Association Between Long-term Blood Pressure Variability and Mortality Among Incident Hemodialysis Patients

Dialysis Association Between Long-term Blood Pressure Variability and Mortality Among Incident Hemodialysis Patients Steven M. Brunelli, MD, MSCE,1,2 Ravi I. Thadhani, MD, MPH,3 Katherine E. Lynch, BA,1 Elizabeth D. Ankers, BA,3 Marshall M. Joffe, MD, PhD,2 Raymond Boston, PhD,2 Yuchaio Chang, PhD,3 and Harold I. Feldman, MD, MSCE1,2 Background: Blood pressure variability (BPV) is one putative risk factor for cardiovascular disease and mortality in hemodialysis patients. The purposes of this study are to identify a suitable metric of long-term BPV in this population and determine whether an association between BPV and all-cause mortality exists. Study Design: Retrospective cohort study. Settings & Participants: Patients from the Accelerated Mortality on Renal Replacement (ArMORR) cohort who were adult, incident to hemodialysis at any Fresenius Medical Care unit between June 2004 and August 2005, and had suitable blood pressure data were studied (n ⫽ 6,961). Predictor: Predialysis blood pressures measured between dialysis days 91 and 180 were used to determine each patient’s absolute level of, trend in (slope over time), and variability in blood pressure. Outcome: All-cause mortality beginning immediately after day 180 and continuing through day 365 or until censoring (median follow-up, 185 days). Results: Of the 4 candidate BPV metrics, only average residual-intercept ratio adequately distinguished BPV from absolute blood pressure level and temporal blood pressure trend. In the primary analysis, each SD increase in systolic and diastolic BPV was associated with adjusted hazard ratios for all-cause mortality of 1.13 (95% confidence interval, 1.03 to 1.23) and 1.15 (95% confidence interval, 1.06 to 1.26), respectively. Results were consistent across multiple sensitivity analyses in which inclusion and exclusion criteria and timing of blood pressure measurements were varied. Limitations: Contingency of results on the validity of mathematic description of BPV; potential for misclassification bias and residual confounding. Conclusions: Provided the mathematical descriptions of BPV are valid, the data suggest that systolic and diastolic BPV is associated with all-cause mortality in incident hemodialysis patients. Additional study is necessary to confirm and generalize findings, assess the interplay between systolic and diastolic BPV, and assess causality. Am J Kidney Dis 52:716-726. © 2008 by the National Kidney Foundation, Inc. INDEX WORDS: Hemodialysis; blood pressure; blood pressure variability; survival.

emodialysis (HD) patients have exceedingly high rates of mortality and, in particular, mortality from cardiovascular (CV) causes.1-6 Blood pressure (BP) variability (BPV) is one putative risk factor for CV disease. In individuals without chronic kidney disease, higher BPV has been associated with CV mor-

bidity and mortality.7-13 Data from animal models support the interpretation that this association may be causal.14-16 Clinical experience dictates that HD patients have greater degrees of BPV than individuals not on HD therapy, suggesting that BPV may contribute to their burden of CV disease. In individuals not on long-term HD therapy, BPV most often is measured using 24-hour ambulatory BP monitoring. However, the study of

1 From the Renal, Electrolyte, and Hypertension Division and 2Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania School of Medicine, Philadelphia, PA; and 3Renal Unit, Massachusetts General Hospital, Harvard Medical School, Boston, MA. Received December 30, 2007. Accepted in revised form April 29, 2008. Originally published online as doi: 10.1053/j.ajkd.2008.04.032 on August 27, 2008. Because an author of this manuscript is an editor for AJKD, the peer-review and decision-making processes were handled entirely by an Associate Editor (Tazeen H. Jafar,

MD, MPH, Aga Khan University) who served as Acting Editor-in-Chief. Details of the journal’s procedures for potential editor conflicts are given in the Editorial Policies section of the AJKD website. Address correspondence to Steven M. Brunelli, MD, MSCE, Center for Clinical Epidemiology and Biostatistics, 109 Blockley Hall, 423 Guardian Dr, Philadelphia, PA 19104. E-mail: [email protected] © 2008 by the National Kidney Foundation, Inc. 0272-6386/08/5204-0013$34.00/0 doi:10.1053/j.ajkd.2008.04.032

Editorial, p. 638

H

716

American Journal of Kidney Diseases, Vol 52, No 4 (October), 2008: pp 716-726

Blood Pressure Variability in Dialysis

BPV in HD patients is complex, given fluctuations that occur during both the diurnal period, as would be captured by ambulatory BP monitoring, and longer periods (such as would result from the combination of waxing and waning states of volume overload, reduced vascular compliance,17,18 and seasonal effects19) that would not. Whereas the effects of shorter term BPV on CV disease have been extensively studied in HD patients,20-23 the effects of longer-term BPV are less well characterized.24 The study of BPV is complicated further because absolute level of BP and temporal trend in BP may each be associated with BPV and with CV disease independent of BPV (ie, serve as confounders).25-29 Therefore, an ideal metric of BPV would be one that measures variability independent of absolute level of and temporal trend in BP. Preliminary evidence suggests that long-term BPV (measured by using the the coefficient of variation of predialysis BPs) is associated with CV and all-cause mortality in HD patients.24 Our pilot work suggests that the coefficient of variation may not be independent of temporal BP trend, and thus the independent association between BPV and outcome is unknown. Therefore, we undertook the present study with a 2-fold purpose: (1) to identify a metric that adequately measures long-term BPV independently of absolute BP level and temporal BP trend, and (2) to determine whether an association between BPV and mortality exists in HD patients.

METHODS The study protocol was approved by the University of Pennsylvania Institutional Review Board, Philadelphia, PA.

Study Design and Setting These analyses were performed retrospectively using patients and data from the Accelerated Mortality on Renal Replacement (ArMORR) cohort; details of the cohort have been previously published.30 ArMORR consists of incident HD patients who began treatment at any Fresenius Medical Care unit between June 2004 and August 2005. Demographic, comorbidity, laboratory, and outcome data were collected at baseline and longitudinally.30 In the present analyses, patients younger than 18 years (n ⫽ 14) were excluded from all analyses.

Participants In the primary analysis, BPs were considered during dialysis days 91 through 180 (the BP exposure widow) to

717

Figure 1. Flow diagram for enrollment in primary cohort and sensitivity analysis 1. Abbreviations: HD, hemodialysis; FMC, Fresenius Medical Care; BP, blood pressure.

allow sufficient time for patients to “acclimate” to dialysis therapy. (Time zero corresponds to a patient’s first day of HD therapy.) Thus, at-risk time began immediately after day 180. To be eligible for inclusion in the primary cohort, patients had to have maintained continuous enrollment at a participating unit until at-risk time began; patients dying, discontinuing HD therapy, or transferring care before day 181 therefore were excluded from the primary cohort. In addition, because the number and timing of BP measurements affects the precision with which BP trend (and to a lesser degree, BPV) are estimable, the primary analysis considered only patients with at least 1 documented BP value in each month of the BP exposure window (ie, at least 1 BP measurement between days 91 and 120, 121 and 150, and 151 and 180; Fig 1). At-risk time concluded upon a patient’s death, transfer of care, discontinuation of dialysis therapy, or as of dialysis day 365 (follow-up was incomplete beyond this point; Fig 2A).

Variables and Measurements BPs were measured with the patient seated and recorded by dialysis unit personnel according to routine clinical practices before each dialysis session. Given the historic nature of the study, no opportunity existed for calibration of BP measurements. All BP metrics were derived using predialysis values. Data were not available for postdialytic or intradialytic BP, ultrafiltration, or interdialytic weight gain. Raw BPs were transformed into 4 candidate metrics of BPV: (1) the SD, (2) the coefficient of variation (SD divided by mean BP), (3) the average residual, and (4) the average residual-intercept ratio. The average residual was estimated for each patient by fitting a mixed-effects linear

718

Brunelli et al included age, sex, race, and body mass index. Comorbid diseases of interest (recorded on or before dialysis day 7) included diabetes, hypertension, coronary artery disease, prior stroke, peripheral vascular disease, and congestive heart failure. Biochemical covariates of interest included serum albumin level, creatinine level, equilibrated Kt/V, hemoglobin level, and phosphate level; these were considered as their averaged observed value during the BP exposure window (days 91 to 180 in the primary analysis). Each was measured at a centralized laboratory. Medications of interest, including adrenergic antagonists, renin-aldosteroneangiotensin blockers, diuretics, calcium channel blockers, and peripheral vasodilators, were considered as of day 14. Time-updated medication data were not available. All deaths were identified prospectively in the context of another study, as previously reported.30 In the primary analysis, deaths were considered during the at-risk period, beginning on dialysis day 181 and ending upon death or censoring.

Statistical Methods

Figure 2. Schematic representation of outcome models for primary analysis and sensitivity analyses 1, 2, 3, and 4. (Figures not drawn to scale.)

regression model of time on observed BP according to the equation: Systolic BPij ⫽ ␣ ⫹ ␤ · timej ⫹ ␥i ⫹ ␦i · timeij ⫹ ␧ij where subscripts i and j represent patient and time, respectively. (A separate analogous procedure was conducted for diastolic BP.) Thus, the systolic BP for the ith patient at time j was predicted as a function of the mean initial systolic BP for the cohort (␣), the mean slope for the cohort (␤ · timej), and additional terms to account for differences between patient i’s initial systolic BP (␥i) and slope (␦i · time) compared with the overall cohort. The residual, ␧ij, is the vertical distance between each observed and predicted systolic BP measurement; larger residuals indicate greater splay in BP measurements, and thus greater BPV. The average residual (candidate BPV metric 3) was the mean of the absolute value of residuals for each patient. (Mixed-effects modeling accounts for differences in precision that would otherwise result in patients with more or fewer BP measurements.) The mixed-effects model also provided information about each patient’s absolute level of BP at the start of the exposure window (␣ ⫹ ␥i; subsequently referred to as BP intercept) and each patient’s temporal BP trend (␤ ⫹ ␦i; subsequently referred to as BP slope). Because patients with higher BPs (large intercepts) may have greater degrees of BP fluctuation, a fourth candidate metric of BPV was identified: the average residual-intercept ratio. Data for other covariates were collected prospectively in the context of another study. Demographic factors of interest

Separate and parallel analyses were conducted for systolic and diastolic BPs. Continuous variables were described graphically and in terms of their mean, SD, median, and interquartile range. Categorical variables were described in terms of their frequency. A priori, we hypothesized that BP intercept and slope would each be associated with survival.25-29 Therefore, to best ascertain the independent association between BPV and survival, we sought to identify candidate BPV metrics that were independent of these other descriptions of BP. Bivariable measures of association between candidate BPV metrics and BP intercept were examined by using scatter plots and Pearson correlation coefficients. Analogous methods were used to examine the relationship between candidate BPV metrics and BP slope. In categorical analysis, BPV and BP slope were categorized into quartiles, and the association with all-cause mortality was measured using Kaplan-Meier curves and log-rank testing. To bear greatest analogy to the published literature, BP intercept was categorized according to clinically relevant thresholds, and its association with all-cause mortality was examined by determining stratumspecific morality rates. In addition, the association between BPV, considered as a linear term, and all-cause mortality was investigated by fitting unadjusted and adjusted Cox proportional hazards models. Adjustment was made for demographic, comorbidity, biochemical, and medication data, as well as for BP intercept and slope. The proportional hazards assumption was tested on bivariable analysis for each predictor variable both graphically and by inclusion of 2-way cross-product terms with time. In the multivariable model, 2-way crossproduct terms with time were included for predictor variables violating the proportional hazards assumption on bivariable analysis. Schoenfeld testing was used to assess the global proportional hazards assumption for all multivariable models. In addition, multiple sensitivity analyses (described more fully in Results and shown in Fig 2) were conducted to

Blood Pressure Variability in Dialysis

719

determine the robustness of findings in the primary analysis. In each of these analyses, BP exposures were reestimated by using mixed-effects models analogous to that used in the primary analysis. Association with outcome was assessed by using survival models analogous to those described. Table 1. Characterization of Primary Cohort Age (y) Mean ⫾ SD Median (interquartile range) Range Sex Men Women Race White Nonwhite Body mass index (kg/m2) ⱕ20 20-25 25-30 30-35 ⬎35 Diabetes Hypertension Arterial disease* Coronary disease Prior stroke Peripheral vascular disease Congestive heart failure Cancer Laboratory results† Albumin (g/dL) Creatinine (mg/dL) Equilibrated Kt/V Urea reduction ratio (%) Hemoglobin (g/dL) Ferritin (ng/mL) Transferrin saturation (%) Calcium (mg/dL) Phosphate (mg/dL) Intact parathyroid hormone (pg/mL) Blood pressure measurements from days 91 to 180 Mean ⫾ SD Median (interquartile range) Range

61.9 ⫾ 15.1 63.0 (51.6-74.0) 18.4-99.6 3,756 (54.0) 3,205 (46.0) 3,966 (57.0) 2,995 (43.0) 701 (10.1) 2,126 (30.5) 2,004 (28.8) 1,091 (15.7) 1,039 (14.9) 1,715 (24.6) 2,801 (40.2) 992 (14.3) 670 (9.6) 220 (3.2) 361 (5.2) 829 (11.9) 206 (3.0) 3.7 ⫾ 0.4 7.3 ⫾ 2.9 1.33 ⫾ 0.27 71.3 ⫾ 6.8 12.5 ⫾ 1.2 475.2 ⫾ 598.5 28.1 ⫾ 9.6 9.0 ⫾ 0.6 5.4 ⫾ 1.4 175.8 ⫾ 160.0 35.9 ⫾ 4.5 38 (35-39) 4-52

Note: n ⫽ 6,961. Values expressed as mean ⫾ SD, number (percent), median (interquartile range), or range. To convert serum creatinine in mg/dL to mol/L, multiply by 88.4; to convert albumin and hemoglobin in g/dL to g/L, multiply by 10; to convert calcium in mg/dL to mmol/L, multiply by 0.2495; to convert phosphate in mg/dL to mmol/L, multiply by 0.3229; parathyroid hormone levels expressed in pg/mL and ng/L are equivalent; ferritin levels expressed in ng/mL and ␮g/L are equivalent. *Categories of arterial disease are not mutually exclusive. †Averaged during dialysis days 91 to 180.

Table 2. Characterization of Antihypertensive Treatment in the Primary Cohort Class

No. (%)

Adrenergic antagonist* ␣-Blocker ␤-Blocker Clonidine/methyldopa Renin-angiotensin-aldosterone blockers* Angiotensin-converting enzyme inhibitor Angiotensin receptor blocker Aldosterone antagonist Calcium channel blockers* Centrally acting Dihydropyridine Diuretics Peripheral vasodilators* Hydralazine Minoxidil Nitrates (oral) Agents (by class) per patient Mean ⫾ SD Median (interquartile range)

3,234 (46.5) 518 (7.4) 2,649 (38.1) 1,021 (14.7) 1,991 (28.6) 1,272 (18.3) 829 (11.9) 51 (0.7) 2,572 (37.0) 431 (6.2) 2,191 (31.5) 1,510 (21.7) 936 (13.5) 455 (6.5) 177 (2.5) 436 (6.3) 1.7 ⫾ 1.6 2 (0-3)

Note: n ⫽ 6,961. *Medications within class are not exclusive.

All analyses were performed using STATA, version 9.0 (Stata Corp, College Station, TX).

RESULTS Participants A flow diagram for enrollment is shown in Fig 1. Of 10,044 patients who began long-term dialysis therapy at any Fresenius Medical Care unit between June 2004 and August 2005, a total of 6,961 qualified for enrollment in the primary cohort. A description of these patients is listed in Table 1. Antihypertensive treatment in patients in the primary cohort is listed in Table 2. The mean number of antihypertensive agents was 1.7 ⫾ 1.6 (SD); 46.5%, 28.6%, 37.0%, 21.7%, and 13.5% of patients were treated with adrenergic antagonists, renin-angiotensin-aldosterone blockers, calcium channel blockers, diuretics, and peripheral vasodilators, respectively. Identifying Metric of BP Variability To determine whether candidate metrics of BPV were independent of absolute BP level, we calculated Pearson correlation coefficients between each patient’s BP intercept and each of the 4 candidate BPV metrics (Table 3). In

720

Brunelli et al

Table 3. Pearson Correlation Coefficients Between Blood Pressure Intercept and Candidate Metrics of BPV

Candidate Metric of BPV

Systolic 1. SD 2. Coefficient of variation 3. Average residual 4. Average residual-intercept ratio Diastolic 1. SD 2. Coefficient of variation 3. Average residual 4. Average residual-intercept ratio

Absolute Blood Pressure (Intercept)

0.44 0.03 0.43 0.01 0.41 ⫺0.09 0.40 ⫺0.12

Abbreviation: BPV, blood pressure variability.

the candidate metrics, SD and average residual showed undesirably high degrees of correlation with BP intercept, whereas the coefficient of variation and average residual-intercept ratio did not, suggesting that the latter 2 metrics better characterized BPV independently of absolute BP level. To determine whether candidate metrics of BPV were independent of the magnitude of BP trend, we analyzed scatter plots of candidate metrics versus the absolute value of BP slope. The absence of data points in the lower right portions of Fig 3A and B (below the line) indicates that the coefficient of variation could not be small when there was a pronounced trend in BP. The flatter nature of the plots in Fig 3C and D indicate that average residual-intercept ratio was more independent of BP slope; thus, this metric (henceforth referred to as BPV) was selected as the most appropriate metric of BP variability. Distributions of BPV, BP intercept, and BP slope are listed in Table 4. Survival Analysis In patients included in the primary cohort, there were 491 deaths in 3,215.5 patient-years of follow up. The mortality rate was 152.6 deaths/ 1,000 patient-years (95% confidence interval [CI], 139.7 to 166.7); median survival time was 185 days. On unadjusted analysis, higher diastolic (P ⬍ 0.001; but not systolic; P ⫽ 0.07) BPV was significantly associated with increased all-cause mortality (Fig 4). Higher systolic and diastolic BP slopes were associated with improved sur-

vival (Fig 5; P ⬍ 0.001 in both cases). Both systolic and diastolic BP intercepts showed Ushaped relationships with all-cause mortality (Fig 6). Unadjusted and adjusted hazard ratios (HRs) for all-cause mortality are listed in Table 5. Considered as a linear term, each SD increase in systolic BPV was associated with an unadjusted HR for all-cause mortality of 1.14 (95% CI, 1.05 to 1.24). For diastolic BPV, the unadjusted HR was 1.27 (95% CI, 1.18 to 1.38). Fully adjusted HRs for all-cause mortality were 1.13 (95% CI, 1.03 to 1.23) and 1.15 (95% CI, 1.06 to 1.26) per SD increase in systolic and diastolic BPV, respectively. Sensitivity Analyses Sensitivity analysis 1 was conducted in patients who had 35 (⬃90% of anticipated) or more BP recordings during the exposure window (Figs 1 and 2A). The rationale for this analysis was 2-fold. First, absent predialysis BP recordings are a proxy for missed HD sessions, which may promote BPV and which themselves are often caused by conditions associated with mortality (eg, hospitalization and nonadherence). Thus, inclusion of patients with very few BP recordings may have biased survival estimates. Second, the precision of BP measures increases in patients with more measurements; therefore, restricting analysis to patients with a large number of observations may reduce exposure misclassification. Adjusted HRs for systolic and diastolic BPV were 1.12 (95% CI, 0.99 to 1.26) and 1.17 (95% CI, 1.04 to 1.31), respectively (Table 6). These estimates were qualitatively similar to those in the primary analysis, but CIs were wider because of diminution in number of outcomes. Because BPV may be expected to affect mortality through CV mechanisms, sensitivity analysis 2 was conducted in the primary cohort in which only CV deaths (as adjudicated by International Classification of Diseases, Ninth Revision code for cause of death; n ⫽ 284) were considered outcome events. Non-CV deaths were treated as censoring events in this analysis. Adjusted HRs for CV mortality were 1.09 (95% CI, 0.98 to 1.23) and 1.17 (95% CI, 1.05 to 1.31) for systolic and diastolic BPV, respectively (Table 6). Again, estimates were qualitatively similar to those in the primary analysis, but CIs were wider because of diminution in number of outcomes.

Blood Pressure Variability in Dialysis

721

Because the choice of assessing BP during days 91 to 180 was arbitrary, sensitivity analysis 3 was conducted in which BP parameters were defined during days 0 to 90, and survival was assessed beginning on day 91 (Fig 2B). Adjusted HRs for all-cause mortality were 1.07 (95% CI, 1.00 to 1.15) and 1.08 (95% CI, 1.00 to 1.16) for systolic and diastolic BPV, respectively (Table 6). Finally, because of the possibility that we were observing a reverse-causal association between BPV and mortality (ie, that BPV increased because of the influence of burgeoning fatal illness), sensitivity analysis 4 was conducted in which BP exposures were defined during days 0 to 90, and survival was assessed beginning on day 121 (Fig 2C), thus allowing for a 30-day lag period between exposure and outcome. Adjusted HRs for all-cause mortality were 1.05 (95% CI, 0.97 to 1.13) and 1.07 (95% CI, 1.00 to 1.16) for systolic and diastolic BPV, respectively (Table 6).

DISCUSSION Mortality, particularly that caused by CV disease, is high in HD patients and exceeds that which would be predicted on the basis of traditional risk factors.1-6 Thus, identifying factors that either predict and/or contribute to CV events in this population is paramount for improving survival and health. In this report, we introduce a novel metric of long-term BPV, the average residual-intercept ratio, and show its ability to characterize variability independently of absolute BP level and temporal BP trend. In addition, we show a strong and independent association between long-term BPV and all-cause mortality in a nationally representative cohort of incident HD patients. BPV was selected for study by analogy to non-HD patients in whom increased BPV has been associated with increased CV mortality,7-9 nonfatal CV events,8 stroke,9 progression of carotid intima-to-medial wall thickness,8,10 increased target-organ damage scores,11-13 and left

4™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™ Figure 3. Scatter plots of the absolute value of blood pressure (BP) slope versus candidate metrics of variability. (A, B) The absence of points in the lower right portion of the plot (below the lines) suggests that the coefficient of variation cannot be small when there is a large upward or downward trend in BP. (C, D) The flatter nature of the plots shows that the average residual-intercept ratio is more independent of BP trend.

722

Brunelli et al Table 4. Distribution of BP Parameters Across the Cohort

Systolic Absolute BP (intercept; mm Hg) BP trend (slope; mm Hg/mo) BPV (average residual: intercept ratio; unitless) Diastolic Absolute BP (intercept; mm Hg) BP trend (slope; mm Hg/mo) BPV (average residual: intercept ratio; unitless)

Mean ⫾ SD

25th Percentile

Median

75th Percentile

150.0 ⫾ 18.9 ⫺0.43 ⫾ 4.7 0.090 ⫾ 0.024

138.0 ⫺3.30 0.073

150.3 ⫺0.60 0.087

163.8 2.41 0.10

77.9 ⫾ 11.3 ⫺0.38 ⫾ 2.4 0.10 ⫾ 0.028

70.0 ⫺1.89 0.082

77.4 ⫺0.42 0.10

85.0 1.06 0.12

Abbreviations: BP, blood pressure; BPV, blood pressure variability.

ventricular mass.13 Studies in animal models have suggested that a causal interpretation of this association is plausible. Rats undergoing sinoaortic denervation showed marked increases in BPV with little change in mean BP, resulting in biventricular hypertrophy,14 and increased atherogenesis,15,16 possibly on the basis of impaired endothelin-dependent relaxation profiles.16 In hemodialysis patients, increased vascular stiff-

ness17,18 and autonomic nervous system dysfunction31-34 may both promote BPV, as well impair autoregulation of cerebral and coronary perfusion. The latter effect would render patients at increased risk of increased endothelial shearing and tissue hypoxic damage (during periods of high and low BP, respectively) that result from long-term variability in BP. In individuals not on long-term HD therapy, BPV has been measured using 24-hour ambulatory BP monitors,7-13 and prior research in HD

Figure 4. Kaplan-Meier survival functions for all-cause mortality according to quartile of (A) systolic and (B) diastolic blood pressure variability (BPV).

Figure 5. Kaplan-Meier survival functions for all-cause mortality according to quartile of (A) systolic and (B) diastolic blood pressure (BP) slope.

Blood Pressure Variability in Dialysis

Figure 6. Unadjusted all-cause mortality rates according to absolute level of (A) systolic and (B) diastolic blood pressure (BP). Abbreviation: pt yrs, patient-years.

patients suggests that such defined BPV bears similar association to CV morbidity and mortality.20-23 However, given that HD patients are predisposed to longer term undulations in BP, we sought to clarify the relationship between longterm BPV and outcome in this study. The only prior study of the effects of longterm BPV in HD patients reported that a high coefficient of variation of predialysis systolic BP was associated with greater all-cause mortality.24 Our study confirms this finding and offers the following advantages: (1) use of a metric of BPV that better distinguishes between variability and temporal BP trend; (2) larger sample size, permitting more precise effect estimates and more extensive adjustment for potential confounders; (3) homogeneity of patients with respect to duration of dialysis (ie, all patients were incident); and (4) multicenter experience. We observed a U-shaped relationship between absolute BP level and mortality. This finding is

723

consistent with multiple prior studies,25-28 which is reassuring about our study’s validity. At the same time, mortality increases at both high and low levels of absolute BP, which limits the potential of interventions directed at altering absolute BP or BP trend. Several limitations of this study should be noted. Our data set did not contain information about and therefore did not permit adjustment for interdialytic weight gain, ultrafiltration, intradialytic BP experience, or adherence to prescribed therapy. Thus, the influence of these variables on the observed degree of BPV and on the BPV– mortality association could not be assessed. In addition, such nonfatal outcomes as hospitalization and nonfatal CV events could not be considered. As with all nonrandomized research, the possibility of selection bias exists. However, given that data were collected in a systematic manner by staff unaware of the study hypothesis and considering the consistency of findings across a number of sensitivity analyses, it seems unlikely that our findings are explained by selection bias. Similarly, the lack of calibration of BP devices raises the possibility of misclassification bias. However, there is no reason to suspect that patients who were likelier to die would have preferentially been misclassified as having greater BPV; thus, the expected direction of the bias Table 5. Hazard Ratios for All-Cause Mortality per SD Increase in BPV

Unadjusted Fully adjusted*

Systolic BPV (/SD change)

Diastolic BPV (/SD change)

1.14 (1.05-1.24) 1.13 (1.03-1.23)

1.27 (1.18-1.38) 1.15 (1.06-1.26)

Note: Values expressed as hazard ratio (95% confidence interval). Abbreviation: BPV, blood pressure variability. *Adjusted for age, sex, race (white, nonwhite), body mass index (ⱕ 20, 20 to 25, 25 to 30, 30 to 35, and ⬎35 kg/m2), diabetes, hypertension, arterial disease (coronary, cerebrovascular, and peripheral vascular), congestive heart failure, cancer, albumin level, creatinine level, equilibrated Kt/V, hemoglobin level, phosphate level, adrenergic antagonist use, renin-angiotensin-aldosterone blocker use, calcium channel blocker use, diuretic use, peripheral vasodilator use, absolute blood pressure (intercept, categorized as in Fig 6), and blood pressure trend (slope). (Because of nonproportional hazards, 2-way cross-products with time were included for race, sex, and body mass index.)

724

Table 6. Adjusted HRs per SD Increase in BPV in Sensitivity Analyses

Sensitivity Analysis

1

2

3

4

Cohort

Entered at-risk time; ⱖ35 dialysis treatments during exposure window (n ⫽ 5,284; total atrisk time ⫽ 2,489.5 patientyears) Entered at-risk time; ⱖ1 blood pressure in each mo of exposure window (n ⫽ 6,961; total at-risk time ⫽ 3,215.5 patient-years) Entered at-risk time; ⱖ1 blood pressure in each mo of exposure window (n ⫽ 8,181; total at-risk time ⫽ 5,263.0 patient-years) Entered at-risk time; ⱖ1 blood pressure in each mo of exposure window (n ⫽ 7,847; total at-risk time ⫽ 4,604.5 patient-years)

Exposure Window (d)

Begin At-Risk Time (d)

91-180

181

91-180

181

0-90

0-90

Adjusted HR* per SD Increase (95% confidence interval) Outcome Considered

Systolic BPV

Diastolic BPV

All-cause mortality (n ⫽ 281)

1.12 (0.99-1.26)

1.17 (1.04-1.31)

Cardiovascular mortality (n ⫽ 284)

1.09 (0.98-1.23)

1.17 (1.05-1.31)

91

All-cause mortality (n ⫽ 863)

1.07 (1.00-1.15)

1.08 (1.00-1.16)

121

All-cause mortality (n ⫽ 754)

1.05 (0.97-1.13)

1.07 (1.00-1.16)

Abbreviations: BPV, blood pressure variability; HR, hazard ratio. *Adjusted for age, sex, race (white, nonwhite), body mass index (ⱕ20, 20 to 25, 25 to 30, 30 to 35, and ⬎35 kg/m2), diabetes, hypertension, arterial disease (coronary, cerebrovascular, and peripheral vascular), congestive heart failure, cancer, albumin level, creatinine level, equilibrated Kt/V, hemoglobin level, phosphate level, adrenergic antagonist use, renin-angiotensin-aldosterone blocker use, calcium channel blocker use, diuretic use, peripheral vasodilator use, absolute blood pressure (intercept, categorized as in Fig 6), and blood pressure trend (slope). (Because of nonproportional hazards, 2-way cross-products with time were included for race, sex, and body mass index.)

Brunelli et al

Blood Pressure Variability in Dialysis

would be toward the null and would not explain the observed associations. As with all observational studies, there is the potential for confounding. We adjusted analyses for factors that we believed to be plausibly related to both BPV and mortality. However, we cannot dismiss the potential for other unmeasured confounders. Moreover, time-updated data for oral medications (most notably antihypertensives) were not available, which may have fostered residual confounding. Finally, all patients in this study were incident patients served by a single dialysis chain. Generalization to prevalent patients or other populations should be undertaken cautiously. In conclusion, provided the mathematical descriptions of BPV are valid, the data suggest that systolic and diastolic BPV, measured by using the average residual-intercept ratio, is associated with all-cause mortality in incident HD patients. Additional studies are necessary to generalize and validate findings, explore the interplay between systolic and diastolic BPV, and clarify the potentially causal nature of the association.

ACKNOWLEDGEMENTS The authors thank R. Localio for insights regarding mixedeffects linear models. This manuscript is dedicated to and in loving memory of Fiore Brunelli. Support: This study was supported in part by American Heart Association Fellow-to-Faculty Transition Award 0775021N and a National Kidney Foundation Postdoctoral Fellowship (both to Dr Brunelli). Dr Brunelli had full access to all data and takes responsibility for the integrity and accuracy of the data analysis. Financial Disclosure: None.

REFERENCES 1. Foley RN, Parfrey PS, Sarnak MJ: Clinical epidemiology of cardiovascular disease in chronic renal disease. Am J Kidney Dis 32:S112-S119, 1998 (suppl 3) 2. Longenecker JC, Coresh J, Powe NR, et al: Traditional cardiovascular disease risk factors in dialysis patients compared with the general population: The CHOICE Study. J Am Soc Nephrol 13:1918-1927, 2002 3. Levey AS, Beto JA, Coronado BE, et al: Controlling the epidemic of cardiovascular disease in chronic renal disease: What do we know? What do we need to learn? Where do we go from here? National Kidney Foundation Task Force on Cardiovascular Disease. Am J Kidney Dis 32:853-906, 1998

725 4. Foley RN, Parfrey PS, Harnett JD, et al: Clinical and echocardiographic disease in patients starting end-stage renal disease therapy. Kidney Int 47:186-192, 1995 5. Herzog CA, Ma JZ, Collins AJ: Poor long-term survival after myocardial infarction among patients on longterm dialysis. N Engl J Med 339:799-805, 1998 6. Harnett JD, Foley RN, Kent GM, Barre PE, Murray D, Parfrey PS: Congestive heart failure in dialysis patients: Prevalence, incidence, prognosis and risk factors. Kidney Int 47:884-890, 1995 7. Kikuya M, Hozawa A, Ohokubo T, et al: Prognostic significance of blood pressure and heart rate variabilites: The Ohasama Study. Hypertension 36:901-906, 2000 8. Sander D, Kukla C, Klingelhöfer J, Winbeck K, Conrad B: Relationship between circadian blood pressure variability and progression of early carotid atherosclerosis: A 3-year follow-up study. Circulation 102:1536-1541, 2000 9. Pringle E, Phillips C, Thijs L, et al, for the Syst-Eur Investigators. Systolic blood pressure variability as a risk factor for stroke and cardiovascular mortality in the elderly hypertensive population. J Hypertens 21:2251-2257, 2003 10. Sander D, Klingelhöfer J: Diurnal systolic blood pressure variability is the strongest predictor of early atherosclerosis. Neurology 47:500-507, 1996 11. Palatini P, Penzo M, Racioppa A, et al: Clinical relevance of nighttime blood pressure and of daytime blood pressure variability. Arch Intern Med 152:18551860, 1992 12. Frattola A, Parati G, Cuspidi C, Albini F, Mancia G: Prognostic value of 24-hour blood pressure variability. J Hypertens 11:1133-1137, 1993 13. Verdecchia P, Borgioni C, Ciucci A, et al: Prognostic significance of blood pressure variability in essential hypertension. Blood Press Monit 1:3-11, 1996 14. Van Vliet BN, Hu L, Scott T, Chafe L, Montani JP: Cardiac hypertrophy and telemetered blood pressure 6 wk after baroreceptor denervation in normotensive rats. Am J Physiol 271:R1759-R1769, 1996 15. Sasaki S, Yoneda Y, Fujita H, et al: Association of blood pressure variability with induction of atherosclerosis in cholesterol-fed rats. Am J Hypertens 7:453-459, 1994 16. Eto M, Toba K, Akishita M, et al: Reduced endothelial vasomotor function and enhanced neointimal formation after vascular injury in a rat model of blood pressure liability. Hypertens Res 26:991-998, 2003 17. Marchais S, Guérin A, Safar M, London G: Arterial compliance in uraemia. J Hypertens 7:S84-S85, 1989 (suppl 1) 18. Kooman JP, Wijnen JA, Draaijer P, et al: Compliance and reactivity of the peripheral venous system in patients treated with chronic intermittent hemodialysis. Kidney Int 41:1041-1048, 1992 19. Brennan PJ, Greenberg G, Miall WE, Thompson SG: Seasonal variation in arterial blood pressure. Br Med J 285:919-923, 1982 20. Covic A, Goldsmith DJ: Ambulatory blood pressure monitoring in nephrology: Focus on BP variability. J Nephrol 12:220-229, 1999 21. Covic A, Goldsmith DJ, Georgescu GC, Ackrill P: Relationships between blood pressure variability and left

726 ventricular parameters in haemodialysis and renal transplant patients. Nephrology 4:87-93, 1998 22. Goldsmith DJ, Covic AC, Venning MC, Ackrill P: Ambulatory blood pressure monitoring in renal dialysis and transplant patients. Am J Kidney Dis 29:593-600, 1997 23. Ertürk S, Ertug˘ AE, Ates¸ K, et al: Relationship of ambulatory blood pressure monitoring data to echocardiographic findings in haemodialysis patients. Nephrol Dial Transplant 11:2050-2054, 1996 24. Tozawa M, Iseki K, Yoshi S, Fukiyama K: Blood pressure variability as an adverse prognostic risk factor in end-stage renal disease. Nephrol Dial Transplant 14:19761981, 1999 25. Port FK, Hulbert-Shearon TE, Wolfe RA, et al: Predialysis blood pressure and mortality risk in a national sample of maintenance hemodialysis patients. Am J Kidney Dis 33:507-517, 1999 26. Zager PG, Nikolic J, Brown RH, et al: “U” Curve association of blood pressure and mortality in hemodialysis patients. Medical Directors of Dialysis Clinic, Inc. Kidney Int 54:561-569, 1998 27. Mazzuchi N, Carbonell E, Fernández-Cean J: Importance of blood pressure control in hemodialysis patient survival. Kidney Int 58:2147-2154, 2000

Brunelli et al 28. Lynn KL, McGregor DO, Moesbergen T, Buttimore AL, Inkster JA, Wells JE: Hypertension as a determinant of survival for patients treated with home dialysis. Kidney Int 62:2281-2287, 2002 29. Charra B, Calemard E, Ruffet M, et al: Survival as an index of adequacy of dialysis. Kidney Int 41:1286-1291, 1992 30. Thadhani R, Tonelli M: Cohort studies: Marching forward. Clin J Am Soc Nephrol 1:1117-1123, 2006 31. Lilley JJ, Golden J, Stone RA: Adrenergic regulation of blood pressure in chronic renal failure. J Clin Invest 57:1190-1200, 1976 32. Friess U, Rascher W, Ritz E, Gross P: Failure of arginine-vasopressin and other pressor hormones to increase in severe recurrent dialysis hypotension. Nephrol Dial Transplant 10:1421-1427, 1995 33. Daul AE, Wang XL, Michel MC, Brodde OE: Arterial hypotension in chronic hemodialyzed patients. Kidney Int 32:728-735, 1987 34. Converse RL Jr, Jacobsen TN, Jost CM, et al: Paradoxical withdrawal of reflex vasoconstriction as a cause of hemodialysis-induced hypotension. J Clin Invest 90:16571665, 1992