the difference between systolic and diastolic blood pressure, is often ignored in hypertension evaluation and treatment guidelines, marked increases in systolic and pulse pressure after midlife, particularly in women, have created an epidemic of often suboptimally controlled predominant or isolated systolic hypertension, which can be succinctly characterized as wide pulse pressure hypertension. ³ Failure to control wide pulse pressure hypertension may result from failure to recognize the underlying pathophysiology. ⁴ Many hypertension drugs target peripheral resistance and mean pressure and may be less effective for reducing wide pulse pressure and treating wide pulse pressure hypertension. Therefore, a quantitative understanding of the pathogenesis of the rise in systolic blood pressure starting in midlife is essential.

In addition, recent reports have shown that projected values for change in blood pressure and various related hemodynamic measures based on cross‐sectional associations can differ from longitudinal assessments of change. ⁵ , ⁶ Discrepancies arise in part from marked nonlinearity of age relations of stiffness measures, which necessitates multiple sequential assessments in order to accurately characterize change. Finally, several studies have shown that systolic and pulse pressure increase more rapidly with age in women after midlife, ⁵ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ possibly because of sex differences in modulation of aortic diameter with advancing age. ¹² , ¹³ However, many of the foregoing observations on sex‐related differences in age relations of systolic and pulse pressure were based on cross‐sectional studies or limited numbers of longitudinal observations, which may not adequately characterize the complex, nonlinear patterns of change in blood pressure and related hemodynamic measures over time.

We hypothesized that sex differences in diameter‐dependent measures of aortic stiffness, such as characteristic impedance of the aorta (Z_c), contribute to the sex differences in age relations and change over time in systolic and pulse pressure. To assess this hypothesis, we perform detailed hemodynamic assessments at 3 points in time over a 14‐year interval in order to assess nonlinearity of and sex differences in change in systolic blood pressure, pulse pressure and aortic stiffness measures in the deeply phenotyped Framingham Heart Study Generation 3, Omni‐2, and New Offspring Spouse cohorts.

Methods

The procedure for requesting data from the Framingham Heart Study can be found at https://framinghamheartstudy.org/.

Participants

The sample for the present study was drawn from the Framingham Third Generation cohort (N=4082), the racially and ethnically diverse Omni‐2 cohort (N=410), and the New Offspring Spouse cohort (N=103). ¹⁴ Participants who attended sequential examination visits 1, 2, or 3 were eligible for inclusion in the present analyses. Of the 4595, 3778, and 3451 eligible individuals who attended the hemodynamic evaluation at examination 1, 2, and 3, respectively, participants were sequentially excluded for missing or incomplete covariate (N=150, 128, and 161) or tonometry (N=184, 91, and 46) data. All protocols were approved by the Boston University Medical Center Institutional Review Board, and participants provided written informed consent.

An expanded version of the Methods is presented in Data S1 under the headings Noninvasive Hemodynamics Assessment and Clinical Evaluation and Definitions. In brief, auscultatory blood pressure was assessed in supine participants after 5 minutes of rest, followed by tonometry of the brachial, radial, femoral, and carotid arteries. Next, limited echocardiography was used to assess left ventricular outflow tract area and flow velocity, which were used to compute the time‐resolved volumetric flow rate into the aorta. Carotid and femoral artery tonometry and the difference in body surface transit distance measurements from suprasternal notch to femoral and carotid sites were used to compute carotid‐femoral pulse wave velocity (CFPWV). Brachial waveforms were calibrated to cuff systolic and diastolic pressure; all other waveforms were calibrated by using integrated mean and diastolic brachial pressure. Calibrated carotid pressure and aortic flow waveforms were used to compute Z_c and perform wave separation analysis. The global reflection coefficient (GRC) was defined as backward wave amplitude divided by forward wave amplitude (FWA).

Statistical Analysis

Demographic and risk factor variables were summarized and tabulated. Variables with skewed distributions, including body mass index, triglyceride levels, fasting blood glucose, and the total‐to‐high‐density lipoprotein cholesterol ratio were transformed by using the natural logarithm, whereas CFPWV was transformed by using the negative inverse value=−1000/CFPWV.

Cross‐sectional relations and longitudinal change in hemodynamic measures were examined using repeated measures linear mixed effects models. Key stiffness measures for primary analyses were negative inverse CFPWV, FWA, and Z_c. We examined 2 types of models, the repeated measures model and the change model. For both types of models, we fitted linear mixed‐effects models using the restricted maximum likelihood method to accommodate data that could be missing at random. We fitted models with an unspecified covariance matrix, which allows variances and correlations of the response variables to differ across visits.

The repeated‐measures model examined relations of the dependent variables with concurrent age and risk factor levels across the 3 repeated measurements. All evaluations at each examination cycle with a valid assessment of CFPWV, Z_c, and covariates were included in these analyses, regardless of whether a participant contributed 1, 2, or 3 observations. Required covariates that were examined in base risk factor models included age (centered at the median value of 47 years), centered age squared, sex, age*sex, age squared*sex, heart rate, height, body mass index, fasting blood glucose, total/high‐density lipoprotein cholesterol cholesterol ratio, triglyceride levels, status of antihypertensive, lipid‐lowering or diabetes medication usage, and status of current smoking or prior cardiovascular disease (16 variables). All models also included a dummy variable for cohort (Generation 3, Omni‐2, or New Offspring Spouse).

The change model directly examined change in hemodynamic measures in participants with complete hemodynamic data at 2 sequential examinations (1–2, 2–3, or both). The dependent variable was the observed value at the subsequent paired visit (examination 2 or 3) adjusted for the value at the initial visit (examination 1 or 2, respectively). Risk factor variables were added to the change model in pairs, consisting of the baseline value at the first examination (1 or 2) and the change (delta) between the 2 exams (1‐2 or 2‐3). All models included terms for centered age and centered age squared, sex, cohort identifier, visit number, and interaction terms for age*sex and age squared*sex. Models were first fitted with the full set of exposure variables defined for the given model. Variables were then eliminated manually in 2 steps. On the first step, all variables with an initial P>0.5 based on a type III F test were removed and models were refitted. On the second step, variables with a new P>0.2 based on a type III F test were removed and models were refitted once again, giving the final model. For the change model, the baseline variable for each baseline‐delta pair of independent variables was retained if the delta variable was retained in the model based on the foregoing thresholds. For models that examined relations of stiffness measures with blood pressure measures, a base model for each blood pressure component was fitted as described above. Key stiffness measures were then entered as a group into the foregoing base models. Our primary analyses included 3 dependent variables and the 16 candidate covariates noted previously. Therefore, an adjusted P value <0.05/(3*16)=0.001 was considered to be significant in those primary models.

Results

Sample Characteristics

Characteristics of the sample at each of the 3 examinations are presented in Table 1, and baseline characteristics in participants according to whether they contributed data at 1, 2, or 3 visits are summarized in Table S1. The cohort was relatively young at examination 1, although with a broad age range, and progressed into midlife, on average, by examination 3, which was approximately 14 years later (Table 1). Participants with missing data were slightly older and heavier with higher lipid levels, higher prevalence of cardiovascular disease, and higher usage of lipid‐lowering, hypertension, and diabetes medications. They were more likely to be currently smoking and had modestly higher systolic, diastolic, and pulse pressures and lower Zc and CFPWV (Table S1). However, patterns of change across the 3 visits were similar when examining only participants who attended all 3 visits (data not shown). The average mean arterial pressure (MAP) of attendees fell between examinations 1 and 2 and then increased by examination 3, noting, however, that the numbers of participants varied by examination. In contrast, brachial and central pulse pressure increased progressively across examinations 1, 2, and 3. CFPWV increased minimally between examinations 1 and 2, when MAP fell, but increased by approximately 10% between examinations 1 and 3. Use of lipid‐lowering, hypertension, and diabetes medications increased while the prevalence of smoking decreased across examinations.

Table 1.

Clinical Characteristics of the Sample by Visit

Variable	Visit 1	Visit 2	Visit 3
Total N (% women)	4261 (53.2)	3559 (53.3)	3244 (54.3)
Age, y	40.7±9.9	46.9±9.6	54.4±9.2
Generation 3 participants, N (%)	3813 (89.5)	3194 (89.7)	2927 (90.2)
Omni‐2 participants, N (%)	365 (8.6)	306 (8.6)	268 (8.3)
New Offspring Spouse participants, N (%)	83 (1.9)	59 (1.7)	49 (1.5)
Height, m	1.70±0.09	1.70±0.09	1.69±0.09
Weight, kg	77±18	80±19	82±19
Body mass index, kg/m2	26 (23, 29)	27 (24, 31)	28 (24, 32)
Total‐to‐high‐density lipoprotein cholesterol cholesterol, ratio	3.5 (2.8, 4.5)	3.1 (2.5, 4.0)	3.2 (2.6, 4.1)
Triglycerides, mg/dL	90 (64, 136)	93 (67, 131)	92 (67, 134)
Fasting blood glucose, mg/dL	93 (87, 99)	93 (88, 100)	96 (90, 103)
Prevalent cardiovascular disease, N (%)	33 (0.8)	49 (1.4)	80 (2.5)
Lipid‐lowering medications, N (%)	314 (7.4)	574 (16.1)	761 (23.5)
Hypertension medications, N (%)	384 (9.0)	606 (17.0)	770 (23.7)
Diabetes medications, N (%)	88 (2.1)	122 (3.4)	191 (5.9)
Current smoking, N (%)	608 (14.3)	335 (9.4)	210 (6.5)
Heart rate, beats/min	62±10	63±10	59±10
Systolic blood pressure, mm Hg	121±14	121±15	127±16
Diastolic blood pressure, mm Hg	67±9	63±9	67±9
Mean arterial pressure, mm Hg	90±11	87±11	92±11
Brachial pulse pressure, mm Hg	54±11	58±12	60±14
Central pulse pressure, mm Hg	52±13	54±14	60±16
Forward pressure wave amplitude, mm Hg	46±11	47±11	48±12
Global reflection coefficient, %	34±6	36±6	38±7
Augmentation index, %	8.4±13.5	9.9±12.5	15.9±11.8
Backward wave transit time, ms	139±23	137±24	134±28
Characteristic impedance, dyne·s/cm5	179±50	197±61	211±74
Peripheral resistance, dyne·s/cm5	1703±354	1752±420	2048±527
Negative inverse carotid‐femoral pulse wave velocity, ms/m	−147±27	−145±26	−131±26

Values are mean±SD, 50th percentile (25th, 75th), or N (%).

Age Relations of Brachial Blood Pressure Across 3 Visits

Figure 1 presents blood pressure and various key hemodynamic measures by age and examination number. Figures S1 through S3 present change between examinations 1 and 2 (panel A) and 2 and 3 (panel B) in men and women. On average, systolic blood pressure changed minimally across the full age range in the 6‐year interval between examinations 1 and 2 and moderately in the 8‐year interval between examinations 2 and 3, particularly in middle‐aged and older women (Table 1; Figure S1). The age relation of systolic blood pressure was relatively flat until midlife and increased thereafter, particularly in women (Figure 1). Mean and diastolic pressures increased with advancing age at each examination through midlife and then plateaued (mean) or fell (diastolic) (see dots in Figure 1). In an unadjusted analysis, MAP fell between examinations 1 and 2 and increased to above the initial baseline between examinations 2 and 3 (line segments in Figure 1 and change plots in Figure S1A and S1B). However, after accounting for explainable linear and quadratic effects of age, sex differences, and the other risk factors included in the repeated measures model, mean and diastolic pressures fell by approximately 6 to 7 mm Hg between examinations 1 and 2 and then increased by 3 and 4 mm Hg between examinations 2 and 3 (see visit coefficients in Table S2), resulting in net reductions of 2.6±0.3 and 3.4±0.2 mm Hg for mean and diastolic pressure, respectively, between examinations 1 and 3 (Table S2). Pulse pressure generally increased between examinations 1 and 2 across the full age range (Figure 1; Figure S1A). Within each examination, pulse pressure fell with age to a nadir in midlife and then increased thereafter, particularly in women (dots in Figure 1). In a model that included age, age squared, sex, and their interactions, cohort, and visit, the average pulse pressure nadir occurred at 37 and 44 years of age in women and men, respectively. In models that further accounted for vascular risk factors, systolic blood pressure increased by approximately 4 mm Hg/decade of age in men and 7 mm Hg/decade in women after the overall median age (47 years) of our sample because of a sex difference in the pulse pressure versus age slope (3.1±0.2 mm Hg/decade higher in women, P<0.001, Table S2).

Figure 1. Age trends and change in key hemodynamic variables across 3 examinations spanning 14 years.

Lightly colored symbols are examination 1, and lightly colored line segments represent examination 1–2 change. Darkly colored symbols and line segments represent examinations 2 and 3 and 2–3 change. Markedly nonlinear age relations and sex differences are apparent for several variables. See text for details. BWTT indicates backward wave transit time; CFPWV, carotid‐femoral pulse wave velocity; GRC, global reflection coefficient; P_b, backward wave amplitude; P_f, forward wave amplitude; and Z_c, characteristic impedance of the aorta.

Age Relations and Change in Stiffness Measures Across 3 Visits

Greater increases in systolic and pulse pressure with age after midlife in women were accompanied by greater increases in FWA and Z_c with age (Table 2). In contrast, augmentation index and GRC increased through midlife and then plateaued or fell thereafter (Figure 1; Table S3). In the repeated measures model, CFPWV increased with advancing age at a modestly higher rate in women than men, although values remained lower in women across the full age range (Table 2, Figure 1). In the multivariable change model, average change in CFPWV was lower in women (Table 3) despite the slightly steeper age slope in women in the repeated measures model (sex*age term, Table 2). In contrast, FWA and Z_c increased at markedly higher rates in women after midlife (Figure 1, Figures S2 and S3, Tables 2 and 3).

Table 2.

Repeated Measures Analysis of Relations of Stiffness Measures With Demographics and Standard Vascular Risk Factors Across 3 Sequential Examinations

	niCFPWV (ms/m)		Forward wave amplitude (mm Hg)		Characteristic impedance (dyne·sec/cm5)
Variable	B±SE	P value	B±SE	P value	B±SE	P value
Intercept	−282±7	<0.0001	−11.9±4.8	0.013	335±26	<0.0001
Age	1.20±0.03	<0.0001	−0.088±0.017	<0.0001	−0.125±0.088	0.16
Age squared	0.005±0.001	<0.0001	0.020±0.001	<0.0001	0.076±0.005	<0.0001
Sex (women)	−9.68±0.49	<0.0001	−0.397±0.385	0.30	7.75±1.97	<0.0001
Sex*age	0.147±0.031	<0.0001	0.187±0.018	<0.0001	0.914±0.098	<0.0001
Sex*age squared	‐	‐	−0.003±0.001	0.016	0.015±0.006	0.008
Visit (Exam 3 is reference)
Exam 1	2.82±0.46	<0.0001	−1.49±0.26	<0.0001	−26.4±1.5	<0.0001
Exam 2	−1.92±0.35	<0.0001	1.66±0.23	<0.0001	−2.81±1.33	0.035
Height	‐	‐	4.77±1.77	0.007	−118±9	<0.0001
Mean arterial pressure	0.762±0.017	<0.0001	0.396±0.010	<0.0001	1.33±0.05	<0.0001
Body mass index†	5.05±1.25	<0.0001	‐	‐	−56.9±3.5	<0.0001
Fasting blood glucose†	6.28±1.34	<0.0001	3.28±0.81	<0.0001	15.1±4.4	0.0006
Total/high‐density lipoprotein cholesterol ratio†	1.21±0.79	0.13	−1.06±0.36	0.0032	‐	‐
Triglycerides†	1.52±0.45	0.0007	‐	‐	2.26±1.11	0.042
Heart rate	0.363±0.019	<0.0001	−0.015±0.011	0.18	0.726±0.059	<0.0001
Medication (Rx) usage
Lipid Rx	‐	‐	−0.466±0.316	0.14	‐	‐
Hypertension Rx	1.39±0.50	0.005	1.24±0.30	<0.0001	‐	‐
Diabetes Rx	4.53±0.96	<0.0001	1.94±0.60	0.0012	14.2±3.3	<0.0001
Current smoking	−1.40±0.57	0.01	0.824±0.324	0.011	‐	‐

^*Multiplication.

^† Natural log transformed. niCFPWV indicates negative inverse carotid‐femoral pulse wave velocity; and Rx, treatment. B represents the regression slope in native units of each dependent variable for a 1‐unit difference in each independent variable; therefore, for log‐transformed independent variables, B/10 is the difference in the dependent variable expected for a 10% higher value of the independent variable. P values were obtained by using t tests with 4504 degrees of freedom.

Table 3.

Longitudinal Change in Stiffness Measures Between Examinations 1 and 2 and Between Examinations 2 and 3

	niCFPWV (ms/m)		Forward wave amplitude (mm Hg)		Characteristic impedance (dyne·sec/cm5)
Variable	B±SE	P value	B±SE	P value	B±SE	P value
Intercept	−126±9	<0.0001	−24.0±5.3	<0.0001	84.8±33.7	0.012
Age	0.393±0.024	<0.0001	0.009±0.020	0.66	0.293±0.111	0.008
Age squared	0.010±0.001	<0.0001	0.015±0.001	<0.0001	0.052±0.007	<0.0001
Sex (women)	−1.82±0.38	<0.0001	0.928±0.316	0.0034	6.47±2.09	0.0020
Sex*age	‐	‐	0.102±0.025	<0.0001	0.576±0.137	<0.0001
Sex*age squared	‐	‐	−0.003±0.002	0.038	0.016±0.009	0.073
Exam 1–2 vs. 2–3	−6.88±0.48	<0.0001	2.37±0.31	<0.0001	9.51±1.86	<0.0001
Height	‐	‐	‐	‐	−73.3±9.6	<0.0001
Baseline dependent variable	0.632±0.010	<0.0001	0.446±0.012	<0.0001	0.537±0.013	<0.0001
Mean arterial pressure	0.358±0.022	<0.0001	0.271±0.015	<0.0001	0.864±0.078	<0.0001
Delta mean arterial pressure	0.748±0.021	<0.0001	0.398±0.014	<0.0001	1.51±0.08	<0.0001
Body mass index†	0.461±1.079	0.67	−2.63±0.75	0.0004	−45.4±4.2	<0.0001
Delta body mass index†	7.26±2.44	0.0030	‐	‐	−17.9±8.9	0.045
Fasting blood glucose†	5.30±1.85	0.0042	7.73±1.20	<0.0001	34.3±6.8	<0.0001
Delta fasting blood glucose†	3.13±1.70	0.067	2.00±1.11	0.073	12.4±6.4	0.052
Total/HDL cholesterol ratio†	‐	‐	0.284±0.453	0.53	4.07±1.39	0.0035
Delta total/HDL†	‐	‐	−1.38±0.62	0.025	‐	‐
Triglycerides†	2.10±0.41	<0.0001	‐	‐	‐	‐
Delta triglycerides†	1.85±0.50	0.0003	‐	‐	‐	‐
Heart rate	0.236±0.021	<0.0001	−0.063±0.013	<0.0001	0.368±0.078	<0.0001
Delta heart rate	0.306±0.026	<0.0001	‐	‐	0.725±0.098	<0.0001
Prior cardiovascular disease	‐	‐	−3.54±1.30	0.007	−17.3±7.4	0.019
Medication (Rx) usage
Prior lipid Rx	‐	‐	0.041±0.426	0.92	−0.344±2.428	0.89
Delta lipid Rx	‐	‐	−1.02±0.43	0.018	−4.54±2.27	0.046
Prior hypertension Rx	0.848±0.568	0.14	1.03±0.42	0.014	4.75±2.27	0.036
Delta hypertension Rx	‐	‐	1.05±0.44	0.018	‐	‐
Prior diabetes Rx	1.53±1.34	0.25	1.97±0.90	0.029	11.7±5.2	0.024
Delta diabetes Rx	1.93±1.23	0.12	‐	‐	‐	‐
Current smoking	−1.33±0.63	0.037	0.656±0.370	0.076	−3.36±2.05	0.10
New smoking	−2.37±0.91	0.009	‐	‐	‐	‐

^*Multiplication.

^† Natural log transformed. See Table 2 for abbreviations and notes on interpretation of B. P values were obtained by using t tests with 3492 degrees of freedom. HDL indicates high‐density lipoprotein; niCFPWV, negative inverse carotid‐femoral pulse wave velocity; and Rx, treatment.

Change in Physiological Components of Blood Pressure Between Visits

The hemodynamic correlates of change in mean and pulsatile components of blood pressure were next examined in sex‐specific models for change in pulse pressure (Tables S4 and Table 4) and MAP (Table S5 and Table 5). Relations of changes in pulse pressure and MAP with age, sex and standard risk factors are presented in Tables S4 and S5, respectively. For pulse pressure, hemodynamic measures added to the model as a group included FWA, GRC, and backward wave transit time to assess effects of forward and reflected waves and CFPWV to assess additional effects of aortic call stiffness. When hemodynamic variables were added to the risk factor‐adjusted model for change in pulse pressure, residual variances were reduced substantially, particularly in women (Table 4). Except for backward wave transit time, each of the individual hemodynamic measures was related to change in pulse pressure (Table 4). In particular, baseline values and change in FWA were associated with marked changes in pulse pressure, whereas relations for GRC and CFPWV were more modest. Similarly, when hemodynamic measures were added to the models for change in MAPs, residual variances were reduced markedly in men and women (Table 5). Backward wave transit time was weakly related to change in MAP, whereas the other measures were strongly related to change in MAP (Table 5).

Table 4.

Relations of Individual Hemodynamic Components With Longitudinal Change in Brachial Pulse Pressure in Men and Women

Variable	Men		Women
	B±SE	P value	B±SE	P value
Forward wave amplitude	6.74±0.22	<0.0001	7.33±0.22	<0.0001
Delta forward wave	11.8±0.1	<0.0001	11.7±0.1	<0.0001
Global reflection coefficient	2.10±0.15	<0.0001	2.00±0.14	<0.0001
Delta global reflection coefficient	3.97±0.13	<0.0001	3.40±0.11	<0.0001
CFPWV	0.691±0.137	<0.0001	0.440±0.136	0.0012
Delta CFPWV	0.811±0.120	<0.0001	0.644±0.108	<0.0001
Backward wave transit time	0.259±0.109	0.018	−0.309±0.113	0.006
Delta backward wave transit time	−0.016±0.103	0.88	−0.292±0.093	0.0018

	Residual variance		BIC	Residual variance		BIC
	Visit 2–1	Visit 3–2		Visit 2–1	Visit 3–2
Base	89	102	21 233	96	109	24 484
Add covariates	57	68	20 156	74	80	23 688
Add vascular measures	37	43	18 909	40	37	21 468

BIC indicates Bayesian information criterion; and CFPWV, carotid femoral pulse wave velocity, transformed using negative inverse. Base model included visit and cohort indicator variables only. Risk factor covariates were added next and are presented in Table S4. Vascular measures were then added as a group of baseline/delta pairs. B represents the regression slope in native units of each dependent variable for a 1‐SD difference in each independent variable. P values were obtained by using t tests with 1633 degrees of freedom in men and 1857 degrees of freedom in women.

Table 5.

Relations of Individual Hemodynamic Components With Longitudinal Change in Mean Arterial Pressure in Men and Women

	Men		Women
Variable	B±SE	P value	B±SE	P value
Peripheral resistance	1.31±0.16	<0.0001	1.31±0.14	<0.0001
Delta peripheral resistance	3.28±0.14	<0.0001	3.25±0.13	<0.0001
Forward wave amplitude	1.34±0.16	<0.0001	2.08±0.15	<0.0001
Delta forward wave	2.03±0.15	<0.0001	4.10±0.13	<0.0001
Global reflection coefficient	1.84±0.20	<0.0001	1.68±0.17	<0.0001
Delta global reflection coefficient	2.16±0.17	<0.0001	2.47±0.15	<0.0001
CFPWV	1.24±0.18	<0.0001	1.41±0.17	<0.0001
Delta CFPWV	2.59±0.15	<0.0001	2.40±0.14	<0.0001
Backward wave transit time	−0.028±0.148	0.85	−0.230±0.145	0.11
Delta backward wave transit time	−0.151±0.139	0.28	−0.431±0.125	0.0005

	Residual variance		BIC	Residual variance		BIC
	Visit 2–1	Visit 3–2		Visit 2–1	Visit 3–2
Base	126	164	22 415	118	176	25 696
Add covariates	96	109	21 623	83	96	24 295
Add vascular measures	23	22	17 303	19	21	19 471

BIC indicates Bayesian information criterion; and CFPWV, carotid femoral pulse wave velocity, transformed using negative inverse. Base model includes visit and cohort indicator variables only. Risk factor covariates were added next and are presented in Table S5. Vascular measures were then added as a group of baseline/delta pairs. B represents the regression slope in native units of each dependent variable for a 1‐SD difference in each independent variable. P values were obtained by using t tests with 1633 degrees of freedom in men and 1857 degrees of freedom in women.

Age Relations and Change in Relative Wave Reflection Between Visits

Table 6 examines relations of risk factors and hemodynamic measures with change in relative wave reflection as assessed by the GRC. GRC was similar in men and women, increased modestly with age through midlife, when pulse pressure, FWA, and Z_c were falling, and then plateaued in older groups, when pulse pressure increased markedly (Figure 1, Table S3). When Z_c was added to a model for change in GRC, the coefficients for initial value and change in Z_c were negative (Table 6, Figure 2).

Table 6.

Longitudinal Change in Global Reflection Coefficient Between Examinations 1 and 2 and Between Examinations 2 and 3

Variable	B±SE	P value
Intercept	69.2±3.3	<0.0001
Age	0.0673±0.0079	<0.0001
Age‐squared	−0.0017±0.0005	0.0002
Sex	0.505±0.179	0.0048
Exam 1–2 vs. 2–3	0.0397±0.1651	0.81
Height	−16.9±1.0	<0.0001
Baseline global reflection coefficient	0.324±0.013	<0.0001
Mean arterial pressure	0.206±0.008	<0.0001
Delta mean arterial pressure	0.212±0.007	<0.0001
Body mass index*	−5.16±0.41	<0.0001
Delta body mass index*	−2.59±0.83	0.0019
Fasting blood glucose*	0.132±0.582	0.82
Triglycerides*	0.371±0.147	0.012
Delta triglycerides*	0.652±0.172	0.0001
Heart rate	−0.217±0.009	<0.0001
Delta heart rate	−0.306±0.009	<0.0001
Medication (Rx) usage
Prior lipid Rx	−0.173±0.224	0.44
Delta lipid Rx	0.085±0.206	0.68
Prior hypertension Rx	−0.586±0.218	0.007
Delta hypertension Rx	−0.765±0.226	0.0007
Prior diabetes Rx	−0.731±0.468	0.12
Current smoking	0.978±0.226	<0.0001
New smoking	0.463±0.308	0.13
Characteristic impedance (Zc)	−0.0274±0.0015	<0.0001
Delta Zc	−0.0376±0.0011	<0.0001

^*Natural log transformed. Rx indicates treatment; and Z_c, characteristic impedance. Global reflection coefficient was scaled as a percentage. See Table 2 for notes on interpretation of B. P values were obtained by using t tests with 3492 degrees of freedom.

Figure 2. Global reflection coefficient relations with characteristic impedance of the aorta.

Residuals from the separate risk factor models for global reflection coefficient and characteristic impedance are plotted by visit. As the aorta stiffens, characteristic impedance increases and global reflection coefficient falls, consistent with the hypothesis that stiffening of the proximal aorta flattens the impedance gradient in the arterial system, reduces global wave reflection, and therefore increases transmission of pulsatile energy into the periphery. Residuals were created by fitting separated models for characteristic impedance and global reflection coefficient using the full list of candidate covariates in both models.

Discussion

To our knowledge, our study is the first to report cross‐sectional associations and change in blood pressure components and detailed hemodynamic phenotypes across 3 sequential examinations spanning 14 years in an initially young‐to‐middle aged cohort, albeit with a broad age range. Our study revealed significant sex differences in age relations and change over time in blood pressure components and various key aortic stiffness‐related measures, including systolic and pulse pressure, FWA, and aortic Z_c. Measures of pressure pulsatility were lower in younger women as compared with younger men, whereas values increased more rapidly in women after midlife, resulting in levels that equaled or exceeded those in men. In contrast, CFPWV was lower in women across the full age range, with a modestly steeper age slope (Table 2) but slower rate of increase between exams in women (Table 3). Despite lower CFPWV, women had earlier return of the backward wave and greater central pressure augmentation. In contrast, relative wave reflection, as assessed by the GRC, was comparable in women and men, indicating that timing rather than magnitude of reflection gave rise to higher augmentation in women. Nevertheless, marked increases in pulse pressure with age and over time in women and men were primarily attributable to greater FWA, with a substantially smaller contribution attributable to relative amplitude of the backward pressure wave.

We observed that global wave reflection increased in young adults, when pulse pressure and Z_c were falling, and fell after midlife, when pulse pressure and Z_c were increasing. The foregoing pattern of change (inverse relation between Z_c and global wave reflection) is consistent with the hypothesis that stiffening of the proximal aorta after midlife increases Z_c and flattens the normally steep impedance gradient between proximal aorta, first‐generation branch vessels, and the periphery, resulting in a loss of protective wave reflection. Wave reflection at the origin of first‐generation branch vessels and beyond normally limits transmission of pulsatile energy into the periphery. ¹⁵ Loss of this protective impedance mismatch and wave reflection exposes the peripheral circulation to greater transmission of higher levels of pulsatile power, which can cause diffuse microvascular remodeling, damage, and dysfunction, with potential adverse implications for organ systems throughout the body. ¹⁶

We observed important sex differences in cross‐sectional relations and longitudinal change in blood pressure components and stiffness measures. Measures related to pressure pulsatility (pulse pressure, FWA) started from a lower value in young women and increased more with advancing age and over time in women. In contrast, average values for CFPWV were lower in women across the full age range and longitudinal change in CFPWV was slightly lower in women as compared with men (Table 3, Figure 1; Figures S1A, S1B and S2A, S2B). Higher Z_c in women is attributable, in part, to smaller body size; however, sex differences persisted in models that adjusted for height and body mass index. Furthermore, measures of pressure pulsatility that result from higher Z_c should not be sensitive to nominal (lean) body size, and yet values for pulse pressure and FWA increased much faster in women after midlife. The combination of a differing rate of increase in pressure pulsatility and Z_c as compared with CFPWV suggests that modulation of aortic diameter may differ between men and women. ⁴ , ¹² , ¹³ Pulse wave velocity is relatively insensitive, whereas Z_c is extremely sensitive, to aortic diameter. ¹⁷ Aortic remodeling, possibly in response to weight gain during early adulthood, may account for the drop in Z_c, FWA, and pulse pressure in young adults through midlife. However, the ability of women to invoke this potentially compensatory remodeling response appears to be limited from midlife onward. ¹² Aortic diameter is a highly heritable trait and women are known to have a smaller proximal aortic diameter adjusted for body size across the full lifespan, ¹² suggesting the women also have a smaller elastic fiber mass, which is established early in life (through toddler stage), before silencing of the elastic fiber gene program. ¹⁸ , ¹⁹ , ²⁰ The resulting reduction in elastic fiber mass in women may constrain adaptive remodeling capacity of the aorta in response to wall stiffening or weight gain.

The remodeling response of the aorta in women to physiological stresses, such as weight gain or aortic wall stiffening, may be further exacerbated by a relative deficiency in natriuretic peptides following menopause. Before midlife, women have lower aortic stiffness and higher natriuretic peptide levels compared with men. ²¹ After midlife, natriuretic peptide levels fall, ²¹ and aortic stiffness increases more in women than men, resulting in much higher aortic stiffness and pulse pressure in women than men after midlife. ¹³ , ¹⁷ , ²² Natriuretic peptides are known to have favorable effects on aortic structure and function. Prior studies in patients with uncomplicated hypertension and heart failure have shown that treatments that increase natriuretic peptide levels can reduce pulse pressure and Z_c of the aorta, particularly in women. ²³ , ²⁴ Additional research is needed to determine whether such treatments should be preferred in older women with a stiffened aorta and wide pulse pressure hypertension, which is a group at high risk for heart failure with preserved left ventricular ejection fraction.

Our results differ in several respects from published observations in other cohorts. As noted previously, we saw minor differences in change in CFPWV in men and women, whereas prior studies have reported more substantial sex differences. For example, in contrast to Lu et al, we found no evidence for a crossover of CFPWV to higher values in women after midlife; this difference may be explained by use in the prior study of brachial‐ankle pulse wave velocity, which includes a considerable contribution from muscular arteries. ⁹ We also did not find evidence of a sex‐specific acceleration of rate of change in CFPWV in our sample. ⁵ The latter difference may relate to our use in statistical models of negative inverse CFPWV, which is highly effective at eliminating the skewed, heteroskedastic distribution of CFPWV values in our sample. However, we plotted CFPWV values in native units and found relatively parallel CFPWV age relations (Figure 1) and similar change (Figure S2) in women and men. Our CFPWV findings were similar to those reported by Scuteri et al ¹⁰ and AlGhatrif et al, ¹¹ although the higher rate of change in CFPWV in men reported by AlGhatrif et al was not evident in our study, whereas the modestly higher rate of increase in CFPWV across multiple visits in women was consistent with Scuteri et al.

We found strongly nonlinear age relations for pulse pressure, FWA, and Z_c in men and women, with values falling with age before midlife and rising thereafter. In the Asklepios cohort, with a more constrained age range, FWA and Z_c fell with age and time, and FWA began to increase in their oldest participants. ⁵ They did not capture the greater acceleration of increase in forward wave with age in women after midlife, although that difference was likely attributable to a lack of older participants in their study as compared with our sample. We saw an increase in systolic pressure after midlife in men and women that attenuated somewhat in the oldest groups, but did not see a drop in systolic or pulse pressure in men after midlife, as was seen in the Sardinia cohort. ¹⁰ Technical differences in blood pressure acquisition (postprandial and seated in Sardinia versus fasting and supine in our study) may have contributed to differing patterns of change in our studies.

Limitations

There are limitations of our study that merit consideration. Our cohort was overwhelmingly composed of White individuals of European ancestry. As a result, our findings may not be generalizable to other racial or ethnic groups. Our study used a prospective design and, in addition to modeling cross‐sectional age relations of blood pressure and vascular measures, we also examined longitudinal change directly; however, our analyses are nevertheless observational and cannot establish causality. Additional known or unknown factors may contribute to or fully explain apparent cross‐sectional and longitudinal associations. In addition, known limitations of noninvasive blood pressure measurement may have weakened the age relations that we and others have observed using measures derived from noninvasive blood pressure recordings. ²⁵ Strengths of our study include a large cohort of deeply phenotyped individuals spanning a broad age range who were examined up to 3 times over a moderately long interval (14 years) using consistent methods, equipment, and personnel.

Conclusions

Aortic stiffness has emerged as a novel and pervasive risk factor for various conditions, including hypertension, diabetes, obesity, dementia, and major cardiovascular disease events, including stroke, myocardial infarction, and heart failure. ²⁶ , ²⁷ , ²⁸ As we have shown, aortic stiffness increases dramatically with advancing age, which has important public health implications in light of aging of the world population. Research and development efforts over the past half century have reduced the incidence and severity of cardiovascular disease; however, those efforts were largely aimed at reducing the impact of atherosclerosis and ischemic events, with substantial gains attributable to major reductions in cholesterol levels made possible by the introduction of statin medications. Substantially less progress has been made in attempts to define and attenuate biological processes that promote arteriosclerosis. As we and others have shown, arteriosclerosis is only modestly related to atherosclerosis. Furthermore, the direction of that association may favor arteriosclerosis having a permissive or facilitating role in atherogenesis in predisposed individuals. Thus, ongoing interventions that target lipid abnormalities are unlikely to modify the highly unfavorable trends in progression of arteriosclerosis that we have identified. ²⁷ The combination of very high prevalence of abnormal aortic stiffness in older individuals with very high relative risk suggests that the burden of cardiovascular disease will shift to stiffness‐related disease over the next decades as the population ages, creating a public health imperative for additional research into pathways and interventions that can favorably modify arteriosclerosis and associated downstream consequences.

Sources of Funding

This study was supported by National Heart, Lung, and Blood Institute contracts N01‐HC‐25195, HHSN268201500001l, and 75N92019D00031 (R.S.V.), and R01‐DK‐080739 (R.S.V.) and R01‐HL‐107385, 1R01HL126136‐01A1, HL93328, HL142983, HL143227 and HL131532 (R.S.V., G.F.M.), and 1RO1‐HL‐70100, R01HL092577, 2U54HL120163, 1R01AG066010 (E.J.B.). R.S.V. was supported in part by the Evans Medical Foundation and the Jay and Louis Coffman Endowment from the Department of Medicine, Boston University Chobanian and Avedisian School of Medicine.

Disclosures

Dr Mitchell is owner of Cardiovascular Engineering, Inc., a company that designs and manufactures devices that measure vascular stiffness. The company uses these devices in clinical trials that evaluate the effects of diseases and interventions on vascular stiffness. Dr Mitchell also serves as a consultant to and receives grants and honoraria from Novartis, Merck, Bayer, Servier, Philips, and deCODE genetics. The remaining authors have no disclosures to report.

Supporting information

Data S1

Tables S1–S5

Figures S1–S3