Visual Overview
Abstract
Background Evidence on the utility of ambulatory BP monitoring for risk prediction has been scarce and inconclusive in patients on hemodialysis. In addition, in cardiac diseases such as heart failure and atrial fibrillation (common among patients on hemodialysis), studies have found that parameters such as systolic BP (SBP) and pulse pressure (PP) have inverse or nonlinear (U-shaped) associations with mortality.
Methods In total, 344 patients on hemodialysis (105 with atrial fibrillation, heart failure, or both) underwent ambulatory BP monitoring for 24 hours, starting before a dialysis session. The primary end point was all-cause mortality; the prespecified secondary end point was cardiovascular mortality. We performed linear and nonlinear Cox regression analyses for risk prediction to determine the associations between BP and study end points.
Results During the mean 37.6-month follow-up, 115 patients died (47 from a cardiovascular cause). SBP and PP showed a U-shaped association with all-cause and cardiovascular mortality in the cohort. In linear subgroup analysis, SBP and PP were independent risk predictors and showed a significant inverse relationship to all-cause and cardiovascular mortality in patients with atrial fibrillation or heart failure. In patients without these conditions, these associations were in the opposite direction. SBP and PP were significant independent risk predictors for cardiovascular mortality; PP was a significant independent risk predictor for all-cause mortality.
Conclusions This study provides evidence for the U-shaped association between peripheral ambulatory SBP or PP and mortality in patients on hemodialysis. Furthermore, it suggests that underlying cardiac disease can explain the opposite direction of associations.
The elevated morbidity and mortality rates in patients with ESRD are mainly driven by the increased prevalence of cardiovascular disease.1,2 Hypertension is also highly prevalent in patients with CKD patients, especially in those with ESRD, undergoing hemodialysis.3,4 Although high BP is common in this population, hypertension control is often inadequate because of several issues, including the complex pathophysiology, the limitations for sodium and water removal caused by the intermittent nature of hemodialysis, the inappropriate use of antihypertensive drugs and others.5
Hypertension guidelines suggest ambulatory BP monitoring (ABPM) as the gold standard for the diagnosis of hypertension6,7; this is also proposed in reviews and consensus statements for patients on dialysis.5,8,9 As recently summarized,5 evidence from prospective studies10–12 suggests that the association of home or ambulatory interdialytic BP with mortality and cardiovascular events is clearer than that of pre- and postdialytic BP recordings. Nevertheless, the evidence on ambulatory BP and risk prediction is still scarce and results are not fully aligned, with more recent studies not confirming an independent association of ambulatory BP with outcomes.13 Furthermore, one can mention various potential limitations of published studies, such as low number of patients,11 results not adjusted,12 different nonlinear associations,10,13 and inclusion of only black patients.10,12
Another important issue in hypertension management and risk prediction in patients on hemodialysis is the number of associated comorbidities patients are suffering from, especially the effect of the complex pattern of cardiac systolic and diastolic dysfunction.14–16 It is now well established that several hemodynamic parameters have a different predictive behavior in patients with heart failure. For example, pulse pressure has an inverse association with risk in patients with heart failure.17–19 Such an inverse association is also documented for systolic BP (SBP) in patients with heart failure.20–22 Furthermore, prevalence of atrial fibrillation (AF) is high and an important risk factor in the hemodialysis population.23 There is limited evidence that there is a U-shaped association of BP and mortality in patients with AF.24,25
Thus, the primary objective of this study was to investigate the nonlinear association between ambulatory BP and mortality in a large cohort of patients on hemodialysis. Furthermore, this study attempted to assess a possible effect of cardiac disease on the nonlinear (i.e., curvilinear, U-shaped) association of BP with survival, as we hypothesize that this is an important moderator that has not previously been taken into account.
Methods
Study Population
All 24-hour BP recordings used in this study were from the rISk strAtification in end-stage Renal disease (ISAR) study (Clinicaltrials.gov identifier: NCT01152892), which is a prospective, longitudinal, observational cohort study aiming at improving cardiovascular risk stratification in patients with ESRD.26 The study protocol was approved by the Ethics Committees of the Klinikum rechts der Isar of the Technical University Munich and of the Bavarian State Board of Physicians. The study adheres to the Declaration of Helsinki. Patients were recruited in 17 dialysis centers in Munich and the surrounding area.26 All patients gave informed consent. Patients were recruited between September of 2010 and January of 2014. Inclusion criteria were age ≥18 years, dialysis vintage ≥90 days, willingness to participate in at least one technical examination (e.g., 24-hour ABPM), and written informed consent.26 Patients were excluded if an ongoing infection, pregnancy, malignant disease with a life expectancy of <24 months was present, or they were unwilling to participate.26
The ISAR cohort included in total 519 patients and 414 thereof agreed to undergo 24-hour ABPM including pulse wave analysis. In total, 344 recordings were included in data analysis after exclusion of patients with too few measurements or too short recording time (n=47), incomplete laboratory values (n=17), or insufficient data quality (n=6) (see Figure 1).
Flowchart of included patients.
Ambulatory BP Measurement
All measurements were obtained with the Mobil-O-Graph 24h PWA Monitor (I.E.M. GmbH, Stolberg, Germany) using validated ARCSolver algorithms (Austrian Institute of Technology GmbH, Vienna, Austria)27–29 within the ISAR hemodialysis study. The Mobil-O-Graph’s brachial BP measurement unit is validated according to standard protocols.30,31 Measurement using appropriate cuffs on the nonfistula arm started before the midweek dialysis session, where volume withdrawal was set according to clinical standard on the basis of the personal dry-weight, and lasted for 24 hours. The monitor was programmed to measure BP and pulse wave analysis parameters every 15 minutes at daytime (8:00 am to 9:00 pm) and every 30 minutes at night (9:00 pm to 8:00 am). BP data were averaged for 24 hours. There was no particular outlier detection and manual data cleaning performed.
Data Collection and Laboratory Measurements
Patients’ clinical characteristics, including regular medical treatment with antihypertensive agents, statins, or oral anticoagulants, were obtained at the time of enrollment from dialysis protocols and medical records. Paroxysmal and permanent AF was determined on the basis of a Holter ECG, if available in the ISAR study, or medical record from dialysis centers. Diagnosis of heart failure was on the basis of medical records from dialysis centers, where information is mainly from echocardiography. Baseline comorbidities were determined by the ISAR Endpoint Committee, a panel of three physicians including a nephrologist and a cardiologist.26
Blood samples were obtained before starting ambulatory BP recording at baseline. High-resolution C-reactive protein was assayed with latex-enhanced reagents (Siemens), on a BN ProSpec analyzer (Siemens), following manufacturer’s instructions. Other serum chemistry values were performed in International Organization for Standardization–certified laboratories in the different dialysis centers.
End points
Patients were followed up until death occurred and were censored after renal transplantation (per transplantation date). The latest follow-up took place between April and September of 2016. The primary end point of the study was all-cause mortality. Cardiovascular mortality (i.e., death from sudden cardiac death, myocardial infarction, congestive heart failure, or stroke, or death after cardiovascular procedure) was a prespecified secondary end point.26 Again, the ISAR Endpoint Committee independently adjudicated all end points on the basis of medical documentation or interviews with the physicians in the dialysis centers.26
Statistical Analyses
Statistical and computational evaluation was performed using Matlab R2014a (The MathWorks, Inc., Natick, MA) and R (version 3.1; R Foundation for Statistical Computing, Vienna, Austria). Continuous data are presented as mean (SD) or median (interquartile range [IQR]), according to the results of a Kolmogorov–Smirnov test for normality, with categorical data as N (percent). Between-group differences of continuous data were examined with a t test or Mann–Whitney U test, as appropriate. For proportions, we used the chi-squared test. In survival analysis, we first used Kaplan–Meier curves comparing quartiles to reveal possible violations of proportional hazard assumptions, thus identifying nonlinear (nonproportional) associations. Then, nonlinear, univariate, and multivariable Cox regression analyses were performed for determining the associations between variables and end points. According to the results of nonlinear and interaction analysis, patients were separated on the basis of the presence/absence of AF or heart failure in two groups to overcome interpretational limitations of nonlinear analysis. Subsequently, these two groups were analyzed with linear, univariate, and multivariable Cox models to correctly determine proportional hazard ratios (HRs) for risk prediction. Adjustment was done for age and sex (model 1), and model 1 plus diabetes mellitus and serum albumin (model 2). Additional adjustment models can be found in the Supplemental Material. Furthermore, interactions between risk predictors and the presence of AF or heart failure were tested using interaction terms in Cox regression analysis. Statistical significance was assumed at a 5% level.
Results
Baseline Characteristics of Study Participants
The study population included 344 patients (234 men and 110 women; median, 69.3; IQR, 55.7–77.2 years of age). Sixty-nine (20%) patients had AF and 62 (18%) had heart failure; in total, 105 (30.5%) of patients had either one or both of the conditions (AF or heart failure group). The median dialysis vintage of all patients was 41.1 (IQR, 22.7–76.6) months. About 92% of all measurements contained a record duration of at least 20 hours. Prevalence of diabetes was 39% and of hypertension 95%, defined as either hypertensive BP values on the 24-hour BP recording or intake of an antihypertensive medication. There was a difference in age, presence of diabetes, smoking status, serum albumin, and use of anticoagulation medication for the two groups (AF or heart failure versus no AF or heart failure). Furthermore, there is a significance difference in systolic, mean, and diastolic BP (e.g., SBP: mean 120 (17.5 SD) versus 125 (15.8 SD) mm Hg; P=0.003), but not pulse pressure, when comparing patients with AF and/or heart failure with patients without AF or heart failure. Detailed baseline characteristics can be found in Supplemental Table 1 and Table 1. Study population was mainly comparable with the excluded patients as defined in Figure 1 (see Supplemental Table 2).
Baseline characteristics
The mean follow-up time for all patients was 37.6 (17.5 SD) months. During follow-up, 115 patients died (59 in AF or heart failure group), of whom 47 were due to cardiovascular reasons (20 in the AF or heart failure group). Reasons for cardiovascular death are presented in Supplemental Table 3. Patients were censored at renal transplantation (n=31), date of moving away (n=2), or loss to follow-up (n=5).
All-Cause Mortality
Kaplan–Meier plots suggest a nonlinear association between SBP/pulse pressure and all-cause mortality (Figure 2A). Figure 2B and Supplemental Figure 1, A–C depict the HRs for systolic/diastolic blood and pulse pressure related to the median value resulting from univariate nonlinear Cox regression analysis in the whole study population. Nonlinear Cox regression analysis revealed a nonlinear behavior for SBP and pulse pressure for all-cause mortality in the whole cohort, i.e., significant nonlinear terms in univariate and multivariable analysis (Supplemental Table 4, Table 2). Diastolic BP turned out to be linearly but inversely associated with mortality.
Nonlinear association of pulse pressure with all-cause mortality. Univariate association of pulse pressure with all-cause mortality in (A) Kaplan–Meier analysis, (B) nonlinear Cox regression analysis using penalized smoothing splines, and proportional hazard models for the (C) AF or HF group and (D) no AF or HF group. HF, heart failure; PP, pulse pressure; Q, quartile.
P values for linear and nonlinear terms for univariate and adjusted nonlinear Cox regression analysis
Interaction analysis underpinned this behavior by significant interaction between predictors and absence of AF and/or heart failure for SBP (P<0.001) and pulse pressure (P<0.001), but not for diastolic pressure (P=0.42). Thus, data were analyzed in subgroups according to absence or presence of AF and/or heart failure. Results from nonlinear Cox regression analysis for the two study subgroups can be seen in Supplemental Figure 2 and Supplemental Tables 5 and 6. Nonlinear effects diminished in the two subgroups. As depicted in Figure 2, C and D, the nonlinear, U-shaped association of SBP and pulse pressure with all-cause mortality in the whole cohort could be the result of opposite associations presented in the two study subgroups.
In Table 3, the results from linear Cox regression analysis are presented for the study subgroups. SBP is an independent risk predictor for and inversely related to all-cause mortality in the AF or heart failure group (univariate HR, 0.97; 95% confidence interval [95% CI], 0.96 to 0.98; P<0.001; multivariable HR, 0.97; 95% CI, 0.96 to 0.99; P≤0.002, independently of chosen adjustment model), but not associated to all-cause mortality in the no AF or heart failure subgroup. Positive association (although not significant) indicates an opposite behavior in the two subgroups. Diastolic BP is negatively associated to all-cause mortality in the whole cohort (HR, 0.95; 95% CI, 0.94 to 0.97; P<0.001), the AF or heart failure group (HR, 0.96; 95% CI, 0.94 to 0.98; P<0.001), and the no AF or heart failure group (HR, 0.97; 95% CI, 0.95 to 0.99; P=0.007), but after adjustment, was not predictive in the no AF or heart failure subgroup. Diastolic BP remains an independent risk predictor in the whole cohort and the AF or heart failure subgroup no matter which adjustment model is chosen. Pulse pressure is negatively associated to all-cause mortality in the AF or heart failure group (HR, 0.97; 95% CI, 0.95 to 1.00; P=0.03) and positively in the no AF or heart failure group (HR, 1.04; 95% CI, 1.02 to 1.07; P<0.001). These associations are independent of other risk predictors. Further adjustment models can be found in Supplemental Table 7.
Univariate and adjusted proportional HRs for all-cause mortality including 95% CIs for systolic/diastolic BP and pulse pressure per mm Hg increase
Cardiovascular Mortality
Kaplan–Meier plots suggest again a nonlinear association between systolic blood and pulse pressure and the study end point (Figure 3A). In Figure 3B and Supplemental Figure 1, D–F, one can see the univariate HRs for systolic/diastolic blood and pulse pressure related to the median value for cardiovascular mortality. The pattern for cardiovascular mortality is even more obvious than that of all-cause mortality. SBP and pulse pressure are nonlinearly associated with cardiovascular mortality in univariate and multivariable analysis (all P values for nonlinear term significant; for linear term nonsignificant; Supplemental Table 4, Table 2). Diastolic BP again turned out to be linearly and inversely associated with cardiovascular mortality. Interaction analysis revealed again interactions for SBP (P<0.001) and pulse pressure (P<0.001) with AF and/or heart failure, but not for diastolic BP (P=0.12). Results from nonlinear Cox regression analysis for the subgroups can be seen in in Supplemental Figure 3 and Supplemental Tables 5 and 6. Again, nonlinear effects diminished in the two subgroups anon. Furthermore, the U-shaped association for SBP and pulse pressure to cardiovascular mortality in the whole cohort can be explained by the opposite, and in each case, linear behavior in the two subgroups.
Nonlinear association of pulse pressure with cardiovascular mortality. Univariate association of pulse pressure with cardiovascular mortality in (A) Kaplan–Meier analysis, (B) nonlinear Cox regression analysis using penalized smoothing splines, and proportional hazard models for the (C) AF or HF group and (D) no AF or HF group. HF, heart failure; PP, pulse pressure; Q, quartile.
In Table 4, the results from linear Cox regression analysis are presented for the two subgroups. SBP is an independent risk predictor for and negatively related to all-cause mortality in the AF or heart failure group (HR, 0.95; 95% CI, 0.93 to 0.98; P<0.001) and positively associated in the no AF or heart failure subgroup (HR, 1.03; 95% CI, 1.00 to 1.05; P=0.02). SBP is an independent risk predictor for cardiovascular mortality independently of the chosen adjustment model. Pulse pressure behaves similarly (HR, 0.93; 95% CI, 0.88 to 0.97; P=0.003 in the AF or heart failure group and HR, 1.06; 95% CI, 1.03 to 1.09; P<0.001 in the no AF or heart failure subgroup). Risk prediction is independent of confounding factors (Supplemental Table 8, Table 4).
Univariate and adjusted proportional HRs for cardiovascular mortality including 95% CIs for systolic/diastolic and pulse pressure per mm Hg increase
Summarized, results for all-cause and cardiovascular mortality are similar. Additional analyses such as Kaplan–Meier curves for different risk factors can be found in Supplemental Figures 4 and 5. Furthermore, incidence rates per person year are presented in Supplemental Table 9. As expected, these are significantly higher for patients with AF and/or heart failure for the whole follow-up period and especially in the first follow-up years.
Discussion
In this study, the associations of ambulatory BP recordings and all-cause and cardiovascular mortality in patients on hemodialysis were evaluated. Furthermore, a possible effect of cardiac disease on the observed associations of SBP and pulse pressure with outcomes was examined. The main findings are a U-shaped association for ambulatory SBP and pulse pressure with all-cause and cardiovascular mortality in the whole cohort, which could be explained by the opposite associations, i.e., linear but negative or positive, for patients with and without AF and/or heart failure. To our knowledge, this is the largest study to demonstrate the nonlinear association of ambulatory systolic BP and pulse pressure with all-cause and cardiovascular mortality, and to provide evidence for the explanation of these associations in white patients on hemodialysis. When comparing results from the study with previous evidence, one has to take into account that in contrast to most other studies, no exclusion criteria regarding heart function were defined in the ISAR study.26 This allows us to study the association of ambulatory BP measurement and mortality in a representative sample of stable European patients on dialysis who are willing to undergo 24-hour APBM.
The results of this study are not in direct contrast with existing evidence from the literature. Amar et al.11 showed the association of elevated 24-hour ambulatory pulse pressure with increased risk of cardiovascular mortality in a relatively small cohort (n=57 with ten cardiovascular events), which could have led to overadjustment. Besides, patients with significant cardiac valvular disease or congestive heart failure with low left ventricular ejection fraction (<40%) were excluded. Alborzi et al.12 described an association between interdialytic ambulatory or home BP with mortality in 150 patients on hemodialysis, but no adjustment for any possible confounding factors was reported. Of note, patients with chronic AF were excluded from analysis. Presented HRs according to quartiles showed a weak U-shaped relationship and authors stated that 115–125 mm Hg by ambulatory BP were associated with the best prognosis. Agarwal10 reported a nonlinear relationship for ambulatory and home SBP with all-cause mortality on the basis of a cohort of mainly black patients; patients with chronic AF were again excluded. Nonlinear analysis revealed a W-shaped curve for home and ambulatory BP, with best prognosis for systolic ambulatory BP between 110 and 120 mm Hg. Recently, Sarafidis et al.13 showed, in a cohort of 170 patients on hemodialysis, that ambulatory pulse wave velocity is a stronger predictor for cardiovascular events and mortality than ambulatory systolic BP in a linear analysis, as SBP showed only a nonsignificant U-shaped trend when comparing quartiles. Chronic AF or other arrhythmia that could interfere with BP measurements and congestive heart failure class III and IV on the basis of the New York Heart Association classification were exclusion criteria in this study.
The results of our study underpin the fact that it is important to keep in mind that “the lower the better” paradigm does not apply to all patients,32 especially when evaluating the association of ambulatory BP with outcomes in patients on hemodialysis. Within the BP in Dialysis pilot study, Miskulin et al.33 reported possible safety issues when intensively treating BP in patients on dialysis, but nevertheless highlighted the need for a full-scale trial to assess intensive hypertension management. The presented results for patients with AF or heart failure confirm the inverse (negative) association between SBP and pulse pressure and mortality for patients with heart failure observed in other populations.17–22 This is in contrast to the observed positive association between SBP and pulse pressure with outcome for patients without AF or heart failure reflecting evidence from the general population.7,34,35 For patients with AF, Badheka et al.24 reported a U-shaped association of SBP and mortality, whereas a statistically significant increase in mortality was seen for SBP <110 and >160 mm Hg. In this study, patients with AF had a mean SBP of 117 mm Hg (17.8 SD) and just one (1.5%) patient had an SBP >160 mm Hg and six (8.7%) had an SBP >140 mm Hg. Thus, the mortality increase for higher pressure is not reflected in the ISAR cohort.
The influence of the heart status delivers a possible explanation for the well known U-shaped associations in patients on hemodialysis found in the data and literature, especially for pre-, post-, or nadir-readings in a dialysis session.5,10,36–38 So far, the origin of this phenomena is not completely understood. Overall, peri-dialytic BP recordings are characterized by poor validity and reproducibility, facts that can greatly interfere with the observed associations with outcome.5 Foley et al.39 stated that low BP is a marker of mortality in patients with ESRD and potentially for cardiac failure anteceding death. Other explanations include survival bias and the reverse epidemiology of other cardiovascular risk factors, e.g., nutrition in patients on dialysis.40,41 Bansal et al.42 raised the hypothesis that it is a sign of relative health if patients on hemodialysis can react by increasing BP as a response to fluid accumulated between hemodialysis sessions. Agarwal10 speculated about the unexpected W-shaped relationship for home and ambulatory BP and mortality to be caused by an updated hypertension treatment using antihypertensive medication or dry-weight reduction on the basis of high ambulatory readings, leading to a modification of the relationship. Our data further support the idea that measurement inadequacies during dialysis can be overcome by ambulatory recordings (either including dialysis session or not) and home recordings. In the ISAR cohort, results for ambulatory BP excluding the dialysis session (data not shown) are in line with and underpin the presented findings. Sarafidis et al.5 summarized that the location (i.e., home or ambulatory) and the timing of the BP recording (i.e., out-of-dialysis unit) determine the strong prognostic significance of interdialytic ambulatory BP measurements observed in previous studies.
Strengths of this study include the sample size, follow-up time, and inclusion of patients with AF and heart failure, which enabled us to do the additional analyses. There was no further preselection of patients besides the mentioned exclusion criteria. An important limitation in this study is that recording was limited to 24 and not 48 hours, thus the dialysis-off day could not be incorporated into the analysis. Another issue could be that heart failure was defined only by medical record criteria; although echocardiography was performed in practically every patient with this diagnosis in the past as part of standard clinical care, it was not repeated at baseline for this specific study. Finally, because the study was limited to participants from centers in Munich and the suburban area (i.e., a mainly white population), generalization of these results to other ethnic groups should be done with caution.
In conclusion, this study provides evidence for the U-shaped association between peripheral ambulatory SBP or pulse pressure and mortality in patients on hemodialysis as a whole. Furthermore, it suggests that the opposed associations can be explained by the underlying patients’ cardiac diseases, as linear relationships were evident in both subgroups representing patients with and without AF or heart failure. Hence, proportional hazard models in subgroups overcome interpretational limitations of nonlinear analysis. The findings of this study therefore support the importance of considering existing cardiac diseases when treating hypertension in patients on hemodialysis.
Disclosures
S.W. and C.C.M. are the inventors (not holders) of a patent that is partly used in the ARCSolver algorithm in the Mobil-O-Graph 24h PWA Monitor. The remaining authors declare no conflicts of interest.
Acknowledgments
We acknowledge the support and assistance of all participating dialysis centers, hospitals, and patients for making this study possible.
Research idea and study design: C.C.M., S.W., P.A.S., M.B., and C.S.; data acquisition: J.M., S.A., G.L., and M.C.B.; data analysis/interpretation: C.C.M., S.H., J.M., P.A.S., and S.W.; statistical analysis: C.C.M. and S.W.; supervision or mentorship: C.S., S.W., and U.H. Each author contributed important intellectual content during manuscript drafting or revision and accepts accountability for the overall work by ensuring that questions pertaining to the accuracy or integrity of any portion of the work are appropriately investigated and resolved.
The direct costs of the rISk strAtification in end-stage Renal disease study at the Klinikum rechts der Isar are internally funded by university grants and by grants from the Else-Kröner-Fresenius Stiftung and Servier.
Footnotes
C.C.M. and J.M. contributed equally to this work.
Published online ahead of print. Publication date available at www.jasn.org.
This article contains supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2018010086/-/DCSupplemental.
- Copyright © 2018 by the American Society of Nephrology