www.medscape.

com

Hemoglobin Concentration and the Risk of Hemorrhagic and Ischemic

Stroke in Patients Undergoing Hemodialysis
The Q-cohort Study

Ryusuke Yotsueda; Shigeru Tanaka; Masatomo Taniguchi; Kiichiro Fujisaki; Kumiko Torisu; Kosuke Masutani; Hideki Hirakata;
Takanari Kitazono; Kazuhiko Tsuruya

Nephrol Dial Transplant. 2018;33(5):856-864.

Abstract and Introduction

Abstract

Background The contribution of the hemoglobin concentration to the incidence of hemorrhagic or ischemic stroke in patients
undergoing hemodialysis is unclear.

Methods In total, 3436 patients undergoing prevalent hemodialysis were followed up for 4 years. The primary outcome was the
first development of hemorrhagic or ischemic stroke. The baseline hemoglobin concentration was divided into quartiles
[hemoglobin (g/dL): Q1, ≤9.7; Q2, 9.8–10.5; Q3, 10.6–11.1; Q4, ≥11.2]. The association between the hemoglobin concentration
and each type of stroke was examined using the Kaplan–Meier method and a Cox proportional hazards model.

Results During the follow-up period, 76 (2.2%) patients developed hemorrhagic stroke and 139 (4.0%) developed ischemic
stroke. The 4-year incidence rate of hemorrhagic stroke was significantly higher in patients with lower hemoglobin concentrations.
Compared with the quartile of patients with the highest hemoglobin concentrations (Q4), the multivariable-adjusted hazard ratios
for hemorrhagic stroke were 1.18 (95% confidence interval, 0.56–2.51), 1.59 (0.82–3.21) and 2.31 (1.16–4.73) in patients in Q3,
Q2 and Q1, respectively. No association was identified between the 4-year incidence rate of ischemic stroke and the hemoglobin
concentration. Compared with the quartile of patients with the lowest hemoglobin concentrations (Q1), the multivariable-adjusted
hazard ratios for ischemic stroke were 1.17 (95% confidence interval, 0.73–1.89), 0.88 (0.51–1.51) and 1.10 (0.66–1.83) in
patients in Q2, Q3 and Q4, respectively.

Conclusions Our results suggest that low hemoglobin concentrations are associated with a high risk of hemorrhagic stroke, but
not of ischemic stroke, in patients undergoing hemodialysis.

Introduction

Stroke is a major cause of death and disability in patients undergoing maintenance hemodialysis (HD).[1–3] Approximately 7% of
deaths in patients undergoing HD in Japan are still caused by stroke despite the fact that these patients have improved longevity
and reduced rates of cardiovascular death.[3] Therefore, identifying the risk factors for stroke is urgently needed to facilitate
improvements in the prognosis of patients undergoing HD.

A close relationship between high hemoglobin concentrations and the incidence of stroke has been highlighted in the field of
chronic kidney disease (CKD), including HD therapy.[4] The Trial to Reduce Cardiovascular Events with Aranesp Therapy
(TREAT) was a large-scale randomized controlled trial involving patients with CKD receiving an erythropoiesis-stimulating agent
(ESA). The TREAT study showed that there was a double risk of stroke in the patient group with higher target hemoglobin
concentrations.[4] High hemoglobin concentrations increase blood viscosity and promote thrombosis,[5] which may at least in part
explain the association between high hemoglobin concentrations and ischemic stroke. However, the TREAT study did not address
the risk of ischemic stroke separate from hemorrhagic stroke.

Low hemoglobin concentrations have also emerged as an established risk factor for the development of cardiovascular disease in
the general population[6] and in patients with CKD.[7,8] The interactions among a poor cardiovascular prognosis, kidney function
and anemia form a vicious circle termed cardiorenal anemia syndrome.[9] Observational studies of the general population have
shown an association between low hemoglobin concentrations and a high incidence of composite stroke.[10,11] In this context, the
Hisayama study, a prospective population-based cohort survey in Japan, recently showed that low hematocrit levels were
associated with a high risk of hemorrhagic stroke in adults in the general population.[12] These findings suggest an association
between anemia and hemorrhagic stroke in the real-world setting. However, the contribution of anemia to the incidence of
hemorrhagic stroke remains unclear in patients with CKD, including those undergoing HD.

Therefore, the present study was performed to determine the association between the hemoglobin concentration and the risk of
hemorrhagic or ischemic stroke separately in a large-scale, prospective cohort study of patients undergoing maintenance HD.

Materials and Methods

Study Cohort

The Q-Cohort Study was a multicenter, prospective observational study of patients undergoing HD at 39 dialysis facilities in
Fukuoka and Saga Prefectures in Kyushu, Japan.[13–16] A total of 3598 outpatients aged >18 years who underwent prevalent HD
participated in the study. All participants were enrolled from December 2006 to December 2007 and followed up until December
2010. Patients without demographic data (n = 65) and patients whose outcomes were missing (n = 97) were excluded. The
remaining 3436 patients were analyzed in this study. All patients provided written informed consent. The Q-Cohort Study was
approved by the Kyushu University Institutional Review Board for Clinical Research (Approval Number: 20–31) and was
registered in the University Hospital Medical Information Network (UMIN) clinical trial registry (UMIN000000556). The study was
performed according to the Ethics of Clinical Research (Declaration of Helsinki) requirements.

Definition of Outcomes

The primary outcomes were the incidence of first development of hemorrhagic or ischemic stroke. The definitions of each type of
stroke were previously published.[15] Briefly, stroke was defined as sudden onset of a neurological deficit persisting for more than
24 h. The types of stroke were determined by the attending physician using imaging of the brain, including computed tomography
and magnetic resonance imaging. All events were collected from the paper medical records by the local physician in each dialysis
facility. Each of these physicians checked the medical records for the occurrence of specified outcomes and reported them to the
steering committee annually. Subarachnoid hemorrhage and hemorrhage after ischemic stroke were excluded from the definition
of hemorrhage stroke. During the follow-up period, three patients developed both hemorrhagic and ischemic stroke. They were
classified as having developed the first type of stroke and were censored for the subsequent stroke.

Measurement of Hemoglobin Concentration

For measurement of the baseline hemoglobin concentration, blood samples were collected from a vascular access in the
recumbent position before dialysis after a 2-day interdialytic interval. The hemoglobin concentration was measured using standard
procedures at multiple facilities depending on the location of each dialysis center.

Measurement of Risk Factors

Details regarding the measurement of risk factors were previously published.[13–16] Briefly, demographic data [e.g. age, sex,
duration of HD, cause of end-stage kidney disease (diabetic nephropathy or other), history of stroke, use of antihypertensive
drugs, use of iron supplementation and dose of ESA] and biochemical parameters [e.g. serum concentrations of albumin, C-
reactive protein (CRP), ferritin, total cholesterol and phosphate; Kt/V; systolic blood pressure (BP); body weight; and
cardiothoracic ratio (CTR)] were collected at baseline by physicians at each dialysis facility. A history of stroke included
hemorrhagic stroke and ischemic stroke. Iron supplementation was provided orally or intravenously. The types of ESAs used in
the present cohort were epoetin alfa, epoetin beta and darbepoetin alfa. For darbepoetin alfa, the ESA dose was calculated by
multiplying the darbepoetin alfa dose (μg) by 200. Dialysis doses were measured by a single-pool Kt/V method. All available data
on BP were measured before dialysis. The CTR was calculated from a chest radiograph in an upright posterior–anterior view. The
horizontal diameter of the heart was divided by the horizontal inner width of the rib cage.

Statistical Analysis

Baseline data according to quartiles (Q) of hemoglobin concentrations are shown as mean ± standard deviation for normally
distributed continuous variables, median (interquartile range) for non-normally distributed continuous variables and number
(percentage) for categorical variables. The distribution of baseline data divided by quartiles of hemoglobin concentrations was
compared using trend analysis. The incidence rates for hemorrhagic and ischemic stroke according to the hemoglobin
concentrations were plotted by the Kaplan–Meier method and compared by the log-rank test. The unadjusted, age- and sex-
adjusted, fully adjusted and parsimoniously adjusted hazard ratios with 95% confidence intervals of hemorrhagic and ischemic
stroke according to the hemoglobin concentrations were calculated using a Cox proportional hazards model. The assumptions of
the proportional hazards were graphically checked using the log cumulative hazard plots for each type of stroke according to the
hemoglobin concentrations. Because an association between a low hemoglobin concentration and hemorrhagic stroke was
assumed, the highest quartile of hemoglobin concentrations was chosen as the reference for the Cox model of hemorrhagic
stroke. Because an association between a high hemoglobin concentration and ischemic stroke was assumed, the lowest quartile
of hemoglobin concentrations was chosen as the reference for the Cox model of ischemic stroke. The fully adjusted model was
adjusted for age, sex, dialysis duration, cause of end-stage kidney disease, history of stroke, use of antihypertensive drugs, use of
iron supplementation, dose of ESA, serum albumin concentration, serum CRP concentration, serum ferritin concentration, serum
total cholesterol concentration, serum phosphate concentration, Kt/V, systolic BP, body weight and CTR. Variables in the fully
adjusted model were based on a priori clinical judgment and previous studies.[5,11,12,15] Variables in the parsimoniously adjusted
model were selected using the Cox proportional hazard model and a stepwise backward method with P < 0.05 for the remaining
variables to determine the risk factors for each type of stroke. Continuous multivariable-adjusted associations of the hemoglobin
concentrations and hazard ratios with 95% confidence intervals for hemorrhagic and ischemic stroke were plotted using restricted
cubic splines. We used four knots located at the 5th, 35th, 65th and 95th percentiles of the hemoglobin concentrations as
previously recommended.[17] The multivariable-adjusted model was adjusted for the same variables as the fully adjusted Cox
model. Age, dialysis duration, dose of ESA, serum albumin concentration, serum CRP concentration, serum ferritin concentration,
serum total cholesterol concentration, serum phosphate concentration, Kt/V, systolic BP, body weight and CTR were treated as
spline terms. The mean hemoglobin concentration (10.5 g/dL) was chosen as the reference for each spline plot. Heterogeneity in
the correlation between subgroups was tested by adding a multiplicative interaction term to the relevant Cox model. In the
subgroup analysis, multivariable-adjusted hazard ratios with 95% confidence intervals of hemorrhagic and ischemic stroke for
every 1 g/dL change in the hemoglobin concentration were calculated. The multivariable-adjusted model was adjusted for the
same variables as the parsimoniously adjusted Cox model for each type of stroke. Statistical calculations were performed using
JMP version 11 for Windows (SAS Institute, Cary, NC, USA), SAS software package version 9.3 (SAS Institute) and R version
3.2.4 (http://www.r-project.org). A difference was considered significant at P < 0.05.

Results
Baseline Characteristics According to Quartiles of Hemoglobin Concentrations

The baseline characteristics of the patients based on the quartiles of hemoglobin concentrations are shown in . Patients with
lower hemoglobin concentrations were older, were more frequently female and had a higher frequency of iron supplementation.
The ESA dosage, serum CRP level, Kt/V and CTR were higher in patients with lower hemoglobin concentrations. In contrast, the
serum albumin level, serum total cholesterol level, serum phosphate level and body weight were lower in patients with lower
hemoglobin concentrations.

Table 1. Baseline characteristics according to quartiles of hemoglobin concentrations

Hemoglobin concentrations
Q1 (n = 775) Q2 (n = 943) Q3 (n = 769) Q4 (n = 949)
Total (n = P for
Variables 3436) ≤9.7 g/dL 9.8–10.5 g/dL 10.6–11.1 g/dL ≥11.2 g/dL trend
Age, years 63.7 ± 12.8 64.8 ± 13.2 64.1 ± 12.7 63.5 ± 12.7 62.4 ± 12.4 <0.001
Female sex, n (%) 1405 (40.9) 373 (48.1) 410 (43.5) 307 (39.9) 315 (33.2) <0.001
Hemodialysis duration, years 5.5 (2.1–11.5) 5.0 (1.9–11.4) 6.1 (2.6–11.6) 5.2 (2.3–11.5) 5.3 (2.0– 0.71
11.6)
Cause of ESKD: diabetes, n 994 (28.9) 218 (28.1) 274 (29.1) 222 (28.9) 280 (29.5) 0.58
(%)
History of stroke, n (%) 535 (15.6) 133 (17.1) 128 (13.6) 132 (17.2) 142 (15.0) 0.63
Use of antihypertensive 2152 (62.6) 466 (60.1) 623 (66.1) 487 (63.3) 576 (60.7) 0.71
agent, n (%)
Use of iron supplementation, 228 (35.7) 320 (41.3) 335 (35.5) 264 (34.3) 309 (32.6) <0.001
n (%)
Dose of ESA, U/week 3000 (1500– 4500 (3000– 3000 (1500– 3000 (1500– 2250 (0– <0.001
4500) 9000) 4500) 4500) 4500)
Serum albumin, g/dL 3.8 ± 0.4 3.6 ± 0.5 3.8 ± 0.4 3.9 ± 0.4 3.9 ± 0.4 <0.001
Serum C-reactive protein, 0.13 (0.06– 0.20 (0.09– 0.15 (0.05– 0.13 (0.05– 0.10 (0.05– <0.001
mg/dL 0.30) 0.53) 0.32) 0.30) 0.26)
Serum ferritin, ng/mL 163 (68–299) 170 (76–325) 163 (70–298) 167 (71–299) 156 (61–291) 0.24
Serum total cholesterol, 155.8 ± 36.6 152.1 ± 36.7 156.1 ± 37.5 156.2 ± 36.5 158.0 ± 35.6 0.002
mg/dL
Serum phosphate, mg/dL 4.9 ± 1.2 4.8 ± 1.2 4.9 ± 1.1 5.0 ± 1.2 5.1 ± 1.2 <0.001
Single-pool Kt/V 1.58 ± 0.28 1.60 ± 0.30 1.60 ± 0.27 1.58 ± 0.28 1.53 ± 0.26 <0.001
Systolic blood pressure, 153.0 ± 23.4 152.3 ± 24.7 154.2 ± 23.5 152.7 ± 21.5 152.8 ± 23.8 0.92
mmHg
Body weight, kg 53.8 ± 11.2 52.1 ± 10.2 53.5 ± 11.3 53.9 ± 10.9 55.4 ± 12.0 <0.001
Cardiothoracic ratio, % 50.0 ± 5.5 51.6 ± 5.8 50.7 ± 5.4 50.3 ± 5.4 49.7 ± 5.3 <0.001

Values are presented as mean ± standard deviation for normally distributed continuous variables, median (interquartile range) for
non-normally distributed continuous variables and number (percentage) for categorical variables.
ESA, erythropoiesis-stimulating agent; ESKD, end-stage kidney disease; Q, quartile.

Association Between Hemoglobin Concentration and Risk of Hemorrhagic Stroke

During a 4-year follow-up period, 76 (2.2%) patients developed hemorrhagic stroke. The incidence rates of hemorrhagic stroke
according to the quartiles of hemoglobin concentrations are shown in Figure 1A. The 4-year incidence rate significantly increased
with lower hemoglobin concentrations (P = 0.03). Patients with the lowest hemoglobin concentrations (Q1) had a 2.31-fold (95%
confidence interval, 1.16–4.73) higher risk of hemorrhagic stroke than those with the highest hemoglobin concentrations (Q4) after
adjustment for full variables (). An inverse association between the quartiles of hemoglobin concentrations and the fully adjusted
hazard ratio for the incidence of hemorrhagic stroke was also evident (P for trend = 0.01). The parsimonious model selected by
the stepwise backward method was adjusted for the sensitivity analysis; this essentially resulted in no changes in the
associations. The continuous multivariable-adjusted association between the hemoglobin concentration and the risk of
hemorrhagic stroke showed a similar relationship (Figure 2A). The risk of hemorrhagic stroke tended to be higher with lower
hemoglobin concentrations. In the upper range of hemoglobin concentrations, the risk of hemorrhagic stroke was significantly
lower than the reference.

Table 2. Hazard ratios for hemorrhagic stroke by quartiles of hemoglobin concentrations

Hemoglobin concentrations
Q1 Q2 Q3 Q4
9.8–10.5 10.6–11.1 P for Every 1 g/dL decrease in
  ≤9.7 g/dL g/dL g/dL ≥11.2 g/dL trend hemoglobin concentration
Number of patients 775 943 769 949    
 Number of events (%) 25 (3.2) 23 (2.4) 14 (1.8) 14 (1.5)    
Unadjusted HR (95% CI) 2.41 1.68 1.23 1.00 0.004 1.31 (1.08–1.58)
(1.27– (0.87– (0.58– (reference)
4.76) 3.34) 2.60)
Age- and sex-adjusted 2.68 1.79 1.28 1.00 0.002 1.35 (1.11–1.63)
HR (95% CI) (1.41– (0.93– (0.61– (reference)
5.33) 3.57) 2.72)
Fully adjusted HR (95% 2.31 1.59 1.18 1.00 0.01 1.31 (1.05–1.63)
CI) (1.16– (0.82– (0.56– (reference)
4.73) 3.21) 2.51)
Parsimoniously adjusted 2.19 1.54 1.21 1.00 0.02 1.26 (1.02–1.56)
HR (95% CI) (1.11– (0.79– (0.57– (reference)
4.48) 3.09) 2.57)

The fully adjusted model is adjusted for age, sex, dialysis duration, cause of end-stage kidney disease, history of stroke,
antihypertensive drug use, iron supplementation use, dose of erythropoiesis-stimulating agent, serum albumin concentration,
serum C-reactive protein concentration, serum ferritin concentration, serum total cholesterol concentration, serum phosphate
concentration, Kt/V, systolic blood pressure, body weight and cardiothoracic ratio. The parsimoniously adjusted model is adjusted
for sex, dose of erythropoiesis-stimulating agent, serum phosphate concentration, Kt/V and systolic blood pressure, which were
selected using the Cox proportional hazard model and a stepwise backward method with P < 0.05 for the remaining variables to
determine the risk factors for the incidence of hemorrhagic stroke.
CI, confidence interval; HR, hazard ratio; Q, quartile.

Figure 1.

Incidence rate of (A) hemorrhagic stroke and (B) ischemic stroke according to quartiles of hemoglobin concentrations during the
4-year follow-up. Q, quartile.
Figure 2.

Multivariable-adjusted restricted cubic spline plots of the hazard ratio for (A) hemorrhagic stroke and (B) ischemic stroke
according to hemoglobin concentrations. The solid lines represent the hazard ratio and the dotted lines represent the 95%
confidence interval. The horizontal gray lines correspond to the normal reference hazard ratio of 1.0. The value of 10.5 g/dL is the
mean hemoglobin concentration and was chosen as the reference. The multivariable-adjusted model is adjusted for age, sex,
dialysis duration, cause of end-stage kidney disease, history of stroke, use of antihypertensive drugs, use of iron supplementation,
dose of erythropoiesis-stimulating agent, serum albumin concentration, serum C-reactive protein concentration, serum ferritin
concentration, serum total cholesterol concentration, serum phosphate concentration, Kt/V, systolic blood pressure, body weight
and cardiothoracic ratio.

Association Between Hemoglobin Concentration and Risk of Ischemic Stroke

During the follow-up period, 139 (4.0%) patients developed ischemic stroke. The incidence rates of ischemic stroke according to
the quartiles of hemoglobin concentrations are shown in Figure 1B. There was no difference in the incidence rate of ischemic
stroke by each quartile according to the hemoglobin concentration. Compared with Q1, the fully adjusted risks of ischemic stroke
were 1.17 (95% confidence interval, 0.73–1.89), 0.88 (0.51–1.51) and 1.10 (0.66–1.83) in patients in Q2, Q3 and Q4, respectively
(). The continuous multivariable-adjusted association between the hemoglobin concentration and the risk of ischemic stroke is
shown in Figure 2B. There was no statistically significant association.

Table 3. Hazard ratios for ischemic stroke by quartiles of hemoglobin concentrations

Hemoglobin concentrations
Q1 Q2 Q3 Q4
9.8–10.5 10.6–11.1 P for Every 1 g/dL increase in
  ≤9.7 g/dL g/dL g/dL ≥11.2 g/dL trend hemoglobin concentration
Number of patients 775 943 769 949    
Number of events (%) 32 (4,1) 43 (4.6) 26 (3.4) 38 (4.0)    
Unadjusted HR (95% CI) 1.00 1.02 0.74 0.88 0.39 0.99 (0.85–1.14)
(reference) (0.65– (0.44– (0.55–
1.62) 1.25) 1.42)
Age- and sex-adjusted 1.00 1.04 0.78 0.97 0.64 1.02 (0.88–1.19)
HR (95% CI) (reference) (0.66– (0.46– (0.60–
1.65) 1.31) 1.56)
Fully adjusted HR (95% 1.00 1.17 0.88 1.10 0.99 1.08 (0.92–1.27)
CI) (reference) (0.73– (0.51– (0.66–
1.89) 1.51) 1.83)
Parsimoniously adjusted 1.00 1.08 0.78 1.00 0.70 1.03 (0.89–1.20)
HR (95% CI) (reference) (0.68– (0.46– (0.62–
1.72) 1.31) 1.61)
The fully adjusted model is adjusted for age, sex, dialysis duration, cause of end-stage kidney disease, history of stroke,
antihypertensive drug use, iron supplementation use, dose of erythropoiesis-stimulating agent, serum albumin concentration,
serum C-reactive protein concentration, serum ferritin concentration, serum total cholesterol concentration, serum phosphate
concentration, Kt/V, systolic blood pressure, body weight and cardiothoracic ratio. The parsimoniously adjusted model is adjusted
for age, cause of end-stage kidney disease, history of stroke and cardiothoracic ratio, which were selected using the Cox
proportional hazard model and a stepwise backward method with P < 0.05 for the remaining variables to determine the risk factors
for the incidence of ischemic stroke.
CI, confidence interval; HR, hazard ratio; Q, quartile.

Subgroup Analysis Stratified by Baseline Clinical Characteristics

The effect of modification of subgroups stratified by baseline covariates was examined. No significant interaction was observed for
any subgroups in relation to the hemoglobin concentrations with the risk of hemorrhagic stroke (). However, there was a
significant interaction between the hemoglobin concentration and use of antihypertensive drugs (P for interaction = 0.04) in
relation to the risk of ischemic stroke (). Higher hemoglobin concentrations were more strongly associated with a risk of ischemic
stroke in patients who did not use antihypertensive drugs than in those who did use such drugs.

Table 4. Multivariable-adjusted hazard ratios for hemorrhagic stroke for every 1 g/dL decrease in hemoglobin concentrations
according to subgroups

P-value
Variables   Events Patients HR (95% CI) Effect Interaction
Age, years <65 40 1779 1.30 (0.97–1.74) 0.08 0.86
  ≥65 36 1657 1.22 (0.90–1.65) 0.20
Sex Male 54 2031 1.16 (0.90–1.50) 0.24 0.11
  Female 22 1405 1.57 (1.06–2.27) 0.03
Hemodialysis duration, years <5.5 33 1705 1.17 (0.84–1.63) 0.34 0.42
  ≥5.5 43 1731 1.34 (1.01–1.76) 0.04
Cause of ESKD Non-DM 50 2442 1.32 (1.01–1.70) 0.04 0.54
  DM 26 994 1.18 (0.81–1.72) 0.38
History of stroke Absence 60 2901 1.21 (0.95–1.53) 0.13 0.41
  Presence 16 535 1.57 (0.97–2.59) 0.07
Use of antihypertensive agent Absence 18 1284 1.40 (0.91–2.12) 0.12 0.57
  Presence 58 2152 1.23 (0.96–1.56) 0.11
Use of iron supplementation Absence 54 2208 1.44 (1.11–1.85) 0.01 0.23
  Presence 22 1228 0.95 (0.64–1.40) 0.79
Dose of ESA, U/week <4500 34 1976 1.49 (1.07–2.05) 0.02 0.39
  ≥4500 42 1460 1.21 (0.91–1.58) 0.19
Serum albumin, g/dL ≥4 32 1301 1.12 (0.79–1.59) 0.54 0.34
  <4 44 2135 1.36 (1.04–1.76) 0.03
Serum C-reactive protein, mg/dL <0.13 32 1526 1.27 (0.91–1.74) 0.16 0.27
  ≥0.13 44 1910 1.29 (0.98–1.70) 0.07
Serum ferritin, ng/dL <300 54 2580 1.25 (0.96–1.61) 0.09 0.91
  ≥300 22 856 1.25 (0.86–1.80) 0.24
Serum total cholesterol, mg/dL <155 38 1831 1.39 (1.05–1.83) 0.02 0.35
  ≥155 38 1605 1.14 (0.83–1.56) 0.43
Serum phosphate, mg/dL <4.9 33 1687 1.13 (0.80–1.57) 0.48 0.63
  ≥4.9 43 1749 1.35 (1.02–1.77) 0.03
Kt/V (single pool) <1.58 50 2155 1.20 (0.93–1.55) 0.15 0.38
  ≥1.58 26 1281 1.44 (0.99–2.05) 0.06
Blood pressure Lower 11 887 1.20 (0.72–1.92) 0.48 0.88
  Higher 65 2549 1.27 (1.01–1.61) 0.04
Body weight, kg <53.8 38 1823 1.40 (1.04–1.86) 0.03 0.12
   ≥53.8 38 1613 1.13 (0.84–1.54) 0.41
Cardiothoracic ratio, % <50 39 1766 1.29 (0.94–1.75) 0.12 0.90
  ≥50 37 1670 1.24 (0.93–1.65) 0.14

The multivariable-adjusted model is adjusted for sex, dose of erythropoiesis-stimulating agent, serum phosphate concentration,
Kt/V and systolic blood pressure, which were selected using the Cox proportional hazard model and a stepwise backward method
with P < 0.05 for the remaining variables to determine the risk factors for the incidence of hemorrhagic stroke. Variables relevant
to the subgroups are excluded from each model. Higher blood pressure is defined as systolic blood pressure of ≥140 mmHg or
diastolic blood pressure of ≥90 mmHg.
CI, confidence interval; DM, diabetes mellitus; ESA, erythropoiesis-stimulating agent; ESKD, end-stage kidney disease; HR,
hazard ratio.

Table 5. Multivariable-adjusted hazard ratios for ischemic stroke for every 1 g/dL increase in hemoglobin concentrations
according to subgroups

P-value
Variables   Events Patients HR (95% CI) Effect Interaction
Age, years <65 45 1779 1.07 (0.83–1.38) 0.59 0.66
  ≥65 94 1657 1.01 (0.84–1.21) 0.94
Sex Male 83 2031 0.99 (0.82–1.21) 0.96 0.72
  Female 56 1405 1.07 (0.85–1.36) 0.56
Hemodialysis duration, years <5.5 80 1705 1.00 (0.82–1.21) 0.98 0.55
  ≥5.5 59 1731 1.09 (0.87–1.37) 0.46
Cause of ESKD Non-DM 80 2442 1.10 (0.90–1.33) 0.35 0.47
  DM 59 994 0.95 (0.76–1.20) 0.67
History of stroke Absence 94 2901 1.03 (0.86–1.23) 0.74 0.82
  Presence 45 535 1.05 (0.80–1.38) 0.71
Use of antihypertensive agent Absence 43 1284 1.29 (0.99–1.65) 0.05 0.04
  Presence 96 2152 0.93 (0.78–1.12) 0.46
Use of iron supplementation Absence 94 2208 1.00 (0.83–1.20) 0.97 0.71
  Presence 45 1228 1.10 (0.85–1.42) 0.48
Dose of ESA, U/week <4500 77 1976 1.09 (0.88–1.34) 0.44 0.67
  ≥4500 62 1460 1.00 (0.80–1.25) 0.98
Serum albumin, g/dL ≥4 36 1301 1.18 (0.85–1.61) 0.32 0.46
  <4 103 2135 1.03 (0.87–1.22) 0.74
Serum C-reactive protein, mg/dL <0.13 57 1526 1.13 (0.89–1.45) 0.31 0.42
  ≥0.13 82 1910 0.99 (0.82–1.20) 0.95
Serum ferritin, ng/mL <300 106 2580 1.05 (0.89–1.24) 0.58 0.51
  ≥300 33 856 0.99 (0.72–1.38) 0.97
Serum total cholesterol, mg/dL <155 84 1831 1.01 (0.85–1.22) 0.88 0.63
  ≥155 55 1605 1.10 (0.86–1.40) 0.47
Serum phosphate, mg/dL <4.9 87 1687 1.01 (0.83–1.23) 0.92 0.63
  ≥4.9 52 1749 1.11 (0.88–1.40) 0.40
Kt/V (single pool) <1.58 87 2155 1.06 (0.88–1.28) 0.54 0.38
  ≥1.58 52 1281 0.96 (0.75–1.24) 0.76
Blood pressure Lower 26 887 1.09 (0.80–1.53) 0.59 0.86
  Higher 113 2549 1.03 (0.87–1.22) 0.74
Body weight, kg <53.8 84 1823 1.03 (0.85–1.25) 0.79 0.76
  ≥53.8 55 1613 1.04 (0.82–1.32) 0.72
Cardiothoracic ratio, % <50 59 1766 1.18 (0.93–1.48) 0.17 0.11
   ≥50 80 1670 0.95 (0.79–1.15) 0.59

The multivariable-adjusted model is adjusted for age, cause of end-stage kidney disease, history of stroke and cardiothoracic
ratio, which were selected using the Cox proportional hazard model and a stepwise backward method with P < 0.05 for the
remaining variables to determine the risk factors for the incidence of ischemic stroke. Variables relevant to the subgroups are
excluded from each model. Higher blood pressure is defined as systolic blood pressure of ≥140 mmHg or diastolic blood pressure
of ≥90 mmHg.
CI, confidence interval; DM, diabetes mellitus; ESA, erythropoiesis-stimulating agent; ESKD, end-stage kidney disease; HR,
hazard ratio.

Discussion

In this cohort of patients undergoing HD, we demonstrated a significant association between lower hemoglobin concentrations
and a higher risk of hemorrhagic stroke. In contrast, we did not identify a relationship between the hemoglobin concentration and
the incidence of ischemic stroke. Our results indicate the novel importance of anemia as a potential risk factor for hemorrhagic
stroke in patients undergoing dialysis.

In the present study, lower hemoglobin concentrations were associated with a higher risk of hemorrhagic stroke. This association
remained unchanged after adjustment for potential confounding factors. Several observational studies have shown a relationship
between anemia and a higher incidence of composite stroke.[10,11] However, these studies were not designed to distinguish
hemorrhagic stroke and ischemic stroke separately as an outcome. The Hisayama study, a population-based cohort study in
Japan, revealed that lower hematocrit levels were associated with a higher risk of hemorrhagic stroke after adjustment for
potential confounders in adults in the general population.[12] This study was the first to suggest an association between anemia
and hemorrhagic stroke. The authors reported that patients with the lowest hematocrits (hematocrit: male, ≤44.7%; female,
≤39.3%) had a 1.91-fold (95% confidence interval, 1.03–3.54) higher risk of hemorrhagic stroke than the reference group
(hematocrit: male, 47.1–49.6%; female, 41.7–43.7%) during the 19-year follow-up period. They also reported a linear inverse
association between the hematocrit level and the hazard ratio for the incidence of hemorrhagic stroke.[12] Our study showed a
similar relationship using the hemoglobin concentration instead of the hematocrit in patients undergoing HD, which is a high-risk
population with cardiovascular disease.

Based on our results, we generated the hypothesis that anemia is associated with the development of hemorrhagic stroke.
Several possible mechanisms support this hypothesis. First, the combination of CKD and anemia is a high risk factor for
cardiovascular disease.[7,8] This association was recently recognized as cardiorenal anemia syndrome.[9] Patients with CKD are
subjected to enhanced oxidative stress.[18,19] A decrease in the blood hemoglobin concentration worsens tissue damage as a
result of decreased oxygen supply.[20] Taken together, these findings suggest that such oxidative stress and hypoxia additively or
synergistically exacerbate a greater arteriosclerotic burden on the cerebral blood vessels. Second, anemia may reflect a decline in
the protective role against bleeding played by red blood cells (RBCs). Ameliorating anemia with RBC transfusions shortens the
bleeding time in uremic conditions.[21,22] Moreover, in experimental models, adenosine triphosphate released from RBCs
facilitates activation of platelets,[23] and RBCs promote thrombin generation.[24,25] Therefore, anemia is potentially associated
with hemorrhagic stroke through a disorder of the coagulation system.

Our subgroup analysis suggested several interesting effect modifiers in relation to the hemoglobin concentration and hemorrhagic
stroke; however, the statistical power was insufficient for this difference to be statistically significant. The effect of anemia on the
risk of hemorrhagic stroke tended to be greater in female than male patients (P for interaction = 0.11). A recent observational
study demonstrated an association between anemia and a high incidence of composite stroke only in female patients,[11]
suggesting a sex-based difference. Moreover, in the present study, the effect of anemia on the risk of hemorrhagic stroke tended
to be greater in patients with a body weight of <53.8 kg than ≥53.8 kg (P for interaction = 0.12). A combination of anemia and a
lean body may indicate malnutrition, which has recently been recognized as a risk factor for atherosclerosis in patients undergoing
HD.[26] Our results may indicate that female or lean patients with anemia are at higher risk of hemorrhagic stroke.

In our study, we did not identify a relationship between the hemoglobin concentration and the incidence of ischemic stroke. Some
studies in the general population have suggested an association between high hemoglobin concentrations and the incidence of
ischemic stroke[12,27,28] but not in patients with CKD, including those undergoing HD. In the TREAT study, the risk of stroke was
increased in the group of patients with higher target hemoglobin concentrations treated by an ESA.[4] However, this previous
study did not show which type of stroke was increased in patients with high hemoglobin concentrations. Moreover, there was no
significant increase in stroke in other randomized controlled trials of patients using an ESA.[29–31] Therefore, whether high
hemoglobin concentrations are correlated with ischemic stroke in patients with CKD remains unclear. Intriguingly, our data
obtained from the subgroup analysis suggested that higher hemoglobin concentrations were associated with a higher risk of
ischemic stroke in patients who did not use antihypertensive drugs than in those who did use such drugs (P for interaction = 0.04).
This result is supported by the brain-protective effect of antihypertensive drugs.[32,33] Moreover, the effect of high hemoglobin
concentrations on the risk of ischemic stroke tended to be greater in patients with a CTR of <50% than ≥50%; however, the
statistical power was insufficient for this finding to achieve statistical significance (P for interaction = 0.11). The CTR reflects the
volume status in patients undergoing HD.[16] A combination of a high hemoglobin concentration and low CTR appears to suggest
hemoconcentration, which is a potential risk factor for ischemic stroke.[5] Well-conducted studies taking these points into
consideration should be performed to elucidate whether high hemoglobin concentrations are indeed associated with ischemic
stroke in patients undergoing HD.

A strength of our study is its large-scale, prospective design and wide-ranging inclusion criteria. Therefore, our results are
generalizable to the real-world HD population. However, some limitations in our study should be noted. First, the hemoglobin
concentrations were measured only at baseline. Second, measurement of baseline parameters might have been insufficient; e.g.
data regarding the history of atrial fibrillation and the use of anticoagulants and antiplatelet drugs were missing. Third, physicians
collected outcomes concerning stroke from the medical records, and these outcomes were not adjudicated by an independent
committee for event evaluation. It is possible that some mild ischemic strokes were overlooked. In addition, a validation study on
the completeness of the record-based data was not performed. Fourth, survival bias may have been present between patients
undergoing prevalent and incident HD. Furthermore, previous studies have shown that high hemoglobin concentrations are
associated with a risk of other cardiovascular events,[4,29–31] leading to the recommendation that hemoglobin concentrations of
patients undergoing HD should not exceed 12 g/dL in Japan.[34] Further, there may have been competing risks for mortality and
cardiovascular (non-stroke) mortality. Further studies are required to address these limitations, which could not be resolved
because of our study design.

In conclusion, our findings suggest that a low hemoglobin concentration is associated with a high risk of hemorrhagic stroke in
patients undergoing HD but that there is no association between the hemoglobin concentration and the risk of ischemic stroke.
These results may provide novel insights into the association between the hemoglobin concentration and stroke. Hemorrhagic
stroke often has a more severe prognosis than ischemic stroke.[35] Therefore, further investigations are required to determine the
optimal range of hemoglobin concentrations and appropriate therapy to maintain them within that range to prevent hemorrhagic
stroke in patients undergoing maintenance HD.

References

1. Power A, Chan K, Singh SK et al. Appraising stroke risk in maintenance hemodialysis patients: a large single-center cohort
study. Am J Kidney Dis 2012; 59: 249–257

2. Doyle EM, Sloan JM, Goodbrand JA et al. Association between kidney function, rehabilitation outcome, and survival in
older patients discharged from inpatient rehabilitation. Am J Kidney Dis 2015; 66: 768–774

3. Masakane I, Nakai S, Ogata S et al. An overview of regular dialysis treatment in Japan (as of 31 December 2013). Ther
Apher Dial 2015; 19: 540–574

4. Pfeffer MA, Burdmann EA, Chen CY et al. A trial of darbepoetin alfa in type 2 diabetes and chronic kidney disease. N Engl
J Med 2009; 361: 2019–2032

5. Rodriguez G, Cordina S, Vazquez G et al. The hydration influence on the risk of stroke (THIRST) study. Neurocrit Care
2009; 10: 187–194

6. Sarnak MJ, Tighiouart H, Manjunath G et al. Anemia as a risk factor for cardiovascular disease in The Atherosclerosis Risk
in Communities (ARIC) study. J Am Coll Cardiol 2002; 40: 27–33

7. Silverberg DS, Wexler D, Blum M et al. The effect of correction of anaemia in diabetics and non-diabetics with severe
resistant congestive heart failure and chronic renal failure by subcutaneous erythropoietin and intravenous iron. Nephrol
Dial Transplant 2003; 18: 141–146

8. Kajimoto K, Sato N, Keida T et al. Associations of anemia and renal dysfunction with outcomes among patients
hospitalized for acute decompensated heart failure with preserved or reduced ejection fraction. Clin J Am Soc Nephrol
2014; 9: 1912–1921

9. Silverberg D, Wexler D, Blum M et al. The cardio-renal anaemia syndrome: does it exist? Nephrol Dial Transplant 2003; 18
(Suppl 8): viii7–vii12

10. Kannel WB, Gordon T, Wolf PA et al. Hemoglobin and the risk of cerebral infarction: the Framingham Study. Stroke 1972;
3: 409–420

11. Panwar B, Judd SE, Warnock DG et al. Hemoglobin concentration and risk of incident stroke in community-living adults.
Stroke 2016; 47: 2017–2024

12. Gotoh S, Hata J, Ninomiya T et al. Hematocrit and the risk of cardiovascular disease in a Japanese community: The
Hisayama Study. Atherosclerosis 2015; 242: 199–204

13. Eriguchi R, Taniguchi M, Ninomiya T et al. Hyporesponsiveness to erythropoiesis-stimulating agent as a prognostic factor
in Japanese hemodialysis patients: the Q-Cohort study. J Nephrol 2015; 28: 217–225

14. Tanaka S, Ninomiya T, Taniguchi M et al. Comparison of oral versus intravenous vitamin D receptor activator in reducing
infection-related mortality in hemodialysis patients: the Q-Cohort Study. Nephrol Dial Transplant 2016; 31: 1152–1160

15. Yamada S, Tsuruya K, Taniguchi M et al. Association between serum phosphate levels and stroke risk in patients
undergoing hemodialysis: The Q-Cohort Study. Stroke 2016; 47: 2189–2196

16. Yotsueda R, Taniguchi M, Tanaka S et al. Cardiothoracic ratio and all-cause mortality and cardiovascular disease events in
hemodialysis patients: the Q-Cohort Study. Am J Kidney Dis 2017; 70: 84–92

17. Harrell FE. General aspects of fitting regression models. In: Harrell FE Jr (ed). Regression Modeling Strategies: With
Applications to Linear Models, Logistic Regression, and Survival Analysis. New York, NY: Springer New York, 2001, 11–40

18. Vaziri ND, Oveisi F, Ding Y. Role of increased oxygen free radical activity in the pathogenesis of uremic hypertension.
Kidney Int 1998; 53: 1748–1754

19. Tepel M, Echelmeyer M, Orie NN et al. Increased intracellular reactive oxygen species in patients with end-stage renal
failure: effect of hemodialysis. Kidney Int 2000; 58: 867–872
 20. Dijk JM, Wangge G, Van Der Graaf Y et al. Hemoglobin and atherosclerosis in patients with manifest arterial disease.
Atherosclerosis 2006; 188: 444–449

21. Livio M, Gotti E, Marchesi D et al. Uraemic bleeding: role of anaemia and beneficial effect of red cell transfusions. Lancet
1982; 2: 1013–1015

22. Fernandez F, Goudable C, Sie P et al. Low haematocrit and prolonged bleeding time in uraemic patients: effect of red cell
transfusions. Br J Haematol 1985; 59: 139–148

23. Carroll JS, Ku CJ, Karunarathne W et al. Red blood cell stimulation of platelet nitric oxide production indicated by
quantitativemonitoring of the communication between cells in the bloodstream. Anal Chem 2007; 79: 5133–5138

24. Peyrou V, Lormeau JC, Herault JP et al. Contribution of erythrocytes to thrombin generation in whole blood. Thromb
Haemost 1999; 81: 400–406

25. Van Der Meijden PE, Van Schilfgaarde M, Van Oerle R et al. Platelet- and erythrocyte-derived microparticles trigger
thrombin generation via factor XIIa. J Thromb Haemost 2012; 10: 1355–1362

26. Kalantar-Zadeh K, Ikizler A, Block G et al. Malnutrition-inflammation complex syndrome in dialysis patients: causes and
consequences. Am J Kidney Dis 2003; 42: 864–881

27. Gagnon DR, Zhang TJ, Brand FN et al. Hematocrit and the risk of cardiovascular disease-the Framingham study: a 34-
year follow-up. Am Heart J 1994; 127: 674–682

28. Di Mascio R, Marchioli R, Vitullo F et al. A positive relation between high hemoglobin values and the risk of ischemic
stroke. Eur Neurol 1996; 36: 85–88

29. Besarab A, Bolton WK, Browne JK et al. The effects of normal as compared with low hematocrit values in patients with
cardiac disease who are receiving hemodialysis and epoetin. N Engl J Med 1998; 339: 584–590

30. Drueke TB, Locatelli F, Clyne N et al. Normalization of hemoglobin level in patients with chronic kidney disease and
anemia. N Engl JMed 2006; 355: 2071–2084

31. Singh AK, Szczech L, Tang KL et al. Correction of anemia with epoetin alfa in chronic kidney disease. N Engl J Med 2006;
355: 2085–2098

32. Schrader J, Luders S, Kulschewski A et al. Morbidity and mortality after stroke, eprosartan compared with nitrendipine for
secondary prevention: principal results of a prospective randomized controlled study (MOSES). Stroke 2005; 36: 1218–
1226

33. Chen GJ, Yang MS. The effects of calcium channel blockers in the prevention of stroke in adults with hypertension: a meta-
analysis of data from 273,543 participants in 31 randomized controlled trials. PLoS One 2013; 8: e57854

34. Yamamoto H, Nishi S, Tomo T et al. 2015 Japanese society for dialysis therapy: guidelines for renal anemia in chronic
kidney disease. Ren Replace Ther 2017; 3: 36

35. Wetmore JB, Phadnis MA, Ellerbeck EF et al. Relationship between stroke and mortality in dialysis patients. Clin J Am Soc
Nephrol 2015; 10: 80–89

Acknowledgements
We appreciate the contribution of the participants in the Q-Cohort Study, members of the Society for the Study of Kidney Disease.
The following personnel (institutions) participated in the study: Takashi Ando (Hakozaki Park Internal Medicine Clinic), Takashi
Ariyoshi (Ariyoshi Clinic), Koichiro Goto (Goto Clinic), Fumitada Hattori (Nagao Hospital), Harumichi Higashi (St Mary's Hospital),
Tadashi Hirano (Hakujyuji Hospital), Kei Hori (Munakata Medical Association Hospital), Takashi Inenaga (Ekisaikai Moji Hospital),
Hidetoshi Kanai (Kokura Memorial Hospital), Shigemi Kiyama (Kiyama Naika), Tetsuo Komota (Komota Clinic), Hiromasa Kuma
(Kuma Clinic), Toshiro Maeda (Kozenkai-Maeda Hospital), Junichi Makino (Makino Clinic), Dai Matsuo (Hirao Clinic), Chiaki
Miishima (Miishima Clinic), Koji Mitsuiki (Japanese Red Cross Fukuoka Hospital), Kenichi Motomura (Motomura Naika Clinic),
Sadatoshi Nakamura, Hidetoshi Nakamura (Kokura Daiichi Hospital), Koichi Nakashima (Ohashi Internal Circulatory Clinic),
Nobumitsu Okita (Shiroishi Kyoritsu Hospital), Shinichiro Osato (Osato Jin Clinic), Sakura Sakamoto (Fujiyamato Spa Hospital),
Keiko Shigematsu (Shigematsu Clinic), Kazumasa Shimamatsu (Shimamatsu Naika Iin), Yoshito Shogakiuchi (Shin-Ai Clinic),
Hiroaki Takamura (Hara Hospital), Kazuhito Takeda (Iizuka Hospital), Asuka Terai (Chidoribashi Hospital), Hideyoshi Tanaka
(Mojiko-Jin Clinic), Suguru Tomooka (Hakozaki Park Internal Medicine Clinic), Jiro Toyonaga (Fukuoka Renal Clinic), Hiroshi
Tsuruta (Steel Memorial Yawata Hospital), Ryutaro Yamaguchi (Shiseikai Hospital), Taihei Yanagida (Saiseikai Yahata General
Hospital), Tetsuro Yanase (Yanase Internal Medicine Clinic), Tetsuhiko Yoshida (Hamanomachi Hospital), Takahiro Yoshimitsu
(Gofukumachi Kidney Clinic, Harasanshin Hospital) and Koji Yoshitomi (Yoshitomi Medical Clinic). We also thank Trish Reynolds,
MBBS, FRACP and Angela Morben, DVM, ELS, from Edanz Group (www.edanzediting.com/ac) for editing drafts of this article.

Funding
This study was supported by the Kidney Foundation (H19 JKFB 07–13, H20 JKFB 08–8, H23 JKFB 11–11) and the Japan Dialysis
Outcome Research Foundation (H19-076-02, H20-003).

Nephrol Dial Transplant. 2018;33(5):856-864. © 2018 Oxford University Press