Klassen PS, Lowrie EG, Reddan DN, DeLong ER, Coladonato JA, Szczech LA, Lazarus JM, Owen, Jr WF. Association Between Pulse Pressure and Mortality in Patients Undergoing
Maintenance Hemodialysis. JAMA. 2002;287(12):1548-1555. doi:10.1001/jama.287.12.1548
Author Affiliations: Duke Institute of Renal Outcomes Research and Health Policy, Department of Medicine, Duke University Medical Center (Drs Klassen, Lowrie, Reddan, Coladonato, Szczech, and Owen) and Duke Clinical Research Institute (Dr DeLong), Durham, NC; and Fresenius Medical Care, North America, Lexington, Mass (Drs Lowrie and Lazarus); Dr Owen now works at Baxter Healthcare International, Waukegan, Ill.
Context Although increased blood pressure is associated with adverse outcomes
in the general population, elevated blood pressure is associated with decreased
mortality in patients with end-stage renal disease undergoing maintenance
hemodialysis. Recent investigations in the general population have demonstrated
the predictive utility of pulse pressure (systolic minus diastolic blood pressure),
a measure reflecting the pulsatile nature of the cardiac cycle.
Objectives To estimate the relationship between pulse pressure and mortality in
patients undergoing maintenance hemodialysis and to test our hypothesis that
an increasing pulse pressure would be associated with increased risk of death
up to 1 year despite the inverse relationship between conventional blood pressure
measures and mortality in patients with end-stage renal disease.
Design, Setting, and Patients Retrospective cohort investigation of patients with end-stage renal
disease undergoing maintenance hemodialysis at 782 hemodialysis facilities
throughout the United States. Of 44 069 eligible patients as of January
1, 1998, 37 069 with complete demographic data were included in the analyses
of clinical and laboratory data collected from October 1 through December
31, 1997. Patients were followed up through December 31, 1998.
Main Outcome Measures The primary study outcome was death at 1 year. A secondary outcome was
the magnitude of the pulse pressure.
Results The final patient cohort was similar to national averages with respect
to age, sex, race, and diabetic status. Mean (SD) pulse pressures before dialysis
were 75.0 (15.0) mm Hg and 66.9 (13.9) mm Hg after dialysis. By the end of
the 1-year follow-up, 5731 patients (18.4%) died. After adjusting for level
of systolic blood pressure, multivariable Cox proportional hazards modeling
showed a direct and consistent relationship between increasing pulse pressure
and increasing death risk. Each incremental elevation of 10 mm Hg in postdialysis
pulse pressure was associated with a 12% increase in the hazard for death
(hazard ratio, 1.12; 95% confidence interval, 1.06-1.18). Postdialysis systolic
blood pressure was inversely related to mortality with a 13% decreased hazard
for death for each incremental elevation of 10 mm Hg (hazard ratio, 0.87;
95% confidence interval, 0.84-0.90). In a multivariable linear regression
model, important variables directly associated with elevated pulse pressure
included age, diabetes, white race, female sex, and number of years receiving
dialysis (all P<.001).
Conclusions Pulse pressure is associated with risk of death in a large, nationally
representative sample of patients undergoing maintenance hemodialysis. The
recognition of pulse pressure as an important correlate of mortality in patients
receiving dialysis highlights the need to investigate the relationship between
potential therapeutic implications of conduit vessel function and clinical
outcomes in patients with end-stage renal disease.
Hypertension is an important predictor of subsequent adverse clinical
outcomes in the general population.1- 3
Information from cross-sectional and longitudinal cohort studies as well as
interventional trials has been used to identify normal and optimal levels
of blood pressure for healthy adults and specific at-risk populations, including
patients with chronic kidney disease.1,3- 7
Application of such outcome-derived management criteria to patients with end-stage
renal disease (ESRD) treated by maintenance dialysis is hindered by counterintuitive
epidemiological relationships between blood pressure and clinical events.
A number of observational cohort studies have reported U-shaped or reverse-J
relationships between conventional blood pressure measures (systolic, diastolic,
and mean arterial) and mortality in patients undergoing hemodialysis.8- 11 These
investigations report that patients receiving dialysis with systolic blood
pressures between 140 and 175 mm Hg have improved survival compared with those
whose systolic pressures are lower than 140 mm Hg.
Investigations of hypertension generally focus on the steady components
of blood pressure, such as mean arterial pressure (MAP), which is determined
by cardiac output and peripheral vascular resistance.12- 14
However, there is increasing evidence that the oscillating pulsatile nature
of the cardiac cycle can also provide important information about the cardiovascular
risk conferred by hypertension, particularly in middle-aged and elderly populations.15- 19
This pulsatile component of blood pressure is governed by the relationship
between ventricular ejection and viscoelastic properties of large arteries
(arterial stiffness), as well as the indirect effect of arterial wave reflection
from the periphery back to central conduit arteries. Pulse pressure (systolic
minus diastolic blood pressures) is an index of the pulsatile component of
the cardiac cycle.14 Data from large prospective
cohort and interventional trials have been used to investigate relationships
between pulse pressure and clinical events in non-ESRD populations. These
studies have correlated increased pulse pressure with heart failure, myocardial
infarction, and all-cause and cardiovascular death.15,17,18,20,21
Patients with ESRD exhibit vascular abnormalities that contribute to
elevated pulse pressure, including increased arterial stiffness, pulse-wave
velocity, and early wave reflection.22,23
These vascular abnormalities have been associated with all-cause and cardiovascular
mortality in small studies using noninvasive vascular imaging and continuous
ambulatory blood pressure monitoring.24,25
Pulse pressure has been related to all-cause and cardiovascular mortality
in nondiabetic Japanese patients receiving dialysis.26
The relatively small population in that study did not exhibit the inverse
relationship between systolic blood pressure and death that is typically seen
in large dialysis cohorts in the United States. Given that epidemiological
associations of conventional blood pressure measures with clinical events
appear to be different between ESRD patients in the United States and the
general population, this study was conducted to examine the relationship between
pulse pressure and mortality in a large, national sample of patients undergoing
The study population included 44 069 patients undergoing maintenance
hemodialysis during 1998 at 782 facilities operated by Fresenius Medical Care,
North America, based in Lexington, Mass.27
Patients with complete data for sex, race, diabetic status, and body size
measures (height and weight) were included (n = 37 069).
Demographic, clinical, and laboratory data were collected on all patients
during the months of October, November, and December 1997. Repeated measures,
including blood pressure, dialysis dose, and laboratory measurements were
averaged over the entire 3 months of data collection. Patients were followed
up until December 31, 1998. Patients who left the facilities or received a
kidney transplantation were censored. Date of death or censoring was recorded
for time-to-event analysis of all-cause mortality.
Demographic and clinical parameters recorded included age, sex, race,
diabetic status (defined either as a comorbid condition or as the etiology
of ESRD), years receiving dialysis prior to entry into the study, height,
and weight before and after dialysis treatment initiation. Dose of hemodialysis
was quantified as the fractional decline in blood urea nitrogen concentration
during a single dialysis treatment (urea reduction ratio). The volume of fluid
removed during dialysis was expressed as a percentage of postdialysis weight
(intradialytic volume change). Systolic and diastolic blood pressures were
recorded both before and after dialysis sessions and were measured, by trained
health care professionals or automated monitors, with patients in a sitting
position. Pulse pressure was quantified as the difference between average
systolic and average diastolic pressures over the data collection period.
Laboratory data included predialysis hematocrit, albumin, creatinine, calcium,
phosphorous, parathyroid hormone (PTH), and ferritin concentrations. All measurements
were determined in a single laboratory (Spectra Laboratories, Rockleigh, NJ).
The clinical and demographic characteristics including measures of systolic,
diastolic, and pulse pressure were described for the cohort overall. Univariate
analyses estimated the associations between pulse pressure and demographic,
clinical, and laboratory parameters, using the t
test and linear regression for categorical and continuous parameters, respectively.
To determine independent associations, multivariable linear regression estimated
the relationship between pulse pressure and the parameters described above.
Associations with mortality to 1 year were estimated using Cox proportional
hazards regression. To confirm that this cohort exhibited the anticipated
inverse relationship between conventional blood pressure measures and mortality,
the association between systolic blood pressure and death was estimated in
an unadjusted model. Pulse pressure was subsequently examined in both unadjusted
and adjusted models built using stepwise methods. Models were also created
to examine other single–blood pressure components and dual–blood
pressure component pairs. Interactions between pulse pressure and age, sex,
race, diabetic status, and level of systolic blood pressure were tested. Finally,
individual subjects were categorized according to level of both systolic blood
pressure and pulse pressure. The percentage of subjects dead at 1 year was
calculated for each category, and the results were graphically displayed.
All P values are 2-sided. Analyses were performed
using SAS software (version 8.1, SAS Institute, Cary, NC).
Of the 37 069 patients included in the analyses, 31 176 completed
1 year of follow-up: 5893 subjects were censored during follow-up because
of either kidney transplantation, transfer to a non-FMC dialysis unit, or
loss to follow-up. Censored patients were more likely to be younger, white,
nondiabetic, and to have been receiving dialysis for fewer years. Demographic
and clinical characteristics of patients in the study population and excluded
subjects are described in Table 1.
During the follow-up year, 5731 patients (18.4%) died. The baseline characteristics
and death rate of the study population are similar to those for the US hemodialysis
population as reported by the United States Renal Data Service and the Centers
for Medicare and Medicaid Services.28,29
The study population had mean (SD) systolic blood pressures of 154.3
(20.4) mm Hg before and 139.6 (19.1) mm Hg after dialysis treatment. Diastolic
blood pressures were 79.3 (11.2) mm Hg before and 72.7 (10.1) mm Hg after
dialysis treatment. Predialysis pulse pressure was 75.0 (15.0) mm Hg and the
postdialysis pulse pressure was 66.9 (13.9) mm Hg.
Results of univariate and multivariable analyses were similar for pulse
pressure before and after dialysis. Univariate analysis showed that postdialysis
pulse pressure rose with increasing age, from 61.7 mm Hg in the lowest quartile
of age to 70.2 mm Hg in the highest quartile; pulse pressure increased by
2.2 mm Hg for every 10-year increase in age (P<.001).
The mean pulse pressure for women was higher than for men (68.7 vs 64.7 mm
Hg, P<.001). Although whites had lower mean systolic
blood pressure than blacks (138.1 vs 141.5 mm Hg, P<.001),
whites also had lower mean diastolic blood pressure (70.7 vs 75.1 mm Hg, P<.001), resulting in a slightly higher mean pulse pressure
in whites than blacks (67.4 vs 66.4 mm Hg, P<.001).
A higher pulse pressure was seen among patients with diabetes mellitus compared
with patients without (70.0 vs 64.0 mm Hg, P<.001).
There was an inverse relationship between duration of dialytic therapy and
pulse pressure, with a drop of 2.7 mm Hg for every 5 years a patient was receiving
dialysis prior to entry into the study (P<.001).
Hematocrit levels were also inversely associated with pulse pressure, with
an increase of 1.7 mm Hg for every 3% drop in hematocrit levels (P<.001). Other univariate associations (body weight; height; intradialytic
volume change; and albumin, calcium, phosphorous and parathyroid hormone levels)
were statistically significant but each contributed less than 1% to the variability
in pulse pressure. Serum ferritin levels did not correlate significantly with
pulse pressure. Variability in pulse pressure between patients was driven
primarily by variations in systolic pressure, evidenced by a large correlation
coefficient for the relationship between systolic and pulse pressure (r2 = 0.73, P<.001)
compared with the relationship between diastolic and pulse pressure (r2 = 0.06, P<.001).
Because systolic blood pressure correlates highly with pulse pressure
in univariate analysis, the level of systolic blood pressure may influence
the associations between pulse pressure and other parameters. Systolic blood
pressure was therefore included as a covariate in a multivariable regression
analysis of pulse pressure (Table 2).
After adjustment for level of postdialysis systolic blood pressure, important
demographic and clinical variables directly associated with elevated pulse
pressure were age (2.1 mm Hg per 10-year increase), the presence of diabetes
mellitus (2.6 mm Hg), white race (2.2 mm Hg), female sex (1.8 mm Hg), and
duration of dialysis prior to entry into the study (0.5 mm Hg per 5-year increase)
(all P<.001). Pulse pressure was also directly
associated with increasing phosphorous (P<.001)
and calcium (P = .04). Pulse pressure was inversely
associated with height, hematocrit, and albumin levels (P<.001). Body weight, percentage of intradialytic volume change,
and ferritin and serum phosphorous levels did not significantly correlate
with pulse pressure in the multivariable model. Multivariable analysis of
predialysis pulse pressure showed similar results.
The results of an unadjusted Cox proportional hazards model relating
systolic blood pressure before and after dialysis to 1-year mortality are
shown in Figure 1. Subjects with
systolic pressures lower than the reference range of 135 to 144 mm Hg had
greater mortality risk; those with systolic pressures higher than the reference
range had lower mortality risk. Similar to a previous report8
of another hemodialysis population, postdialysis systolic blood pressure exhibited
more of a U-shape or Reverse-J curve than predialysis systolic pressure. The
unadjusted Cox model relating pulse pressure before and after dialysis to
1-year mortality is shown in Figure 1B.
The curves for unadjusted pulse pressure were similar to the systolic blood
pressure curves; postdialysis pulse pressure exhibited a more pronounced U-shape.
Compared with subjects in the reference category of 50 to 59 mm Hg, patients
with higher postdialysis pulse pressures experienced increasing mortality.
Because pulse pressure has a high correlation with systolic blood pressure,
examining the independent association between pulse pressure and mortality
requires an adjustment for level of systolic pressure. After controlling for
the level of systolic blood pressure, a direct and consistent association
is seen between increasing pulse pressure and increasing risk for mortality
(Figure 2A). In this model pulse
pressure is a function solely of diastolic blood pressure. For any given level
of systolic pressure, lower diastolic blood pressure is associated with increased
risk of death (Figure 2B).
The results of a full mortality model adjusting for other variables
known to influence either pulse pressure or death are shown in Table 3. After adjustment for systolic blood pressure and 14 other
demographic and clinical parameters, postdialysis pulse pressure remained
significantly associated with mortality to 1 year; each incremental elevation
of 10 mm Hg in pulse pressure was associated with a 12% increased hazard of
death (95% CI, 1.06-1.18; P<.001). Systolic blood
pressure was inversely related to mortality (13% decreased death hazard for
each elevation of 10 mm Hg in systolic pressure; 95% CI, 0.84-0.91; P<.001). Mortality risk increased with age (hazard ratio
[HR], 1.30; 95% CI, 1.26-1.33), years receiving dialysis (HR, 1.04; 95% CI,
1.02-1.06), and the presence of diabetes mellitus (HR, 1.42; 95% CI, 1.32-1.51)
(all P<.001). Mortality decreased with higher
albumin levels (HR, 0.33; 95% CI, 0.31-0.36), weight (HR, 0.92; 95% CI, 0.91-0.93),
and hematocrit levels (HR, 0.93; 95% CI, 0.90-0.96) (all P<.001). Female sex (HR, 0.80; 95% CI, 0.74-0.85), and black race
(HR, 0.82; 95% CI, 0.77-0.88) were also associated with lower mortality (both P<.001). Multivariable analysis of predialysis pulse
pressure showed similar relationships (pulse pressure HR, 1.07; 95% CI, 1.02-1.13).
To further examine the relationships between postdialysis blood pressure
components and mortality, each was entered into the adjusted mortality model
as single and paired covariates (Table 4). When examined as isolated variables, systolic, diastolic, and
pulse pressure are each inversely associated with mortality (models 1-3).
Adjusting for systolic blood pressure changes the association between pulse
pressure and death to a direct relationship (model 4). Adjusting for diastolic
blood pressure causes both pulse pressure (model 5) and systolic pressure
(model 6) to become nonsignificant contributors to the mortality models (Table 4).
The risk attributed to pulse pressure in the full mortality model differed
among individuals with elevated and nonelevated blood pressure (P<.001 for the interaction term between systolic pressure and pulse
pressure). Subjects were then divided into 5 categories of systolic blood
pressure: <120, 120 to 139, 140 to 159, 160 to 179, and ≥180 mm Hg.
The full mortality model was tested for each blood pressure category and the
results are shown in Table 5.
Pulse pressure was significantly associated with mortality only in subjects
with systolic pressures less than or equal to 140 mm Hg. The relationship
between systolic blood pressure, pulse pressure, and death is graphically
demonstrated in Figure 3. As pulse
pressure increased within each category of systolic blood pressure, the percentage
of subjects who died at 1 year also increased. As systolic blood pressure
increased within each category of pulse pressure, the death percentages decreased
until pressures higher than 165 mm Hg, at which point some groups displayed
an increase in death (reverse-J curve).
The interaction between pulse pressure and age was significant (P<.001). The risk associated with pulse pressure in
the younger half of the cohort (age <62 years) was approximately 2 times
the risk associated with pulse pressure in the older half (HR, 1.24 vs 1.12).
Similar interactions for race, diabetes mellitus, and sex were tested but
did not meet statistical significance.
To our knowledge, these analyses are the first to demonstrate the association
between pulse pressure and an increased death risk in a large, Western national,
representative sample of patients receiving hemodialysis. In an adjusted model,
an incremental increase of 10 mm Hg in postdialysis pulse pressure was associated
with a 12% increase in the hazard for death. The amount of variability in
mortality accounted for by pulse pressure was similar to that seen for race,
hematocrit level, years receiving dialysis, or parathyroid hormone level.
The association between pulse pressure and mortality was examined both with
and without adjustment for level of systolic blood pressure. Since variability
of pulse pressure in the study population is largely accounted for by variation
in systolic blood pressure, a model examining mortality and pulse pressure
alone could be confounded by the independent relationship between mortality
and systolic pressure. By adjusting for systolic pressure in the final model,
the question effectively posed is, for a given level of systolic blood pressure,
does the magnitude of the pulse pressure matter? These results suggest that
it does. Moreover, although pulse pressure is more likely to be elevated in
patients with elevated systolic blood pressure, the risk attributed to pulse
pressure is actually driven by patients with systolic pressures below 140
As Figure 2B illustrates,
when adjusting for level of systolic blood pressure the magnitude of the pulse
pressure is determined solely by the diastolic blood pressure. In essence,
this indicates that for any given level of systolic blood pressure, the lower
the diastolic pressure, the greater the risk of death. When systolic and diastolic
blood pressure are jointly entered into a mortality model, it is diastolic
blood pressure, and not systolic, that is significantly associated with death
(model 6 in Table 5).
Pulse pressure is a simple mathematical combination of systolic and
diastolic blood pressure and is therefore an easily measured correlate of
pulsatile hemodynamic load during the cardiac cycle. It reflects dynamic stress
that is caused by large-artery stiffness and early, reflected arterial pressure
waves from the periphery back to central, conduit vessels. Early return of
reflected arterial waves increases afterload at end-systole and decreases
coronary perfusion pressure during diastole.30
This may be particularly important when underlying cardiac dysfunction is
present,31 as is often the case in ESRD.32,33 Studies in non-ESRD populations have
shown that death and cardiac disease may be related more to pulsatile stress
than to the steady-state stress that is associated with small-vessel resistance
and reflected in static measures of blood pressure, such as systolic, diastolic,
or mean arterial pressure.19,21
End-stage renal disease is associated with elevated arterial stiffness,
pulse-wave velocity, and early wave reflections compared with age- and blood
pressure–matched controls with normal renal function.22,23
The mechanisms underlying these changes are incompletely described but may
be a reflection of abnormal vascular biology seen in chronic kidney disease.
The atherogenic environment of chronic kidney disease includes contributions
from hypertension, hypervolemia, qualitative and quantitative lipid abnormalities,
divalent ion changes, aberrant inflammatory responses, and hyperhomocysteinemia,
among other putative disturbances.34 Pulse
pressure was extremely high in the majority of patients undergoing hemodialysis
in this current cohort, with fewer than 10% of subjects having pulse pressures
lower than 50 mm Hg, the mean value found in population-based samples.35 The pathobiologic correlate of this finding is that
the preponderance of patients undergoing hemodialysis may experience a substantial
burden of increased afterload and decreased coronary perfusion pressure. It
is conceivable that this may contribute to the high prevalence of left ventricular
hypertrophy and cardiovascular death seen in the ESRD population.
Static measures of blood pressure in patients with ESRD have relationships
with mortality that appear to be quite different than in the general population.
Epidemiologic investigations have reported that higher blood pressures are
associated with improved survival compared with lower pressures.8- 11
A potential explanation for this phenomenon is that cardiac dysfunction in
the ESRD population confers an increased risk of death and is reflected by
normal or reduced blood pressures. This may partially explain the finding
that the risk associated with pulse pressure is significant and of greatest
magnitude in subjects with blood pressure in the normal range, lower than
140 mm Hg. In its simplest form, pulse pressure may serve as an additional
risk-stratification tool to identify patients with ESRD at high risk for death
in a more clinically comprehensible form than what systolic or diastolic blood
pressure measurements alone would indicate. Further studies are required to
determine the efficacy of using pulse pressure and other measures of conduit
vessel function as predictors of risk and as targets for primary and secondary
The therapeutic implications of this association between pulse pressure
and mortality are difficult to define because of the observational nature
of the study, which limits implications of causality. However, abnormalities
in conduit vessel stiffness are modifiable by exercise and reduced sodium
intake.36,37 Similarly, antihypertensive
agents have differential effects on vascular compliance. In particular, angiotensin-converting
enzyme inhibition and diuretics decrease arterial wave reflection and pulse-wave
velocity and increase arterial compliance in certain populations.38- 40 β-Blocker monotherapy
may increase vessel stiffness and magnitude of arterial reflected waves.41,42 Calcium channel blockade produces
varying results.43,44 Further
studies are needed to determine (1) the effect of diet and/or pharmacotherapy
on conduit vessel compliance and (2) whether targeted modification of vessel
compliance produces improvements in clinical outcomes for patients with ESRD.
The principle strength of this investigation is that the study population
is a large, contemporary, and representative sample of hemodialysis patients
in the United States. Multiple blood pressures were collected to represent
true blood pressure load and minimize outlying values. A large number of demographic,
clinical, and laboratory parameters were available, allowing appropriate adjustments
to isolate the relationships between blood pressure and mortality.
This investigation is limited by several design issues germane to observational
registry research. First, this study was limited to data collected during
routine provision of hemodialysis, and blood pressure measurements were not
carried out using standardized protocols. This limitation is offset by the
large number of observations recorded for each subject (approximately 36 measurements
for times before and after dialysis). Moreover, the large number of participants
minimizes bias introduced by a limited number of treatment centers or providers
collecting blood pressure measurements. Second, in this analysis, we have
assumed that pulse pressure is reflective of decreased vascular compliance.
Because increased pulse pressure may be observed in other clinical conditions,
such as aortic valve insufficiency, this assumption may not be true in some
patients. Moreover, it is unclear what effect arteriovenous vascular anastomoses
like autologous fistulae or prosthetic grafts, which are the principle form
of hemodialysis access, have on pulse pressure. Third, pulse pressure is a
crude measure of conduit vessel function compared with more direct measures
such as pulse-wave velocity, proximal aortic compliance, characteristic impedance,
and wave-form morphology. Although such measurements may better estimate the
extent of the relationship between vessel characteristics and clinical outcomes,
the feasibility of these investigations is often limited by their resource
intensity, particularly in large study populations. Finally, this investigation
was unable to precisely characterize the causes of mortality into cardiovascular
vs noncardiovascular etiologies. Despite this limitation, the size of this
data set and the representative nature of the final patient cohort make it
unlikely that the causes of death differ from US ESRD population. Based on
national ESRD registry information, cardiovascular disease accounts for more
than half the reported deaths in the American ESRD patients.28
Pulse pressure is associated with mortality in patients undergoing maintenance
hemodialysis. Considered individually, both systolic and diastolic blood pressure
have counterintuitive, inverse associations with mortality. This may be the
result of widespread cardiac dysfunction in the ESRD population. It may also
point to the need for careful consideration of what constitutes appropriate
levels of blood pressure control. Addressing these questions will likely require
prospective, interventional trials examining different levels of blood pressure
control. Similarly, the recognition of pulse pressure as an important correlate
of mortality requires prospective confirmation and should be incorporated
in future interventional trials to investigate the chain of causality between
conduit vessel function and both cardiovascular events and all-cause death.