Use of vasopressor for dialysis-related hypotension is a risk factor for death in hemodialysis patients: Nationwide cohort study

Because hypotension during hemodialysis (HD) makes continuation of HD difficult and is associated with mortality, pressor approaches are necessary for patients with hypotension. However, the relationships between the pressor approaches and the risk of death have not been clarified yet. We analyzed data from a nationwide prospective cohort study of the Japanese Society for Dialysis Therapy Renal Data Registry (n = 29,309). The outcome was all-cause one-year death. The association between the use of pressor approaches and the outcome was examined using Cox proportional hazards models adjusted for baseline characteristics, propensity score matched analysis and Bayesian networks. The background features of the patients were as follows: male, 59.6%; average age, 64.5 ± 12.5 years; and patients with diabetes mellitus, 31.5%. The pressor group showed a higher risk of the outcome than the control group [adjusted hazard ratio (aHR) 1.33 (95% CI: 1.21, 1.47), p = 0.0001]. Propensity score matched analysis also showed that the matched-pressor group had a higher risk of the outcome than the matched-control group [aHR 1.30 (95% CI: 1.17, 1.45), p = 0.0001]. Moreover, the Bayesian network showed a direct causal relationship from the use of pressor approaches to the outcome. The use of oral vasopressors [aHR 1.20 (95% CI: 1.07, 1.35), p = 0.0018], intravenous injection of vasopressors [aHR 1.54 (95% CI: 1.32, 1.79), p = 0.0001] and normal saline [aHR 1.18 (95% CI: 1.05, 1.33), p = 0.0066] were associated with a high risk of the outcome. In conclusion, this study showed that the use of pressor approaches during HD may be an independent risk factor for death.

treatment 9 . In Japan, the most commonly used vasopressors per os (po) were midodrine hydrochloride, amezinium metilsulfate, etilefrine hydrochloride, and droxidopa. Etilefrine hydrochloride is also used as a vasopressor administered by intravenous injection (iv). Catecholamine preparations are used as vasopressors (iv), but at a low frequency. And a systematic review showed that low-temperature dialysis improves blood pressure during HD and reduces the rate of IDH 10 . However, a cohort study have reported that midodrine is associated with high risk of death 11 . The effects of the treatment of IDH may not be always beneficial.
Although pressor approaches are necessary for patients with hypotension, its effect on HD patients' prognosis has not been clarified yet as far as our literature research has shown. Because a randomized controlled trial can hardly show which combination of pressor approaches is effective in preventing hypotension and improving HD patients' prognosis, there is no evidence of how to use pressor approaches. Therefore, the aims of this study were to investigate the relationships between the use of pressor approaches and one-year all-cause death, and to determine the appropriate methods to use such approaches on the basis of JRDR data.

Results
Baseline characteristics. From the JRDR data, pressor approaches included low-temperature dialysis, the vasopressors (po), and the intravenous injection of medicines [normal saline, high-concentration sodium chloride solution, glycerin, vasopressors (iv)]. The subjects were categorized into the pressor and nonpressor groups on the basis of the use of pressor approaches. The baseline characteristics including biochemical data are shown in Table 1.
The pressor group showed larger numbers of females, subjects with CVD, and subjects with diabetes mellitus (DM) as the cause of end-stage renal disease (ESRD); older age; higher body mass index (BMI); lower serum www.nature.com/scientificreports www.nature.com/scientificreports/ albumin and creatinine levels; higher serum C-reactive protein (CRP) level; and higher fluid removal rate. The mean pre-HD, post-HD blood pressures and minimum systolic and diastolic blood pressures during HD were lower in the pressor group than in the nonpressor group.
The multivariate logistic regression model showed that the pressor group tended to include a high rate of females, older age, CVD, DM, long vintage, high BMI, low serum albumin and creatinine levels, high serum CRP level, and high hemoglobin level, and high fluid removal rate, and low blood pressures ( Table 2).
Risk of death and use of pressor approaches. All-cause, CVD-caused, and infection-caused deaths were more frequently observed in the pressor group than in the nonpressor group (Table 1). U-shaped relationships were observed between blood pressure and the risk of all-cause death ( Supplementary Fig. S1). Kaplan-Meier analysis showed that the pressor group had a higher mortality rate than the nonpressor groups ( Fig. 1). Cox proportional hazards models (PHMs) and adjusted Cox PHMs showed that the pressor group showed a high risk of all-cause death (Table 3). Competing risk regression models showed that the risks of CVD-and infection-caused deaths in the pressor group were higher than those in the nonpressor group (Table 4).
propensity score-matched analysis and Bayesian network (BN). There were no significant differences between the baseline characteristics of the matched pressor and matched nonpressor groups except for post-HD and minimum systolic and diastolic blood pressures (Supplementary Table 1). Kaplan-Meier analysis showed that the matched pressor group had a higher mortality rate than the matched nonpressor groups ( Supplementary Fig. S2). Cox PHMs showed that the use of pressor approaches was independently associated with a higher risk of the all-cause death (Supplementary Table 2).
Using the dataset of all subjects, BN including all blood pressures showed causal relationships between variables ( Fig. 2). All-cause death was affected by the use of pressor approaches; serum albumin, creatinine, and CRP levels; BMI; and CVD. The use of pressor approaches were affected by the minimum systolic blood pressure, being male, older age, DM, use of depressors, CVD, and fluid removal rate. Each BN including each blood pressure showed similar relationships between blood pressure, pressor approaches, and all-cause death (results not shown).
BN also showed the factors associated with the use of pressor approaches such as gender, old age, a history of CVD, DM, high fluid removal rate, and low blood pressure. types of pressor approaches and risk of all-cause death. There were weak differences in the baseline characteristics between the uses of pressor approaches ( Table 5). The groups of glycerin, vasopressors (iv) and (po) showed older age. The groups of normal saline and high-concentration sodium chloride showed smaller numbers of subjects with CVD. The glycerin group showed smaller numbers of subjects with DM and lower fluid removal rate. The vasopressor (iv) group showed lower blood pressures. And the groups of glycerin and vasopressors (iv) showed higher risk of all-cause and CVD-caused death. www.nature.com/scientificreports www.nature.com/scientificreports/ The relationship between the risk of all-cause death and the type of pressor approaches was evaluated by multivariate Cox PHMs using the dataset of all subjects ( Table 6). The use of normal saline, vasopressor (iv), and vasopressor (po) was associated with a high risk of all-cause death.

Discussion
This study using large-scale cohort data, showed that the pressor approaches were independent risk factors for all-cause death. Because observational analyses of pressor approaches and outcomes are subject to bias owing to unmeasured confounders, propensity score matched analysis was used in this study to minimize the bias, which showed that the pressor group had high risk of all-cause death. BN suggested that all-cause death was directly affected by the use of pressor approaches. From these results, it was suggested that the use of pressor approaches was an independent risk factor for death. There has been no report on the harmful effects of pressor approaches on patients' prognosis as far as we searched the literature of prospective interventional studies.
IDH is caused by many factors, such as low dry weight, excessive ultrafiltration, decrease in osmolality, and autonomic neuropathy 5,12,13 . Pressor approaches are usually administered to prevent and control IDH 5,13 . In this study, the harmful effects of administration of normal saline, vasopressors (iv), and vasopressors (po) on patients' prognosis were observed. On the other hand, low-temperature dialysis, and injection of high-concentration sodium chloride solution and high-concentration glycerin were not associated with the risk of all-cause death. Normal saline is usually given to replace intravascular volume as a method of acute management of IDH, and to effectively maintain blood pressure 14 . The harmful effects of sodium loading have been reported 15 . Some of the  www.nature.com/scientificreports www.nature.com/scientificreports/ harmful effects of normal saline are the acute change in circulating plasma volume, poor cardiac function, and complications of IDH. However, considering that the use of high-concentration sodium chloride solution and high-concentration glycerin was not associated with the risk of all-cause death, intravascular volume loading may have a strong effect on the risk of all-cause death, and the effect of sodium loading may not be strong.
Our study showed the relationship between the use of vasopressors and the risk of all-cause death. A cohort study showed that the use of midodrine is associated with risk of death 11 . There were also reports that midodrine hydrochloride and amezinium metilsulfate might have worsened the leg ulcers in HD patients 16,17 . IDH causes hypoperfusion in organs causing myocardial infarction, stroke, and bowel ischemia, and leads to myocardial fibrosis and cardiac remodeling 18 . HD-induced myocardial stunning is associated with the development of heart failure and increased risk of all-cause death in HD patients [19][20][21] . IDH is associated with low cardiac index and high peripheral resistance 22 . Considering the mechanism of action of vasopressors, vasoconstriction induced by the vasopressors may decrease the blood supply to peripheral arteries, and worsen HD patients' prognosis. If an HD patient's blood pressure can be maintained during HD by pressor approaches excluding the use of vasopressors, it would be better to avoid using vasopressors as much as possible.
Low-temperature dialysis is used for IDH by increasing peripheral vascular resistance and improving cardiac output. In our study, low-temperature dialysis was not associated with a risk of all-cause death. A systematic review of 26 studies showed that low-temperature dialysis reduces the rate of IHD and increases intradialytic mean arterial pressure 10 . The candidate mechanisms underlying the favorable effects of low-temperature dialysis on HD are the improvement of intradialytic hemodynamics by the preservation of cardiac output and central blood volume, and the slowing of the progression of HD-associated cardiomyopathy, brain protection, and improvement of nocturnal sleep [23][24][25][26][27] . These effects are different from those of vasopressors, and may affect patients' prognosis. Although discomfort is more often observed in low-temperature dialysis than in normal-temperature dialysis, low-temperature dialysis can be a candidate therapy to prevent IDH. Low-temperature dialysis is also recommended by the guidelines 5,13 .
In this study, logistic regression models and BN were used to investigate the factors associated with the use of pressor approaches. Logistic regression models and BN showed the common factors such as gender, old age, a history of CVD, DM, high fluid removal rate, and low blood pressure. These characteristics were in accordance with previous studies 28,29 . Patients with severe hypotension that is difficult to control have a high risk of all-cause death. According to previous studies, comorbid conditions such as coronary artery disease, systolic dysfunction, left ventricular hypertrophy, and autonomic dysfunction are observed in patients with IDH 18,28-31 . These factors indicate the difficulty of fluid removal, and hidden risks of cardiac dysfunction in such patients.
No relationship between serum albumin level and the use of pressor approaches was observed. However, this does not suggest that a low serum albumin level is not an important risk factor, but means that hypotension due to hypoalbuminemia may be difficult to control using pressor approaches. Albumin preparations are blood products and are difficult to use for daily HD. Hypoalbuminemia is caused by not only malnutrition but also inflammation and comorbid conditions 32 . To control hypoalbuminemia, it is important to identify its underlying causes and treat them.  www.nature.com/scientificreports www.nature.com/scientificreports/ In this study, the numbers of the subjects using vasopressor (po) and normal saline were higher than those using other pressor approaches. This finding might reflect a tendency that in Japan, a vasopressor (po) is commonly used for IDH as the first choice, and that normal saline is used when a patient's blood pressure suddenly decreases. However, considering these results of our study and the JSDT and K/DOQI guidelines, the acute management of IDH by increasing intravascular volume using normal saline and use of vasopressors should be avoided 5,11,13 . Moderate fluid removal should be a key therapeutic strategy, for which reassessment of dry weight, regulation of ultrafiltration rate, extended hours of HD, and changes to other modes such as extracorporeal ultrafiltration method, hemodiafiltration, daily HD, and nocturnal HD may be effective 33,34 .
This study has several limitations. First, because of the observational nature of this study, the results may be biased by unmeasured confounders. To overcome this limitation, propensity matching analysis was conducted. However, JRDR data did not include the baseline characteristics of HD patients at the time of their first use of pressor approaches. Thus, to examine the causal relationship between vasopressors and the risk of all-cause death, BN was conducted, and these analyses showed similar results. Although propensity matching analysis and BN were conducted, there remains some bias. For example, pressor approaches were often used for patients with high risk of death. Second, we did not include the patients with missing data in this study, which might have caused selection bias. Third, the JRDR data did not standardize the methods of blood pressure measurement. However, the methods were almost similar throughout Japan. Therefore, despite the difference in the method of measuring blood pressure, the relationship between pressor approaches and the risk of all-cause death is considered robust. Fourth, in this study, the use of high-concentration sodium chloride and normal saline, which increase intravascular volume, tended to be avoided in patients with CVD. There were biases in the indication of the use of pressor approaches in this study, and biases in the difference in sample size between pressor approaches. However, although the glycerin group showed a higher risk of death than other groups, the Cox proportional hazard models showed no statistically significant relationship between the use of glycerin and the risk of death. www.nature.com/scientificreports www.nature.com/scientificreports/ The indication bias might not be strong and might be adjusted in the models. Because the details on the use of pressor approaches were not included in the JRDR data, we were unable to evaluate the effects of the patterns of the use of pressor approaches. Randomized controlled trials are needed to evaluate the effects of the use of pressor approaches on dialysis patients' mortality. Fifth, there was some bias due to the medical situations with regard to the use of pressor approaches in Japan such as their coverage by national health insurance and health care policies at facilities. Because there may be differences in IDH treatment between countries, to show external validity, comparison of the use of pressor approaches between countries is needed. Sixth, although IDH is pathophysiologically associated with cardiac output, arterior vasoconstrictrion, and autonomic system, in this study, we lacked pathophysiological data or details on CVD death such as myocardial infarction and arrythmia 4 . More studies are needed to investigate the pathophysiological mechanism of IDH and medication. Seventh, there were some differences in the factors associated with the use of pressor approaches between the logistic regression models and BN such as increases in BMI and hemoglobin level. The logistic regression models might not be appreciable to the complex relationships between variables. More details on medication are needed to determine the mechanisms of development of IDH. Vintage (years) 7.2 ± 5.7 5.7 (3, 9.5) 7.9 ± 6.7 5.7 (3, 10.5) 8 ± 6.7 5.9 (3.1, 10.8) 7.9 ± 6.6 6.1 (2.9, 11) 7.4 ± 6.3 5.4 (2.9, 9.6) 7.4 ± 6.2 5.5 (3, 9.   www.nature.com/scientificreports www.nature.com/scientificreports/ In conclusion, this study showed that the HD patients for whom pressor approaches were used during HD have high risk of death. In particular, the use of normal saline, and vasopressors is associated with a poor prognosis. To improve HD patients' prognosis, it may be effective to prevent a situation that requires acute management of IDH.

Methods
Dataset. This is a prospective cohort study of maintenance HD patients using JRDR data. JSDT has been conducting annual year-end surveys of dialysis facilities in Japan since 1968. The JRDR data from 2005 to 2006 were used in this study. This study was approved by the ethics committee of JSDT and was exempt from the need to obtain informed consent from participants (JSDT No. 18). The study was performed in accordance with the relevant guidelines and the Declaration of Helsinki.
The subjects of this cohort study were the 245,441 patients. The exclusion criteria were as follows: patients younger than twenty years; patients on hemodiafiltration, hemofiltration, or peritoneal dialysis; patients with missing values or outlier values of laboratory data; patients who had a limb amputated; and patients with a HD vintage of less than one year. Thus, 29,309 subjects were included in the analysis. The sample size was evaluated to maximize statistical power.
The baseline data were as follows: gender; age; history of CVD; DM as a cause of ESRD; vintage; post-HD BMI; serum albumin, creatinine, and CRP levels; hemoglobin level; and fluid removal rate. The laboratory data were measured before HD. Fluid removal rate (%) was calculated as follows: (post-HD weight − pre-HD weight)/ pre-HD weight × 100. Data on Pre-HD, post-HD blood pressures and minimum systolic and diastolic blood pressures during HD were collected. In this study, methods of blood pressure measurement were not standardized, because JDRD data were collected by survey of dialysis facilities. However, the general method of blood pressure measurement in Japan at that time was as follows 5 . Blood pressure was measured under fixed conditions in either the seated or supine position depending on the setup at each facility. Pre-HD blood pressure was measured before the start of dialysis. Post-HD blood pressure was measured just before returning the blood at the end of dialysis. The lowest blood pressure was recorded as the minimum blood pressure during HD. Information on depressors [angiotensin-converting-enzyme inhibitors (ACEIs) and angiotensin II receptor blockers (ARBs)] used was collected. Information on methods used for raising blood pressure (pressors) was also collected, namely, low-temperature dialysis, vasopressors (po), and medicines for intravenous injection (normal saline, high-concentration sodium chloride solution, glycerin, vasopressors iv).
The primary outcome was all-cause death within one year. CVD-and infection-caused deaths within one year were also evaluated. Subjects were categorized on the basis of the use of pressor approaches: the pressor and nonpressor groups. statistical analyses. Normally distributed variables are presented as mean ± standard deviation; otherwise, the median and interquartile ranges are presented. Highly skewed variables were transformed with the natural logarithm function prior to their use in models [ln(vintage), ln(CRP)]. Intergroup comparisons of parameters were performed using the chi-square test, t-test, and Mann-Whitney U test as appropriate after the F-test of equality of variances between the groups.
Using the Kaplan-Meier method, we compared the pressor and nonpressor groups in terms of the incidence of the primary outcome, and statistical significance was evaluated using the log-rank and Wilcoxon tests. A multivariate Cox PHM was used to evaluate the relationships of the pressor and nonpressor groups with mortality. This multivariate Cox PHM was adjusted for baseline characteristics [gender, age, CVD, DM, ln(vintage), BMI, serum albumin, and creatinine levels, ln(CRP), hemoglobin level, and fluid removal rate, and use of depressors] and blood pressures using the splines of the blood pressures. The results were presented as HRs with 95% confidence intervals (CIs). In the analysis of competing risks of cause-specific death, Fine and Gray competing risk regression models adjusted for the baseline characteristics were also examined.
A propensity score-matched cohort for the use of pressor approaches was conducted. Possible confounders were chosen for baseline characteristics, including baseline characteristics [gender, age, CVD, DM, ln(vintage), BMI, serum albumin and creatinine levels, ln(CRP), hemoglobin level, fluid removal rate, and use of depressors], and pre-HD systolic blood pressure. The predicted probability of the use of pressor approaches was calculated by applying a logistic regression model including these possible confounders. Next, we performed propensity score matching using the following algorithm: 1:1 nearest neighbor match within a caliper width defined as ±0.2 of a standard deviation (SD) of the logit of the propensity score and no replacement. Comparisons of baseline characteristics were based on the chi-square test, t-test, Mann-Whitney U, and standardized differences. The incidence of the primary outcome was compared between the matched pressor and nonpressor groups using the Kaplan-Meier method. A Cox PHM was used to evaluate the association between the matched pressor and nonpressor groups and the primary outcome. Adjustment in the multivariate Cox PHM was demonstrated for variables that showed absolute standardized differences (10% in baseline characteristics) between the matched pressor and nonpressor groups. The relationship between the types of pressor approaches and the risk of all-cause death was also evaluated using multivariate Cox PHMs adjusted for baseline characteristics and blood pressures.
BN is a kind of probabilistic graphical model that shows variables and their causal relationships via a directed acyclic graph, and represents the probabilistic relationships between diseases and symptoms. BN was used to evaluate the relationship between the variables described above. The incremental association Markov blanket method was used for structure learning algorithm for BN. The resulting directed acyclic graph was interpreted as the causal BN. Continuous variables were discretized by cutoff levels determined by receiver operating characteristic curves for the prediction of the primary outcome. Blood pressures were categorized as follows: systolic blood pressures, 100 mmHg, 150 mmHg, and 200 mmHg; diastolic blood pressures, 50 mmHg, 100 mmHg. These