Trajectory classes identified for mid–blood pressure, their pattern by age, and number of CARDIA participants in each class.
Trajectory classes identified for systolic and diastolic blood pressure, their pattern by age, and number of CARDIA participants in each class.
eAppendix. Trajectory Modeling and References
eTable 1. Adjusted Odds Ratios and 95% Confidence Intervals of the Association of Mid-Blood Pressure Trajectory Groups with Coronary Artery Atherosclerosis > 0
eTable 2. Percent (%) Missing Mid-BP by Trajectory Group and CARDIA Exam
eTable 3. Average Systolic and Diastolic Blood Pressure and Percentage on Medication at Each CARDIA Exam by SBP Trajectory Group
eTable 4. Average Systolic and Diastolic Blood Pressure and Percentage on Medication at Each CARDIA Exam by DBP Trajectory Group
eTable 5. Demographic Characteristics and Cardiovascular Risk Factors by SBP Trajectory Group
eTable 6. Demographic Characteristics and Cardiovascular Risk Factors by DBP Trajectory Group
eFigure. Proportion of DBP Trajectories by SBP Trajectory Groups in CARDIA
Customize your JAMA Network experience by selecting one or more topics from the list below.
Allen NB, Siddique J, Wilkins JT, et al. Blood Pressure Trajectories in Early Adulthood and Subclinical Atherosclerosis in Middle Age. JAMA. 2014;311(5):490–497. doi:10.1001/jama.2013.285122
Single measures of blood pressure (BP) levels are associated with the development of atherosclerosis; however, long-term patterns in BP and their effect on cardiovascular disease risk are poorly characterized.
To identify common BP trajectories throughout early adulthood and to determine their association with presence of coronary artery calcification (CAC) during middle age.
Design, Setting, and Participants
Prospective cohort data from 4681 participants in the CARDIA study, who were black and white men and women aged 18 to 30 years at baseline in 1985-1986 at 4 urban US sites, collected through 25 years of follow-up (2010-2011). We examined systolic BP, diastolic BP, and mid-BP (calculated as [SBP+DBP]/2, an important marker of coronary heart disease risk among younger populations) at baseline and years 2, 5, 7, 10, 15, 20, and 25. Latent mixture modeling was used to identify trajectories in systolic, diastolic, and mid-BP over time.
Main Outcomes and Measures
Coronary artery calcification greater than or equal to Agatston score of 100 Agatston units (AU) at year 25.
We identified 5 distinct mid-BP trajectories: low-stable (21.8%; 95% CI, 19.9%-23.7%; n=987), moderate-stable (42.3%; 40.3%-44.3%; n=2085), moderate-increasing (12.2%; 10.4%-14.0%; n=489), elevated-stable (19.0%; 17.1%-20.0%; n=903), and elevated-increasing (4.8%; 4.0%-5.5%; n=217). Compared with the low-stable group, trajectories with elevated BP levels had greater odds of having a CAC score of 100 AU or greater. Adjusted odds ratios were 1.44 (95% CI, 0.83-2.49) for moderate-stable, 1.86 (95% CI, 0.91-3.82) for moderate-increasing, 2.28 (95% CI, 1.24-4.18), for elevated-stable, and 3.70 (95% CI, 1.66-8.20) for elevated-increasing groups. The adjusted prevalence of a CAC score of 100 AU or higher was 5.8% in the low-stable group. These odds ratios represent an absolute increase of 2.7%, 5%, 6.3%, and 12.9% for the prevalence of a CAC score of 100 AU or higher for the moderate-stable, moderate-increasing, elevated-stable and elevated-increasing groups, respectively, compared with the low-stable group. Associations were not altered after adjustment for baseline and year 25 BP. Findings were similar for trajectories of isolated systolic BP trajectories but were attenuated for diastolic BP trajectories.
Conclusions and Relevance
Blood pressure trajectories throughout young adulthood vary, and higher BP trajectories were associated with an increased risk of CAC in middle age. Long-term trajectories in BP may assist in more accurate identification of individuals with subclinical atherosclerosis.
Blood pressure (BP) represents a major modifiable risk factor for cardiovascular disease (CVD). Current risk prediction models take into account BP level only at the time of risk prediction, usually in middle or older age, and do not consider the potential effect of BP levels earlier in life or the changes in BP levels over time. Time-averaged and cumulative BP among adults have also been shown to predict CVD risk in several large prospective investigations.1-4 For nearly everyone, at least in populations who consume large amounts of salt,5 systolic BP (SBP) increases with age, and these increases during middle age have been associated with CVD risk.6,7 It was recently demonstrated that changes in BP throughout middle and older age are significantly associated with lifetime risk of CVD, and the larger the changes the greater the lifetime risk.8 However, the patterns of BP change may differ among individuals; thus, a life-course perspective when evaluating the effects of BP is essential. Long-term BP trajectory patterns from young adulthood to middle age and their effect on CVD risk remain unknown. Therefore, the aims of this study were (1) to identify subgroups of individuals with similar trajectories in BP from young adulthood through middle age; (2) to characterize those participants and BP trajectories; and (3) to determine the independent association of BP trajectories from young adulthood to middle age with the presence of subclinical CVD as measured by coronary artery calcification (CAC) during middle age. We hypothesized that multiple trajectory patterns exist within the Coronary Artery Risk Development in Young Adults (CARDIA) population and that in comparison with a trajectory in which individuals maintain ideal BP levels throughout young adulthood, a portion of the population experiences higher levels of BP and/or faster rates of increase in BP that are associated with increased subclinical atherosclerosis in year 25 of follow-up.
The CARDIA study is a prospective cohort study designed to investigate the development of cardiovascular risk and disease. In 1985-1986, 5115 black and white men and women aged 18 to 30 years were recruited from 4 urban sites across the United States, including Birmingham, Alabama; Chicago, Illinois; Minneapolis, Minnesota; and Oakland, California. CARDIA participants have been followed up for more than 25 years with detailed demographic and clinical data, including self-reported anthropometric and laboratory measures and assessment for subclinical atherosclerosis. Race was determined by self-report at year 0 and verified at year 2. Participant data have been collected across 8 examination cycles including baseline and at years 2, 5, 7, 10, 15, 20, and 25. Retention rates have been high throughout the 25 years of follow-up (90%, 86%, 81%, 77%, 74%, 72%, and 72%, respectively). A more thorough description of the study design and conduct has been previously published.9-11 The study was approved by the institutional review board at all study sites and all participants provided written informed consent. The current analyses include CARDIA participants with BP measures available at 3 or more examinations.
Blood pressure was measured at each of the 8 CARDIA examination cycles. At each examination, trained technicians used either a random zero sphygmomanometer (years 0-20) or an Omron sphygmomanometer (years 20-25) to measure participants’ BP. Resting SBP and diastolic BP (DBP) was measured 3 times at 1-minute intervals. We used the average of the second and third BP measurements. In our main analysis we examined trajectories in SBP and mid-BP, defined as the mean of the SBP and DBP measurements. In addition to SBP, mid-BP was chosen a priori because it has been shown to have the greatest predictive utility for CVD compared with other single measures of BP (SBP or DBP, pulse pressure, or mean arterial pressure).12,13 In addition, mid-BP allowed us to take into account the effects of both SBP and DBP jointly. This is of particular importance among this younger CARDIA population because DBP has been shown to play a larger role in CHD risk among young to middle-aged adults.14-17 Separate trajectories in SBP and DBP were also examined.
Subclinical coronary atherosclerosis was quantified using CAC measured at the year 25 examination. Calcified coronary artery plaque measurement was done with an electrocardiographically gated multidector computed tomography scanner with a standard phantom for calibration using a standardized protocol18 with published accuracy, comparability, and reproducibility.19,20 Briefly, scans were obtained and image analysts blinded to participant characteristics calculated a total coronary artery calcium score using a modified Agatston method,21 with select overreading by a physician expert in cardiovascular imaging. Our outcome in this study was the presence of CAC greater than or equal to an Agatston score of 100 Agatston units (AU). A CAC score of 100 AU or greater has been identified as a marker of high risk of coronary events22 and has been used often in previous consensus statements.23,24 In addition, we examined CAC scores greater than 0 in sensitivity analyses.
Blood pressure trajectories were modeled among all 4681 CARDIA participants with BP measured at 3 or more examinations. Only 3442 participants with CAC data available at year 25 were included in the models examining the association between BP and CAC score of 100 AU or greater. We used latent class models to identify subgroups within the CARDIA cohort that share a similar underlying trajectory in BP. These models were fit using SAS Proc Traj.25-27 The eAppendix in the Supplement contains details on the trajectory modeling process. To estimate the association of trajectory group on subclinical atherosclerosis, trajectory group membership was included as an independent variable in a logistic regression model examining predictors of the presence of CAC Agatston score of 100 AU or greater at examination year 25. The adjusted model included baseline age, race, sex, highest level of education, antihypertensive medication use at each examination, total cholesterol level per year, cumulative number of years with diabetes, cumulative number of years as a current smoker, and body mass index (calculated as weight in kilograms divided by height in meters squared) at both baseline and year 25. As in other studies, to account for the uncertainty in BP trajectory group assignment the posterior probability of group membership, we generated 20 multiple imputations of trajectory membership using the posterior probabilities.28 We then fit the logistic regression model for each of the 20 data sets and combined our results across the multiple imputations. Model calibration was examined using the Hosmer-Lemeshow goodness-of-fit χ2 statistic. We compared the predictive utility of trajectory group compared with other longitudinal BP measures (single baseline BP measurements, multiple BP measurements [baseline and year 25], and cumulative BP [mm Hg × years]) using the logistic C statistic and the discrimination (D) statistic29 of the models for each of these BP measures individually and jointly. The D statistic represents the difference in the average of event probabilities between the groups who experience the event and those who do not. The Wilcoxon test was used to detect significant location shifts and differences in the distributions of the predicted event probabilities. To quantify how the model reclassifies individuals with the addition of trajectory group, we calculated the integrated discrimination improvement index by calculating the difference in the D statistics (discrimination slopes) between the models including only baseline BP or year 25 BP and those including trajectory group. All analyses were completed using SAS software version 9.3 (SAS Institute Inc). Two-sided P<.05 was considered statistically significant.
The CARDIA cohort enrolled 5115 individuals. Trajectories in mid-BP were examined among 4681 CARDIA participants with 3 or more BP measurements. Of these, 3442 participants had CAC measurements available at year 25. Five discrete trajectories in mid-BP from young adulthood to middle age were identified (Figure 1): 21.8% (95% CI, 19.9%-23.7%; n = 987) of participants maintained low BP throughout follow-up (low-stable group); 42.3% (95% CI, 40.3%-44.3%; n = 2085) of participants experienced moderate BP levels (moderate-stable group); 12.2% (95% CI, 10.4%-14.0%; n = 489) started with moderate levels and experienced a rapid increase in BP beginning at an average age of 35 years (moderate-increasing group); 19.0% (95% CI, 17.1%-20.0%; n = 903) had relatively elevated BP levels throughout (elevated-stable group); and 4.8% (95% CI, 4.0%-5.5%; n = 217) had high BP levels (elevated-increasing group).
In general, BP for each trajectory group tracked from age 25 over time and experienced small, steady increases in SBP and DBP, as did the proportion of participants taking antihypertensive medications; however, 2 groups (the moderate-increasing and elevated-increasing groups) experienced large and more rapid increases in BP over time (Table 1). The moderate-increasing and elevated-increasing groups experienced the largest increases in SBP (mean increases were 30.2 mm Hg and 21.2 mm Hg, respectively) and in DBP (mean increases, 20.9 mm Hg and 11.5 mm Hg, respectively).
Individuals in the low-stable group were the most likely to be female, white, more highly educated, and more likely to have fewer concurrent cardiovascular risk factors (Table 2). African Americans were more likely to experience rapid increases in BP, with the moderate-increasing group having the largest proportion of African American women and the elevated-increasing group having the largest proportion of African American men. In addition, these rapid increases in BP were associated with higher rates of smoking and occurred alongside increases in body mass index (mean body mass index increased by 7.4-7.6 among the increasing and high groups but only by 4.5-5.7 among the groups with more stable BP levels).
The prevalence of a CAC score of 100 AU or greater varied from 4.0% in the low-stable BP trajectory group to 25.4% in the elevated-increasing BP trajectory group. Within each trajectory group, individuals taking antihypertensive medications had a higher prevalence of CAC compared with individuals within the same trajectory group who were not taking antihypertensive medications. Even among groups that started at similar baseline BP levels, such as the moderate-stable and moderate-increasing groups, participants who experienced steeper increases in BP had a higher prevalence of a CAC score of 100 AU or greater (7.9% vs 10.1%, respectively; P<.001 for trend). In comparison with individuals in the low-stable group, those in trajectory groups with patterns of higher mid-BP had increasingly greater odds of having a CAC score of 100 AU or greater. Adjustment for individual demographic characteristics and education only slightly attenuated the odds ratios. Additional adjustment for other cardiovascular risk factors and BP medication resulted in larger attenuations of the odds ratios for the higher-risk groups, especially the elevated-stable and elevated-increasing groups. Elevated BP trajectory groups were independently associated with CAC scores of 100 AU or greater at year 25 (Table 3) even after adjustment for demographic characteristics, other cardiovascular risk factors, and antihypertensive medication use in model 1. Adjusted odds ratios were 1.44 (95% CI, 0.83-2.49) for moderate-stable, 1.86 (95% CI, 0.91-3.82) for moderate-increasing, 2.28 (95% CI, 1.24-4.18) for elevated-stable, and 3.70 (95% CI, 1.66-8.20) for elevated-increasing groups. The adjusted prevalence of a CAC score of 100 AU or greater was 5.8% in the low-stable group. These odds ratios represent absolute increases of 2.7%, 5%, 6.3%, and 12.9% for the prevalence of a CAC score of 100 AU or greater in the moderate-stable, moderate-increasing, elevated-stable, and elevated-increasing groups, respectively, compared with the low-stable group. Findings were similar for the outcome of a CAC score greater than 0 (eTable 1 in the Supplement). In sensitivity analyses, the BP trajectory patterns were similar and their association with a CAC score of 100 AU or greater was essentially unchanged when restricted to individuals who did not take BP medication at any time during follow-up (n = 3507).
Odds ratios for CAC for different trajectory groups were not significantly altered even with additional adjustment for baseline SBP and DBP (model 2) or year 25 SBP and DBP (model 3). All models fit the data well with Hosmer-Lemeshow χ2 statistics with P>.05.The average predicted probability of having a CAC score of 100 AU or greater in the adjusted model including trajectory group for CARDIA participants with a CAC score of 100 AU or greater was 25.3% compared with 8.1% among those who did not have a CAC score of 100 AU or greater. The discrimination of the model and its ability to correctly identify individuals with CAC scores of 100 AU or greater were similar to models with either baseline or year 25 BP alone. The C statistic for the unadjusted model was 0.66, which increased to 0.84 in models 1, 2, and 3. Similarly, the D statistic for the adjusted model of trajectory groups was only slightly higher (0.172) compared with models of baseline BP and year 25 BP (D statistics of 0.167 and 0.169, respectively; P<.001); however, integrated discrimination improvement was only 0.5 and 0.3, respectively.
Overall patterns were similar when we examined the trajectories of SBP and DBP separately (Figure 2 and eTables 2-6 in the Supplement). Five discrete trajectory groups were identified for long-term patterns in SBP: 24.6% (95% CI, 22.5%-26.8%; n = 1113) of participants maintained low SBP throughout follow-up (low-stable group); 44.1% (95% CI, 42.1%-46.1%; n = 2167) of participants experienced moderate SBP levels (moderate-stable group); 9.8% (95% CI, 7.9%-10.7%; n = 363) started at moderate levels and experienced a rapid increase in SBP starting at an average age of 35 years (moderate-increasing group); 18.8% (95% CI, 16.6%-21.0%; n = 884) had relatively elevated SBP levels throughout (elevated-stable group); and 3.3% (95% CI, 2.6%-4.0%; n = 154) had high SBP levels (elevated-increasing group). Systolic BP trajectory groups had increasing prevalence of a CAC score of 100 AU or greater with increasing SBP trajectory group (eFigure in the Supplement). After adjustment for demographics, long-term cardiovascular risk factors, and BP medication, these SBP trajectory groups were significantly associated with prevalence of a CAC score of 100 AU or greater; odds ratios were 1.56 (95% CI, 0.96-2.53) for the moderate-stable group, 1.71 (95% CI, 0.82-3.54) for the moderate-increasing group, 2.16 (95% CI, 1.23-3.83) for the elevated-stable group, and 5.24 (95% CI, 2.21-12.44) for the elevated-increasing group compared with the low-stable group (Table 3). The adjusted prevalence of a CAC score of 100 AU or greater at year 25 was 6.4% among the low-stable group and increased by 2.5%, 3.3%, 5.9%, and 14.3% for the moderate-stable, moderate-increasing, elevated-stable, and elevated-increasing groups, respectively.
Five discrete trajectory groups were identified for long-term patterns in DBP: 22.5% (95% CI, 20.5%-24.6%; n = 1024) of participants maintained low DBP throughout follow-up (low-stable group); 42.6% (95% CI, 40.4%-44.8%; n = 2108) of participants experienced moderate DBP levels (moderate-stable group); 11.4% (95% CI, 9.6%-13.2%; n = 448) started at moderate levels and experienced a rapid increase in DBP starting at an average age of 35 years (moderate-increasing group); 19.2% (95% CI, 17.2%-21.2%; n = 909) had relatively elevated DBP levels throughout (elevated-stable group); and 4.4% (95% CI, 3.6%-5.1%; n = 192) had high DBP levels (elevated-increasing group). The adjusted associations (Table 3) for DBP trajectory groups were less strongly associated with a CAC score of 100 AU or greater. Systolic and diastolic trajectory groups were moderately to highly correlated with mid-BP trajectory groups (Pearson correlation coefficient of 0.83 for DBP and 0.63 for SBP) but only moderately correlated with each other, with a correlation coefficient of 0.5.
In this study, we found heterogeneous trajectories in mid-BP over a 25-year span from young adulthood to middle age. We identified 5 unique trajectories in mid-BP and SBP that were significantly associated with the presence of subclinical atherosclerosis later in middle age. CARDIA participants who exhibited elevated BP levels throughout and those who had increases in BP levels over time had the greatest odds of having a CAC score of 100 AU or greater. Importantly, the majority of these participants had BP levels within the range of prehypertension. Although the finding was not statistically significant, even individuals in the moderate-stable group, who on average maintained BP levels within recommended clinical ranges but were not in the group exhibiting the lowest levels of BP (low-stable group), tended to have almost twice the odds of having a CAC score of 100 AU or greater. Membership in a trajectory group with elevated BP represented an independent predictor of CAC beyond either baseline BP in young adulthood or year 25 BP during middle age and may help identify individuals with prevalent levels of CAC that have been shown to be associated with increased CVD risk.
More than 33% of US adults have hypertension30; however, much of the understanding of BP levels comes from cross-sectional data. Many longitudinal studies that have examined changes in BP over time often had fewer than 5 years of follow-up and focused mainly on elderly individuals. Blood pressure levels in midlife have been shown to predict stroke risk31 and cardiovascular mortality more than 25 years later32 and may be more strongly associated with coronary heart disease mortality than BP levels measured at a time closer in proximity to the time of event.33 Additionally, although BP lowering due to pharmacological treatment may be achieved quickly, measures of cardiovascular function (eg, left ventricular hypertrophy, increased left ventricular mass, and degree of carotid stenosis) tend to remain at abnormal levels after exposure to long-term periods of elevated BP.34-37
The findings from the current study provide a unique insight into long-term patterns of BP change and highlight that within the CARDIA population there are heterogeneous patterns in BP trajectories. Latent class modeling, as used in these current analyses, has allowed identification of different patterns of BP change as separate trajectory groups, thus providing a more realistic understanding of lifetime trends in BP compared with population mean levels. Among CARDIA participants, Pletcher and colleagues previously demonstrated that cumulative years of prehypertension prior to age 35 years was associated with prevalent coronary calcium in middle age.4 The current study extends those findings to demonstrate that not only is cumulative prehypertension important but certain populations such as African Americans and smokers are more likely to experience rapid increases in BP during middle age, placing them at higher risk. Consistent with our long-term findings, recent data from the United Kingdom have also demonstrated important heterogeneity in short-term BP trajectories in midlife.38 Our findings further suggest that these trajectory groups provide important additional information regarding the presence of significant levels of subclinical atherosclerosis, which has been shown to be a strong indicator of future cardiovascular risk.22 This understanding of the effect of change or timing of change in BP on subclinical atherosclerosis may be important for risk stratification in the future.
The strengths and limitations of this investigation are worth noting. This study applied innovative statistical methods to examine patterns of BP in a large, well-characterized cohort of black and white Americans. The longitudinal nature of CARDIA and phenotyping at each of the 8 examinations provides detailed long-term patterns of BP. Although CARDIA is a racially and geographically diverse cohort, the trajectory groups identified may not be generalizable to other populations. Not all CARDIA participants had BP information available at all examination periods. However, missing BP measurements are unlikely to have altered our findings because the mean number of BP measurements was 7 and did not differ by trajectory group. In sensitivity analyses, BP was imputed at each examination for all surviving CARDIA participants. Using these imputed BPs, trajectory group assignment was consistent and did not result in any changes in the association between trajectory group and CAC. In addition, CARDIA participants were aged 18 to 30 years at baseline; thus, we have no information on BP patterns prior to their entry into CARDIA. Coronary artery calcification score at year 25 was missing for 28% of CARDIA participants. Participants who did not attend more recent examinations were more likely to be African American and of lower socioeconomic status and to have a greater burden of cardiovascular risk factors; nevertheless, these subgroups are well represented in CARDIA attendees, and by adjusting for these risk factors, CAC is assumed to be missing at random and thus results are unlikely to be biased.
Although BP has been a well-known risk factor for CVD for decades, these findings suggest that an individual’s long-term patterns of change in BP starting in early adulthood may provide additional information about his or her risk of development of coronary calcium. In particular, prehypertension at a young age followed by long-term exposure to BP levels in the prehypertension range or higher was strongly associated with a CAC score of 100 AU or greater. Additional research is needed to examine the utility of specific BP trajectories in risk prediction for clinical CVD events and to explore the effect of lifestyle modification, treatment, and timing of intervention on lifetime trajectories in BP and outcomes.
Corresponding Author: Norrina B. Allen, PhD, MPH, Department of Preventive Medicine, Feinberg School of Medicine, Northwestern University, 680 N Lake Shore Dr, Ste 1400, Chicago, IL 60611 (email@example.com).
Author Contributions: Dr Allen had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Allen, Wilkins, Shay, Goff, Jacobs, Lloyd-Jones.
Acquisition of data: Lewis, Jacobs, Liu.
Analysis and interpretation of data: Allen, Siddique, Wilkins, Jacobs, Liu, Lloyd-Jones.
Drafting of the manuscript: Allen, Shay, Lloyd-Jones.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Allen, Siddique, Shay, Jacobs.
Obtained funding: Lewis, Liu, Lloyd-Jones.
Administrative, technical, and material support: Wilkins, Shay, Lewis, Lloyd-Jones.
Study supervision: Allen, Lloyd-Jones.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Goff reports receipt of grants/grants pending from Merck, payment for lectures/speakers bureaus from Merck, and membership on a data safety and monitoring board for Takeda.
Funding/Support: The CARDIA study is conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with the University of Alabama at Birmingham (grants HHSN268201300025C and HHSN268201300026C), Northwestern University (grant HHSN268201300027C), University of Minnesota (grant HHSN268201300028C), Kaiser Foundation Research Institute (grant HHSN268201300029C), and Johns Hopkins University School of Medicine (grant HHSN268200900041C). CARDIA is also partially supported by the Intramural Research Program of the National Institute on Aging (NIA), by an intra-agency agreement between NIA and NHLBI (grant AG0005), and by grant RO1 HL098445 from the National Heart, Lung, and Blood Institute.
Role of the Sponsors: This article has been reviewed by CARDIA for scientific content. The funding agencies had no other role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.
Correction: This article was corrected online March 25, 2014, for an error in the name of the units for the coronary artery calcification score and for an omitted grant.