Association between blood pressure and clinical outcomes in patients undergoing maintenance hemodialysis

Abstract

Background

Despite the clinical importance of blood pressure control in patients undergoing hemodialysis (HD), evidence on the blood pressure ranges associated with favorable outcomes remains limited and inconsistent.

Methods

We retrospectively analyzed the data from the HD quality assessments in South Korea. Systolic blood pressure (SBP) was classified into 6 groups: VL-sys group (< 100, n = 230), L-sys (100–120, n = 2,272), R-sys (120–140, n = 17,004), H-sys (140–160, n = 17,493), VH-sys (160–180, n = 4,627), and EH-sys (≥ 180, n = 632). Diastolic blood pressure (DBP) was divided into 6 groups: EL-dia (< 60, n = 1,349), VL-dia (60–70, n = 5,460), L-dia (70–80, n = 13,361), R-dia (80–90, n = 17,442), H-dia (90–100, n = 4,211), and VH-dia (≥ 100, n = 435). Outcomes include all-cause mortality, cardiovascular events (CVE), dementia, atrial fibrillation (Afib), and fracture risk.

Results

Higher SBP was associated with increased risks of all-cause mortality, CVE, dementia, and fractures compared to those of the R-sys group. Conversely, lower SBP was associated with reduced risks for all-cause mortality and CVE. The L-sys group had an adjusted hazard ratio of 0.92 (95% confidence interval [CI], 0.86–0.99) for all-cause mortality and 0.84 (95% CI, 0.76–0.93) for CVE compared with the R-sys group. Regarding DBP, in univariable analysis, survival rate was higher in the VH-dia group compared to R-dia group, but contradictive results were shown in multivariable analysis. Overall, in multivariable analysis, higher levels were associated with increased risks of all-cause mortality, CVE, and dementia, while lower DBP, particularly in the VL-dia and L-dia groups, was associated with reduced risks.

Conclusions

In our study, a SBP of 100–119 mmHg and a DBP of 60–79 mmHg were associated with improved overall survival, reduced incidence of CVE, and lower risk of dementia and fractures in patients undergoing HD.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12882-026-04954-6.

Introduction

Hemodialysis (HD) is the most commonly utilized form of renal replacement therapy for patients with end-stage kidney disease. Research shows that patients undergoing HD face substantially higher mortality risk than that of normal individuals and those with chronic conditions (such as diabetes or arterial hypertension who are not on dialysis [1, 2]. Therefore, identifying and effectively modifying risk factors in this population is critical. Arterial hypertension, which is highly prevalent among patients with HD, is associated with various adverse clinical outcomes.

Despite the clinical importance of blood pressure (BP) control in patients undergoing HD, evidence on the BP range associated with favorable outcomes remains limited and inconsistent. While the link between arterial hypertension and adverse outcomes is well established, data supporting lower BP targets remain limited [3–7]. Epidemiological studies report potential risks associated with low BP; however, the SPRINT trial—demonstrating the benefits of strict BP control in the general population—prompts consideration of lower BP targets not only for the general population but also in patients undergoing HD [8, 9]. Nevertheless, research supporting the safety and effectiveness of lower BP thresholds in patients undergoing HD remains insufficient, particularly regarding the identification of tolerable lower limits that do not increase clinical risk. Furthermore, most prior studies primarily focus on all-cause mortality, with limited investigation into other clinically relevant outcomes. Therefore, this study aims to identify the BP range associated with favorable clinical outcomes in a population-based cohort of patients undergoing HD by categorizing pre-dialysis systolic blood pressure (SBP) and diastolic blood pressure (DBP) in 20 mmHg and 10 mmHg intervals, respectively. Beyond all-cause mortality, we evaluated the associations between BP levels and cardiovascular events (CVE), dementia, atrial fibrillation (Afib), and fracture risk.

Materials and methods

Data source and study population

In the Republic of Korea, regular HD quality assessments were conducted to ensure quality control [10]. The fourth (July–December 2013) and fifth (July–December 2015) HD quality assessments targeted adult patients (≥ 18 years) undergoing maintenance HD for at least 3 months, twice weekly. We retrospectively analyzed the data from the Health Insurance Review and Assessment Service (HIRA), including HD quality assessments, claims, and mortality records. All patients who participated in the 4th and 5th HD quality assessment programs were included (n = 21,846 and 35,538 in the 4th and 5th assessments, respectively) (Figure S1). Of the 57,384 participants, 13,870 patients were included in both the 4th and 5th assessments. For these patients, only data from the 4th assessment were retained to avoid duplication. Accordingly, 13,870 duplicate records were excluded. Additionally, we removed those who underwent catheter-based HD (n = 1,107), or extreme SBP or DBP values (≤ 0.1 or ≥ 99.9th percentile for either measure). The final cohort consisted of 42,258 patients. The institutional review board of Yeungnam University Medical Center (approval no. YUMC 2023-12-012) approved the protocol. Informed consent was waived as records and information were anonymized and de-identified before the analysis.

Exposure

SBP and DBP were measured pre-dialysis. The pre-dialysis BP was recorded during the last 3 months (October–December) of each HD quality assessment period. Pre-dialysis BP was defined as the mean of three measurements obtained once monthly over a 3-month period, representing a short-term, time-averaged baseline BP. SBP was classified into six groups based on 20 mmHg intervals: VL-sys group (< 100 mmHg), L-sys (100 ≤ SBP < 120 mmHg), R-sys (120 ≤ SBP < 140 mmHg), H-sys (140 ≤ SBP < 160 mmHg), VH-sys (160 ≤ SBP < 180 mmHg), and EH-sys (≥ 180 mmHg). DBP was divided into six groups based on 10 mmHg intervals: EL-dia (< 60 mmHg), VL-dia (60 ≤ DBP < 70 mmHg), L-dia (70 ≤ DBP < 80 mmHg), R-dia (80 ≤ DBP < 90 mmHg), H-dia (90 ≤ DBP < 100 mmHg), and VH-dia (≥ 100 mmHg). In this study, the reference group comprised patients with a SBP of 120 − 140 mmHg and a DBP of 80 − 90 mmHg. These ranges were selected as they represent clinically moderate and widely accepted BP levels in both general and dialysis populations, avoiding extreme categories potentially affected by reverse epidemiology or acute hemodynamic instability. Additionally, these categories included a significant proportion of patients, ensuring statistical stability for group comparisons. We evaluated BP variability using the coefficient of variation (CV) calculated from three measurements, defined as follows: SBP-CV = 100 × (SD of SBP / mean of SBP) and DBP-CV = 100 × (SD of DBP / mean of DBP) [11].

Study variables

Baseline data, including BP measurements, were collected from July–December 2013 for patients included in the 4th assessment and from July–December 2015 for those in the 5th assessment. We compiled data on several variables, such as age, sex, HD vintage (months), etiology of end-stage kidney disease, and vascular access type. Clinical parameters documented during the evaluation included measurements of blood hemoglobin levels (g/dL); serum levels of albumin (g/dL), calcium (mg/dL), phosphorus (mg/dL), and creatinine (mg/dL); Kt/V_urea; and ultrafiltration volume (L/session). The data were collected monthly, and laboratory values were averaged across these months. Kt/V_urea was calculated using the Daugirdas equation [12].

Medications, such as anti-hypertensive drugs, aspirin, clopidogrel, and statins, were evaluated using the medication codes provided (Table S1). Medication use was defined as ≥ 1 prescription identified during an HD quality assessment program. Before the HD quality assessments, comorbidities were evaluated over a year and defined using the Charlson comorbidity index (CCI), which encompasses 17 different conditions [13–15]. Furthermore, CCI scores were computed for all patients. Myocardial infarction (MI), congestive heart failure (CHF), and cerebrovascular accidents (CVA) were identified based on ICD-10 codes.

Outcomes

Follow-up for outcomes began on January 1, 2014, for the 4th assessment cohort and on January 1, 2016, for the 5th assessment cohort. Patients were followed until June 2024. The primary outcome was all-cause mortality; secondary outcomes were CVE, Afib, dementia, and fractures. Incidences and dates of death were collected from the HIRA. CVE (MI, stroke, and revascularization) were ascertained regardless of survival status, as previously described [15]. Dementia, Afib, and fractures were identified using ICD-10 codes (F00-03, G30-G31 for dementia; I48 for Afib; S22, S32, S52, S62, S72 for fracture). For each outcome, patients with CVE, Afib, dementia, or fracture during the 6 months of the assessment and the 6 months prior to the assessment (a total of 1 year) were excluded from corresponding analysis.

The exclusion window was implemented to determine incident events after cohort entry while minimizing misclassification of prevalent cases. Therefore, follow-up began on January 1, 2014, for patients included in the 4th assessment and on January 1, 2016, for those in the 5th assessment. To capture only incident events, a 1-year washout period was applied prior to follow-up, defined as the 6-month assessment period and the preceding 6 month. Patients with diagnostic or procedure codes related to the specific outcome during this period were excluded as likely pre-existing (prevalent) conditions. Particularly, owing to the fact that this washout period occurred entirely before the start of follow-up, no immortal time was introduced into the risk period. Time-at-risk commenced after the defined follow-up start date, and outcome events were ascertained prospectively. Patients were censored at transfer to peritoneal dialysis or at kidney transplantation if no outcome had occurred.

Statistical analyses

The data were analyzed using SAS Enterprise Guide v7.1 and R v3.5.1. Categorical variables are presented as frequencies and percentages, while continuous variables are presented as means ± SD. Categorical differences were assessed using Pearson’s χ² or Fisher’s exact test. Continuous differences were assessed using one-way analysis of variance followed by Tukey’s post hoc test.

Survival curves were estimated using the Kaplan–Meier curves. P-values for the comparison of survival curves were determined using the log-rank test. Hazard ratios (HRs) and confidence intervals (CIs) were calculated using Cox proportional-hazards regression. Multivariable Cox models were adjusted for age, sex, vascular access type, CCI score, HD vintage, ultrafiltration volume, Kt/V_urea, blood levels of hemoglobin, albumin, creatinine, phosphorus, and calcium, use of anti-hypertensives, aspirin, clopidogrel, or statins, MI or CHF, and CVA. Multivariable Cox regression analyses were performed using the enter method. The proportional hazard assumption was evaluated using the Schoenfeld residual for all variables included in the Cox regression models. Correlation coefficients (rho) ranged from − 0.1 to 0.1, indicating minimal correlation with time and supporting the validity of the assumption. Furthermore, visual inspection of the plots revealed no systematic time-dependent patterns.

To explore non-linear associations between BP and outcomes, we applied restricted cubic spline models adjusted for the same covariates included in the multivariable Cox models. Measures of dry weight were not available in this dataset; however, available data remain relevant for associations with BP and clinical outcomes. To better account for volume-related confounding and interindividual differences in body size, we performed sensitivity analyses using ultrafiltration volume normalized to post-dialysis body weight rather than including ultrafiltration volume as a covariate. Additional sensitivity analyses were conducted adjusting for β-blocker use or BP variability, given its potential role as a crucial confounder in the association between BP and clinical outcomes. Statistical significance was set at P < 0.05.

Results

Baseline characteristics

The patient numbers in VL-sys, L-sys, R-sys, H-sys, VH-sys, and EH-sys were 230, 2,272, 17,004, 17,493, 4,627, and 632, respectively (Table 1). For EL-dia, VL-dia, L-dia, R-dia, H-dia, and VH-dia, the numbers were 1,349, 5,460, 13,361, 17,442, 4,211, and 435, respectively (Table S2). The R-sys group was older than the other groups. VL-sys had the lowest proportions of males, diabetes, anti-hypertensive or statin use, MI or CHF, and CVA. Additionally, it had the lowest body mass index, CCI score, and ultrafiltration volume, and the greatest HD vintage and Kt/V_urea. VL-sys and L-sys groups had higher hemoglobin levels and lower serum albumin levels than those of the other groups. Among the DBP-based groups, EL-dia had the oldest patients and the lowest proportion of males. It also had the highest CCI score and Kt/V_urea, and the lowest ultrafiltration volume, serum albumin, and serum creatinine. VH-dia had the lowest proportion of diabetes, the use of aspirin, clopidogrel and statin, and the prevalence of CVA. Table S3 presents the CV values based on the group. For SBP, CV values were highest in the L-sys group. For DBP, CV values were highest in the VL-sys, L-sys, and EH-sys groups.

Table 1.

Clinical characteristics of the patients

	VL-sys (n = 230)	L-sys (n = 2,272)	R-sys (n = 17,004)	H-sys (n = 17,493)	VH-sys (n = 4,627)	EH-sys (n = 632)	P
Age (years)	55.7 ± 13.5	58.4 ± 14.0a	60.2 ± 13.1ab	59.7 ± 12.6abc	59.1 ± 12.3ac	57.5 ± 12.3cde	< 0.001
Sex (male, %)	69 (30%)	986 (43.4%)	9742 (57.3%)	10,992 (62.8%)	2851 (61.6%)	346 (54.7%)	< 0.001
Body mass index (kg/m2)	21.5 ± 3.7	22.1 ± 3.4	22.3 ± 3.3ab	22.3 ± 3.3ab	22.5 ± 3.6abcd	22.9 ± 3.8abcd	< 0.001
HD vintage (months)	126 ± 105	85 ± 85a	64 ± 67ab	64 ± 63ab	64 ± 58ab	60 ± 49ab	< 0.001
Underlying causes of ESKD							< 0.001
Diabetes mellitus	22 (9.6%)	451 (19.9%)	6426 (37.8%)	8068 (46.1%)	2601 (56.2%)	437 (69.1%)
Hypertension	39 (17.0%)	584 (25.7%)	4740 (27.9%)	4636 (26.5%)	11,115 (24.1%)	111 (17.6%)
Glomerulonephritis	51 (22.2%)	406 (17.9%)	2111 (12.4%)	1779 (10.2%)	318 (6.9%)	26 (4.1%)
Others	55 (23.9%)	406 (17.9%)	1714 (10.1%)	1138 (6.5%)	203 (4.4%)	17 (2.7%)
Unknown	63 (27.4%)	425 (18.7%)	2013 (11.8%)	1872 (10.7%)	390 (8.4%)	41 (6.5%)
CCI score	5.7 ± 2.6	6.1 ± 2.9	6.6 ± 2.8ab	6.7 ± 2.7abc	6.9 ± 2.7abcd	7.2 ± 2.6abcd	< 0.001
Arteriovenous fistula	199 (86.5%)	1951 (85.9%)	14,547 (85.6%)	14,995 (85.7%)	3902 (84.3%)	526 (83.2%)	0.115
Kt/Vurea	1.60 ± 0.30	1.58 ± 0.27	1.53 ± 0.25ab	1.50 ± 0.24abc	1.49 ± 0.23abc	1.51 ± 0.24ab	< 0.001
UFV (L/session)	2.13 ± 0.93	2.12 ± 0.99	2.21 ± 0.95b	2.37 ± 0.96abc	2.55 ± 0.97abcd	2.72 ± 1.03abcde	< 0.001
Hemoglobin (g/dL)	10.9 ± 1.1	10.9 ± 1.0	10.7 ± 0.8ab	10.6 ± 0.8abc	10.5 ± 0.9abcd	10.5 ± 0.8abcd	< 0.001
Serum albumin (g/dL)	3.90 ± 0.40	3.93 ± 0.35	3.97 ± 0.34ab	4.00 ± 0.34abc	4.01 ± 0.35abc	4.02 ± 0.32abc	< 0.001
Serum phosphorus (mg/dL)	4.93 ± 1.47	4.83 ± 1.30	4.88 ± 1.31	5.05 ± 1.36bc	5.17 ± 1.43bcd	5.28 ± 1.47abcd	< 0.001
Serum calcium (mg/dL)	8.93 ± 0.84	8.93 ± 0.82	8.91 ± 0.83	8.95 ± 0.84c	8.97 ± 0.86c	9.02 ± 0.85c	< 0.001
Serum creatinine (mg/dL)	9.70 ± 2.87	9.43 ± 2.85	9.42 ± 2.78	9.74 ± 2.72bc	9.80 ± 2.64bc	9.80 ± 2.49c	< 0.001
Use of antihypertensive drugs
RASB	18 (7.8%)	630 (27.7%)	10,276 (60.4%)	12,017 (68.7%)	3363 (72.7%)	490 (77.5%)	< 0.001
Others except RASB	25 (10.9%)	332 (14.6%)	2699 (15.9%)	2669 (15.3%)	662 (14.3%)	71 (11.2%)	< 0.001
Use of aspirin	25 (10.9%)	247 (10.9%)	2148 (12.6%)	2149 (12.3%)	567 (12.3%)	76 (12.0%)	0.263
Use of clopidogrel	11 (4.8%)	152 (6.7%)	1307 (7.7%)	1317 (7.5%)	351 (7.6%)	42 (6.6%)	0.294
Use of statins	65 (28.3%)	817 (36.0%)	6687 (39.3%)	6749 (38.6%)	1779 (38.4%)	267 (42.2%)	< 0.001
MI or CHF	60 (26.1%)	739 (32.5%)	6447 (37.9%)	6532 (37.3%)	1811 (39.1%)	266 (42.1%)	< 0.001
Cerebrovascular accidents	36 (15.7%)	430 (18.9%)	3778 (22.2%)	4053 (23.2%)	1105 (23.9%)	161 (25.5%)	< 0.001

Data are presented as mean ± standard deviation for continuous variables and as numbers (percentages) for categorical variables. P-values were tested using one-way analysis of variance, followed by Tukey’s post hoc test for continuous variables, and Pearson’s χ² test for categorical variables

Abbreviations: CCI, Charlson Comorbidity Index; CHF, congestive heart failure; ESKD, end-stage kidney disease; HD, hemodialysis; MI, myocardial infarction; RASB, renin-angiotensin system blockade; SBP, systolic blood pressure; UFV, ultrafiltration volume

SBP categories; EH-sys, SBP ≥ 180 mmHg; H-sys, 140 ≤ SBP < 160 mmHg; L-sys, 100 ≤ SBP < 120 mmHg; R-sys, 120 ≤ SBP < 140 mmHg; VH-sys, 160 ≤ SBP < 180 mmHg; VL-sys, SBP < 100 mmHg

Post hoc comoparison (Tukey’s test):

^aP < 0.05 vs. VL-sys; ^bP < 0.05 vs. L-sys; ^cP < 0.05 vs. R-sys; ^dP < 0.05 vs. H-sys; ^eP < 0.05 vs. VH-sys

Clinical outcomes according to SBP groups

The follow-up durations in VL-sys, L-sys, R-sys, H-sys, VH-sys, and EH-sys were 80 ± 42, 78 ± 41, 76 ± 39, 75 ± 39, 71 ± 40, and 69 ± 40 months, respectively. The 5-year survival rates in the VL-sys, L-sys, R-sys, H-sys, VH-sys, and EH-sys were 72.5%, 70.6%, 68.0%, 66.1%, 61.3%, and 60.9%, respectively; CVE-free rates were 82.8%, 81.7%, 75.4%, 72.8%, 67.8%, and 69.2%; Afib-free rates were 88.5%, 87.4%, 84.0%, 83.7%, 83.5%, and 83.4%; dementia-free rates were 92.4%, 92.3%, 91.1%, 90.4%, 89.9%, and 89.0%; and fracture-free rates were 76.7%, 76.7%, 77.6%, 76.5%, 75.4%, and 71.5% (Fig. 1).

Fig. 1.

Kaplan–Meier curves of clinical outcomes based on SBP groups. Patient survival (A), CVE-free (B), Afib-free (C), dementia-free (D), and fracture-free (E). P-values for pairwise comparisons, determined with log-rank tests, are presented at the bottom of the graph. Abbreviations: Afib, atrial fibrillation; CVE, cardiovascular event; SBP, systolic blood pressure. SBP categories: EH-sys, SBP ≥ 180 mmHg; H-sys, 140 ≤ SBP < 160 mmHg; L-sys, 100 ≤ SBP < 120 mmHg; R-sys, 120 ≤ SBP < 140 mmHg; VH-sys, 160 ≤ SBP < 180 mmHg; VL-sys, SBP < 100 mmHg

Multivariable Cox regression analyses revealed that the L-sys group was associated with lower all-cause mortality and CVE risk, while groups with higher SBP were associated with a higher risk of both than those of the R-sys group (Table 2). The L-sys group was associated with lower Afib risk, and the VH-sys and EH-sys groups were associated with greater Afib risk than that of the R-sys group. Dementia and fracture risks were also associated with greater in groups with higher SBP than in the R-sys group.

Table 2.

Systolic blood pressure and clinical outcomes

	Univariate		Multivariate
	HR (95% CI)	P	HR (95% CI)	P
All-cause mortality
VL-sys	0.83 (0.69–1.00)	0.051	1.12 (0.93–1.36)	0.237
L-sys	0.82 (0.77–0.87)	< 0.001	0.92 (0.86–0.99)	0.016
H-sys	1.09 (1.06–1.12)	< 0.001	1.10 (1.07–1.14)	< 0.001
VH-sys	1.30 (1.24–1.35)	< 0.001	1.33 (1.28–1.39)	< 0.001
EH-sys	1.34 (1.21–1.48)	< 0.001	1.47 (1.33–1.63)	< 0.001
CVE
VL-sys	0.56 (0.41–0.78)	< 0.001	0.76 (0.55–1.06)	0.111
L-sys	0.72 (0.65–0.79)	< 0.001	0.84 (0.76–0.93)	< 0.001
H-sys	1.14 (1.09–1.19)	< 0.001	1.11 (1.06–1.15)	< 0.001
VH-sys	1.40 (1.32–1.49)	< 0.001	1.32 (1.24–1.41)	< 0.001
EH-sys	1.30 (1.11–1.52)	< 0.001	1.21 (1.04–1.42)	0.016
Afib
VL-sys	0.90 (0.66–1.23)	0.509	1.01 (0.74–1.39)	0.943
L-sys	0.75 (0.67–0.84)	< 0.001	0.83 (0.74–0.93)	0.002
H-sys	1.02 (0.97–1.07)	0.395	1.00 (0.95–1.05)	0.910
VH-sys	1.12 (1.04–1.20)	0.003	1.09 (1.01–1.17)	0.029
EH-sys	1.19 (1.00–1.42)	0.046	1.21 (1.01–1.44)	0.036
Dementia
VL-sys	0.80 (0.52–1.25)	0.331	1.04 (0.67–1.63)	0.853
L-sys	0.94 (0.82–1.08)	0.371	1.01 (0.88–1.16)	0.870
H-sys	1.06 (0.99–1.13)	0.066	1.11 (1.04–1.18)	0.002
VH-sys	1.12 (1.02–1.24)	0.021	1.20 (1.08–1.32)	< 0.001
EH-sys	1.23 (0.98–1.55)	0.077	1.39 (1.10–1.75)	0.006
Fracture
VL-sys	0.99 (0.77–1.28)	0.952	0.99 (0.76–1.27)	0.913
L-sys	1.04 (0.96–1.13)	0.376	1.04 (0.96–1.14)	0.332
H-sys	1.03 (0.99–1.08)	0.103	1.06 (1.02–1.11)	0.005
VH-sys	1.11 (1.04–1.18)	< 0.001	1.14 (1.07–1.22)	< 0.001
EH-sys	1.25 (1.08–1.45)	0.003	1.30 (1.12–1.51)	< 0.001

Multivariate analysis was adjusted for age, sex, vascular access type, hemodialysis vintage, underlying cause of end-stage kidney disease, Charlson Comorbidity Index score, Kt/Vurea, ultrafiltration volume, hemoglobin, serum albumin, serum creatinine, serum phosphorus, serum calcium, use of antihypertensive drugs, statins, clopidogrel or aspirin, the presence of myocardial infarction or congestive heart failure, and cerebrovascular events. The analysis was performed using the enter method. The reference group consisted of patients with 120 ≤ systolic blood pressure < 140 mmHg

Abbreviations: Afib, atrial fibrillation; CI, confidence interval; CVE, cardiovascular event; HR, hazard ratio; SBP, systolic blood pressure. SBP categories: EH-sys, SBP ≥ 180 mmHg; H-sys, 140 ≤ SBP < 160 mmHg; L-sys, 100 mmHg ≤ SBP < 120 mmHg; VH-sys, 160 ≤ SBP < 180 mmHg; VL-sys, SBP < 100 mmHg

Clinical outcomes according to DBP groups

The follow-up durations in EL-dia, VL-dia, L-dia, R-dia, H-dia, and VH-dia were 64 ± 38, 70 ± 39, 75 ± 39, 76 ± 40, 78 ± 41, and 82 ± 38 months, respectively. The 5-year survival rates in the EL-dia, VL-dia, L-dia, R-dia, H-dia, and VH-dia were 53.6%, 59.8%, 66.5%, 68.4%, 70.8%, and 77.9%, respectively; CVE-free rates were 71.4%, 69.6%, 74.7%, 74.2%, 75.3%, and 72.0%; Afib-free rates were 78.5%, 80.9%, 84.0%, 84.8%, 86.0%, and 84.1%; dementia-free rates were 90.0%, 88.3%, 90.7%, 90.9%, 92.5%, and 94.0%; and fracture-free rates were 65.6%, 72.2%, 76.3%, 78.3%, 80.5%, and 82.1% (Fig. 2).

Fig. 2.

Kaplan–Meier curves of clinical outcomes based on DBP groups. Patient survival (A), CVE-free (B), Afib-free (C), dementia-free (D), and fracture-free (E). P-values for pairwise comparisons, determined with log-rank tests, were presented at the bottom of the graph. Abbreviations: Afib, atrial fibrillation; CVE, cardiovascular event; DBP, diastolic blood pressure. DBP categories: EL-dia, DBP < 60 mmHg; H-dia, 90 ≤ DBP < 100 mmHg; L-dia, 70 ≤ DBP < 80 mmHg; R-dia, 80 ≤ DBP < 90 mmHg; VH-dia, DBP ≥ 100 mmHg; VL-dia, 60 ≤ DBP < 70 mmHg

Multivariable Cox regression analyses revealed that the VL-dia and L-dia groups were associated with lower all-cause mortality, while groups with higher DBP were associated with higher all-cause mortality than that of the R-dia group (Table 3). Groups with lower DBP were associated with lower CVE risk, whereas those with higher DBP had higher CVE risk than that of the R-dia group. Dementia risk was associated with lower in groups with lower DBP and higher in the H-dia group than in the R-dia group. Kaplan–Meier analysis and univariable analyses revealed higher survival rates in the VH-dia group compared with reference group. In the multivariable analysis, the highest HR was observed in the VH-dia group, while the VL-dia and L-dia groups were associated with lower mortality. The apparently contradictory results for the VH-dia group may reflect confounding by age or comorbidity.

Table 3.

Diastolic blood pressure and clinical outcomes

	Univariate		Multivariate
	HR (95% CI)	P	HR (95% CI)	P
All-cause mortality
EL-dia	1.71 (1.60–1.83)	< 0.001	0.96 (0.89–1.02)	0.193
VL-dia	1.37 (1.32–1.42)	< 0.001	0.89 (0.86–0.93)	< 0.001
L-dia	1.09 (1.06–1.12)	< 0.001	0.93 (0.90–0.96)	< 0.001
H-dia	0.92 (0.88–0.97)	0.001	1.12 (1.06–1.17)	< 0.001
VH-dia	0.69 (0.60–0.81)	< 0.001	1.19 (1.02–1.38)	0.029
CVE
EL-dia	1.15 (1.02–1.29)	0.024	0.82 (0.73–0.93)	0.002
VL-dia	1.17 (1.10–1.24)	< 0.001	0.91 (0.86–0.97)	0.005
L-dia	0.98 (0.94–1.03)	0.446	0.90 (0.86–0.94)	< 0.001
H-dia	0.99 (0.93–1.06)	0.760	1.08 (1.01–1.16)	0.019
VH-dia	0.92 (0.76–1.11)	0.386	1.23 (1.01–1.48)	0.035
Afib
EL-dia	1.37 (1.20–1.54)	< 0.001	1.07 (0.94–1.22)	0.289
VL-dia	1.23 (1.15–1.32)	< 0.001	1.03 (0.96–1.10)	0.440
L-dia	1.06 (1.01–1.12)	0.021	1.01 (0.96–1.06)	0.732
H-dia	0.97 (0.89–1.05)	0.381	1.05 (0.97–1.14)	0.215
VH-dia	0.90 (0.72–1.13)	0.362	1.19 (0.94–1.50)	0.140
Dementia
EL-dia	1.16 (0.97–1.38)	0.100	0.66 (0.55–0.78)	< 0.001
VL-dia	1.29 (1.18–1.41)	< 0.001	0.84 (0.77–0.92)	< 0.001
L-dia	1.05 (0.98–1.13)	0.131	0.89 (0.83–0.96)	0.001
H-dia	0.91 (0.81–1.01)	0.068	1.11 (1.00–1.24)	0.049
VH-dia	0.66 (0.47–0.93)	0.017	1.13 (0.80–1.59)	0.490
Fracture
EL-dia	1.59 (1.44–1.75)	< 0.001	1.08 (0.98–1.19)	0.134
VL-dia	1.31 (1.24–1.39)	< 0.001	0.99 (0.93–1.05)	0.665
L-dia	1.10 (1.05–1.15)	< 0.001	0.98 (0.94–1.03)	0.462
H-dia	0.91 (0.85–0.97)	0.007	1.03 (0.96–1.10)	0.418
VH-dia	0.74 (0.60–0.91)	0.004	1.01 (0.82–1.25)	0.892

Multivariate analysis was adjusted for age, sex, vascular access type, hemodialysis vintage, underlying cause of end-stage kidney disease, Charlson Comorbidity Index score, Kt/Vurea, ultrafiltration volume, hemoglobin, serum albumin, serum creatinine, serum phosphorus, serum calcium, use of antihypertensive drugs, statins, clopidogrel, or aspirin, the presence of myocardial infarction or congestive heart failure, and cerebrovascular events. The anlaysis was performed using the enter method. The reference group consisted of patients with 80 ≤ diastolic blood pressure < 90 mmHg

Abbreviations: Afib, atrial fibrillation; CI, confidence interval, CVE, cardiovascular event; DBP, diastolic blood pressure; HR, hazard ratio. DBP categories: EL-dia, DBP < 60 mmHg; H-dia, 90 ≤ DBP < 100 mmHg; L-dia, 70 ≤ DBP < 80 mmHg; VH-dia, DBP ≥ 100 mmHg; VL-dia, 60 ≤ DBP < 70 mmHg

Spline curves between systolic/diastolic blood pressure and clinical outcomes

Table S4 presents the number of events at the end-point of follow-up. SBP levels below the median were associated with lower all-cause mortality and CVE than those of the median (Fig. 3), while levels above the median were associated with higher all-cause mortality, CVE, dementia, and fractures. DBP levels above the median were associated with higher all-cause mortality, CVE, and dementia, while DBP levels below the median value correlated with lower dementia risk than that of the median value. For all-cause mortality and CVE, DBP levels below the median were associated with lower risk, but as DBP decreased beyond a specific threshold, the protective effect diminished and was no longer statistically significant.

Fig. 3.

Spline curves illustrating hazard ratios and 95% confidence intervals for clinical outcomes according to blood pressure (A-E, SBP; F-J, DBP). Patient survival (A, F), CVE-free (B, G), Afib-free (C, H), dementia-free (D, I), and fracture-free (E, J). The reference points were established at 140 mmHg for SBP and 80 mmHg for DBP. Data were plotted using a multivariable model, with adjustments for the following factors: age, sex, vascular access type, hemodialysis vintage, underlying cause of end-stage kidney disease; Charlson Comorbidity Index score, Kt/V_urea, ultrafiltration volume, hemoglobin, serum albumin, serum creatinine, serum phosphorus, serum calcium, use of anti-hypertensive drugs, statins, clopidogrel or aspirin, and the presence of myocardial infarction or congestive heart failure, and cerebrovascular events. Abbreviations: Afib, atrial fibrillation; DBP, diastolic blood pressure; SBP, systolic blood pressure

Sensitivity analyses

We performed sensitivity analyses using ultrafiltration volume normalized to post-dialysis body weight rather than including ultrafiltration volume as a covariate. The results of these analyses were consistent with the primary analysis (Table S5). Additional sensitivity analyses were conducted adjusting for β-blocker use, given its potential role as a crucial confounder in the association between BP and Afib (Table S6). The findings were largely consistent with those of the primary multivariable analyses, and adjustment for β-blocker use did not materially alter the overall pattern of associations. Futhermore, sensitivity analyses were performed incorporating SBP-CV and DBP-CV (Table S7). Despite differences in the CV of BP across groups, multivariate analyses including both CV measures showed trends consistent with the primary analyses.

Discussion

Our study examined the association between SBP or DBP and clinical outcomes in a cohort of patients undergoing HD in South Korea. Higher SBP, whether analyzed as a grouped or continuous variable, was associated with increased risks of all-cause mortality, CVE, dementia, and fractures compared to those of the reference group. Conversely, lower SBP was associated with reduced risks for all-cause mortality and CVE. Regarding DBP, higher levels were associated with increased risks of all-cause mortality, CVE, and dementia, while lower DBP, particularly in the VL-dia and L-dia groups, was associated with reduced risks. No significant associations were observed between DBP and Afib or fractures.

Recent UK guidelines recommend a pre-dialysis SBP of 140–164 mmHg and a DBP of 60–100 mmHg in patients undergoing HD [6]. In contrast, the KDIGO guidelines set a target SBP of < 120 mmHg for patients with non-dialysis chronic kidney disease, but offer no explicit recommendations for those undergoing HD [5]. The higher SBP targets in patients undergoing HD stem from studies demonstrating reverse epidemiology, where lower BP paradoxically correlates with increased mortality [16]. The elevated mortality risk with low SBP is attributed to factors such as baseline low cardiac function and increased risk of intradialytic hypotension [17–19]. While AHA and ESC guidelines for the general population recommend targeting < 130/80 mmHg and 120–129 mmHg for SBP, respectively, these targets may not be suitable for patients undergoing HD [20, 21]. Lower BP may offer favorable or neutral effects in some populations; however, evidence supporting safe lower thresholds in patients undergoing HD remains limited and requires further investigation.

The 2005 KDOQI guidelines recommended a target pre-dialysis BP of < 140/90 mmHg and a post-dialysis BP of < 130/80 mmHg [22]. However, these recommendations were classified as weak because of the limited availability of randomized controlled trial data in the HD population at that time. In subsequent updates, specific target BP levels remain unclear, reflecting ongoing uncertainty and the limited availability of high-quality interventional evidence in this population [23]. The KDIGO guidelines focused on BP targets primarily in patients with non-dialysis chronic kidney disease, incorporating evidence from major trials, such as SPRINT [5, 8]. These recommendations generally support more intensive BP control in selected populations. However, the applicability of these targets to patients undergoing maintenance HD remains uncertain, as robust, dialysis-specific randomized evidence is limited. In this context, these findings may help inform the balance between earlier, more conservative BP targets and the recent trend toward tighter BP control. Historically, strict BP reduction in patients undergoing dialysis has raised concerns regarding potential harm, leading to relatively higher target levels. However, current perspectives increasingly support achieving BP levels closer to the normal range when clinically feasible. Nevertheless, high-quality evidence in the HD population remains limited. Owing to its large sample size and national, real-world cohort design, this study provides clinically relevant observational evidence that may inform the design of future randomized controlled trials. These findings should not be interpreted as establishing definitive treatment targets but as hypothesis-generating and a potential framework to guide future interventional studies aimed at defining optimal BP goals in patients undergoing maintenance HD.

Studies in HD populations primarily focus on the increased mortality risk associated with lower SBP [18, 24–26]. For instance, Kim et al. analyzed the Korean registry data to examine the association between pre-dialysis BP and all-cause mortality from 2001 to 2020 [24]. Although the study confirms the elevated mortality risk associated with SBP < 120 mmHg, all patients with SBP < 120 mmHg were grouped into a single category, limiting detailed evaluation of outcomes in the 100–120 mmHg range. Moreover, the study reports a weak association between DBP and mortality, likely due to methodological limitations, including voluntarily submitted data, potential measurement errors, and reliance on single rather than averaged BP values. Similarly, Li et al. analyzed prevalent patients undergoing HD from the Fresenius Medical Care database (1997–2001) and observed increased mortality with SBP < 120 mmHg; however, no subgroup analyses were performed within this cohort [25]. Earlier studies, such as those by Zager et al., report that SBP < 130 mmHg is associated with higher mortality than that of a reference group with SBP 140–149 mmHg [18]. More recently, Jhee et al. analyzed prospective cohort data, showing that SBP < 110 mmHg is associated with increased mortality, whereas SBP 110–129 mmHg exhibits a neutral effect compared to that of the reference group (130–139 mmHg) [26]. However, their study population was relatively limited.

While substantial evidence links low BP to risks in patients undergoing HD, data on stratified outcomes within the lower SBP range remain limited. In this study, we subdivided SBP < 120 mmHg into < 100 mmHg and 100–119 mmHg categories, using 120–139 mmHg as the reference. Our findings suggest that, even within the lower SBP range, a reduction to 100 mmHg may be clinically beneficial if well tolerated. This provides new insights into the potential benefits of tighter BP control for improving clinical outcomes in patients undergoing HD. However, as a retrospective analysis, our study cannot establish causality and lacks data on potentially confounding variables such as post- and intradialytic BP, heart rate, and other hemodynamic parameters. Therefore, future studies incorporating these variables and adjusting for their potential confounding effects will be necessary to validate and expand our findings.

One strength of our study is that, in addition to evaluating all-cause mortality, we assessed several key outcomes, including CVE, Afib, dementia, and fractures. Compared to the reference group, higher SBP and DBP levels were associated with worse survival, whereas modestly lower BP (SBP 100–119 mmHg or DBP 60–79 mmHg) correlated with reduced mortality risk. CVE followed a similar trend to all-cause mortality. We also explored the association between BP and dementia. While higher BP was associated with increased dementia risk, the protective effect of lower BP was relatively minimal. For Afib, no strong association was observed with SBP or DBP. This may be due to the widespread use of antihypertensive medications, particularly beta-blockers, in patients with high BP or cardiovascular disease. These confounding factors may have influenced the incidence of Afib in our cohort.

While the relationship between intradialytic hypotension and the risks of falls and fractures is mechanistically explained and has been investigated in multiple studies, the biological mechanisms linking arterial hypertension to fracture risk are poorly defined and supporting evidence remains limited. Nevertheless, two potential hypotheses may be proposed. First, arterial hypertension may be associated with overt or subclinical neurological abnormalities that increase fracture risk. Patients with arterial hypertension are more likely to develop small-vessel disease, microinfarctions, or other cerebrovascular injuries, which potentially impair neurological or neuromuscular coordination. These impairments may reduce protective responses during falls, disrupt bone–muscle interactions, and ultimately elevate fracture risk. Second, an arterial hypertension–vascular calcification–bone pathology pathway may be considered. Arterial hypertension is closely associated with vascular calcification, which is linked to bone pathologies, including osteoporosis [27]. Through shared mineral and metabolic pathways, these vascular alterations may reduce bone strength and increase fracture risk. While this mechanism differs in directionality from the chronic kidney disease–mineral and bone disease pathway typically observed in patients undergoing dialysis, similar vascular–bone interactions are proposed to explain the association between arterial hypertension and fracture risk in non-dialysis populations. A previous study shows that arterial calcification represents a particularly harmful form of ectopic calcification, sharing common molecular mechanisms with osteoporosis [28]. Overall, mechanistic studies evaluating the association between arterial hypertension and fracture risk remain limited, particularly in patients undergoing dialysis. Further research is required to elucidate these potential biological pathways.

These findings should be interpreted with caution for several reasons. First, most baseline characteristics differed significantly across the SBP and DBP groups. While we performed multivariable adjustment to account for these differences, complete adjustment is not possible in an observational study, and residual confounding may persist. Therefore, the potential influence of residual confounding should be carefully considered when interpreting these findings. Specifically, the very low SBP (VL-sys) group comprised a relatively small number of patients and was younger (mean age, approximately 55 years), approximately 5 years younger than those in the reference group. Given these differences, the multivariable estimates for this subgroup may be particularly susceptible to residual confounding. Consequently, the attenuated, reversed, or non-significant associations observed in this group should be interpreted with caution. Given the inherent limitations of this study, these findings should not be applied as definitive targets for all patients undergoing HD. Further validation through well-designed prospective studies with more rigorously controlled baseline characteristics is necessary to define clearer and more definitive BP targets in this population.

In our cohort, substantial differences in baseline characteristics were observed across SBP and DBP groups. Given the retrospective study design and the relatively small sample size in the extreme BP groups, findings of these subgroups should be interpreted with caution. Therefore, conclusions should not reply on a single analytical result but rather reflect a comprehensive evaluation of univariate analyses, multivariable models, spline analyses, statistical significance, and the sample size and clinical characteristics of each group. Regarding the multivariable analyses (Tables 2 and 3), the hazard ratios suggest a U-shaped association for SBP and a more linear relationship for DBP. However, when the adjusted spline curves, statistical significance, and differences between univariate and multivariate models are considered, the trends observed in the extremely low BP groups—particularly EL-dia and VL-sys—should be interpreted with caution. For example, the EL-dia group included a substantially higher proportion of older patients than the other groups. In univariate analysis, this group demonstrated higher all-cause mortality than the reference group; however, after multivariable adjustment and spline analyses, the association was attenuated and no longer statistically significant, with a trend toward lower mortality. These findings may reflect the influence of marked baseline imbalances, and over-adjustment or residual confounding cannot be excluded. Therefore, definitive conclusions regarding these extreme categories cannot be reached. Nevertheless, compared to the reference group, the L-dia, VL-dia, and L-sys groups consistently demonstrated lower mortality across multivariable Cox models and spline analyses. The consistency of these findings across analytical approaches suggests that moderately lower BP ranges may be associated with reduced all-cause mortality in this population. However, these associations should not be interpreted as causal and warrant further validation in prospective studies.

The inclusion of dementia and fracture as clinical outcomes constitutes a strength of the study; however, these findings should be interpreted with caution. Both outcomes were identified exclusively using ICD-10 diagnostic codes, without verification through detailed clinical records or objective assessments, introducing the potential for misclassification bias. Furthermore, significant baseline differences across groups, along with the absence of key fracture-related variables—such as vitamin D status, bone mineral density, and intact parathyroid hormone levels—limit comprehensive adjustment for residual confounding. Therefore, the observed associations should be interpreted with caution. However, because misclassification bias typically attenuates association toward the null, the statistically significant link between BP and dementia in this study warrants further investigation. Prospective studies incorporating more comprehensive clinical assessments are needed to clarify potential causal relationships.

In this study, the association between BP and Afib was not clearly established. For SBP, the L-sys group exhibited a lower risk of Afib compared to the reference group, whereas the VH-sys and EH-sys groups demonstrated an increased risk. However, these associations were borderline statistically significant, and no significant differences were observed in the VL-sys or H-sys groups. Furthermore, spline curve analyses did not reveal a clear dose–response relationship between SBP and Afib incidence. For DBP, no significant associations with Afib were identified across groups. The absence of a definitive association may reflect residual or unmeasured confounding factors, such as structural cardiac abnormalities, intradialytic hypotension, concomitant medications, and lifestyle factors not fully captured in the dataset. Future studies incorporating detailed cardiovascular and hemodynamic data may further clarify the relationship between BP and Afib risk in patients undergoing HD.

Our study has some limitations. First, this study is a retrospective analysis of data derived from prevalent patients undergoing HD. Therefore, causal inferences cannot be established, and the findings should be interpreted as associative rather than causal. Selection bias may have occurred because of baseline differences between groups, and despite multivariable adjustment, residual confounding cannot be fully excluded. Additionally, unmeasured variables, such as nutritional status, dialysis vintage, volume status, and frailty, may have influenced the observed associations. Therefore, the observed intergroup differences in these variables may have influenced the associations, and complete adjustment may not have been achievable. Measures of dry weight were unavailable in this dataset. The sensitivity analyses confirmed trends observed in the primary analysis, but information regarding dry weight and the extent to which it was achieved was not available in this dataset. Future prospective studies incorporating objective volume status assessments and repeated BP measurements at multiple time points are necessary to address this limitation and further clarify the relationship between BP and clinical outcomes in patients undergoing dialysis. Furthermore, as this study included only Korean patients undergoing HD, potential differences related to ethnicity, national background, or healthcare systems could not be assessed. Therefore, caution is warranted when generalizing these findings to other racial or ethnic populations or healthcare settings. Second, no standardization is present for measuring BP, which may have resulted in inter-center variability in BP measurements. BP was assessed using pre-dialysis measurements, defined as the mean of three readings. While averaging multiple measurements may improve reliability, pre-dialysis BP does not fully capture intradialytic hypotension, interdialytic or ambulatory BP patterns, or BP variability during dialysis sessions, and may be influenced by volume status or white-coat effects. Therefore, these factors limit the generalizability of our findings, and the observed associations should be interpreted within the context of routinely collected clinical BP measurements rather than as definitive indicators of overall BP burden. Third, we only used diagnostic codes and procedural information to assess the effects of CVE, Afib, dementia, and fractures. More detailed data, such as echocardiographic results, electrocardiograms, questionnaires, and image studies could provide a clearer association with these outcomes. Fourth, the dataset only included all-cause mortality data with limited information on specific causes of death, such as cachexia, infection, and cardiovascular diseases. Such information would have been valuable for a deeper understanding of the relationship between BP and cause-specific mortality. Additionally, all BP measurements in our study were taken immediately before dialysis. Further measurements, such as those for intradialytic hypotension, post-dialysis BP, or heart rate, were not evaluated, and these factors may act as confounders. Fifth, the number of patients with extremely high or low BP was significantly lower than that of other groups. Therefore, caution is needed in interpreting the results of multivariable analyses for those patients. In particular, the low statistical significance observed in the VL-sys and EL-dia was likely due to the limited sample size. Sixth, we applied a consistent set of covariates across outcomes. Different outcomes have distinct pathophysiological mechanisms, and outcome-specific covariate selection may be theoretically appropriate. In the primary analyses, a consistent set of core covariates was applied across all outcomes to maintain comparability and adjust for major confounders commonly associated with both BP and adverse clinical events in patients undergoing dialysis. These covariates included demographic factors, comorbidities, dialysis-related variables, laboratory parameters, and medication use. Additionally, certain outcome-specific factors—such as educational level and inflammatory markers for dementia, or intact parathyroid hormone levels and prior fracture history for fracture risk—may further refine risk estimation; however, these variables were unavailable in the dataset. While a consistent set of covariates was applied across outcomes to maintain comparability, outcome-specific risk factors may differ. The absence of certain outcome-specific variables may have introduced residual confounding. Future studies incorporating more granular, outcome-specific covariates are warranted to further clarify these associations.

In conclusion, a SBP of 100–119 mmHg was associated with improved overall survival and reduced incidence of CVE, whereas a DBP of 60–79 mmHg was associated with improved survival and lower risks of CVE and dementia in this observational study. These findings suggest that lower BP ranges were correlated with more favorable outcomes in the dataset. However, given study limitations, these findings should not be interpreted as definitive BP targets for all patients undergoing HD. These associations should be not interpreted as causal or as justification for immediate changes to current guideline-recommended targets. Rather, the findings should be considered hypothesis-generating. Further validation through well-designed prospective studies with more rigorously controlled baseline characteristics is needed to establish clearer and more definitive BP targets in this population.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

Afib

Atrial fibrillation

BP

Blood pressure

CCI

Charlson comorbidity index

CHF

Congestive heart failure

CI

Confidence interval

CVA

Cerebrovascular accidents

CVE

Cardiovascular events

DBP

Diastolic blood pressure

HD

Hemodialysis

HIRA

Health Insurance Review and Assessment Service

HR

Hazard ratio

MI

Myocardial infarction

SBP

Systolic blood pressure

Author contributions

SHK conceptualized and designed the study, performed the analysis and interpretation of data. SHK and JYD wrote the manuscript. SYP, YJL, JYC, and BYK generated and collected the data. JYD drafted and revised the manuscript. All authors approved the final version of the manuscript.

Funding

This work was supported by the Medical Research Center Program through the National Research Foundation (NRF) of Korea funded by the Ministry of Science, ICT, and Future Planning (2022R1A5A2018865), and the Basic Science Research Program through the NRF of Korea, funded by the Ministry of Education (2022R1I1A3072966).

Data availability

The raw data were generated by the Health Insurance Review and Assessment Service. The database can be requested from the Health Insurance Review and Assessment Service by sending a study proposal including the purpose of the study, study design, and duration of analysis through the web site (https://www.hira.or.kr/main.do). The authors cannot distribute the data without permission.

Declarations

Ethical approval and consent to participate

The study was conducted ethically in accordance with the World Medical Association Declaration of Helsinki. The study was approved by the institutional review board (IRB) of the Yeungnam University Medical Center (approval no. YUMC 2023-12-012). Informed consent was not obtained from the patients since the records and information of the participants were anonymized and de-identified before the analysis. The IRB also waived the need for obtaining informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.