Comparative Effectiveness and Safety of Roxadustat versus Erythropoiesis-Stimulating Agents in Patients Receiving Maintenance Hemodialysis: A Real-World Cohort Study

Abstract

Introduction

Renal anemia, a common complication of chronic kidney disease (CKD), is traditionally managed with erythropoiesis-stimulating agents (ESAs), which carry cardiovascular risks. Roxadustat, an oral hypoxia-inducible factor prolyl hydroxylase inhibitor, offers a novel mechanism by enhancing erythropoietin production and iron metabolism. While randomized controlled trials demonstrate its efficacy, real-world data on long-term safety and effectiveness remain limited.

Methods

This retrospective study analyzed 6,414 hemodialysis patients with renal anemia from a single center (December 2018–December 2023), comparing roxadustat (n = 3,184) and ESA (n = 3,230) groups. Propensity score matching was used to balance the baseline characteristics. Efficacy outcomes included hemoglobin (Hb) changes (baseline to months 6–12 and 18–30) and Hb response rates. Safety endpoints assessed major adverse cardiovascular events (MACE), heart failure hospitalization (HHF), thromboembolism, and all-cause mortality. Sensitivity analyses addressed treatment crossover and confounding.

Results

Roxadustat showed significantly greater Hb increases versus ESA at 6–12 months (least-squares mean difference: 0.46 g/dL, p = 0.04) and 18–30 months (0.26 g/dL, p = 0.01). Hb response rates were higher with roxadustat (84.0% vs. 76%, p < 0.01). No significant differences were observed in MACE (HR: 1.08, p = 0.12), HHF (HR: 0.88, p = 0.25), thromboembolism (HR: 1.05, p = 0.34), or mortality (HR: 0.94, p = 0.29). Subgroup analyses suggested that roxadustat elevated MACE risk in patients with baseline hypertension or cardiovascular history, but sensitivity analyses nullified this association.

Conclusion

Roxadustat demonstrated superior efficacy in elevating and sustaining Hb levels compared to ESA over a longer observation period, with comparable cardiovascular safety in hemodialysis-dependent CKD patients. Roxadustat represents a viable alternative to ESA for renal anemia management, though long-term multicenter studies are needed to validate safety and optimize clinical use.

Introduction

Renal anemia is a prevalent and severe complication of chronic kidney disease (CKD), contributing to increased cardiovascular morbidity, mortality, and diminished quality of life [1–3]. It results from dysregulation of erythropoietin (EPO) production due to impaired oxygen-sensing mechanisms in the kidneys, compounded by chronic inflammation, iron-restricted erythropoiesis driven by elevated hepcidin levels, and shortened red blood cell (RBC) survival [4–6]. Traditional treatments for renal anemia include erythropoiesis-stimulating agents (ESAs), iron supplementation, and red blood cell (RBC) transfusion [7]. While RBC transfusions provide rapid hemoglobin (Hb) correction, they carry risks of iron overload, transfusion reactions, and infections [8, 9]. Similarly, intravenous iron therapy, though effective in replenishing iron stores, has been linked to oxidative stress, cardiovascular toxicity, and impaired immune function. Additionally, iron therapy may compromise phagocytic function, leading to a greater susceptibility to infections [10, 11]. ESAs have emerged as an ideal option, offering sustained Hb elevation without transfusion-related risks and significantly improving patient-reported outcomes [12]. However, there are some limitations of ESA, including dose-dependent hypertension, thrombotic events, and hyporesponsiveness – often exacerbated by iron deficiency or inflammation – remain persistent concerns [13–15], highlighting the need for safer and more effective alternatives.

The advent of hypoxia-inducible factor prolyl hydroxylase inhibitors (HIF-PHIs) represents a paradigm shift in anemia management. Roxadustat, a first-in-class oral HIF-PHI, stabilizes hypoxia-inducible factor (HIF) to mimic physiological hypoxia, stimulating endogenous EPO production, enhancing iron absorption and mobilization via hepcidin suppression, and improving iron utilization efficiency [16]. Several clinical trials have demonstrated non-inferiority of roxadustat to ESAs in Hb correction among both non-dialysis and dialysis-dependent CKD patients [17–19], with additional benefits such as reduced transfusion requirements and improved lipid profiles [20]. However, randomized clinical trials (RCTs) inherently exclude clinically complex populations – such as elderly patients, those with advanced comorbidities, or individuals with labile Hb levels – limiting generalizability to real-world practice [19, 21]. Long-term safety data, particularly regarding cardiovascular outcomes and iron homeostasis, remain limited.

Real-world data (RWD) provide an opportunity to evaluate the effectiveness and safety of roxadustat by capturing heterogeneous patient populations, routine clinical practices, and longer-term outcomes. This study aimed to evaluate the comparative effectiveness and safety of roxadustat versus ESAs, with a focus on Hb trajectory, response rates, and safety outcomes. By bridging the gap between RCTs and real-world practice, the findings of this study could offer valuable insights into the role of roxadustat in anemia management and inform clinical decision-making in CKD care.

Methods

Study Design and Population

This retrospective study was conducted using electronic medical data (EMR) in the renal division of West China Hospital, Sichuan University (WCH-SCU), southwestern China. The renal division of WCH-SCH consists of 177 hemodialysis machines from 2 hospital-based hemodialysis centers that undertake treatment and follow-up of maintenance hemodialysis (MHD) patients, and the total number of patients reached 9,374 from December 2018 to December 2023. EMR were collected by integrating multisource data, including demographic characteristics, medical history, examinations, laboratory measurements, and prescription records.

We enrolled patients who had undergone regular MHD for at least 3 months, were diagnosed with renal anemia, and were treated with ESAs or roxadustat from December 2018 to December 2023 at WCH-SCU (Chictr.org.cn, No. ChiCTR2000034325). This study was approved by the Ethics Committee of West China Hospital of Sichuan University (HX-IRB-AF-2020-03) and adhered to the Declaration of Helsinki. Patients had to meet the following inclusion criteria: (1) patients aged at least 18 years with anemia of CKD; (2) must have been receiving HD at least 3 months and three times weekly at screening; (3) the mean of the two most recent Hb values ≤10.0 g/dL before the first prescription of roxadustat or ESA [22, 23]; (4) were iron replete as determined by transferrin saturation ≥20% or serum ferritin ≥100 ng/mL at screening [7, 20]; (5) did not receive ESAs or roxadustat within 4 weeks before enrollment [24]. The following exclusion criteria were applied: (1) anemia unrelated to kidney disease; (2) had shown severe compliance with roxadustat treatment; (3) had undergone kidney transplantation or was on the waiting list for transplantation; (4) combined with active underlying infection, severe heart failure (NYHA grade IV), severe liver malfunction, malignant tumor, malignant hypertension, autoimmune disease, blood system disease, or other active bleeding disorders at baseline; (5) with missing data at baseline.

We followed patients from the index date until the earliest occurrence of the following events, including the end of the observation period (December 31, 2023), loss to follow-up, death, or occurrence of the study outcomes. We considered patients to be exposed to the index drug until 28 days after the end of the days’ supply of the last prescription, which was consistent with the RCT studies [25, 26].

Treatments and Outcomes

Roxadustat and ESA treatment groups were derived from the entire study sample according to anemia medication they were taking at baseline. The date of an individual’s first prescription with roxadustat or ESA was defined as the index date. Roxadustat was orally administered three times per week with the initial dose of 70 mg or 100 mg based on the patient’s weight (weight ≥45.0–70.0 kg, roxadustat 70 mg; weight >70.0–160.0 kg, roxadustat 100 mg) [7]. The initial treatment goal was an increased Hb level of 1–2 g/dL per month, and subsequent adjustments would be made according to the patient’s Hb level, speed of Hb change, and treatment response. [24, 27]. ESAs were subcutaneously administered twice or three times per week by nurses during the in-center HD process, and the dose was adjusted based on package insert guidelines. In general, the target Hb level is 10–12 g/dL.

The efficacy outcomes included the mean change in Hb level and Hb response rate. The mean change in Hb level was defined as the absolute difference from baseline to average level during months 6–12 and during months 18–30 (if maximum follow-up reaches ≥18th month) [25, 28]. The Hb response rate was defined as the proportion of patients with Hb ≥11.0 g/dL, or an increase from baseline of ≥1.0 g/dL (baseline Hb value <8.0 g/dL), or an Hb increase from baseline of ≥2.0 g/dL (baseline Hb value <8.0 g/dL) in the first 24 weeks [29, 30]. The safety outcomes were the incidence of major adverse cardiovascular events (MACE), thromboembolic events (including arterial thrombosis, deep venous thrombosis, pulmonary embolism, and vascular access thrombosis), hospitalization for heart failure (HHF), and all-cause death. MACE was defined as a composite of death from any cause, nonfatal myocardial infarction, or nonfatal stroke.

Covariates

Covariates were selected based on clinical knowledge and biological plausibility, including patient demographic characteristics (age and sex), medical history, body mass index (BMI), blood pressure at baseline, estimated glomerular filtration rate (eGFR), C-reactive protein (CRP), blood lipid, and iron parameters, including iron, ferritin, transferrin saturation (TSAT), and total iron-binding capacity (TIBC). Diabetes mellitus was defined as a fasting plasma glucose of 126 mg/dL or higher, a non-fasting plasma glucose level of 200 mg/dL or higher, or self-reported use of glucose-lowering medication [31]. Cardiovascular disease (CVD) was defined by any history of coronary artery disease, revascularization, heart failure, stroke, or peripheral artery disease [22]. eGFR was calculated based on serum creatinine concentration using the CKD Epidemiology Collaboration (CKD-EPI) equation [32].

Statistical Analyses

We estimated the propensity score (PS) to control confounding when matching roxadustat and ESA treatment groups on a ratio of 1:1 [33]. PS was calculated by the probability of initiating roxadustat vs. each comparator, using multivariable logistic regression analysis, conditional on the following covariates: age; sex; BMI; baseline blood pressure, baseline Hb, eGFR; creatinine; urea; CRP; albumin; iron metabolism parameters; and history of hypertension, cardiovascular diseases, and thromboembolic events. The matching process was processed based on the nearest-neighbor algorithm without replacement, with a specified maximum caliper of 0.1 on the PS scale [34]. Additionally, we estimated and matched PS separately within each prespecified subgroup analysis. Standardized mean differences (SMD) were used to assess covariate balance after PS matching. SMD ≤0.1 was considered acceptable, or covariates were adjusted for in the regression models [35]. As patients with missing data were excluded during cohort selection, we used complete case analysis for all the following statistical analyses.

We summarized continuous variables as means (standard deviations [SDs]) or medians (interquartile ranges [IQRs]), and compared them with parametric or nonparametric tests as appropriate. Categorical variables were summarized with counts (percentages) and compared using chi-square tests.

We used a linear mixed multivariable model of repeated Hb measurements to compare the mean change of Hb between the roxadustat and ESA groups. The model was adjusted for comorbidities (including hypertension, cardiovascular disease, and thromboembolic events), baseline Hb, in-person study visits, and treatment-by-visit interaction. Hb response rate analysis was performed using the Miettinen-Nurminen approach [36], with a two-sided 95% CI adjusted for stratification factors. The risk of safety outcomes over the follow-up period was described by cumulative incidence function plots using the Kaplan-Meier method and further calculated using time-to-event analyses. Treatment effects were estimated using hazard ratios (HRs) from Cox proportional hazards regression models [37]. The proportional hazard assumption was tested on the basis of Schoenfeld residuals [38].

To mitigate potential informative censoring, we additionally performed intention-to-treat (ITT) analyses without censoring for treatment discontinuation or switching and followed patients until the date of event, death, or the last available observation, whichever occurred first [39]. Subgroup analyses were performed based on relevant clinical conditions and parameters: (1) age (≤65 vs. >65 years on the cohort entry date), (2) sex (male vs. female), (3) BMI (≤25 vs. >25 to ≤30 vs. >30 kg/m²), (4) baseline history of CVD vs. no baseline history of CVD, (5) presence vs. absence of baseline hypertension, (6) presence vs. absence of baseline diabetes, (7) baseline Hb (≤10.5 vs. >10.5 g/dL), and (8) CRP (≤3 vs. >3 mg/L). The heterogeneity of treatment effect across subgroups was detected using the Wald test for homogeneity [40].

As sensitivity analyses, we conducted a per-protocol set (PPS) analysis by censoring patients at the point at which they switched. Also, we repeated our analyses by excluding patients who crossed over between the two treatments. Moreover, we repeated our primary ITT analyses using inverse probability of treatment weighting, and average treatment effect weights, to compare the consistency of our findings with that from PS matching. Lastly, in order to further adjust for potential confounding factors, we developed a new propensity score matching model that incorporated additional covariates, including dialysis duration, vascular access type, and cardiovascular medication use. p < 0.05 was considered statistically significant for all 2-sided tests.

Results

Patient Characteristics

We finally included 6,414 MHD patients, and details of the patient selection process were shown in Figure 1. Baseline characteristics of the 6,414 patients across roxadustat and ESA treatment groups are given in Table 1, stratified by before and after PS matching. In the overall patient population, the mean (SD) age was 50.7 (16.5) years, and 42.6% were female. 3,230 patients received ESAs, and 3,184 received roxadustat. In total, 2,463 (38.4%) patients had a history of CVD, and 655 (10.2%) patients had a history of thromboembolic events. After PS matching, both groups were balanced across all covariates, with 2,443 patients in each group. In the roxadustat vs. ESA group retrospectively, the mean (SD) age was 50.5 (16.1) vs. 50.6 (16.0) years, and the proportion of female patients was 41.7% vs. 41.1%. Mean baseline Hb levels were 8.5 and 8.6 g/dL in the roxadustat and ESA groups, respectively. Of note, oral iron supplements were taken during treatment by 1,586 of 2,443 (64.9%) patients in the roxadustat group and 1,602 of 2,443 (65.6%) patients in the ESA group. The proportions of patients who had red blood cell transfusion for rescue therapy were 4.9% and 5.1% in the roxadustat and ESA groups, respectively.

Fig. 1.

Flowchart of the selection process.

Table 1.

Baseline characteristics of study participants before and after matching

Characteristic	Before PSM			After PSMa
	ESA (N = 3230b)	roxadustat (N = 3184b)	SMDc	ESA (N = 2443b)	roxadustat (N = 2443b)	SMDc
Age, years	53.4 (16.5)	47.9 (16.0)	0.34	50.6 (16.0)	50.5 (16.1)	0.01
Female, N (%)	1,402 (43.4)	1,332 (41.8)	0.03	1,005 (41.1)	1,018 (41.7)	0.01
Weight, kg	59.6 (13.1)	59.5 (12.9)	0.01	60.2 (13.2)	59.8 (12.9)	0.03
BMI, kg/cm2	22.3 (4.0)	21.9 (3.9)	0.09	22.3 (4.0)	22.1 (3.9)	0.05
Baseline blood pressure, mm Hg
Systolic	142.5 (23.1)	139.4 (22.4)	0.14	141.3 (22.6)	141.2 (22.2)	0.00
Diastolic	87.2 (15.6)	87.3 (15.5)	0.00	87.4 (15.4)	87.2 (15.3)	0.02
Duration of HD, months	4.8 (3.5, 8.3)	4.6 (3.3, 7.8)	0.06	4.6 (3.3, 7.9)	4.5 (3.3, 7.8)	0.01
Vascular access type, N (%)
Central venous catheter	585 (18.1)	562 (17.7)	0.01	452 (18.5)	437 (17.9)	0.02
Arteriovenous fistula	2,138 (66.2)	2,054 (64.5)	0.04	1,571 (64.3)	1,554 (63.6)	0.05
Arteriovenous graft	148 (4.6)	137 (4.3)	0.02	103 (4.2)	105 (4.3)	−0.01
Unknown	359 (11.1)	431 (13.5)	−0.07	317 (13.0)	347 (14.2)	−0.03
Most likely cause of CKDd, N (%)
Chronic GN	265 (8.2)	243 (7.6)	0.02	216 (8.8)	180 (7.4)	0.05
Chronic interstitial nephritis	24 (0.7)	18 (0.6)	0.02	17 (0.7)	14 (0.6)	0.02
Cystic kidney disease	38 (1.2)	18 (0.6)	0.07	28 (1.1)	16 (0.7)	0.05
Diabetic nephropathy	548 (17)	418 (13)	0.11	401 (16)	358 (15)	0.05
FSGS	203 (6.3)	180 (5.7)	0.03	156 (6.4)	140 (5.7)	0.03
Ischemic/hypertensive nephropathy	27 (0.8)	23 (0.7)	0.01	16 (0.7)	20 (0.8)	0.02
IgA nephropathy	132 (4.1)	192 (6.0)	0.09	120 (4.9)	137 (5.6)	0.03
Lupus nephritis	119 (3.7)	83 (2.6)	0.06	73 (3.0)	64 (2.6)	0.09
Minimal change	6 (0.2)	5 (0.2)	0.01	5 (0.2)	2 (<0.1)	0.03
Membranous nephropathy	29 (0.9)	21 (0.7)	0.01	25 (1.0)	18 (0.7)	0.02
Chronic pyelonephritis	16 (0.5)	28 (0.9)	0.05	12 (0.5)	19 (0.8)	0.04
Unknown	1,823 (56.4)	1,955 (61.4)	0.10	1,346 (55.1)	1,386 (56.7)	0.08
Hypertension, N (%)	2,618 (81.1)	2,869 (90.1)	0.26	2,016 (82.5)	2,097 (85.8)	0.06
Diabetes, N (%)	828 (25.6)	734 (23.1)	0.06	607 (24.8)	623 (25.5)	0.02
History of cardiovascular diseasese, N (%)
CHD	675 (20.9)	601 (18.9)	0.05	482 (19.7)	506 (20.7)	0.02
Heart failure	863 (26.7)	820 (25.8)	0.02	641 (26.2)	679 (27.8)	0.04
Angina pectoris	65 (2.0)	65 (2.0)	0.00	43 (1.8)	54 (2.2)	0.03
Stroke	208 (6.4)	161 (5.1)	0.06	147 (6.0)	140 (5.7)	0.01
Myocardial infarction	150 (4.6)	195 (6.1)	0.07	120 (4.9)	153 (6.3)	0.06
Atrial fibrillation	71 (2.2)	73 (2.3)	0.01	45 (1.8)	63 (2.6)	0.05
TIA	20 (0.6)	14 (0.4)	0.02	13 (0.5)	13 (0.5)	0.00
Peripheral vascular diseasef, N (%)	408 (12.6)	339 (10.6)	0.06	289 (11.8)	275 (11.3)	0.02
Revascularization, N (%)	57 (1.8)	55 (1.7)	0.00	42 (1.7)	47 (1.9)	0.02
Thromboembolic eventg, N (%)	378 (11.7)	277 (8.7)	0.10	246 (10.1)	225 (9.2)	0.03
Hb, g/dL
Median(IQR)	8.7 (7.5, 10.0)	8.4 (7.1, 9.8)	0.07	8.5 (7.2, 10.0)	8.5 (7.1, 9.8)	−0.01
Mean(SD)	8.7 (2.0)	8.6 (2.2)	0.07	8.6 (1.9)	8.5 (2.1)	0.02
RBC, 1012/L	2.9 (0.7)	2.8 (0.8)	0.01	2.8 (0.7)	2.8 (0.8)	−0.03
eGFR, mL/min/1.73 m2	6.3 (4.5, 10.3)	8.3 (5.2, 11.5)	−0.44	7.0 (4.8, 10.2)	7.2 (4.8, 10.6)	−0.06
Scr, μmol/L	697.6 (369.4)	574.8 (372.9)	0.33	646.7 (344.4)	642.9 (365.6)	0.01
Urea, mmol/L	20.8 (10.5)	19.2 (10.8)	0.15	20.4 (10.6)	20.4 (10.6)	0.00
CRP, mg/L	5.4 (2.1, 15.7)	6.4 (2.3, 20.3)	−0.04	5.4 (1.9, 15.3)	6.5 (2.3, 21.6)	−0.08
Albumin, g/dL	3.4 (0.5)	3.2 (0.7)	0.07	3.3 (0.6)	3.2 (0.5)	0.05
Iron, µmol/L	11.4 (6.7)	12.0 (7.4)	−0.08	11.5 (6.7)	11.9 (7.3)	−0.06
Ferritin, ng/mL	410.6 (445.0)	423.5 (453.0)	−0.03	398.0 (455.6)	430.5 (470.4)	−0.07
TSAT, %	29.5 (17.8)	29.9 (18.8)	−0.03	29.3 (17.9)	29.9 (18.6)	−0.03
TIBC, µmol/L	40.6 (10.4)	42.0 (11.9)	−0.12	41.4 (10.4)	41.4 (11.5)	0.00
Total cholesterol, mmol/L	3.8 (1.3)	3.8 (1.3)	0.01	3.8 (1.3)	3.8 (1.3)	0.02
Triglyceride, mmol/L	1.7 (1.3)	1.7 (1.2)	−0.02	1.7 (1.2)	1.7 (1.2)	−0.03
LDL, mmol/L	2.0 (0.9)	2.0 (1.0)	0.01	2.0 (1.0)	2.0 (1.0)	0.02
HDL, mmol/L	1.1 (0.4)	1.1 (0.4)	−0.01	1.1 (0.4)	1.1 (0.4)	0.02
Medicatione, N (%)
ACE-I/ARBS	1,269 (39.3)	1,372 (43.1)	−0.15	965 (39.5)	1,050 (43.0)	−0.14
CCB	1,247 (38.6)	1,137 (35.7)	0.07	911 (37.3)	867 (35.5)	0.02
Beta-blockers	1,543 (47.8)	1,420 (44.6)	0.04	1,106 (45.3)	1,097 (44.9)	0.01
Diuretics	180 (5.6)	251 (7.9)	−0.11	132 (5.4)	198 (8.1)	−0.13
Nitrates	136 (4.2)	207 (6.5)	−0.12	115 (4.7)	154 (6.3)	−0.10
Oral anticoagulants	271(8.4)	280 (8.8)	−0.01	208 (8.5)	225 (9.2)	−0.03
Antiplatelets	391 (12.1)	328 (10.3)	0.06	286 (11.7)	266 (10.9)	0.02
Lipid-lowering drugs	665 (20.6)	595 (18.7)	0.05	474 (19.4)	452 (18.5)	0.03

BMI, body mass index; CKD, chronic kidney disease; GN, glomerulonephritis; FSGS, focal segmental glomerulosclerosis; CHD, coronary heart disease; TIA, transient ischemic attack; RBC, red blood cell; eGFR, estimated glomerular filtration rate; Scr, serum creatinine; CRP, C-reactive protein; TSAT, transferrin saturation; TIBC, total iron-binding capacity; LDL, low-density lipoprotein; HDL, high-density lipoprotein; PSM, propensity score matching; ESA, erythropoiesis-stimulating agent; SMD, standardized mean difference; IQR, interquartile range; SD, standard deviation.

^aPSM was performed on age at index, sex, BMI, baseline blood pressure, baseline Hb, CRP, iron metabolism, eGFR, history of cardiovascular diseases, and history of hypertension.

^bEntries are mean (SD), number (percentage), or median (IQR), depending on the data type and distribution characteristics.

^cStandardized mean difference <0.1 indicates good balance between 2 groups.

^dInferred by the patients’ past medical records or pathological diagnosis.

^ePatients could have been included in more than one category.

^fPeripheral vascular diseases include arteriosclerosis obliterans, thromboangiitis obliterans, varicose veins, Raynaud’s disease, acute limb ischemia, lymphedema, and aneurysms.

^gThromboembolic events include pulmonary embolism, deep-vein thrombosis, retinal-vein occlusion, arteriovenous graft thrombosis, arteriovenous fistula thrombosis, and central venous catheter thrombosis.

Mean Hb Change and Hb Response Rate

Over the time from the 6th to 12th month, the adjusted least-squares mean (LSM) change from baseline Hb was significantly 0.46 g/L greater in the roxadustat compared with the ESA group (3.17 vs. 2.71, p = 0.04). Similar results were found for the time window from the 18th to 30th month, with Hb moderately higher in the roxadustat group compared with the ESA group (3.34 vs. 3.08, p = 0.01). Mean Hb levels in both groups increased during the first 24 weeks of treatment and remained relatively stable between 11.5 and 12.5 g/dL for the rest of the evaluation period (Fig. 2). The roxadustat group achieved a significantly higher Hb response rate (84.0%) compared to the ESA group (76%) during the first 24 weeks, with a higher Hb response rate of 8% (p < 0.01). Both groups achieved Hb response rates above 50% by week 8 and ultimately stabilized around 80% (online suppl. Fig. S1; for all online suppl. material, see https://doi.org/10.1159/000548711).

Fig. 2.

Hb levels by visit in the propensity score matching cohort.

Safety Outcomes

For MACE and thromboembolic events, the estimate of effect indicated potential risks in the roxadustat group compared to the ESA group but without significance (MACE: HR 1.08, p = 0.12; thromboembolic events: HR 1.05, p = 0.34). For HHF and all-cause death, the estimate of effect indicated potential protection from roxadustat but still without significance (HHF: HR 0.88, p = 0.25; all-cause death: HR 0.94, p = 0.29) (Table 2; Fig. 3). We further explored the risk of safe outcomes in different subgroups. There were no significant interaction effects across subgroups for age, sex, BMI, baseline Hb, or CRP (Fig. 4). However, for the subgroup with hypertension at baseline, roxadustat was associated with a higher risk of MACE compared to ESA (HR: 1.10, 95% CI: [1.01, 1.20]). Similar results were observed in the subgroup with CVD history (HR: 1.09, 95% CI: [1.02, 1.15]). There were no significant differences between groups across subgroups for other safety outcomes (online suppl. Fig. S2–S4).

Table 2.

Prespecified efficacy and safety end points in ITT analyses

Outcome	Roxadustat				ESA				Difference in LSM changes (95% CI)/HR (95% CI)c	p value
	patients, Na	baseline value	final value	LSM change/adjusted LSM/event rateb	patients, Na	baseline value	final value	LSM change/adjusted LSM/event rateb
Patients after PSM regardless of treatment discontinuation (ITT)
Change in Hb from BL to mean during months 6–12, g/dL	2,113	8.52	11.92	LSM change: 3.17	2,101	8.55	11.68	LSM change: 2.71	Difference in LSM changes: 0.46 (0.16–0.75)	0.04
Change in Hb from BL to mean during months 18–30, g/dL	1,825	8.61	12.17	LSM change: 3.34	1,811	8.64	12.06	LSM change: 3.08	Difference in LSM changes: 0.26 (0.07–0.46)	0.01
Proportion with Hb response during the first 24 weeks	2,443	NA	NA	Adjusted LSM: 0.84	2,443	NA	NA	Adjusted LSM: 0.76	Difference in LSM: 0.08 (0.03–0.13)	<0.01
Event rate for MACE	2,443	NA	NA	Event rate: 21.37	2,443	NA	NA	Event rate: 19.79	HR: 1.08 (0.97–1.18)	0.12
Event rate for HHF	2,443	NA	NA	Event rate: 15.35	2,443	NA	NA	Event rate: 17.33	HR: 0.88 (0.69–1.07)	0.25
Event rate for all-cause death	2,443	NA	NA	Event rate: 8.03	2,443	NA	NA	Event rate: 8.63	HR: 0.94 (0.84–1.05)	0.29
Event rate for thromboembolism	2,443	NA	NA	Event rate: 12.42	2,443	NA	NA	Event rate: 11.81	HR: 1.05 (0.94–1.15)	0.34

ITT, intention-to-treat; ESA, erythropoiesis-stimulating agent; LSM, least-squares mean; HR, hazard ratio; CI, confidence interval; Hb, hemoglobin; BL, baseline; MACE, major adverse cardiovascular events; HHF, heart failure hospitalization; NA, not applicable.

^aNumbers are the same for baseline and final value unless otherwise specified.

^bRate of prespecified event per 100 patient-years.

^cHR (95% CI) comparing risk of prespecified event for roxadustat vs. ESA.

Fig. 3.

a–d Cumulative incidence of safety outcomes.

Fig. 4.

Subgroup analyses of MACE outcome in propensity score-matched patients.

We observed considerable crossover from ESA to roxadustat, with a median duration of ESA exposure of 53 (34–72) days. There were 171 of 2,443 patients (7.0%) who crossed over within 30 days of initiating ESA. When excluding patients who crossed over between the 2 treatments (934 who initiated ESA and 390 who initiated roxadustat) or censoring those at the time of medication switch (PPS dataset), there were still no statistically significant associations between roxadustat and the safety outcomes. Based on the elevated risk of MACE observed in the hypertensive and CVD subgroups within our ITT analysis, we conducted a subsequent subgroup analysis in the PPS dataset. However, this association was not replicated in the PPS analysis (online suppl. Fig. S5). The results of the inverse probability of treatment weighting analyses for safety endpoints were also consistent with the primary analyses (Table 3). After additionally adjusting for dialysis duration, vascular access type, and use of cardiovascular medicine in the propensity score matched population, the efficacy and safety outcomes remained consistent with the primary analysis (online suppl. Table. S1).

Table 3.

End points in PPS and sensitivity analyses

Outcome	Roxadustat				ESA				Hazard ratio (95% CI)a	p value
	patients, N	events, N	person-years	event rate, per 100 person-years	patients, N	events, N	person-years	event rate, per 100 person-years
Patients who crossed over between 2 treatments censored when they switched (PPS)
Event rate for MACE	2,443	320	1,392.81	22.98	2,443	305	1,272.71	23.96	0.96 (0.77–1.15)	0.57
Event rate for HHF	2,443	255	1,337.18	19.07	2,443	268	1,236.16	21.68	0.88 (0.74–1.05)	0.21
Event rate for all-cause death	2,443	144	1,301.27	11.06	2,443	160	1,267.68	12.62	0.87 (0.64–1.09)	0.24
Event rate for thromboembolism	2,443	211	1,311.83	16.08	2,443	191	1,259.89	15.16	1.06 (0.89–1.22)	0.47
Patients who crossed over between treatments excluded
Event rate for MACE	1,119	187	834.36	22.41	1,119	116	573.22	20.24	1.11 (0.95–1.26)	0.38
Event rate for HHF	1,119	142	836.94	16.97	1,119	109	574.31	18.98	0.85 (0.62–1.08)	0.56
Event rate for all-cause death	1,119	65	829.91	7.83	1,119	46	571.50	8.05	0.97 (0.84–1.11)	0.67
Event rate for thromboembolism	1,119	101	840.36	12.02	1,119	67	581.24	11.53	1.04 (0.85–1.23)	0.72
IPTW instead of propensity-score matching
Event rate for MACE	3,184	589	3,099.86	19.00	3,230	518	2,873.67	18.03	1.05 (0.89–1.19)	0.54
Event rate for HHF	3,184	437	3,088.78	14.15	3,230	447	2,934.69	15.23	0.93 (0.81–1.07)	0.31
Event rate for all-cause death	3,184	234	3,054.14	7.66	3,230	254	2,995.71	8.48	0.90 (0.59–1.21)	0.50
Event rate for thromboembolism	3,184	355	3,123.24	11.35	3,230	324	3,053.47	10.61	1.07 (0.98–1.16)	0.12

PPS, per-protocol set; ESA, erythropoiesis-stimulating agent; CI, confidence interval; MACE, major adverse cardiovascular events; HHF, heart failure hospitalization.

^aHR (95% CI) comparing risk of prespecified event for roxadustat vs. ESA.

Discussion

This real-world cohort study demonstrated that roxadustat achieved superior efficacy in elevating and sustaining Hb levels compared to ESAs in hemodialysis-dependent CKD patients, with comparable long-term cardiovascular safety. The significantly greater Hb increments observed with roxadustat at 6–12 months (LSM difference: 0.46 g/dL) and 18–30 months (LSM difference: 0.26 g/dL), alongside higher Hb response rates (84% vs. 76%). In previous RCT studies [20, 25, 27], roxadustat showed promising efficacy for the treatment of CKD patients with anemia, which is consistent with our current study. Recently, some observational studies compared the treatment effectiveness of roxadustat and ESA in the real world. A single-center study involving 209 MHD patients showed that roxadustat was superior to ESA treatment in elevating Hb levels, particularly during the first month [41]. Similarly, a hospital-based cohort study evaluated the efficacy of roxadustat compared to ESA in chronic dialysis subjects [42]. They found that the change of Hb from baseline and the Hb response rate were numerically greater in the roxadustat group over an observation period of 6 months. Jin et al. [43] conducted a retrospective cohort study of 790 consecutive patients with renal anemia treated with roxadustat and ESA and found similar results. The studies mentioned above all suggested that roxadustat offers greater advantages over ESA in treating dialysis patients with renal anemia, particularly during the initial treatment phase (<6 months), which was in accordance with our study. However, our study further revealed that roxadustat can maintain this therapeutic advantage over a longer period (up to 2 years). This difference may be attributed to our larger sample size and longer follow-up duration. Additionally, to avoid overestimating the efficacy of the investigational drug, we employed ITT analyses, which enhanced the reliability of the study conclusions.

In CKD patients, anemia treatment with ESAs may be associated with an increased risk of cardiovascular events [44]. Similar relationships cannot be excluded with the use of HIF-PHIs. Considering the close interplay between the cardiovascular and renal systems and the elevated risk of cardiovascular disease in CKD patients, the safety of roxadustat has been a key focus of interest. A pooled analysis of dialysis patients from multiple phase 3 studies found that roxadustat and ESA (epoetin alfa and DA) had similar safety profiles, with comparable rates of death, MACE, and MACE plus (including hospitalization for heart failure or unstable angina) [22]. The authors of this study observed a lower incidence of MACE associated with roxadustat use compared with ESA in the incident dialysis subgroup (<4 months) but showed the opposite results in the stable dialysis group (≥4 months) (incident dialysis: HR: 0.83 [95% CI: 0.61–1.13]; stable dialysis: HR: 1.18 [95% CI: 1.00–1.38]). A meta-analysis of 8806 CKD patients found no significant increase in cardiac or kidney adverse events with roxadustat compared to placebo in both NDD-CKD and DD-CKD [45]. Similarly, a systematic review and meta-analysis involving 12,821 dialysis patients assessed the cardiovascular safety of HIF-PHI compared to ESA treatment. The analysis found no significant differences in the incidence of first MACE, myocardial infarction, stroke, or thrombosis between the two groups, indicating comparable cardiovascular outcomes [46]. These literature data indicate that roxadustat and ESAs appear to have comparable cardiovascular safety profiles.

Using these RWD in our study, we found no significant difference in the risk of cardiovascular endpoints (including MACE, HHF, and all-cause death) between patients taking roxadustat vs. those taking ESA in our primary ITT analyses. We observed a high crossover rate of treatment, which is a common characteristic of real-world studies in dialysis populations, reflecting clinical practice where physicians adapt therapy based on individual patient response, adverse events, patient compliance, drug availability, and changing clinical guidelines. This crossover can potentially bias the ITT effect estimates toward the null as patients are no longer exclusively representative of their initial treatment group. However, the results remained consistent in subsequent sensitivity analyses after censoring or excluding those patients who crossed over between the two treatments. While we did observe roxadustat to be associated with a significantly higher risk of MACE outcome in the subgroup of patients with baseline hypertension or CV history, there were no statistically significant differences between roxadustat and ESA across other subgroups. However, when the subgroup analysis was repeated in the PPS dataset, this association was no longer present. It is likely that residual confounding due to patient and prescriber behaviors could have contributed to the observed findings, especially the increased risk of MACE outcome, in ways that would not occur in an RCT [47, 48]. Another possible reason is that in the ITT analysis set, some patients in both groups experienced treatment switches during the follow-up period [49]. This provides a plausible basis for a potentially design-related bias that may have impacted the difference in outcomes between the ESA- and roxadustat-treated patients. As for thromboembolic events, some RCT studies found that there was an increase in the rate of deep-vein thrombosis and vascular access thrombosis with roxadustat compared with ESA [22, 25, 30]. However, in real-world settings, we found that this difference did not reach statistical significance. In brief, our study suggested the safety profile of roxadustat was similar to that of ESA with respect to MACE, HHF, and all-cause death, as well as thromboembolic events.

It is imperative to consider some limitations of our study. First, this is a single-center study, lacking data from multiple centers for validation. The iron parameters, hepcidin and transferrin, were not routinely performed in our nephrology department. Therefore, we included only patients with baseline iron parameters but did not evaluate iron parameters in relation to the outcomes due to the lack of follow-up data. Additionally, we did not incorporate heart ultrasound findings or other cardiovascular parameters to comprehensively assess the cardiovascular outcomes of our patients. The switching between study drugs is unavoidable in real-world settings. Although we conducted a sensitivity analysis, nearly 40% of patients in the ESA group switched to roxadustat treatment during the follow-up period in the primary analysis, which may have contributed to residual confounding and highlights some of the challenges of using RWD to emulate RCTs. Another important limitation is that the safety outcomes in this study were determined based on EMR data, which may introduce measurement bias. First, we might have underestimated the absolute incidence of adverse events, particularly those that were mild or asymptomatic. Thus, we focused on robust, hard endpoints (e.g., MACE, mortality) that are reliably captured in EMRs due to their severity and the consequent medical attention they require. Second, although we made every effort to confirm primary endpoint events through manual medical record review, inaccuracies in diagnostic codes could still lead to misclassification of some events.

In conclusion, among patients with anemia of CKD on maintenance hemodialysis, treatment with oral roxadustat increased Hb as effectively as ESA. The safety profile of roxadustat was comparable with ESA. The results obtained in this study indicate that roxadustat might be superior to ESAs without increasing the risk of cardiovascular events. To carefully study the currently unclear effects and the long-term use benefits of roxadustat, it is essential to expand the sample size and include additional observation parameters.

Acknowledgments

The authors would like to thank the staff of the Division of Nephrology, Institute of Kidney Diseases, and Department of West China Hospital of Sichuan University.

Statement of Ethics

The study was reviewed and approved by the Ethics Committee of West China Hospital of Sichuan University (HX-IRB-AF-2020-03). This research was conducted in accordance with the World Medical Association Declaration of Helsinki. Written informed consent was obtained for participation in this study.

Conflict of Interest Statement

The authors declare that they have no conflict of interest.

Funding Sources

This study was supported by the following grants: (1) grants from the 1.3.5 project for disciplines of excellence from the West China Hospital of Sichuan University (ZYGD23015); (2) grants from the National Natural Science Foundation of China (82300814); and (3) grants from the Science and Technology Department of Sichuan Province (2024NSFSC1498).

Author Contributions

Z.Z. and J.L. designed the study. Z.Z. drafted the manuscript and processed statistical data analysis. Y.K. and Y.L. polished the manuscript. J.L. and P.F. helped critically review the manuscript. All authors participated in material preparation, data collection, results interpretation, polishing the manuscript, and final approval.

Funding Statement

This study was supported by the following grants: (1) grants from the 1.3.5 project for disciplines of excellence from the West China Hospital of Sichuan University (ZYGD23015); (2) grants from the National Natural Science Foundation of China (82300814); and (3) grants from the Science and Technology Department of Sichuan Province (2024NSFSC1498).

Data Availability Statement

The data that support the findings of this study are not publicly available to protect the privacy of research participants but can be obtained from the author Z.Z. (carloszhang@126.com) upon reasonable request.