Association of HA130/KHA80 hemoperfusion combined with hemodialysis with survival outcomes in patients receiving maintenance hemodialysis: a time-dependent and propensity score-matched analysis Authors: Yan Liu, Wei Gang, Wei Chen, Limei Yang, Rui Dong, Zhiyong Guo Published: 2026 ABSTRACT AbstractObjectivePatients receiving maintenance hemodialysis (MHD) experience a high mortality burden, with cardiovascular disease remaining a major cause of death. This study aimed to evaluate the association between adjunctive hemoperfusion combined with conventional hemodialysis (HD + HP) and long-term survival outcomes in patients receiving MHD.MethodsThis single-center retrospective cohort study was conducted at Shanghai Changhai Hospital. Patients with end-stage renal disease who received MHD between January 2015 and December 2023 were retrospectively identified from the center’s hemodialysis records. Among the 640 eligible patients, 419 (65.5%) received HD during the study period, whereas 221 (34.5%) received adjunctive HP in addition to HD. HP exposure was modeled as a time-varying covariate in a multivariable time-dependent Cox proportional hazards model to account for treatment timing. A marginal structural model (MSM) using stabilized inverse probability of treatment weighting was further constructed to address measured time-dependent confounding. Propensity score matching was performed as a sensitivity analysis using 1:1 nearest-neighbor matching without replacement, with a caliper width of 0.02. Survival outcomes in the matched cohort were evaluated using Simon–Makuch curves and Mantel–Byar tests.ResultsAmong the 640 included MHD patients, the median age was 67.5 years, 66.9% were male, and the median dialysis vintage was 48.0 months. During follow-up, 348 patients died, corresponding to all-cause mortality proportion of 54.4% in the study sample. Diabetic kidney disease was the most common primary renal disease (30.9%), followed by chronic glomerulonephritis (29.4%) and hypertensive nephropathy (18.6%). Among the 348 deaths, cardiovascular disease was the leading cause of death (32.8%), followed by multiple organ dysfunction syndrome (21.6%). In both the multivariable time-dependent Cox model and the MSM, HD + HP was consistently associated with a lower risk of all-cause mortality compared with HD, with HRs of 0.69 (95% CI, 0.52–0.90; P = 0.009) and 0.73 (95% CI, 0.55–0.97; P = 0.030), respectively. After 1:1 propensity score matching, 173 patients remained in each group with improved baseline balance. In the propensity score-matched cohort, Simon-Makuch analyses further supported a protective association of HD + HP with all-cause and cardiovascular mortality compared with HD (Mantel-Byar test, P = 0.032 and P = 0.003, respectively).ConclusionIn this single-center retrospective cohort of patients receiving MHD, cardiovascular disease was the leading cause of death. HD + HP therapy was associated with lower all-cause mortality in time-dependent Cox regression and MSM-based weighted analyses, and with lower all-cause and cardiovascular mortality in propensity score-matched Simon–Makuch analyses. These findings suggest a potential survival benefit of adjunctive HP.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12882-026-05124-4.Keywords: Hemoperfusion, Maintenance hemodialysis, All-cause mortality, Cardiovascular mortality, Time-dependent cox regression, Marginal structural model, Propensity score matching, Simon-Makuch curve Introduction End-stage renal disease (ESRD) represents a significant global health burden, with maintenance hemodialysis (MHD) serving as the primary renal replacement therapy [1, 2]. MHD accounts for approximately 69% of all renal replacement therapies and 89% of dialysis treatments. Despite continuous improvements in dialysis technology and clinical management, patients receiving MHD still carry an unacceptably high mortality risk, with cardiovascular disease being the leading cause of death [3]. Conventional hemodialysis (HD) effectively removes small water-soluble uremic toxins through diffusion, but its capacity to clear middle-to-large molecular weight and protein-bound uremic toxins remains limited [4]. Accumulation of protein-bound uremic toxins has been associated with cardiovascular complications and mortality in patients receiving MHD [5–7]. Hemoperfusion (HP) is an important blood purification technique that effectively removes middle-to-large molecular weight and protein-bound uremic toxins via adsorbent materials within an extracorporeal circuit [8, 9]. The addition of HP to conventional HD, hereafter referred to as HD + HP therapy, may theoretically complement conventional HD by enhancing the removal of middle-to-large molecular weight and protein-bound uremic toxins [10]. Nevertheless, real-world evidence on the long-term survival outcomes associated with adjunctive HP in patients receiving MHD remains limited. Therefore, this single-center retrospective cohort study aimed to describe the clinical characteristics and distribution of causes of death among patients receiving MHD, identify factors associated with all-cause mortality, and evaluate the association between HD + HP therapy and the risks of all-cause and cardiovascular mortality while accounting for treatment timing and potential confounding. Methods Study design and setting This single-center retrospective cohort study was conducted at Shanghai Changhai Hospital, The First Affiliated Hospital of Naval Medical University, a high-volume tertiary referral center. Eligible patients with end-stage renal disease receiving MHD between January 2015 and December 2023 were retrospectively identified from routinely collected hemodialysis records at our center. The date of MHD initiation for each patient was retrieved individually from medical records and the hemodialysis management system. Therefore, the study population included both patients who initiated MHD during the 2015–2023 study period and those who had initiated MHD before 2015 but continued regular treatment and follow-up at our center during the study period. For patients who initiated MHD before January 2015, the cohort entry date was defined as the first documented regular HD treatment at our center during the study period. Patients were categorized into the HD and HD + HP groups according to whether adjunctive HP was initiated during follow-up as a component of their maintenance blood purification regimen, with the treatment modality remaining fixed after HP initiation. In subsequent survival analyses, HP exposure was modeled as a time-varying covariate, whereby patients contributed person-time to the HD exposure state before HP initiation and to the HD + HP exposure state thereafter. Patients in the HD group received conventional HD three times weekly using bicarbonate-based dialysate [potassium (K), 2.0 mmol/L; calcium (Ca), 1.25–1.75 mmol/L; sodium, 138 mmol/L; magnesium, 0.5 mmol/L]. Blood flow and dialysate flow rates were maintained at 200–300 mL/min and 500 mL/min, respectively. Patients in the HD + HP group received conventional HD combined with HP using specific HP cartridges, including HA130 and KHA80. The HP cartridge was connected in series before the dialyzer, removed after 2 h of treatment, and the total session duration was maintained at 3.5–4 h. Blood flow was controlled at 180–250 mL/min, and the dialysate composition was identical to that used in the HD group. In both groups, vascular access was established via either an arteriovenous fistula (AVF) or a tunneled cuffed catheter (TCC). HP initiation in clinical practice At our center, the decision to add HP to conventional HD was primarily guided by the Blood Purification Standard Operating Procedures issued by the National Health Commission of the People’s Republic of China, including the 2010 Edition and, after its release, the 2021 Edition. In routine practice, adjunctive HP was considered for patients with dialysis-related indications, including refractory uremic symptoms, elevated β2-microglobulin (β2-MG) levels, suspected dialysis-related amyloidosis, persistent inflammatory status, or other manifestations suggesting inadequate toxin clearance with conventional HD. Once initiated, adjunctive HP was regularly incorporated into the patient’s maintenance blood purification regimen, with a treatment frequency of once weekly or once every two weeks. The treatment modality generally remained fixed thereafter. Thus, HP initiation reflected routine real-world clinical practice based on standardized indications, individualized assessment, and informed patient preference. Although this clinical decision-making process may introduce confounding by indication, subsequent analyses were designed to account for treatment timing and measured confounders. Eligibility and exclusion criteria A total of 1,084 patients with ESRD who received MHD at our center between January 2015 and December 2023 were retrospectively identified from hemodialysis records. Patients were excluded if they met any of the following criteria: age < 18 years, MHD duration < 3 months, conversion to peritoneal dialysis during follow-up, kidney transplantation, transfer to other hospitals, loss to follow-up, severely incomplete clinical records, or irregular blood purification regimens or repeatedly changed treatment modalities during follow-up. Specifically, 21 patients were excluded because they were aged < 18 years, 207 because they had received MHD for < 3 months, 7 because they converted to peritoneal dialysis during follow-up, 33 because they underwent kidney transplantation, 111 because they transferred to other hospitals, 24 because they were lost to follow-up, 23 because they had severely incomplete clinical records, and 18 because they had irregular blood purification regimens or repeatedly changed treatment modalities during follow-up, which precluded clear exposure classification. Ultimately, 640 patients met the inclusion criteria and were included in the final statistical analysis (Fig. 1). Flowchart of study population inclusion and exclusion Data collection and clinical variables Demographic characteristics, past medical history, primary renal disease and laboratory results were retrieved from the hospital’s inpatient record system and hemodialysis data-management system, and were manually verified against physician-recorded discharge summaries, hospital chart abstracts and diagnostic coding sheets. The presence of comorbidities was determined based on International Classification of Diseases, 10th Revision (ICD-10) diagnostic codes assigned at discharge and confirmed by supporting clinical documentation in the medical records. Collected clinical data included: sex, age, dialysis initiation date, date of death (for deceased patients), primary renal disease [diabetic kidney disease (DKD), chronic glomerulonephritis (CGN), hypertensive nephropathy (HN) and other causes], comorbidities [hypertension, diabetes mellitus, cardiovascular disease], vascular access type (AVF or TCC), and cause of death (for deceased patients). Laboratory parameters included albumin (ALB), hemoglobin (HB), red blood cell distribution width (RDW), parathyroid hormone (PTH), prealbumin (PAB), total protein (TP), chloride (Cl), creatine kinase-MB (CK-MB), and other routinely measured indicators. Calculation Dialysis vintage \:V(in months) and age \:A(in years) were defined as follows: Where: and are the dialysis vintage and age for patient \:i, respectively. and are the dialysis vintage and age for patient \:i, respectively. denotes the floor function, used here to calculate completed months or years. denotes the floor function, used here to calculate completed months or years. Definitions Cardiovascular death was defined as death caused by acute myocardial infarction, ischemic heart disease, cardiomyopathy, fatal arrhythmia, cardiac arrest, congestive heart failure, or other cardiovascular conditions. For survival analyses, time zero was defined as the cohort entry date. Cohort entry was defined as the date of MHD initiation for patients who started MHD during the study period and as the first documented regular HD treatment at our center during the study period for patients who had initiated MHD before January 2015. The date of first MHD initiation was used to calculate baseline dialysis vintage. To reduce immortal time bias, HP exposure was modeled as a time-varying covariate. Patients who had initiated HP before cohort entry were classified as HP-exposed from cohort entry onward, whereas patients who initiated HP after cohort entry contributed unexposed person-time before the first HP session and HP-exposed person-time thereafter. Follow-up continued until all-cause death or December 31, 2023, whichever occurred first. Statistical analyses Statistical analyses were performed using R software (version 4.5.2; http://www.R-project.org) and SPSS software (version 27.0). A two-sided P < 0.05 was considered statistically significant. Data description and group comparisons: Continuous variables were tested for normality using the Shapiro–Wilk test. The distribution of continuous variables was assessed separately for the overall cohort and for each comparison subgroup when descriptive statistics were generated. Normally distributed variables are presented as mean ± standard deviation (M ± SD) and were compared using independent-sample t-tests. Non-normally distributed variables are expressed as median [interquartile range, IQR] and were compared using the Mann–Whitney U test. Categorical variables are described as number (%) and were compared using χ² tests or Fisher’s exact test as appropriate. Survival outcomes were analyzed using time-to-event methods. All-cause mortality was defined as the primary outcome. Patients were followed from cohort entry until death or December 31, 2023, whichever occurred first. HP exposure was treated as time-varying: patients who had initiated HP before cohort entry were classified as exposed from cohort entry onward, whereas patients who initiated HP after cohort entry contributed unexposed person-time before the first HP session and exposed person-time thereafter. HP exposure was modeled as a time-varying covariate in the Cox proportional hazards model. To further address time-varying HP exposure and measured time-dependent confounding, a marginal structural model (MSM) was constructed using stabilized inverse probability of treatment weighting (IPTW). The association between HD + HP therapy and all-cause mortality was estimated using an IPTW-weighted survival model and reported as hazard ratios (HRs) with 95% confidence intervals (CIs). To reduce potential confounding due to baseline imbalance, propensity score matching was performed as a sensitivity analysis using 1:1 nearest-neighbor matching without replacement, with a caliper width of 0.02. The propensity score model included baseline covariates comprising sex, age, primary renal disease, vascular access, and comorbidities, including diabetes mellitus, hypertension, and cardiovascular disease. Covariate balance after matching was evaluated using standardized mean differences (SMDs), with an absolute SMD < 0.1 considered indicative of acceptable balance. After matching, continuous variables were compared between groups using paired tests as appropriate: the paired t-test for approximately normally distributed data and the Wilcoxon signed-rank test for non-normally distributed data. Because 1:1 nearest-neighbor matching without replacement was used, all matched individuals contributed equally to the analysis, and no additional matching weights were applied. In the matched cohort, survival was displayed using Simon–Makuch curves and compared between groups using the Mantel–Byar test, which are appropriate for time-to-event analyses with time-varying treatment exposure. Results A total of 640 patients undergoing MHD were included in this study. The median age of the study population was 67.5 years, with 66.9% being male. The median dialysis vintage was 48.0 months. Among the included patients, 348 (54.4%) comprised the deceased group, while 292 (45.6%) comprised the survivor group. DKD was the most common primary renal disease in this study (30.9%), followed by CGN (29.4%) and HN (18.6%). Among patients aged ≥ 60 years, DKD was the most frequent etiology (149 patients, 32.6%), whereas among those aged < 60 years, CGN was predominant (65 patients, 35.5%). For vascular access, AVF was used in 71.9% of patients and TCC in 28.1%. Regarding comorbidities, hypertension had the highest prevalence (83.9%), followed by diabetes mellitus (38.1%) and cardiovascular disease (19.5%). Among the 640 eligible patients, 419 (65.5%) received HD during the study period, whereas 221 (34.5%) received adjunctive HP in addition to HD. The clinical and laboratory characteristics of patients who survived and those who died are presented and compared in Table 1. Compared with survivors, patients who died were significantly older (70.0 [62.8, 79.0] vs. 64.0 [51.8, 70.0] years; P < 0.001), had different distributions of primary renal disease, with higher proportions of HN (23.0% vs. 13.4%) and DKD (33.3% vs. 28.1%), and were more likely to use a TCC for vascular access (33.6% vs. 21.6%; P = 0.001). The distribution of hypertension also differed significantly between the deceased and survivor groups (79.3% vs. 89.4%; P = 0.001). The deceased group had a significantly higher proportion of patients receiving HD alone (81.9% vs. 45.9%; P < 0.001). Table 1Baseline clinical, laboratory, and demographic characteristics of MHD patientsCharacteristicOverallSurvival GroupDeath GroupP value(N = 640)(N = 292)(N = 348)DemographicsSex, n (%) Male428 (66.9)195 (66.8)233 (67.0)0.963 Female212 (33.1)97 (33.2)115 (33.0)Age, years67.5 [58.0, 75.0]64.0 [51.8, 70.0]70.0 [62.8, 79.0]< 0.001Primary Renal Disease, n (%) DKD198 (30.9)82 (28.1)116 (33.3) HN119 (18.6)39 (13.4)80 (23.0)< 0.001 CGN188 (29.4)108 (37.0)80 (23.0) Other135 (21.1)63 (21.6)72 (20.7)Vascular Access, n (%) AVF460 (71.9)229 (78.4)231 (66.4)0.001 TCC180 (28.1)63 (21.6)117 (33.6)Comorbidities, n (%) hypertension537 (83.9)261 (89.4)276 (79.3)0.001 diabetes mellitus244 (38.1)109 (37.3)135 (38.8)0.765 cardiovascular disease125 (19.5)65 (22.3)60 (17.2)0.134Dialysis ParametersDialysis vintage, months48.0 [20.0, 96.0]44.0 [17.8, 98.3]50.0 [23.0, 95.0]0.343Treatment Modality, n (%) HD419 (65.5)134 (45.9)285 (81.9)< 0.001 HD + HP221 (34.5)158 (54.1)63 (18.1)Laboratory ParametersCardiac biomarkers CK, U/L74.00 [46.00, 133.50]86.50 [58.00, 137.00]63.00 [40.00, 124.00]< 0.001 CK-MB, U/L11.00 [7.00, 13.50]10.00 [7.00, 13.00]11.00 [8.00, 15.00]0.004 LDH, U/L207 [175, 258.50]207 [175, 257]206.50 [176, 259]0.90Lipid profile TG, mmol/L1.22 [0.83, 1.98]1.40 [0.91, 2.15]1.07 [0.76, 1.68]< 0.001 TC, mmol/L3.57 [2.95, 4.29]3.68 [3.23, 4.35]3.35 [2.80, 4.19]0.004 HDL-C, mmol/L1.11 [0.84, 1.38]1.11 [0.86, 1.38]1.12 [0.80, 1.37]0.368Nutritional markers ALB, g/L39 [36, 42]41 [38, 43]37 [35, 40]< 0.001 TP, g/L72 [68, 75]73 [69, 76]71 [67, 74]0.017 PAB, mg/L268.50 [220, 318]293.50 [251, 346]235.50 [195, 298]< 0.001Hematological indices HB, g/L110 [99, 120]113 [101, 123]106 [96, 115]< 0.001 RDW, %14.40 [13.40, 15.65]13.85 [13.20, 14.70]15.10 [13.80, 16.00]< 0.001 PLT, ×10⁹/L181.00 [143.50, 227.50]183.00 [156.00, 229.00]175.50 [131.00, 226.00]0.098Electrolytes & metabolism Glu, mmol/L7.30 [5.80, 9.90]6.85 [5.70, 8.90]7.50 [5.90, 10.90]0.021 K, mmol/L4.50 [4.00, 4.95]4.60 [4.00, 5.00]4.50 [3.80, 4.90]0.104 Cl, mmol/L98 [95, 100]98 [96, 101]97 [94, 100]0.006 Ca, mmol/L2.35 [2.22, 2.53]2.37 [2.24, 2.52]2.33 [2.18, 2.53]0.495Middle-molecular uremic toxins β2-MG, mg/L42.22 [35.59, 45.77]42.87 [36.56, 45.98]41.61 [35.37, 45.37]0.261 PTH, pg/mL269.00 [155.50, 471]269 [146, 500]268 [158, 452]0.833Kidney function indicators Urea, mmol/L23.6 [18.9, 28.0]23.9 ± 7.0023.9 ± 6.240.202 Scr, µmol/L863 [686,1030]763 [627,948]927 [793,1090]< 0.001 eGFR, mL/min/1.73 m²5.20 [4.40,6.53]5.80 [4.80,6.98]4.70 [4.10,5.70]< 0.001Note: Continuous variables are presented as mean ± SD or median [IQR] according to the distribution of the corresponding dataset. For variables with different distributional patterns between the overall cohort and subgroups, descriptive statistics were reported separately according to the respective normality test results. P < 0.05 was considered statistically significant. Abbreviations: DKD, diabetic kidney disease; HN, hypertensive nephropathy; CGN, chronic glomerulonephritis; Other, other causes; AVF, arteriovenous fistula; TCC, tunneled cuffed catheter; HD, hemodialysis; HD + HP, hemodialysis combined with hemoperfusion; β2-MG, β2-microglobulin; CK, creatine kinase; CK-MB, creatine kinase-MB; LDH, lactate dehydrogenase; TG, triglycerides; TC, total cholesterol; HDL-C, high-density lipoprotein cholesterol; ALB, albumin; TP, total protein; PAB, prealbumin; HB, hemoglobin; RDW, red blood cell distribution width; PLT, platelet count; Glu, glucose; K, potassium; Cl, chloride; Ca, calcium; Scr, serum creatinine; eGFR, estimated glomerular filtration rate; PTH, parathyroid hormone Baseline clinical, laboratory, and demographic characteristics of MHD patients Note: Continuous variables are presented as mean ± SD or median [IQR] according to the distribution of the corresponding dataset. For variables with different distributional patterns between the overall cohort and subgroups, descriptive statistics were reported separately according to the respective normality test results. P < 0.05 was considered statistically significant. Abbreviations: DKD, diabetic kidney disease; HN, hypertensive nephropathy; CGN, chronic glomerulonephritis; Other, other causes; AVF, arteriovenous fistula; TCC, tunneled cuffed catheter; HD, hemodialysis; HD + HP, hemodialysis combined with hemoperfusion; β2-MG, β2-microglobulin; CK, creatine kinase; CK-MB, creatine kinase-MB; LDH, lactate dehydrogenase; TG, triglycerides; TC, total cholesterol; HDL-C, high-density lipoprotein cholesterol; ALB, albumin; TP, total protein; PAB, prealbumin; HB, hemoglobin; RDW, red blood cell distribution width; PLT, platelet count; Glu, glucose; K, potassium; Cl, chloride; Ca, calcium; Scr, serum creatinine; eGFR, estimated glomerular filtration rate; PTH, parathyroid hormone Causes and time trends of mortality in MHD patients Among deceased MHD patients treated at this hemodialysis center between January 2015 and December 2023, cardiovascular disease was the leading cause of death (32.8%), followed by multiple organ dysfunction syndrome (MODS, 21.6%). Mortality causes in the HD group (n = 285) reflected the overall pattern, with cardiovascular disease (33.7%) and MODS (21.8%) as the predominant causes. In the HD + HP group (n = 63), cardiovascular disease (28.6%) and cerebrovascular disease (22.2%) were the most frequent causes of death (Fig. 2). The 3-year numbers of deaths and cause distributions showed distinct temporal trends. Cardiovascular events remained the primary cause of death in each 3-year period. Deaths related to infection and MODS progressively increased over time: infection-related deaths rose from 7 cases in 2015–2017 to 16 cases in 2018–2020 and 22 cases in 2021–2023, while MODS-related deaths increased from 16 to 28 and 31 cases across the same periods. By 2021–2023, MODS had become the second most common cause of death, accounting for 21.5% of all deaths. Deaths due to cerebrovascular disease, malignancy, and other causes peaked in 2018–2020, representing 13.8% (19/137), 5.1% (7/137), and 16.0% (22/137) of deaths in that period, respectively. Other recorded causes included sudden death and traffic accidents (Fig. 3). Distribution of causes of death by treatment modality. Abbreviations: MODS, multiple organ dysfunction syndrome Distribution of causes of death among MHD patients by 3-year period, 2015–2023. Abbreviations: MODS, multiple organ dysfunction syndrome Distribution of primary renal disease and dialysis vintage among deceased MHD patients During the 9-year study period, 348 MHD patients died at this hemodialysis center. Among the deceased, DKD was the most common primary renal disease, followed by HN, CGN, and other causes, accounting for 116, 80, 80, and 72 cases, respectively. The distribution of dialysis vintage varied significantly across primary disease categories. Mortality peaked earliest in patients with DKD, with 41.4% of deaths occurring within the first three years of dialysis. In contrast, dialysis vintage among deceased patients with CGN was more uniformly distributed and tended to be longer overall. The proportion of patients with dialysis vintage exceeding 10 years was substantially higher in the CGN group (28.8%) than in the HN (13.8%) and DKD (4.3%) groups (Fig. 4). Distribution of dialysis vintage by primary renal disease among deceased MHD patients HP exposure and all-cause mortality in time-dependent cox and marginal structural model During follow-up, 54.4% of MHD patients died (Table 1). To account for the timing of HP initiation during follow-up, HP exposure was first modeled as a time-varying covariate in a multivariable time-dependent Cox proportional hazards model. In this model, HD + HP was associated with a lower risk of all-cause mortality compared with HD (HR, 0.69; 95% CI, 0.52–0.90; P = 0.009) (Fig. 5). Multivariable time-dependent Cox regression analysis for all-cause mortality, with HP exposure modeled as a time-varying covariate. The model included HP exposure, sex, age, diabetic kidney disease (DKD), hypertensive nephropathy (HN), chronic glomerulonephritis (CGN), arteriovenous fistula (AVF), hypertension (HTN), β2-microglobulin (β2-MG), parathyroid hormone (PTH), calcium (Ca), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), albumin (ALB), prealbumin (PAB), total protein (TP), red blood cell distribution width (RDW), platelet count (PLT), glucose (Glu), potassium (K), creatine kinase (CK), creatine kinase-MB isoenzyme (CK-MB), chloride (Cl), and lactate dehydrogenase (LDH) Several clinical covariates were also associated with all-cause mortality in the time-dependent Cox model. Age ≥ 60 years was associated with a higher risk of all-cause mortality compared with age < 60 years (HR, 2.22; 95% CI, 1.56–3.16; P < 0.001). DKD was associated with a markedly higher mortality risk compared with non-DKD primary renal disease (HR, 4.01; 95% CI, 2.90–5.54; P < 0.001). In contrast, AVF use was associated with a lower risk of all-cause mortality compared with non-AVF vascular access (HR, 0.65; 95% CI, 0.46–0.93; P = 0.02). HB ≥ 100 g/L was also associated with a lower mortality risk compared with HB < 100 g/L (HR, 0.68; 95% CI, 0.47–0.97; P = 0.03). Chloride (Cl) ≥ 107 mmol/L was associated with a higher mortality risk compared with Cl < 98 mmol/L (HR, 5.22; 95% CI, 1.18–23.01; P = 0.03). Because HP initiation may be influenced by patients’ evolving clinical conditions during follow-up, we further constructed an MSM using stabilized IPTW to address time-varying HP exposure and measured time-dependent confounding. In the IPTW-weighted MSM, HD + HP therapy remained associated with a lower risk of all-cause mortality compared with HD (HR, 0.73; 95% CI, 0.55–0.97; P = 0.030) (Fig. 6). Hazard ratio estimates for the association between HP exposure and all-cause mortality in the marginal structural model and time-dependent cox model. Abbreviations: HD, hemodialysis; HP, hemoperfusion; MSM, marginal structural model; IPTW, inverse probability of treatment weighting; HR, hazard ratio; CI, confidence interval Taken together, the time-dependent Cox model and the MSM consistently showed that HD + HP therapy was associated with a lower risk of all-cause mortality among patients receiving MHD after accounting for treatment timing and relevant clinical covariates. Baseline characteristics before and after propensity score matching Baseline characteristics before and after propensity score matching are presented in Table 2. Before matching, the HD and HD + HP groups showed substantial differences in several baseline characteristics, most notably age and vascular access. Patients in the HD group were older than those in the HD + HP group (median age, 70.0 vs. 64.0 years, P < 0.001; SMD = -0.688). In addition, the distribution of vascular access differed significantly between groups (P < 0.001; SMD = -0.222), with a higher proportion of AVF use in the HD + HP group. Table 2Baseline characteristics of patients by treatment modality before and after propensity score matching (1:1 matching; caliper = 0.02)Total sampleMatchedHD (N = 419)HD + HP (N = 221)P valueSMDHD (N = 173)HD + HP (N = 173)P valueSMDAge, years70.0[61.0,78.0]64.0[52.0,69.0]< 0.001−0.68866.0[58.0,71.0]65.0[58.0,70.0]0.33−0.08Sex, n (%)male276 (65.9)152 (68.8)0.55−0.029120 (69.4)116 (67.1)0.730.023female143 (34.1)69 (31.2)53 (30.6)57 (32.9)Primary Renal Disease, n (%)DKD134 (32.0)64 (29.0)0.18−0.0354 (31.2)53 (30.6)0.97−0.006HN82 (19.6)37 (16.7)−0.02828 (16.2)31 (17.9)0.017CGN111 (26.5)77 (34.8)0.08359 (34.1)56 (32.4)−0.017Other92 (22.0)43 (19.5)−0.02532 (18.5)33 (19.1)0.006Vascular Access, n (%)AVF269 (64.2)191 (86.4)< 0.001−0.222150 (86.7)143 (82.7)0.370.040TCC150 (35.8)30 (13.6)23 (13.3)30 (17.3)Comorbidities, n (%)diabetes mellitusY164 (39.1)80 (36.2)0.370.02962 (35.8)66 (38.2)0.74−0.023N255 (60.9)141 (63.8)111 (64.2)107 (61.8)hypertensionY350 (83.5)187 (84.6)0.57−0.011147 (85.0)147 (85.0)10N69 (16.5)34 (15.4)26 (15.0)26 (15.0)cardiovascular diseaseY83 (19.8)42 (19.0)0.990.00834 (19.7)37 (21.4)0.79−0.017N336 (80.2)179 (81.0)139 (80.3)136 (78.6)Abbreviations: DKD, diabetic kidney disease; HN, hypertensive nephropathy; CGN, chronic glomerulonephritis; Other, other causes; AVF, arteriovenous fistula; TCC, tunneled cuffed catheter; HD, hemodialysis; HD + HP, hemodialysis combined with hemoperfusion Baseline characteristics of patients by treatment modality before and after propensity score matching (1:1 matching; caliper = 0.02) Abbreviations: DKD, diabetic kidney disease; HN, hypertensive nephropathy; CGN, chronic glomerulonephritis; Other, other causes; AVF, arteriovenous fistula; TCC, tunneled cuffed catheter; HD, hemodialysis; HD + HP, hemodialysis combined with hemoperfusion After 1:1 propensity score matching based on sex, age, primary renal disease, vascular access, and comorbidities (diabetes mellitus, hypertension, and cardiovascular disease), 173 patients remained in each group. In the matched cohort, baseline characteristics were well balanced between the HD and HD + HP groups. The median age was 66.0 years in the HD group and 65.0 years in the HD + HP group (P = 0.33; SMD = -0.08). No significant between-group differences were observed in sex, primary renal disease, vascular access, or comorbidities (all P > 0.05), and all post-matching SMDs were small, indicating improved covariate balance and comparability between the two groups. In addition, post-matching comparisons of clinically relevant laboratory parameters showed no statistically significant differences between the HD and HD + HP groups, further supporting the comparability of the matched cohort (Supplementary Table 1). All-cause mortality after propensity score matching After propensity score matching, the overall distribution of causes of death was not significantly different between the two groups (Table 3). To account for the time-varying nature of treatment exposure, all-cause mortality was further evaluated using a Simon-Makuch curve. This analysis showed a significantly higher risk of all-cause mortality in the HD group than in the HD + HP group (Mantel-Byar test, P = 0.032) (Fig. 7). Table 3Distribution of causes of death by treatment modality after propensity score matchingCause of deathTotalHDHD + HPP value(n = 165)(n = 111)(n = 54)Cardiovascular disease60 (36.4)46 (41.4)14 (25.9)0.051Cerebrovascular disease26 (15.8)12 (10.8)14 (25.9)MODS32 (19.4)21 (18.9)11 (20.4)Infection15 (9.1)10 (9.0)5 (9.3)Cancer6 (3.6)6 (5.4)0 (0.0)Other26 (15.8)16 (14.4)10 (18.5)Abbreviations: Data are presented as n (%). MODS, multiple organ dysfunction syndrome Distribution of causes of death by treatment modality after propensity score matching Abbreviations: Data are presented as n (%). MODS, multiple organ dysfunction syndrome Simon–Makuch curve for all-cause mortality by treatment group. The shaded areas represent the 95% confidence intervals. The risk of all-cause mortality was significantly higher in the HD group (Mantel–Byar test, P = 0.032) Cardiovascular mortality after propensity score matching After propensity score matching, cardiovascular disease represented a major cause of death in the matched cohort (Table 3). To account for the time-varying nature of treatment exposure, cardiovascular mortality was further assessed using a Simon-Makuch curve. This analysis showed a significantly higher risk of cardiovascular mortality in the HD group than in the HD + HP group (Mantel-Byar test, P = 0.003) (Fig. 8). Simon–Makuch curve for cardiovascular mortality by treatment group. Shaded areas represent 95% confidence intervals. The risk of cardiovascular mortality was significantly higher in the HD group (Mantel–Byar test, P = 0.003) Discussion Chronic kidney disease (CKD) represents a major and growing global health challenge. In 2023, an estimated 788 million (95% CI 743–843 million) adults aged 20 years and older were living with CKD, a marked increase from 378 million (95% CI 354–407 million) in 1990. MHD remains the predominant renal replacement therapy for patients with ESRD worldwide, yet mortality in this population remains unacceptably high [11]. Consistent with previous reports [12, 13], our single-center retrospective study showed that cardiovascular disease was the leading cause of death in patients undergoing MHD, accounting for 32.8% of all-cause mortality. These findings underscore the need for effective strategies to improve outcomes in this high-risk population. From a clinical perspective, adjunctive HP may serve as a complementary blood purification strategy to address residual risk not fully controlled by conventional HD. In this real-world retrospective cohort, HP was initiated after the start of MHD in a subset of patients; therefore, the primary analysis used a time-dependent Cox model with HP as a time-varying covariate to account for treatment timing and reduce immortal time bias. In the fully adjusted model, HD + HP remained associated with lower all-cause mortality (HR = 0.69, 95% CI 0.52–0.90, P = 0.009). To further address measured time-dependent confounding, we additionally constructed a MSM using stabilized inverse probability of treatment weighting. The protective association of HD + HP persisted (HR = 0.73, 95% CI 0.55–0.97, P = 0.030), indicating robustness after adjustment for time-varying clinical factors. To further reduce confounding from baseline imbalances, we also performed 1:1 propensity score matching using key demographic and clinical characteristics, which resulted in well-balanced matched groups. In the matched cohort, Simon–Makuch curves and Mantel–Byar tests continued to show lower risks of both all-cause mortality (P = 0.032) and cardiovascular mortality (P = 0.003) in the HD + HP group. Given that cardiovascular disease was the leading cause of death in our cohort, this concordant signal is clinically meaningful. Our findings are broadly consistent with the recently published randomized, open-label, multicenter trial by Lu et al., conducted in 11 HD centers in Shanghai, China, which evaluated hemoadsorption combined with HD versus HD alone in patients with ESRD. In that trial, 1,362 patients receiving MHD were randomly assigned to hemoadsorption plus HD or HD alone. After a mean follow-up of 39.5 months, hemoadsorption plus HD significantly reduced all-cause mortality compared with HD alone (17.1% vs. 21.2%; HR, 0.778; 95% CI, 0.609–0.994; P = 0.045), with a significant reduction also observed in cardiovascular mortality (HR, 0.659; 95% CI, 0.481–0.901; P = 0.009). The consistency between that randomized trial and our real-world findings supports the potential survival benefit of adjunctive HP in patients receiving MHD [14]. Importantly, our study provides exploratory real-world evidence for a clinically relevant question that is not fully addressed by randomized treatment allocation. MHD is a chronic and long-term treatment process, and some patients may not receive or accept adjunctive HP immediately after initiating maintenance dialysis because of treatment burden, limited acceptance, cost, scheduling constraints, or individualized clinical considerations. In routine practice, HP may be introduced after a period of conventional HD rather than at the very beginning of MHD. Therefore, our study explored whether initiating adjunctive HP during ongoing MHD was still associated with potential survival benefit. By modeling HP exposure as a time-varying covariate and applying complementary analytical approaches, our study better reflects this real-world treatment scenario and extends the clinical relevance of existing trial evidence. The potential survival benefit of adjunctive HP may be biologically plausible. As discussed by Lu et al., adjunctive hemoadsorption may favorably influence several cardiovascular and metabolic risk-related factors, including blood pressure control, left ventricular mass index, erythropoietin requirement, Hb level, microinflammatory status, and advanced glycation end products [14]. These factors are closely related to atherosclerosis, heart failure, arrhythmia, angina, and other cardiovascular complications in patients receiving dialysis [14]. Therefore, the lower cardiovascular mortality observed in our cohort may be partly explained by enhanced toxin removal, attenuation of inflammation, and improved control of anemia-related and cardiovascular stress. Nevertheless, because these intermediate indicators were not comprehensively and longitudinally assessed in the present study, this explanation should be interpreted as mechanistic plausibility rather than direct causal evidence. Among these potential mechanisms, enhanced removal of protein-bound and medium-to-large uremic toxins may be particularly relevant. Conventional HD has limited efficacy in clearing protein-bound solutes such as indoxyl sulfate (IS) and p-cresyl sulfate (PCS) [15], both of which have been linked to all-cause and cardiovascular mortality [6, 7, 16, 17]. Mechanistically, IS and PCS can activate nicotinamide adenine dinucleotide phosphate oxidase, increase reactive oxygen species generation, and induce oxidative stress in vascular endothelial cells, thereby promoting endothelial injury, vascular calcification, coronary artery disease, and other cardiovascular events [18–24]. Consistent with these experimental findings, clinical studies have shown a positive correlation between serum IS levels and carotid intima-media thickness in MHD patients [25], whereas PCS has been associated with increased arterial stiffness [16]. In addition, IS and PCS may activate the nuclear factor-kappa B pathway, stimulate the production of pro-inflammatory cytokines, and further amplify systemic inflammation and oxidative stress [26, 27]. This inflammatory and oxidative milieu may aggravate endothelial dysfunction and vascular injury, thereby contributing to cardiovascular disease progression and mortality [16, 25–30]. Although specific toxin levels were not measured in the present study, the lower cardiovascular mortality observed in the HD + HP group is mechanistically consistent with the superior adsorptive capacity of HP for these pathogenic molecules [31–33]. A second plausible mechanism involves attenuation of the chronic microinflammatory state characteristic of ESRD and MHD. Persistent low-grade inflammation is closely associated with atherosclerosis, protein-energy wasting, and poor outcomes in patients receiving MHD [16, 34]. HP cartridges can adsorb a broad range of inflammatory mediators, including cytokines such as interleukin-6 and tumor necrosis factor-α [35–37]. Given the central role of inflammation in cardiovascular injury and malnutrition-inflammation-atherosclerosis syndrome, modulation of this inflammatory milieu may partly contribute to the lower cardiovascular mortality observed in the HD + HP group. A third, potentially related mechanism involves anemia, a major and multifactorial determinant of adverse outcomes in patients with ESRD. In the present study, higher Hb levels were associated with better survival. In the time-dependent Cox model, Hb ≥ 100 g/L was associated with a lower risk of all-cause mortality than Hb < 100 g/L (HR, 0.68, 95% CI, 0.47–0.97, P = 0.03). Beyond reduced iron availability and erythropoietin deficiency, chronic inflammation and retained uremic solutes may increase erythropoietin resistance and impair erythropoiesis [38–40]. Mechanistically, IS may directly suppress erythropoiesis by inhibiting erythroid progenitor cell function [38]. In parallel, elevated hepcidin-25, a key regulator of iron metabolism, can restrict iron availability and contribute to inflammation-related anemia; notably, HP has been reported to adsorb hepcidin-25 [16, 41]. Based on these findings, we speculate that adjunctive HP may facilitate anemia control by reducing inflammatory burden, decreasing retained uremic solutes, and mitigating inhibitors of erythropoiesis. Consistent with this hypothesis, clinical studies have reported improvements in Hb levels after HD + HP treatment [32, 42]. Several limitations should be acknowledged. First, although adjunctive HP was implemented in routine clinical practice according to standardized blood purification procedures and individualized clinical assessment, the treatment modality was not assigned within a randomized study design. Therefore, residual confounding, including unmeasured confounding, cannot be completely excluded. Although we used time-dependent Cox regression to account for treatment timing, a MSM to address measured time-dependent confounding, and propensity score matching with Simon–Makuch/Mantel–Byar analyses to improve comparability, these methods cannot fully substitute for randomization. Therefore, the observed association between HD + HP therapy and lower mortality should be interpreted cautiously and should not be regarded as definitive evidence of causality. Second, mechanistic interpretation was limited by the lack of serial measurements of protein-bound uremic toxins and incomplete longitudinal data on inflammation- and anemia-related biomarkers. Third, several potentially relevant variables, including smoking status, dyslipidemia, body mass index, dialysis adequacy indicators such as Kt/V or urea reduction ratio, medical costs, and standardized safety outcomes, were not routinely recorded. Future multicenter prospective studies incorporating standardized criteria for HP initiation, serial biomarker monitoring, systematic adverse event surveillance, and formal cost-effectiveness analyses are warranted to validate and extend these findings. Conclusions This single-center retrospective cohort study showed that cardiovascular disease remained the leading cause of death among patients receiving MHD, accounting for 32.8% of all-cause mortality. HD + HP therapy was associated with a significantly lower risk of all-cause mortality in the time-dependent Cox model accounting for treatment timing. This association remained significant in the MSM after adjustment for relevant clinical covariates and was further supported by propensity score-matched Simon–Makuch and Mantel–Byar analyses, which showed lower risks of both all-cause and cardiovascular mortality in the HD + HP group. Nevertheless, because this was a retrospective observational study, residual confounding and unmeasured bias cannot be excluded, and causal inference remains limited. Future multicenter prospective studies are warranted to confirm these survival associations and to further clarify the optimal indications, mechanisms, safety, and cost-effectiveness of HD + HP therapy. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements Not required. Abbreviations Chronic kidney disease End-stage renal disease Maintenance hemodialysis Hemodialysis Hemoperfusion Hemodialysis combined with hemoperfusion Arteriovenous fistula Tunneled cuffed catheter Diabetic kidney disease Chronic glomerulonephritis Hypertensive nephropathy Multiple organ dysfunction syndrome International Classification of Diseases, 10th Revision Albumin Hemoglobin Red blood cell distribution width Parathyroid hormone Prealbumin Creatine kinase Creatine kinase-MB Lactate dehydrogenase Triglycerides Total cholesterol High-density lipoprotein cholesterol Total protein Platelet count Glucose Potassium Chloride Calcium Serum creatinine Estimated glomerular filtration rate β2-microglobulin Standardized mean difference Marginal structural model Inverse probability of treatment weighting Hazard ratio Confidence interval Interquartile range Standard deviation Indoxyl sulfate P-cresyl sulfate Author contributions Y.L. and W.G. contributed equally to this work as co-first authors. R.D. and Z.G. are co-corresponding authors. Conceptualization: Y.L., R.D., Z.G.; Methodology: Y.L., W.G., R.D., W.C.; Investigation: Y.L., W.G.; Formal Analysis: Y.L., W.G.; Writing – Original Draft: Y.L.; Writing – Review & Editing: All authors; Supervision: R.D., Z.G.; Project Administration: Z.G.; Resources: Z.G.; Funding Acquisition: Z.G. All authors have read and agreed to the final version of the manuscript. Funding This work was supported by the Shanghai Xinxin Medical Technology Development Foundation (grant number: KY-HXZ-20240511-001). Data availability The data used in the study are available upon reasonable request to the corresponding author. Declarations Ethics approval and consent to participate This clinical study was approved by the Shanghai Changhai Hospital Ethics Committee (Ethics Approval No. CHEC2022-207, October 28, 2022). Ethics committee approval was obtained prior to data extraction and analysis; given the study’s retrospective, non-interventional nature, the requirement for written informed consent was waived. All procedures and data handling were compliant with the ethical principles of the Declaration of Helsinki and institutional data protection policies. Patient information was anonymized prior to extraction, and only routinely recorded clinical data were analyzed. Consent for publication Not required. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Yan Liu and Wei Gang contributed equally to this work. Contributor Information Rui Dong, Email: drdongr@163.com. Zhiyong Guo, Email: drguozhiyong@163.com. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data used in the study are available upon reasonable request to the corresponding author.