Statistics Major only (please dont waste my time by bidding)

ContentServer.pdf

Home >Mathematics homework help >Statistics homework help >Statistics Major only (please dont waste my time by bidding)

Systolic Blood Pressure, Diastolic Blood Pressure, and Pulse Pressure: An Evaluation of Their Joint Effect on Mortality Roberto Pastor-Barriuso, PhD; José R. Banegas, MD, PhD; Javier Damián, MD, PhD; Lawrence J. Appel, MD, MPH; and Eliseo Guallar, MD, DrPH

Background: The relative importance of blood pressure compo- nents (systolic blood pressure, diastolic blood pressure, and pulse pressure) on cardiovascular risk is currently being debated. Many studies, however, are limited by inadequate statistical methods to separate these effects.

Objective: To evaluate the joint effect of blood pressure com- ponents on all-cause and cardiovascular mortality by using non- parametric and change point models.

Design: Prospective cohort study.

Setting: 15-year mortality follow-up of participants in the Second National Health and Nutrition Examination Survey.

Participants: 7830 white and African-American men and women 30 to 74 years of age, apparently free of cardiovascular disease at baseline.

Measurements: Baseline blood pressure, corrected for measure- ment error.

Results: Of the 1588 patients who died, 582 died of cardiovas- cular disease. Systolic blood pressure was linearly related to all- cause and cardiovascular mortality in younger and elderly partic- ipants. The association of diastolic blood pressure with all-cause and cardiovascular mortality was hockey stick–shaped (flat then increasing) in younger participants and J-shaped in elderly partic- ipants. Increased pulse pressure was associated with increased risk, decreased risk, or no change in risk depending on age and systolic and diastolic blood pressure.

Conclusions: On the basis of these and previous data, the evidence for a monotonic association of systolic blood pressure with all-cause and cardiovascular mortality is compelling, but a J-shaped association for diastolic blood pressure may develop at older age. The complexity of the association of pulse pressure with mortality discourages its use for prognostic or therapeutic decisions.

Ann Intern Med. 2003;139:731-739. www.annals.org For author affiliations, see end of text.

Two aspects of the effect of blood pressure on mortalityhave been controversial. The first is the relative impor- tance of blood pressure components (systolic blood pres- sure, diastolic blood pressure, and pulse pressure) as deter- minants of risk (1–5), and it has been suggested that beyond middle age, pulse pressure is a more important determinant of risk than systolic or diastolic blood pressure (2). The second is the possibility of a J-shaped relationship between blood pressure and mortality (4, 6, 7), with the concern that lowering blood pressure below a threshold of lowest risk would not be justified and could even be harmful.

Evaluating these 2 issues, however, is methodologically complex. First, systolic blood pressure, diastolic blood pressure, and pulse pressure are linearly dependent, that is, knowledge of any 2 of them determines the third. Second, because of the high correlation between systolic and dia- stolic blood pressures, interpreting the effect of blood pres- sure components is difficult. For instance, if only 1 com- ponent (such as diastolic blood pressure) is used to model risk, it is unclear what part of the effect is due to its cor- relation with systolic blood pressure (4). However, if sys- tolic and diastolic blood pressure are introduced in a re- gression model, the interpretation of both coefficients is uncertain because the coefficients partly reflect the effect of pulse pressure. Third, most studies addressing the J-shape phenomenon have used analysis strategies that did not ac- count for the complex relationships between blood pres- sure components and mortality.

We examined the relationship of blood pressure com- ponents with all-cause and cardiovascular disease (CVD)

mortality in a representative population– based prospective study. We used generalized additive models to estimate the effect of the joint distribution of blood pressure compo- nents, avoiding model assumptions about the underlying dose–response relationship (8, 9). These methods take ad- vantage of the high correlation between systolic and dia- stolic blood pressures to estimate the risk surface for com- binations of blood pressure measures. We also discuss how to interpret the relative importance of each blood pressure component based partly on graphical presentation of re- sults.

METHODS Study Sample

The study sample consisted of 9250 participants in the Second National Health and Nutrition Examination Sur- vey (NHANES II) Mortality Study (10), a prospective co- hort study that followed participants 30 to 74 years of age. The complex survey, conducted by the National Center for Health Statistics between 1976 and 1980, used a stratified, multistage sampling design to obtain a representative sam- ple of the noninstitutionalized U.S. population (11). The response rate among persons selected for the examination was 73%.

Baseline Assessment During the NHANES II baseline examination (1976

to 1980), a trained physician took 3 blood pressure mea- surements by using a calibrated mercury sphygmomanom- eter on the right arm, according to the American Heart

Article

Association guidelines. One measurement was taken at the beginning of the physical examination with the participant in the sitting position, and 2 more measurements were taken at the end of the physical examination, 1 with the participant in the supine position and the other with the participant in the sitting position (11). We used these 3 measurements to estimate each participant’s underlying systolic and diastolic blood pressure (see Statistical Analysis).

Participants’ age, sex, race or ethnicity, smoking status, and years of education were obtained by interview. Weight and height were measured to calculate the body mass in- dex. Total serum cholesterol levels were obtained from standard blood assays. Use of antihypertensive medication and previous diagnosis of diabetes were identified through the medical history questionnaire. Evidence of CVD at baseline was defined as a positive response on a modified Rose angina questionnaire or a medical history of heart attack or stroke (11).

Mortality Ascertainment during Follow-up Mortality status was ascertained from 1976 to 1992 by

searching the National Death Index and the Social Security Administration Death Master File (10). Participants not found to be deceased by 31 December 1992 were assumed to be alive. The causes of death were coded by nosologists according to the International Classification of Diseases, Ninth Revision (ICD-9). Deaths were ascribed to CVD if any of the following conditions were coded as the under- lying cause of death: hypertensive heart disease (402.0 to 402.9); ischemic heart disease (410.0 to 414.9); cardiac arrest (427.5); unspecified heart failure (428.9); unspecified CVD (429.2); cerebrovascular disease (430.0 to 438.9); or dis- eases of the arteries, arterioles, and capillaries (440.0 to 444.9).

Statistical Analysis Our main objective was to estimate the joint effect of

systolic and diastolic blood pressure on all-cause and CVD mortality, that is, the risk for death for each combination of systolic and diastolic blood pressure. This is in contrast to usual analyses, which estimate the marginal rather than the joint effects. The marginal effect of diastolic blood pressure, for instance, is the effect at each level of diastolic blood pressure “averaged” across all values of systolic blood pressure. To estimate the joint effect of systolic and dia- stolic blood pressure, we simultaneously fitted 2 nonpara- metric terms (locally weighted smoothers with 40% to 50% bandwidth) for each blood pressure component in a generalized additive logistic model (8). This technique es- timates the smooth surface of the relative risk for death as a function of both systolic and diastolic blood pressure. It does not make assumptions about the shape of the associ- ation, and it takes advantage of the correlation between systolic and diastolic blood pressure to obtain more precise estimates of risk over the joint distribution of observed values.

To compare our results with previous analyses, we also estimated the marginal effect of blood pressure compo- nents by using Cox proportional hazards models and non- parametric logistic regression to explore the continuous risk trend without imposing any assumption on the shape of the association (8). In addition, we performed statistical tests for detecting the existence of threshold effects or points of abrupt risk change (12). When a change point was detected, transition logistic models were fitted to esti- mate its location (12). Transition models are similar to spline models (13), but in the former the change point is estimated from the data rather than fixed in advance. In contrast to standard analysis of the J-shape phenomenon based on log-linear or categorical models with limited abil- ity to detect threshold effects, nonparametric regression and change point models correctly identify nonlinear ef- fects and threshold levels (12–15).

Since we detected that the proportional hazards as- sumption was applicable only after the initial 2-year follow-up, and because of the few deaths and the possibil- ity that early deaths at low blood pressure reflect underly- ing disease or frailty conditions (reverse causation bias), the first 2 years of follow-up (122 all-cause deaths, 44 of which were CVD deaths) were analyzed separately from the deaths after 2 years of follow-up (1466 all-cause deaths, 538 of which were CVD deaths).

To correct for regression dilution bias due to random within-participant variability in blood pressure (16 –18), each participant’s underlying blood pressure was estimated from the observed measurements by using a random-effects model for each sex (17). These underlying blood pressure values were used subsequently in all analyses. Since the relationship of blood pressure with CVD risk may change with age (2), we conducted all analyses separately for 2 age strata (�65 and �65 years). All models presented were

Context

Researchers often debate relationships between various blood pressure components and risk for death.

Contribution

This careful analysis from a large cohort study confirmed linear relationships between increasing systolic blood pres- sure and increasing risk for death and, depending on age, either hockey stick–shaped or J-shaped relationships be- tween diastolic blood pressure and mortality. Relationships between pulse pressure and mortality depended on whether increased pulse pressure was due to increased systolic or decreased diastolic blood pressure.

Implications

Pulse pressure alone, without appropriate attention to sys- tolic and diastolic blood pressure components, is an inade- quate indicator of mortality risk.

–The Editors

Article Blood Pressure and Mortality

732 4 November 2003 Annals of Internal Medicine Volume 139 • Number 9 www.annals.org

adjusted for age, sex, race or ethnicity, smoking status, total serum cholesterol level, body mass index, education, anti- hypertensive treatment, and history of diabetes mellitus. Statistical analyses were performed by using S-plus 2000 (Mathsoft, Seattle, Washington) (19).

Role of the Funding Source The funding source had no role in the choice of topic;

design, analysis, or interpretation of the data; or the deci- sion to submit the manuscript for publication.

RESULTS From the initial 9250 participants in the NHANES II

Mortality Follow-up Study, we excluded 1124 participants who had evidence of CVD at baseline, 146 participants with missing blood pressure values, and 150 participants who were not white or African American. Thus, the final study sample included 7830 individuals. Table 1 describes the characteristics of the study participants. Systolic blood pressure was highly correlated with diastolic blood pressure and pulse pressure in participants younger than 65 years of age and participants 65 years of age and older (Pearson correlation coefficients, 0.66 to 0.85), while diastolic blood pressure was weakly correlated with pulse pressure (Pearson correlation coefficients, 0.20 and 0.17 for participants �65 and �65 years of age, respectively). Follow-up extended from enrollment in 1976 to 1980 through 31 December 1992, with an average follow-up of 14.9 years among sur- vivors (range, 12.8 to 16.9 years), and a total of 106 387 person-years of follow-up, 1588 all-cause deaths, and 582 CVD deaths.

Early Mortality Findings During the first 2 years of follow-up, the association of

systolic blood pressure with all-cause mortality was U-shaped (Table 2), and the risk for death was lowest for participants with systolic blood pressure between 120 and 149 mm Hg (P for quadratic trend � 0.03). For diastolic blood pres-

sure, the lowest risk for death during the first 2 years of follow-up was observed at 90 to 94 mm Hg, but the over- all U-shaped trend was not significant (P for quadratic trend � 0.2). This pattern was similar after stratifying by age younger than 65 years or 65 years and older. The small number of CVD deaths during the first 2 years of fol- low-up (n � 44) precluded a corresponding dose–response analysis for this outcome.

Joint Effect of Blood Pressure Components Figure 1 shows the joint effect of systolic and diastolic

blood pressure on mortality beyond 2 years of follow-up. In younger participants, systolic blood pressure was linearly related to all-cause and CVD mortality for all diastolic

Table 1. Baseline Characteristics of Participants in the Second National Health and Nutrition Examination Survey Mortality Study by Age Group*

Characteristic Participants < 65 Years of Age (n � 5775)

Participants > 65 Years of Age (n � 2055)

All Participants (n � 7830)

Age, y 48.0 � 11.3 69.0 � 2.8 53.5 � 13.4 Women, % 52.6 56.3 53.6 White, % 89.3 89.0 89.2 Systolic blood pressure, mm Hg 128.8 � 18.2 143.4 � 19.9 132.6 � 19.7 Diastolic blood pressure, mm Hg 81.3 � 10.5 82.3 � 10.5 81.5 � 10.5 Pulse pressure, mm Hg 47.5 � 12.3 61.1 � 15.3 51.1 � 14.5 Smoking status, %

Never 37.7 51.5 41.4 Past 24.8 29.3 25.9 Current 37.5 19.2 32.7

Total serum cholesterol level, mg/dL (mmol/L) 222.1 � 47.9 (5.7 � 1.2) 234.8 � 48.7 (6.1 � 1.3) 225.4 � 48.5 (5.8 � 1.3) Body mass index, kg/m2 25.9 � 4.9 26.2 � 4.9 26.0 � 4.9 Education � high school, % 37.9 59.4 43.5 Antihypertensive treatment, % 12.1 29.5 16.7 History of diabetes mellitus, % 3.8 8.4 5.0

* Values with plus/minus signs are means � SDs.

Table 2. Blood Pressure and Risk for All-Cause Mortality during First 2 Years of Follow-up in the Second National Health and Nutrition Examination Survey Mortality Study*

Blood Pressure, mm Hg

Deaths, n Relative Risk (95% CI)†

P Value‡

Systolic 0.03 �120 21 1.00 120–129 18 0.63 (0.33–1.20) 130–139 19 0.65 (0.34–1.23) 140–149 20 0.67 (0.35–1.29) 150–159 17 0.79 (0.39–1.59) �160 27 1.06 (0.56–2.00)

Diastolic �0.2 �80 46 1.00 80–84 28 1.27 (0.78–2.04) 85–89 21 1.21 (0.71–2.07) 90–94 7 0.73 (0.32–1.64) 95–99 9 1.29 (0.62–2.71) �100 11 1.82 (0.88–3.76)

* Results for cardiovascular disease mortality are not presented because of the small number of events during first 2 years of follow-up. † Relative risk from Cox proportional hazards model adjusted for age, sex, race or ethnicity, smoking, total cholesterol level, body mass index, education, use of antihypertensive medications, and history of diabetes. ‡ Tests for quadratic risk trends were performed by including an ordinal variable with quadratic scores across blood pressure categories.

ArticleBlood Pressure and Mortality

www.annals.org 4 November 2003 Annals of Internal Medicine Volume 139 • Number 9 733

blood pressure levels. This can be appreciated by noting the increasing slope of the risk surface with increasing sys- tolic blood pressure when the surface is followed along lines parallel to the systolic blood pressure axis. Along these lines, the surface represents the effect of increasing systolic blood pressure for each fixed value of diastolic blood pres- sure, also known as the conditional effect of systolic blood pressure. This effect can be interpreted as the effect of increasing pulse pressure by increasing systolic blood pres-

sure and holding each value of diastolic blood pressure constant.

The relationship of diastolic blood pressure and mor- tality in younger participants was hockey stick–shaped, with a flat region below 80 mm Hg and then a sharp increase in risk. The effect of diastolic blood pressure for a fixed systolic blood pressure can be evaluated by following the risk surface along lines parallel to the diastolic blood pressure axis. The flat surface below 80 mm Hg of diastolic

Figure 1. Risk surfaces for all-cause and cardiovascular mortality as a function of systolic and diastolic blood pressure by age group.

Surfaces were obtained by fitting simultaneous nonparametric terms for systolic and diastolic blood pressure in generalized additive logistic models. Breaks in gray scale are determined by the risk levels labeled in each vertical axis. Adjusted for age, sex, race or ethnicity, smoking, total cholesterol level, body mass index, education, use of antihypertensive medications, and history of diabetes. The surfaces represent the relative risk for all-cause and cardiovascular mortality for each combination of systolic and diastolic blood pressure. For a fixed diastolic blood pressure, the effect of systolic blood pressure can be evaluated by following the risk surface along lines parallel to the systolic blood pressure axis. This effect, known as the conditional effect of systolic blood pressure, can also be interpreted as the effect of increasing pulse pressure by increasing systolic blood pressure and holding diastolic blood pressure constant. Similarly, for a fixed systolic blood pressure, the conditional effect of diastolic blood pressure can be evaluated by following the risk surface along lines parallel to the diastolic blood pressure axis. This effect corresponds to the effect of decreasing pulse pressure by increasing diastolic blood pressure and holding systolic blood pressure constant.

Article Blood Pressure and Mortality

734 4 November 2003 Annals of Internal Medicine Volume 139 • Number 9 www.annals.org

blood pressure implies that increasing pulse pressure by decreasing diastolic blood pressure below 80 mm Hg does not increase risk, while decreasing pulse pressure by in- creasing diastolic blood pressure above 80 mm Hg in- creases risk for death.

In elderly participants, systolic blood pressure also showed essentially a linear increase in risk, while diastolic blood pressure showed a J-shaped relationship for all-cause and CVD mortality (Figure 1). For a fixed diastolic blood pressure, increasing systolic blood pressure (and thus pulse pressure) was associated with increased risk, while for a fixed systolic blood pressure, increasing diastolic blood pressure (and thus decreasing pulse pressure) was associated with decreased risk for diastolic blood pressure below 80 to 90 mm Hg and with increased risk for diastolic blood pressure above 80 to 90 mm Hg.

Marginal Effects of Blood Pressure Components and Threshold Detection

In marginal analysis, participants younger than 65 years of age showed increasing risk for all-cause mortality with increasing systolic, diastolic, and pulse pressure (Fig- ure 2), with no evidence of threshold effects for any of

these blood pressure measures (P � 0.2 for the existence of change points). The risk trends for CVD and all-cause mortality were similar (Figure 2), also without evidence of change points (P � 0.2 for the existence of change points).

The marginal relationship of blood pressure with all- cause mortality differed in participants 65 years of age and older. Although the trend was also linear for systolic blood pressure (P � 0.21 for the existence of a change point), there was a J-shaped pattern for diastolic blood pressure and pulse pressure (Figure 3). The existence of a change point was marginally significant for diastolic blood pressure (P � 0.09) and highly significant for pulse pressure (P � 0.003). The change point estimate for diastolic blood pressure was 79.0 mm Hg (95% CI, 68.6 to 88.7 mm Hg). For pulse pressure, the change point estimate was 41.8 mm Hg (CI, 34.3 to 48.5 mm Hg). There was no evidence of change points for the association of systolic or diastolic blood pressure and CVD mortality (P � 0.2 for both com- parisons). However, for diastolic blood pressure, the non- parametric analysis suggested a flat association below 80 to 90 mm Hg. The relationship of pulse pressure with CVD mortality found a significant change point (P � 0.03), but

Figure 2. Blood pressure and relative risk for all-cause and cardiovascular mortality among participants younger than 65 years of age.

Risk trends were estimated from linear (solid lines) and nonparametric logistic models (dashed lines) adjusted for age, sex, race or ethnicity, smoking, total cholesterol level, body mass index, education, use of antihypertensive medications, and history of diabetes. The bars represent the frequency distribution of systolic, diastolic, and pulse pressure among study participants younger than 65 years of age.

ArticleBlood Pressure and Mortality

www.annals.org 4 November 2003 Annals of Internal Medicine Volume 139 • Number 9 735

it was estimated at a very low level (35.5 mm Hg, corre- sponding to the 2.3rd percentile of the pulse pressure dis- tribution).

DISCUSSION Table 3 summarizes the association of blood pressure

components with mortality for the different analytical models. Systolic blood pressure showed a consistent linear increase in long-term all-cause and CVD mortality in younger and elderly participants in all analyses. The rela- tionship of diastolic blood pressure with mortality was hockey stick–shaped in younger participants but was J- shaped in elderly participants. This dose–response relation- ship was evident when both systolic and diastolic blood pressures were modeled simultaneously, but it was unclear in marginal analyses of diastolic blood pressure. The asso- ciation of pulse pressure with mortality was complex. In joint analyses, increasing pulse pressure by increasing sys- tolic blood pressure was consistently associated with in- creased risk, while increasing pulse pressure by decreasing diastolic blood pressure could be associated with increased risk, decreased risk, or no change in risk depending on age and blood pressure level. In our view, this result does not

support using pulse pressure as a single risk indicator for prognostic and therapeutic decisions.

Our study has several limitations. Blood pressure was based on 3 measurements taken on a single day. Even though we tried to control for short-term variability by estimating each participant’s underlying blood pressure, more measurements taken over a longer period probably would have provided more precise estimates. The effect of these measurement errors on complex risk surfaces is un- known, although there is evidence that measurement error limits detecting threshold effects and biases their location (20). Estimating change points is also limited by the size of the study and the location of the change point, particularly if change points are close to the extremes of the data dis- tribution. We identified a significant change point in the relationship of pulse pressure with CVD mortality at a very low pulse pressure (35.5 mm Hg, the 2.3rd percentile of the pulse pressure distribution), but its clinical implications are uncertain because there are not enough observations with which to reliably estimate the risk trend below the change point. In addition, we did not have follow-up data on blood pressure and medication patterns or data on more newly identified risk factors, which may modify the effect

Figure 3. Blood pressure and relative risk for all-cause and cardiovascular mortality among participants 65 years of age or older.

Risk trends were estimated from linear or transition logistic models (solid lines) and nonparametric logistic models (dashed lines) adjusted for age, sex, race or ethnicity, smoking, total cholesterol level, body mass index, education, use of antihypertensive medications, and history of diabetes. The bars represent the frequency distribution of systolic, diastolic, and pulse pressure among study participants 65 years of age or older.

Article Blood Pressure and Mortality

736 4 November 2003 Annals of Internal Medicine Volume 139 • Number 9 www.annals.org

of blood pressure on mortality. Finally, we also did not have data on nonfatal events, which may show a different rela- tion to blood pressure components than mortality end points.

Evaluating the joint effect of correlated continuous variables is sometimes based on estimating the risk for each combination of prespecified categories of the variables and presenting the results in 3-dimensional bar plots (3). This approach is useful if there are enough cases to define nar- row categories, and it assumes that the cutoffs for the cat- egories correctly identify points of risk change. For com- plex dose–response relationships, such as blood pressure and mortality, smooth surfaces are a flexible alternative that avoid categorization and make no assumption about the shape of the association (8, 9). However, modeling smooth surfaces is a complex process, with several method- ologic alternatives that may affect the results. Furthermore, the risk surface may be difficult to interpret quantitatively (8). In this paper, we used additive surfaces, obtained by fitting simultaneous nonparametric terms for systolic and diastolic blood pressure and adding them to form the smooth surface. Although we did not include an interac- tion term in the model, additive surfaces adequately esti- mated the joint effect of blood pressure components be- cause of their high correlation (8, 9). To avoid overfitting, we restricted the estimation of smooth surfaces to 8 to 9 degrees of freedom, which, in terms of the effective number of parameters, corresponds approximately to a 3- dimensional bar plot with 3 categories for systolic and diastolic blood pressure.

A linear, monotonically increasing association of sys- tolic blood pressure to mortality has been identified in a pooled data analysis of 61 prospective observational studies

totaling 12.7 million person-years at risk (4) and in the 22-year follow-up of 342 815 participants of the Multiple Risk Factor Intervention Trial (3). These results are com- patible with our analysis of the marginal effect of systolic blood pressure in NHANES II and are supported by our analyses of the joint effect of systolic and diastolic blood pressure. In contrast, Port and colleagues (6) reassessed the Framingham data and, contrary to the graded increase in risk obtained from previous analyses of this cohort (2, 21, 22), found no increased risk for death for systolic blood pressure below certain prespecified age- and sex-dependent thresholds. Unfortunately, they used strong model assump- tions (linear–linear splines with the left slope constrained to be 0) with apparently arbitrary location of the points of risk change.

A J-shaped relationship between diastolic blood pres- sure and CVD risk or mortality is evident in some (7, 23) but not all prospective studies in the elderly. Several stud- ies, however, show a flat association with low diastolic blood pressure that may extend up to a diastolic blood pressure of 80 to 85 mm Hg (24, 25). In fact, although the pooled analysis of 61 prospective studies described the marginal association of diastolic blood pressure with coro- nary and stroke mortality as linear (4), their graphs of the marginal effect of diastolic blood pressure are similar to our marginal analysis of diastolic blood pressure in NHANES II—that is, the association of diastolic blood pressure and mortality for low diastolic blood pressure became flat with advancing age. As we have shown in this paper, this flat marginal association may be reflecting a J-shaped relation- ship for diastolic blood pressure once its correlation with systolic blood pressure is accounted for. In addition, a

Table 3. Conclusions of the Association of Blood Pressure Components with Long-Term Mortality for the Different Analytical Models*

Blood Pressure Joint Analysis Marginal Analysis

Parametric Nonparametric

Age � 65 y Systolic blood pressure Linear Linear Linear Diastolic blood pressure Hockey stick–shaped Linear Hockey stick–shaped for all-cause

mortality and linear for cardiovascular mortality

Pulse pressure Increased risk for increasing pulse pressure by increasing systolic blood pressure; decreased risk or no change in risk for increasing pulse pressure by decreasing diastolic blood pressure

Linear Linear

Age � 65 y Systolic blood pressure Linear Linear Linear Diastolic blood pressure J-shaped J-shaped for all-cause mortality

and linear for cardiovascular mortality

J-shaped for all-cause mortality and hockey stick–shaped for cardiovascular mortality

Pulse pressure Increased risk for increasing pulse pressure by increasing systolic blood pressure; increased risk, decreased risk, or no change in risk for increasing pulse pressure by decreasing diastolic blood pressure

J-shaped J-shaped

* The conclusions apply both to all-cause and cardiovascular mortality, unless explicitly stated.

ArticleBlood Pressure and Mortality

www.annals.org 4 November 2003 Annals of Internal Medicine Volume 139 • Number 9 737

meta-analysis of 8 trials reported that diastolic blood pres- sure in elderly control patients with isolated systolic hyper- tension was inversely correlated with mortality (26), a find- ing compatible with the J-shape hypothesis. In this age group, the J-shape phenomenon for diastolic blood pres- sure may be related to the progressive stiffening of the elastic arteries, which reduces subepicardial coronary flow, particularly in the presence of coronary atherosclerosis. In these patients, low diastolic blood pressure may precipitate coronary events and increase mortality.

The short-term, U-shaped association of blood pres- sure with mortality in our data is consistent with previous studies that observe either the same pattern (27) or the disappearance of the increased mortality in the low blood pressure groups after controlling for concurrent illnesses (28). This is also evident in studies that update blood pres- sure measurements periodically instead of using baseline measurements for risk analysis (29, 30). Despite the exclu- sion of participants with self-reported CVD at baseline, our short-term results might be affected by other condi- tions (such as noncardiovascular diseases and frailty) or occult CVD. This short-term, U-shaped relationship is probably another source of variability in the association of blood pressure with mortality across studies.

Our findings may have important clinical conse- quences for risk stratification (Table 4). Several studies have focused on identifying the main predictor of risk among systolic blood pressure, diastolic blood pressure, or pulse pressure. However, both systolic and diastolic blood pressures (and implicitly pulse pressure) must be accounted for to correctly identify the risk surface (3). Moreover, in- terpreting pulse pressure as a standalone measurement is complex: Although reducing pulse pressure by decreasing systolic blood pressure is consistently associated with de- creased risk, the overall effect of pulse pressure will depend on both systolic and diastolic blood pressures. Clinical de- cision making should then be based on combinations of systolic and diastolic blood pressures rather than systolic, diastolic, or pulse pressure alone. With current technology, it is feasible to incorporate these complex risk surfaces into

Framingham-type risk scores to obtain a more accurate risk profile.

The evidence for a monotonic association of systolic blood pressure with mortality is compelling, but the changes in the risk shape and the J-shaped curve for dia- stolic blood pressure with advancing age are still open to debate. We also show that the effect of pulse pressure is complex, and we do not recommend using this measure alone for prognostic or therapeutic decisions. The shape of the association between blood pressure and mortality is critical for individual treatment and public policy strate- gies. Our results indicate that a simultaneous, rather than marginal, analysis of systolic and diastolic blood pressure is needed to properly assess blood pressure–related risk.

From National Center for Epidemiology, Instituto de Salud Carlos III, Madrid, Spain; Universidad Autónoma de Madrid, Madrid, Spain; and Johns Hopkins Medical Institutions, Baltimore, Maryland.

Grant Support: By a grant from the Instituto de Salud Carlos III (EPY 1261/02) (R. Pastor-Barriuso).

Potential Financial Conflicts of Interest: None disclosed.

Requests for Single Reprints: Eliseo Guallar, MD, DrPH, Department of Epidemiology and Welch Center for Prevention, Epidemiology and Clinical Research, Johns Hopkins Medical Institutions, 2024 East Monument Street, Room 2-639, Baltimore, MD 21205; e-mail, eguallar @jhsph.edu.

Current author addresses and author contributions are available at www .annals.org.

References 1. Dart AM, Kingwell BA. Pulse pressure—a review of mechanisms and clinical relevance. J Am Coll Cardiol. 2001;37:975-84. [PMID: 11263624] 2. Franklin SS, Larson MG, Khan SA, Wong ND, Leip EP, Kannel WB, et al. Does the relation of blood pressure to coronary heart disease risk change with aging? The Framingham Heart Study. Circulation. 2001;103:1245-9. [PMID: 11238268] 3. Domanski M, Mitchell G, Pfeffer M, Neaton JD, Norman J, Svendsen K, et al. Pulse pressure and cardiovascular disease-related mortality: follow-up study of the Multiple Risk Factor Intervention Trial (MRFIT). JAMA. 2002;287:2677- 83. [PMID: 12020303] 4. Lewington S, Clarke R, Qizilbash N, Peto R, Collins R. Age-specific rele- vance of usual blood pressure to vascular mortality: a meta-analysis of individual data for one million adults in 61 prospective studies. Lancet. 2002;360:1903-13. [PMID: 12493255] 5. White WB. The systolic blood pressure versus pulse pressure controversy [Editorial]. Am J Cardiol. 2001;87:1278-81. [PMID: 11377354] 6. Port S, Demer L, Jennrich R, Walter D, Garfinkel A. Systolic blood pressure and mortality. Lancet. 2000;355:175-80. [PMID: 10675116] 7. Benetos A, Thomas F, Safar ME, Bean KE, Guize L. Should diastolic and systolic blood pressure be considered for cardiovascular risk evaluation: a study in middle-aged men and women. J Am Coll Cardiol. 2001;37:163-8. [PMID: 11153732] 8. Hastie TJ, Tibshirani RJ. Generalized Additive Models. London: Chapman and Hall; 1990. 9. Harrell FE Jr. Regression Modeling Strategies. New York: Springer-Verlag; 2001. 10. Loria CM, Sempos CT, Vuong C. Plan and operation of the NHANES II

Table 4. Key Summary Points

Systolic blood pressure, diastolic blood pressure, and pulse pressure are highly correlated blood pressure components.

Rather than assessing the effects of each blood pressure component separately, it is more effective and appropriate to evaluate the joint effect of all components on clinical outcomes.

In joint analysis: 1) systolic blood pressure is linearly related to all-cause and cardiovascular mortality at all ages; 2) diastolic blood pressure shows a “flat then increasing” risk trend in younger participants and a J-shaped trend in elderly participants; and 3) increased pulse pressure is associated with increased risk, decreased risk, or no change in risk depending on age and systolic and diastolic blood pressure levels.

The complexity of the association of pulse pressure with mortality limits its use for prognostic or therapeutic decisions.

Article Blood Pressure and Mortality

738 4 November 2003 Annals of Internal Medicine Volume 139 • Number 9 www.annals.org

Mortality Study, 1992. Vital Health Stat 1. 1999;38:1-16. [PMID: 10464470] 11. McDowell A, Engel A, Massey JT, Maurer K. Plan and operation of the Second National Health and Nutrition Examination Survey, 1976-1980. Vital Health Stat 1. 1981;Series 1:1-144. [PMID: 7344293] 12. Pastor-Barriuso R, Guallar E, Coresh J. Transition models for change-point estimation in logistic regression. Stat Med. 2003;22:1141-62. [PMID: 12652559] 13. Greenland S. Dose-response and trend analysis in epidemiology: alternatives to categorical analysis. Epidemiology. 1995;6:356-65. [PMID: 7548341] 14. Goetghebeur E, Pocock SJ. Detection and estimation of J-shaped risk-re- sponse relationships. Journal of the Royal Statistical Society A. 1995;158:107-21. 15. Pastor R, Guallar E. Use of two-segmented logistic regression to estimate change-points in epidemiologic studies. Am J Epidemiol. 1998;148:631-42. [PMID: 9778169] 16. MacMahon S, Peto R, Cutler J, Collins R, Sorlie P, Neaton J, et al. Blood pressure, stroke, and coronary heart disease. Part 1, Prolonged differences in blood pressure: prospective observational studies corrected for the regression di- lution bias. Lancet. 1990;335:765-74. [PMID: 1969518] 17. Hughes MD, Pocock SJ. Within-subject diastolic blood pressure variability: implications for risk assessment and screening. J Clin Epidemiol. 1992;45:985- 98. [PMID: 1432027] 18. Clarke R, Shipley M, Lewington S, Youngman L, Collins R, Marmot M, et al. Underestimation of risk associations due to regression dilution in long-term follow-up of prospective studies. Am J Epidemiol. 1999;150:341-53. [PMID: 10453810] 19. Mathsoft. S-Plus 2000 User’s Guide. Seattle: Mathsoft; 1999. 20. Küchenhoff H, Carroll RJ. Segmented regression with errors in predictors: semi-parametric and parametric methods. Stat Med. 1997;16:169-88. [PMID: 9004390] 21. Kannel WB, Gordon T, Schwartz MJ. Systolic versus diastolic blood pres- sure and risk of coronary heart disease. The Framingham study. Am J Cardiol.

1971;27:335-46. [PMID: 5572576]

22. Kannel WB. Blood pressure as a cardiovascular risk factor: prevention and treatment. JAMA. 1996;275:1571-6. [PMID: 8622248]

23. Taylor JO, Cornoni-Huntley J, Curb JD, Manton KG, Ostfeld AM, Scherr P, et al. Blood pressure and mortality risk in the elderly. Am J Epidemiol. 1991; 134:489-501. [PMID: 1897505]

24. Kannel WB, D’Agostino RB, Silbershatz H. Blood pressure and cardiovas- cular morbidity and mortality rates in the elderly. Am Heart J. 1997;134:758-63. [PMID: 9351745]

25. Psaty BM, Furberg CD, Kuller LH, Cushman M, Savage PJ, Levine D, et al. Association between blood pressure level and the risk of myocardial infarction, stroke, and total mortality: the cardiovascular health study. Arch Intern Med. 2001;161:1183-92. [PMID: 11343441]

26. Staessen JA, Gasowski J, Wang JG, Thijs L, Den Hond E, Boissel JP, et al. Risks of untreated and treated isolated systolic hypertension in the elderly: meta- analysis of outcome trials. Lancet. 2000;355:865-72. [PMID: 10752701]

27. Vatten LJ, Holmen J, Krüger O, Forsén L, Tverdal A. Low blood pressure and mortality in the elderly: a 6-year follow-up of 18,022 Norwegian men and women age 65 years and older. Epidemiology. 1995;6:70-3. [PMID: 7888450]

28. Guo Z, Viitanen M, Winblad B. Low blood pressure and five-year mortality in a Stockholm cohort of the very old: possible confounding by cognitive impair- ment and other factors. Am J Public Health. 1997;87:623-8. [PMID: 9146442]

29. Hansson L, Zanchetti A, Carruthers SG, Dahlöf B, Elmfeldt D, Julius S, et al. Effects of intensive blood-pressure lowering and low-dose aspirin in patients with hypertension: principal results of the Hypertension Optimal Treatment (HOT) randomised trial. HOT Study Group. Lancet. 1998;351:1755-62. [PMID: 9635947]

30. Boutitie F, Gueyffier F, Pocock S, Fagard R, Boissel JP. J-shaped relation- ship between blood pressure and mortality in hypertensive patients: new insights from a meta-analysis of individual-patient data. Ann Intern Med. 2002;136:438- 48. [PMID: 11900496]

ArticleBlood Pressure and Mortality

www.annals.org 4 November 2003 Annals of Internal Medicine Volume 139 • Number 9 739

Current Author Addresses: Drs. Pastor-Barriuso and Damián: Epide- miology and Biostatistics Section, National Center for Epidemiology, Instituto de Salud Carlos III, Sinesio Delgado 6, 28029 Madrid, Spain. Dr. Banegas: Department of Preventive Medicine and Public Health, Universidad Autónoma de Madrid, 28029 Madrid, Spain. Dr. Appel: Welch Center for Prevention, Epidemiology and Clinical Research, Johns Hopkins Medical Institutions, 2024 East Monument Street, Room 2-630, Baltimore, MD 21205. Dr. Guallar: Welch Center for Prevention, Epidemiology and Clinical Research, Johns Hopkins Medical Institutions. 2024 East Monument Street, Room 2-639, Baltimore, MD 21205.

Author Contributions: Conception and design: R. Pastor-Barriuso, J.R. Banegas, J. Damián, E. Guallar. Analysis and interpretation of the data: R. Pastor-Barriuso, J.R. Banegas, J. Damián, L.J. Appel, E. Guallar. Drafting of the article: R. Pastor-Barriuso, J.R. Banegas, J. Damián, L.J. Appel, E. Guallar. Critical revision of the article for important intellectual content: R. Pas- tor-Barriuso, J.R. Banegas, J. Damián, L.J. Appel, E. Guallar. Final approval of the article: R. Pastor-Barriuso, J.R. Banegas, J. Damián, L.J. Appel, E. Guallar. Statistical expertise: R. Pastor-Barriuso, E. Guallar. Obtaining of funding: R. Pastor-Barriuso. Administrative, technical, or logistic support: E. Guallar.

E-740 Annals of Internal Medicine Volume • Number www.annals.org