Blood Pressure Tracking Over the Adult Life Course: Patterns and Correlates in the Framingham Heart Study

Abstract

The extent to which select vascular risk factors differentially influence blood pressure (BP) is incompletely understood. Thus, we used multilevel modeling to analyze serial BP measurements using 21,732 person-observations obtained on Framingham Heart Study participants (mean age 38 years, 52% women; 4,993 unique individuals) over a 28-year period. We related longitudinal tracking of each BP measure (systolic BP [SBP], diastolic BP [DBP], mean arterial pressure [MAP], and pulse pressure [PP]) to age, sex, body mass index (BMI), smoking, diabetes, total/high-density lipoprotein (HDL) cholesterol ratio, and heart rate. In multivariable-adjusted analyses, we observed that older age, male sex, greater BMI, and higher heart rate were positively associated with increase in all BP measures (p<0.0001). Notably, higher total/HDL cholesterol ratio was associated with greater MAP (p<0.01). Conversely, diabetes and smoking were associated with higher PP (p<0.01). We also observed effect modification by sex: the increase in PP with age and BMI was more pronounced in women compared to men (p<0.0001). All BP measures tracked at higher levels in both men and women with multiple vascular risk factors. Taken together, our longitudinal observations of a large community-based sample demonstrate a greater pulsatile load in women than men with increasing age. We also observed a differential impact of select vascular risk factors on the individual components of BP, underscoring distinct regulation of these measures over the life course.

BACKGROUND

Elevated blood pressure (BP) remains a widely prevalent contributor to cardiovascular risk.^1,2 In the industrialized world, it has been long known that BP steadily increases with aging such that, for much of the last century, a progressively rising BP was thought to be part of the `normal' aging process.^3,4 Evidence now indicates that age-related increases in BP are not obligatory. Indeed, little to no age-related increase in BP is seen in non-acculturated populations such as the Yanomamo and Xingu Indians of Brazil, who live without exposure to vascular risk factors.^5,6 Because differences in the distributions of BP are observed between populations that are genetically similar but geographically separate,^7,8 the primary determinants of age-related changes in BP are likely due to environmental and lifestyle influences operating at least partly through traditional vascular risk factors. Nonetheless, the relationship between risk factors and longitudinal BP tracking over the life course is not well understood. Limited data suggest that the presence of risk factors in childhood may influence BP in young adulthood,⁹ but similar data in middle-aged and older adults are lacking. Such knowledge may be critical for preventing age-related increases in BP and the development of systolic hypertension in older individuals.

Previous work has established that arterial pressure can be divided into 2 main hemodynamic components: a steady state load represented by mean arterial pressure (MAP) and a pulsatile load represented by pulse pressure (PP). Whereas hypertension in younger adults commonly presents with increased diastolic BP (DBP) and MAP, hypertension in adults over age 50 is predominantly characterized by elevations in systolic BP (SBP) and PP, frequently presenting as isolated systolic hypertension.^10–12 Accordingly, the components of BP appear to change with age in a non-linear fashion.^10,13 However, the pathophysiology underlying these non-linear changes in BP remains unknown. Although obesity,^14–16 dyslipidemia,^15–17 diabetes^18,19 have been associated with various forms of hypertension, the extent to which these risk factors contribute to trajectories of BP over the adult life course is unclear.

We hypothesized that increases in the steady-state and pulsatile load components of BP with aging are influenced differentially by traditional vascular risk factors. We tested this hypothesis by using multiple serial BP measurements obtained on a large community-based sample to examine how standard risk factors influence longitudinal tracking of BP and its components over the time.^20–23 The ability to analyze repeated BP measures from longitudinal data offers the opportunity not only to estimate longitudinal BP tracking with relative precision, but also to evaluate the influence of covariates on such tracking. Thus, we also examined differences in longitudinal tracking of BP between men and women, and determined the extent to which these sex-related differences are influenced by the presence or absence of standard vascular risk factors.¹⁴

METHODS

Study Sample

The Framingham Offspring Study was initiated in 1971 with the enrollment of 5124 individuals comprising the children of the original Framingham Heart Study cohort and the spouses of the children.²⁴ Participants of the Offspring cohort undergo routine examinations approximately every 4 years, including a standardized assessment of cardiovascular risk factors. A total of 5,124 unique participants (24,208 observations) attended Offspring cohort examination cycles 1 (1971–1975), 2 (1979–1983), 3 (1983–1987), 4 (1987–1991), 5 (1991–1995), and 6 (1995–1998). These examination cycles capture participants in young-to-mid adulthood during a contemporary time period when use of antihypertensive medications was low to moderate,^25,26 thereby allowing us to elucidate BP tracking patterns minimally influenced by BP treatment. Of this sample, we excluded observations for individuals who were <25 or ≥75 years old at the time of any given examination (N=59; 1025 observations): a total of 59 individuals were excluded due to not meeting the age criteria at any attended examination; a total of 1025 observations were excluded, including for individuals who met the age criteria at ≥1 but not all examinations. In similar fashion, we also excluded from the relevant examination any observations for individuals with prevalent myocardial infarction or heart failure (N=39; 664 observations); those with missing data on the main outcome variables, SBP or DBP (N=0; 12 observations); and, those missing with data on any of the main clinical covariates for analyses (N=33; 775 observations). Thus, individuals could contribute observations to some examinations and not to others, e.g. participants who developed myocardial infarction or heart failure during the follow-up period could contribute observations prior to the event but not afterwards. In total, 4,993 unique individuals providing 21,732 person-examinations were included in the present analyses.

The study protocols for all examinations were approved by the Institutional Review Board at the Boston University Medical Center. All attendees at each examination provided written informed consent.

Clinical Assessment

Cardiovascular risk factors, including BP, were assessed at each examination according to a standardized protocol,²⁷ as described in detail in the Supplemental Methods. PP was calculated as SBP minus DBP; MAP was calculated as DBP plus 1/3 PP. Hypertension was defined as SBP ≥140 mmHg or DBP ≥90 mmHg or taking anti-hypertensive therapy. Diabetes was defined as fasting blood glucose ≥126 mg/dL or the use of hypoglycemic medications.

Statistical Analyses

Multilevel statistical modeling allows for the analyses of data that vary at multiple levels.²⁸ The application of multilevel modeling is particularly well-suited to analyzing longitudinal data where repeated measurements, such as BP, are collected at different time points (1st level) within the same individual (2nd level). Multilevel models allow for estimating the overall pattern of BP measurements over time, using serially collected BP measurement data, in addition to assessing the effect of clinical covariates on temporal patterns of BP change. Unlike traditional regression models, the multilevel analytical approach provides the advantage of accommodating individuals with non-attendance at some of eligible examinations and, thus, facilitates analyses that can include the maximal number of observations in a longitudinal investigation.²⁹

We used multilevel statistical modeling to estimate growth curves for each BP measure (SBP, DBP, MAP, and PP; separate models for each) using SAS PROC MIXED, with an unstructured correlation matrix. For individuals taking anti-hypertensive therapy, the previously described method²⁷ of adding 10 mmHg to SBP and 5 mmHg to DBP measures was used; MAP and PP were calculated based on these imputed SBP and DBP values using formulae noted above. To graphically demonstrate the tracking of BP measures over time, separate graphs for each BP measure were constructed for men and women.

Multilevel modeling was used to analyze the associations of each BP measure with clinical covariates that have previously been reported to be related to variation and/or change in BP: age, sex, body mass index (BMI), smoking status, diabetes mellitus, total/high-density lipoprotein (HDL) cholesterol ratio, and resting heart rate. Because all clinical covariates were assessed both at the baseline examination and at each subsequent examination, information regarding all covariate measures (including serially updated values and the timing of their measurement) were incorporated in all multivariable-adjusted multilevel models. To account for variation in BP measures between examinations, the examination cycle was included as a covariate in all analyses. Random intercepts and random effects of age were examined for all models to reflect different starting values and different slopes for age for each BP measure in each participant. To investigate possible non-linear relations of age with each BP measure, we also examined the quadratic effect of age in all analyses (fitting an age-squared term). We fit a series of pre-specified models, constructed based on biologically plausible relationships between clinical covariates and BP measures, with direct entry of candidate variables. In addition, biologically plausible interactions between age, sex, and other clinical risk factors were also investigated using corresponding interaction terms.

To illustrate the association of overall vascular risk factor burden on the patterns of BP tracking, we generated plots for both sexes based on representative `high' versus `low' risk factor status (using results of the final multivariable models). A representative `high risk' status was defined as presence of all of the following: BMI = 30 kg/m<sup>2</sup>, heart rate = 80 beats per minute,<sup>30</sup> presence of diabetes, total/HDL cholesterol ratio = 5,<sup>31</sup> and being an active smoker. Similarly, a representative `low risk' status was presence of all of the following characteristics: BMI = 25 kg/m², heart rate = 60 beats per minute,³⁰ absence of diabetes, total/HDL cholesterol ratio = 4,³¹ and being a non-smoker.

Secondary Analyses

We repeated all multivariable analyses in the subset of 4,898 individuals without anti-hypertensive therapy (18,658 observations). We also repeated main analyses while additionally adjusting for triglycerides, alcohol use, and physical activity index, in the subset of 4,280 individuals (16,381 total observations) in whom these clinical data were available.

All analyses were performed using SAS version 9.2 and a two-tailed P value of <0.05 was considered significant. S-PLUS and Excel were used to create the graphical displays.

RESULTS

The baseline and final examination characteristics of the study sample are shown in Table 1. Women compared to men had lower values of all BP measure at both baseline and final examinations, with the exception of PP; at the final examination, PP measures were similar between sexes. Measurements of BP were collected from a total of 21,732 observations over 28 years of follow up (Figure S1).

Table 1.

Clinical characteristics of the study participants at the first and last attended examinations

	Baseline Examination		Final Examination
Clinical Characteristics	Men (N=2408)	Women (N=2585)	Men (N=2408)	Women (N=2585)
Age, years	38±9	38±9	55±11	55±11
Body mass index, kg/m2	26.8±3.7	24.4±4.7	28.1±4.4	26.9±5.8
Systolic blood pressure, mmHg	127±16	118±17	133±20	129±22
Diastolic blood pressure, mmHg	82±11	76±11	80±11	76±10
Mean arterial pressure, mmHg	97±12	90±12	98±12	93±13
Pulse pressure, mmHg	45±10	42±10	53±15	53±17
Antihypertensive treatment, %	3	2	25	21
Hypertension, %	26	13	43	36
Smoking, %	51	44	26	25
Diabetes, %	4	2	14	8
Total cholesterol, mg/dL	202.6±38.9	193.4±38.9	203.0±40.6	210.5±40.2
HDL cholesterol, mg/dL	44.4±12.0	56.4±14.6	43.5±12.6	56.6±16.2
Total/HDL cholesterol	4.9±1.7	3.6±1.2	5.0±2.0	4.0±1.4
Triglycerides, mg/dL†*	151.9±104.5	103.3±74.8	154.4±164.3	132.2±80.1
Lipid lowering treatment, %	0.5	0.4	9.6	6.8
Alcohol use, %*†	28	19	19	15
Heart rate, beats per minute	72±13	78±14	65±12	67.6±11.8
Physical activity index*	35.9±6.5	34.0±4.9	37.1±7.7	36.2±6.0

Values shown are means±standard deviations or percents.

^*Triglycerides, alcohol use, and physical activity index were assessed in a subsample of 2051 men and 2229 women.

^†Alcohol intake of >14 drinks per week in men or >7 drinks per week in women.

Based on serial BP measurement data collected on the same study participants followed over time, the mean values of BP recorded at each age were plotted to depict longitudinal BP tracking in men and women separately (Figure). As shown in unadjusted plots, SBP increased steadily with advancing age in both sexes; DBP increased and then peaked during the fifth decade and then progressively decreased in both women and men. These longitudinal profiles of SBP and DBP were accompanied by a steady increase in MAP until approximately the seventh decade, after which there was a relative plateau in MAP with increasing age. In contrast, PP remained relatively constant until the fifth decade, after which PP steeply increased in both sexes.

Figure.

Unadjusted mean SBP, DBP, MAP, and PP values with increasing age for men and women.

Clinical Correlates of BP Measures

Over the total follow-up period, several clinical factors were positively related to progressively increasing SBP: age, male sex, BMI, diabetes, and heart rate (Table 2). On the other hand, smoking status was inversely associated with SBP. In the absence of any significant interaction terms, each of these clinical correlates in the multivariable model was evaluated with respect to the relative strength of its association with longitudinal tracking of SBP. The risk factors that demonstrated the strongest associations with increasing SBP were (in decreasing order, based on values of F statistic in the model): older age, greater BMI, higher heart rate, male sex, and diabetes.

Table 2.

Clinical correlates of longitudinal tracking of change in SBP, DBP, MAP, and PP.

					Steady-State Load		Pulsatile Load
BP Components	SBP		DBP		MAP		PP
Baseline Covariates	Coefficient (SE)	P value	Coefficient (SE)	P value	Coefficient (SE)	P value	Coefficient (SE)	P value
Age (per 10 years)	6.566 (0.182)	<0.0001	4.159 (0.357)	<0.0001	4.132 (0.126)	<0.0001	4.656 (0.160)	<0.0001
Age-squared	—		−0.838 (0.039)	<0.0001	—		—
Male Sex	6.425 (0.350)	<0.0001	4.129 (0.223)	<0.0001	4.743 (0.256)	<0.0001	6.241 (1.219)	<0.0001
Age * Male Sex	—		—		—		−1.310 (0.193)	<0.0001
BMI (per 5 kg/m2)	5.084 (0.155)	<0.0001	3.362 (0.097)	<0.0001	4.212 (0.110)	<0.0001	1.739 (0.134)	<0.0001
Age * BMI	—		−0.422 (0.064)	<0.0001	—		—
Male Sex * BMI	—		—		—		−0.993 (0.219)	<0.0001
Current smoking	−0.724 (0.249)	0.004	−1.191 (0.156)	<0.0001	−0.965 (0.173)	<0.0001	0.557 (0.184)	0.003
Diabetes	2.145 (0.534)	<0.0001	−1.124 (0.311)	0.0003	−0.676 (0.350)	0.054	4.571 (0.406)	<0.0001
Total/HDL cholesterol (per 2 units)	−0.142 (0.078)	0.069	0.469 (0.095)	<0.0001	0.274 (0.105)	0.009	−0.719 (0.115)	<0.0001
Heart rate (per 10 bpm)	1.655 (0.085)	<0.0001	1.100 (0.053)	<0.0001	1.135 (0.058)	<0.0001	0.795 (0.064)	<0.0001

SBP, systolic blood pressure; DBP, diastolic blood pressure; PP, pulse pressure; MAP, mean arterial pressure; BMI, body mass index; HDL, high-density lipoprotein; bpm, beats per minute. Age was centered at the mean of all participants at all exams (49 years) to reduce multi-collinearity between regression coefficients. All models also adjusted for examination cycle. The regression coefficients represent the change in mean SBP, DBP, PP, or MAP (in mmHg) per corresponding unit increase in the continuous covariates (or presence versus absence of categorical covariates). Note that main effects of covariates are not directly interpretable in the presence of cross-product terms evaluating interactions.

Increasing DBP was positively associated with total/HDL cholesterol along with many of the same covariates associated with SBP, including: age, male sex, BMI, and heart rate. Inverse correlates of DBP included diabetes as well as smoking. In addition, DBP was also inversely associated with age-squared (p_adjusted<0.0001), reflecting a non-linear (inverted U-shaped) pattern of DBP change with advancing age. The interaction between age and BMI was also statistically significant for DBP (p_adjusted<0.0001), such that the relation of BMI to DBP was more prominent in younger compared to older individuals (Figure S2).

The clinical correlates of MAP generally resembled those of DBP, with the exception that MAP was not significantly associated with diabetes (Table 2). Correlates of PP were more similar to those of SBP, with some noteworthy exceptions. Unlike for SBP, there was a statistically significant interaction between age and sex for PP, such that the increase in PP with age was greater women than in men (p<0.0001). A significant interaction between BMI and sex was also observed for PP, where the positive association of BMI with PP was greater in women compared to men (Figure S2). In addition, a higher total/HDL cholesterol ratio was associated with a lower PP, and smoking was positively associated with PP.

Association of Risk Factor Burden on Longitudinal Tracking of BP

Consistent with the associations described above, a representative category of `high' clinical risk status (i.e. greater BMI, higher heart rate, presence of diabetes, active smoking history, and higher total/HDL cholesterol ratio) compared to a representative `low' clinical risk status (i.e. normal BMI, normal heart rate, absence of diabetes and smoking, and lower total/HDL cholesterol ratio) was associated with higher BP levels over the life course (Figure S3). Both women and men with representative `high' compared to `low' risk status demonstrated higher levels of SBP, DBP, MAP, and PP from young adulthood to older age. Notably, the greater rate of rise in PP among women compared to men was similar for individuals with either high or low clinical risk burden.

Secondary Analyses

In multivariable analyses repeated in the subset of individuals not taking anti-hypertensive therapy, results were similar (Table S1). In multivariable analyses that included adjustment for additional clinical covariates, higher triglyceride levels were significantly associated with longitudinal increase in BP measures (all βs are in mmHg per SD increment in triglycerides): SBP (β=0.877; p<0.0001), DBP (β=0.178; p=0.017), MAP (β=0.307; p=0.0002), and PP (β=0.511; p<0.0001). Moderate or greater alcohol consumption use was also associated with increase in all BP outcomes (βs in mmHg): SBP (β=2.816; p<0.0001), DBP (β=1.471; p<0.0001), MAP (β=2.028; p<0.0001), and PP (β=1.385; p<0.0001). Physical activity was not significantly related to either SBP (p=0.39) or MAP (p=0.26). However, physical activity was modestly related to increased DBP (β=0.174 mmHg per 1-SD increment in physical activity index; p=0.015), in a direction that was consistent with the proportionately greater decrease in PP that was observed in association with physical activity (β=−0.297 mmHg per 1-SD increment in physical activity; p=0.001).

DISCUSSION

Prior investigations of age-related trends in BP have been limited to cross-sectional or group averaged data.^5,10,13 By applying a multilevel modeling approach to analyze both intra- and inter-individual BP measurements over serial time points, we were able to delineate BP tracking over the life course and also identify risk factor correlates. Our data confirm an overall pattern of age-related change in BP that is characterized by steady increases in SBP and DBP in early adulthood, followed by steeper increases in SBP and marked decrease in DBP in later life. Accordingly, we observed that MAP steadily increases throughout adulthood until reaching a relative plateau in the seventh decade of life, somewhat later than suggested by prior studies;¹⁰ conversely, PP begins to increase at approximately age 40 years, after which PP continues to steeply rise throughout the remaining life course.

Several established risk factors for hypertension were associated with longitudinal increase in all BP measures. These risk factors included increased heart rate^32,33 and greater BMI,^34,35 with similar results observed for triglycerides^36,37 and moderate alcohol consumption in the subgroup with available data on these variables.^38,39 We observed that age and sex influenced the association of BMI with select BP indices. Notably, the association of greater BMI with higher DBP was more prominent in younger compared to older adults, raising the possibility that larger body size in early life is associated with small vessel resistance in the absence of the large vessel remodeling that more typically occurs in older age.⁴⁰ The relationship between BMI and PP was more prominent in women than men, consistent with cross-sectional reports⁴¹ and potentially due to sex-based anatomic limits in the extent to which arterial diameter can increase along with BMI, reflecting the arterial remodeling responses to excess adiposity.

Not all risk factors were associated with similar increases in all BP outcomes over the life course. Diabetes was associated with measures of pulsatile rather than steady state arterial load. Cross-sectional studies have related diabetes to arterial stiffness^42,43 and the preferential stiffening of the central over peripheral arteries in particular.⁴⁴ Diabetes may promote large artery stiffening through the formation of advanced glycation end-products, causing cross-linking of collagen along the arterial wall.^43,45 Although some cross-sectional studies have suggested that the relationship between diabetes and BP indices may be more pronounced in women than in men,⁴³ we did not observe a sex-interaction in our longitudinal analyses. In contrast to diabetes, total/HDL cholesterol was associated with BP measures of steady state but inversely with pulsatile load. Despite being a recognized risk factor for hypertension,⁴⁶ the extent to which hypercholesterolemia affects the hemodynamic components of BP has previously been unclear. Studies exploring the possibility that serum cholesterol is a stronger determinant of small versus large vessel resistance have thus far produced conflicting results.^47–50

Longitudinal patterns of BP tracking were consistently more pronounced in women than in men. Women experienced an earlier rise in SBP and PP with age and a slightly later decrease in DBP with a corresponding later plateau in MAP. Notably, age-related increase in PP was higher in women than men, corresponding to differences in PP that were observed between women and men at the baseline but not at the final examinations. Sexual dimorphism in BP tracking is likely due to factors that precede as well as accompany the menopausal transition.^51,52 Small observational and experimental studies indicate that young, premenopausal women have lower autonomic tone^53,54 and baroreceptor response⁵⁴ than men of similar age. Additionally, women have higher forward and reflected wave amplitude than men on arterial tonometry, manifesting more frequently as aortic stiffness.⁵⁵ Together with prior work, our findings suggest that although women may initially have lower BP measures than men overall, they tend to experience greater pulsatile load over time. This effect could be exacerbated by greater coupling of ventricular and vascular stiffening with aging and in the setting of smaller-sized ventricles in women compared to men.⁵⁶ Notably, a higher versus lower clinical risk profile, defined by the presence of traditional determinants of hypertension, conferred similar increases in all BP measures that were comparable between the sexes.

Several limitations of our investigation merit consideration. Because data regarding serum triglycerides, alcohol use, and physical activity were available for only a subset of the total sample, these covariates were included in secondary analyses only. Given the long follow-up duration required for our study design, baseline data were collected up to several decades ago; thus, the generalizability of our findings to individuals with elevated BP in the contemporary era and to recent birth cohorts (with different sets of environmental exposures) is not known. Our study sample was predominantly of European ancestry and, thus, the extent to which our findings are generalizable to other racial/ethnic groups is also unknown. Notwithstanding these limitations, our study had several strengths. Our investigation included a large, community-based sample followed closely with serial BP measurements over three decades. This longitudinal design permitted the use of a multilevel modeling analytical approach, which facilitates the evaluation of serial BP measures within and between individuals as well as the correlates of longitudinal alterations in BP indices. Because our findings are based on observational data, they may be considered hypothesis generating.

NOVELTY AND SIGNIFICANCE.

What Is New?

• Although it is well known that blood pressure (BP) steadily increases with age, the extent to which risk factors influence age-related elevations in BP is unclear.

What is Relevant?

• Elevated BP remains a major lifetime contributor to cardiovascular disease.

• Understanding how risk factors may impact trajectories of BP over the life course is needed to prevent age-related increases in BP.

Summary

• Select vascular risk factors differentially impact the longitudinal tracking of individual components of BP.

• Further research is needed to investigate the mechanisms that regulate BP components over the life course, and their potential as targets for preventing age-related BP progression.

Perspectives.

We observed that longitudinal BP changes with aging are characterized by distinctive patterns of increase in SBP, DBP, MAP, and PP. These patterns begin in young adulthood, span several decades, and are influenced to varying degrees by the presence of select vascular risk factors. Although many risk factors are related to longitudinal increase in all BP measures, dyslipidemia is particularly related to increase in steady state load, whereas diabetes and smoking are more strongly associated with elevations in pulsatile load. Additionally, women are more predisposed than men to higher pulsatile load over the life course. Further research is needed to investigate the mechanisms underlying these observations and, in turn, the extent to which targeted interventions can attenuate the typical trajectory of BP progression with aging.