Abstract

Background:

Roxadustat is an oral hypoxia-inducible factor prolyl hydroxylase inhibitor used to treat anemia in patients with chronic kidney disease (CKD); however, evidence from randomized trials has not fully clarified its associations with thrombotic outcomes and all-cause mortality.

Methods:

We searched PubMed, Embase, Web of Science, and the Cochrane Library from inception to 21 August 2025 for randomized controlled trials comparing roxadustat with placebo or erythropoiesis-stimulating agents (ESAs) in adults with CKD. The primary outcome was vascular access thrombosis (VAT), while the secondary outcomes were all-cause mortality, any venous thromboembolism (VTE), and adverse events (AEs) leading to treatment discontinuation. We used the random-effects model for primary analyses and conducted sensitivity analyses using the leave-one-out and fixed-effects models. Furthermore, certainty of evidence was assessed using the GRADE framework.

Results:

Twenty randomized comparisons (involving 11,418 participants) were included in this study. Roxadustat was associated with higher odds of VAT (odds ratio (OR): 1.50; 95% confidence interval (CI): 1.06–2.12) and all-cause mortality (OR: 1.14; 95% CI: 1.06–1.22) with minimal heterogeneity; it was also found to increase AEs leading to discontinuation (OR: 1.76; 95% CI: 1.48–2.10). For any VTE, the estimate was imprecise and had a wide CI including the null value (OR: 3.69; 95% CI: 0.71–19.14). Subgroup analyses showed no evidence of effect modification for mortality by dialysis status or comparator type. For VAT, the subgroup estimates were directionally adverse for both the dialysis-dependent (DD) and non-dialysis-dependent (NDD) populations. However, the primary clinical interpretation of VAT pertained to the DD population because the majority of NDD patients lacked established vascular access; thus, the NDD findings warrant cautious interpretation. Discontinuation due to AEs increased in both the DD and NDD trials, with larger effects in the DD and ESA-controlled trials. Certainty of evidence was moderate for mortality and VAT but low for any VTE and AE-related discontinuation.

Conclusion:

In this meta-analysis of anemia in CKD, roxadustat was found to be associated with higher odds of VAT and all-cause mortality, along with increased AEs leading to treatment discontinuation, whereas the effects on any VTE remained uncertain because of imprecise results. These findings support careful selection of patients, close surveillance of hemoglobin levels after treatment initiation or dose adjustment, and continued monitoring of vascular access when using roxadustat in dialysis settings.

1 Introduction

Anemia is a common and clinically consequential complication of chronic kidney disease (CKD) (Iatridi et al., 2025; Zheng et al., 2025), with a prevalence that increases as kidney function declines, exceeding 90% among patients receiving renal replacement therapy (Nakhoul et al., 2016). In CKD populations, anemia is associated with impaired functional status and poorer health-related quality of life. It is also linked to higher transfusion requirements and increased risks of hospitalization and death (Hörl, 2013; Hanna et al., 2021). Current management strategies for anemia primarily integrate iron therapy and erythropoiesis-stimulating agents (ESAs), while transfusion of red blood cells is reserved only for select clinical situations (Locatelli et al., 2013). However, randomized trials pursuing higher hemoglobin targets with ESAs have reported greater cardiovascular and thrombotic risks, informing more conservative treatment targets and sustained interest in alternative approaches (Besarab et al., 1998; Singh et al., 2006; Pfeffer et al., 2009; Palmer et al., 2010).

Roxadustat is an oral hypoxia-inducible factor prolyl hydroxylase inhibitor (HIF-PHI) developed to increase endogenous erythropoietin production, coordinate iron handling, and lower circulating hepcidin levels (Provenzano et al., 2016a; Besarab et al., 2015). Phase 2 studies in non-dialysis and dialysis populations of CKD have shown dose-dependent hemoglobin responses and suggested favorable effects on iron indices, thereby supporting longer-term evaluations (Provenzano et al., 2016a; 2016b; Chen et al., 2017). Subsequent phase 3 trials, which expanded exposure across non-dialysis and dialysis populations using placebo or ESA comparators, have confirmed hemoglobin efficacy over clinically relevant follow-up periods (Chen et al., 2019a; Charytan et al., 2021; Chen et al., 2019b; Fishbane et al., 2022; Provenzano et al., 2021a).

Although several meta-analyses have shown improvements in hemoglobin and iron indices with roxadustat (Liu et al., 2020; Abdelazeem et al., 2022; Tang et al., 2021), safety estimates, particularly regarding cardiovascular signals, have been less consistent across analyses and populations (Abdelazeem et al., 2022; Tang et al., 2021; Li et al., 2024). Thrombotic outcomes are particularly sensitive to endpoint definition and ascertainment; as these events are uncommon, they are often captured through serious adverse event (SAE) reporting using the preferred terms of the Medical Dictionary for Regulatory Activities (MedDRA) rather than adjudicated clinical endpoints (Chen et al., 2019a; Charytan et al., 2021; Fishbane et al., 2022). In dialysis-dependent (DD) CKD, vascular access thrombosis (VAT) is a clinically important concern as it can lead to missed dialysis, hospitalization, urgent intervention, and loss of access; therefore, accurate quantification is central to the benefit-risk assessment (Quencer et al., 2017; Roy-Chaudhury et al., 2003). Earlier pooled analyses of roxadustat and recent meta-analyses, including network meta-analyses, have generally evaluated broader cardiovascular or composite safety outcomes rather than VAT as the central clinical endpoint (Barratt et al., 2021a; Provenzano et al., 2021b; Ren et al., 2024).

Hypoxia and HIF activation intersect with thrombosis biology through their effects on coagulation programs and inflammatory thrombosis pathways, providing a biological rationale for scrutinizing thrombotic safety signals under sustained HIF pathway modulation (Evans, 2019). Vascular endothelial growth factor (VEGF) has also been linked to risk of venous thromboembolism (VTE) in genetic analysis, further supporting systematic evaluation of thrombotic outcomes with HIF-PHIs (Zhang et al., 2022). Against this background, we conducted a systematic review and meta-analysis of randomized trials comparing roxadustat with placebo or ESAs by designating VAT as the primary outcome to provide a more dialysis-relevant and clinically actionable assessment focused on access patency and treatment continuity. We also quantified all-cause mortality, any VTE, and treatment discontinuation, along with assessing the certainty of evidence using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework (Evans, 2019; Zhang et al., 2022; Guyatt et al., 2008).

2 Methods2.1 Data sources and search strategy

Our systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and was registered with PROSPERO (registration ID: CRD420251274247; available at https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD420251274247). The eligibility criteria and outcomes were specified in the registered PROSPERO record, while the relevant protocol details and post-registration clarifications are summarized in Supplementary Table S2. We searched records regarding roxadustat in PubMed, Embase, Web of Science, and the Cochrane Library from database inception up to 21 August 2025. The search strategy combined database-specific controlled vocabulary (e.g., MeSH and Emtree) with free-text terms for roxadustat and its development code (FG-4592), along with terms related to anemia, CKD, dialysis, and randomized controlled trials (RCTs). The complete database-specific search strategy is outlined in Supplementary Table S3. We also manually screened the reference lists of relevant reviews and eligible reports to identify additional studies.

2.2 Eligibility criteria

The eligibility criteria were defined a priori using the PICOS framework. We included parallel-group RCTs that enrolled adults (≥18 years) with CKD comprising dialysis-dependent (DD) and non-dialysis-dependent (NDD) populations. The eligible studies were required to randomize participants to roxadustat and a clearly defined control group. The comparators included a placebo (with permission for protocol-specified ESA rescue) or active ESAs (e.g., epoetin alfa or darbepoetin alfa). The included trials were required to report extractable data for at least one pre-specified outcome. The primary outcome of our analysis was VAT, while the secondary outcomes included all-cause mortality, any VTE, and adverse events (AEs) leading to discontinuation of the study treatment.

The data on VAT and any VTE were derived from SAE tables based on MedDRA preferred terms. VAT was defined using terms related to arteriovenous fistula (AVF) and/or arteriovenous graft (AVG) thrombosis. When both categories were reported in a trial, the event counts were summed; if only one category was used, then its values were used as is. Similarly, VTE was derived as a composite of deep-vein thrombosis (DVT) and pulmonary embolism (PE). We acknowledge that summing up the counts of the preferred terms may overestimate participant-level incidence because of potential lack of mutual exclusivity (i.e., double counting); however, this approach was used to maximize event capture, given the incomplete reporting of composite endpoints. We excluded non-randomized or single-arm studies, pediatric populations, studies of non-CKD indications, and reports in which data could not be isolated for roxadustat-specific randomized comparisons.

2.3 Selection process and data extraction

The study selection and data extraction were performed independently by two investigators (JT and YZe). Screening was conducted in two stages, namely an initial review of titles and abstracts, followed by a full-text assessment of potentially eligible reports. The data were extracted using a pre-specified standardized extraction template. Disagreements were resolved through discussion to achieve consensus, and a third investigator (XL) served as an adjudicator when necessary.

2.4 Risk of bias assessment

Two independent reviewers (JT and YZe) assessed the risk of bias for the included RCTs using the revised Cochrane Risk of Bias tool. This assessment covered five core domains as follows: (1) randomization process; (2) deviations from intended interventions; (3) missing outcome data; (4) measurements of the outcomes; and (5) selection of the reported results. Discrepancies between the reviewers were resolved through discussion or adjudication by a third investigator.

2.5 Data analysis and sensitivity assessment

The data analyses were conducted within a frequentist framework using the “meta” package in R (version 4.3.2). Dichotomous outcomes were summarized in terms of odds ratios (ORs) with 95% confidence intervals (CIs). We used the random effects model as the primary approach to estimate between-study variance (τ2) via the restricted maximum likelihood method and calculated the CIs using the Hartung–Knapp–Sidik–Jonkman method (IntHout et al., 2014). Heterogeneity was quantified using I2 and τ2, and the 95% CIs were reported to estimate the expected range of effects in future settings (Higgins et al., 2009). When handling sparse data, a continuity correction of 0.5 was applied only to studies that reported zero events in a single arm. Trials with zero events in both arms were excluded from the primary pooling of Ors, as they provide no information on the relative effect magnitude in this framework (Sweeting et al., 2004; Friedrich et al., 2007). Sensitivity analyses were conducted using the fixed effects (common effects) model and by assessing the influence of individual studies using the leave-one-out approach.

Subgroup analyses were pre-specified by dialysis status (DD vs. NDD) and comparator type (placebo vs. ESA), and the differences were assessed using the χ2 statistic. The dialysis status analysis included only trials enrolling exclusively DD or NDD populations, or those reporting stratified data. Similarly, the comparator analysis was restricted to trials with a single randomized control group (placebo or ESA). Trials reporting more than one independent randomized cohort were extracted and analyzed as separate comparisons. Publication bias was assessed by visually inspecting the funnel plots and using Egger’s regression test for outcomes with ten or more studies (Egger et al., 1997; Sterne et al., 2011).

2.6 Evidence certainty assessment

The certainty of evidence was assessed using the GRADE framework. Here, two independent reviewers (JT and YZe) evaluated each outcome against the five core domains of risk of bias, inconsistency, indirectness, imprecision, and publication bias. Evidence derived from the randomized trials was initially classified as high certainty and was downgraded by one or two levels based on the presence of serious or very serious limitations in any domain. Discrepancies were resolved through discussion or adjudication by a third investigator (XL).

3 Results3.1 Study selection and characteristics

A total of 19 trial reports comprising 20 randomized comparisons involving 11,418 participants were included in the meta-analysis (Figure 1) (Besarab et al., 2015; Provenzano et al., 2016b; Chen et al., 2017; Chen et al., 2019a; Charytan et al., 2021; Chen et al., 2019b; Fishbane et al., 2022; Provenzano et al., 2021a; Fishbane et al., 2021; Shutov et al., 2021; Barratt et al., 2021b; Coyne et al., 2021; Akizawa et al., 2021; Akizawa et al., 2019; Csiky et al., 2021; Akizawa et al., 2020; Hou et al., 2022; Tan et al., 2024; Wu et al., 2024). The detailed study characteristics are summarized in Table 1. One publication reported two independent randomized cohorts, including an NDD placebo-controlled cohort and a DD ESA-controlled cohort (Chen et al., 2017).

PRISMA flow diagram of the study selection process.

Author, yearTrial IDCountryMean age (years)Female% (Rox/Ctrl)Hb (g/dL)Dialysis statusIntervention (starting dose)Comparatorn (Rox)n (Ctrl)Follow-up durationFishbane et al. (2021)NCT02174627Global62.458/589.1NDD70 mg TIWPlacebo1393138816.4 monthsShutov et al. (2021)NCT01887600Europe63.856/559.1NDDWeight-based TIWPlacebo278278104 weeksBarratt et al. (2021b)NCT02021318Europe66.554/539.5NDDWeight-based TIWDarbepoetin alfa323293104 weeksCoyne et al. (2021)NCT01750190Global6458/599.1NDDWeight-based TIWPlacebo616305104 weeksChen et al. (2019b)NCT02652819China5848/368.9NDDWeight-based TIWPlacebo1015026 weeksAkizawa et al. (2021)NCT02988973Japan68.948/4610.3NDD50/70/100 mg TIWDarbepoetin alfa13213152 weeksBesarab et al. (2015)NCT00761657USA6266/549.7NDD0.7–2.0 mg/kg TIWPlacebo88284 weeksAkizawa et al. (2019)NCT01964196Japan67.540/379.8NDD50/70/100 mg TIWPlacebo802724 weeksChen et al. (2017)NCT01599507China51.442/458.8NDD1.1–2.25 mg/kg TIWPlacebo61308 weeksFishbane et al. (2022)NCT02174731Global54.843/429.6DD (HD)70–100 mg or ESA-basedEpoetin alfa10681065104 weeksProvenzano et al. (2021a)NCT02052310Global5644/458.3DD (Inc)70–100 mg TIWEpoetin alfa522517104 weeksCharytan et al. (2021)NCT02273726USA/Global55.745/489.6DD (HD)ESA-based TIWEpoetin alfa370370104 weeksCsiky et al. (2021)NCT02964936Europe62.540/379.7DDESA-based TIWESA (Var)415421104 weeksChen et al. (2019a)NCT02652806China4938/438.7DD100/120 mg TIWEpoetin alfa20410126 weeksAkizawa et al. (2020)NCT02952092Japan61.632/3710.9DD (HD)70/100 mg TIWDarbepoetin alfa15015224 weeksHou et al. (2022)ChiCTR2000035054China5342/538DD (PD)100/120 mg TIWESA864324 weeksTan et al. (2024)ChiCTR2000041202China51.240/4910.5DD (HD)Titrated TIWrHuEPO575724 weeksWu et al. (2024)NCT04655027China55.130/418.7DD/NDDStandard TIWrHuEPO13122 weeksProvenzano et al. (2016b)NCT01147666USA53.845/3910.2DD1.0–2.0 mg/kg TIWEpoetin alfa832319 weeksChen et al. (2017)NCT01596855China5135/4110.7DD1.1–2.3 mg/kg TIWEpoetin alfa65226 weeks

Baseline characteristics of the included studies.

Values are presented as the mean or percentage. Abbreviations: Rox, roxadustat; Ctrl, control; DD, dialysis-dependent; ESA, erythropoiesis-stimulating agent; Hb, hemoglobin; HD, hemodialysis; Inc, incident dialysis; NDD, non-dialysis-dependent; PD, peritoneal dialysis; rHuEPO, recombinant human erythropoietin; TIW, three times weekly; Var, variable ESA type (epoetin alfa or darbepoetin alfa).

3.2 Risk of bias

The risk of bias assessments are summarized in Figure 2. The majority of domains across the included trials were judged as low risk. High-risk designations were concentrated in domains that deviated from the intended interventions or had missing outcome data, whereas ratings of “some concerns” were primarily attributable to the randomization process. Overall, the majority of comparisons were classified as low risk, while a few subsets exhibited some concerns or were classified as high risk.

Assessment of bias for inclusion as a randomized controlled trial in the meta-analysis.

3.3 Primary outcome

The primary analysis of VAT included seven trials (Figure 3A); here, roxadustat was associated with significantly higher odds of VAT than the controls (OR: 1.50; 95% CI: 1.06–2.12), with no evidence of statistical heterogeneity (I2 = 0.0%, τ2 < 0.0001). Notably, the 95% CI (1.05–2.13) excluded unity, suggesting that this increased risk is likely to be a consistent finding across comparable future study populations. The pooled estimates remained stable in the leave-one-out analyses (Supplementary Figure S1A), and the fixed effects modeling yielded directionally consistent results (Supplementary Figure S2A). The certainty of evidence for VAT was graded as moderate (Table 2).

Forest plots of the thrombotic outcomes, mortality, and discontinuation for trials involving anemia in chronic kidney disease (CKD) that compared roxadustat with a control group: (A) vascular access thrombosis (VAT); (B) any venous thromboembolism (VTE); (C) all-cause mortality; and (D) adverse events (AEs) leading to treatment discontinuation.

Certainty assessmentNo. of patientsEffectsCertaintyImportance№ of studiesStudy designRisk of biasInconsistencyIndirectnessImprecisionOther considerations[Intervention][Comparison]Relative (95% CI)Absolute (95% CI)All-cause mortality16Randomized trialsSeriousaNot seriousNot seriousNot seriousNone5,339/10,078 (53.0%)4,739/10,078 (47.0%)OR: 1.14 (1.06–1.22)10 more per 1,000 (from 4 more to 16 more)b⊕⊕⊕○ ModerateaCRITICALVAT7Randomized trialsSeriouscNot seriousNot seriousNot seriousNone4,242/8,155 (52.0%)3,913/8,155 (48.0%)OR: 1.50 (1.06–2.12)33 more per 1,000 (from 4 more to 73 more)d⊕⊕⊕○ ModeratecCRITICALAEs leading to treatment discontinuation15Randomized trialsVery seriouseNot seriousNot seriousNot seriousNone4,995/9,125 (54.7%)4,130/9,125 (45.3%)OR: 1.76 (1.48–2.10)36 more per 1,000 (from 23 more to 52 more)f⊕⊕○○ LoweCRITICALAny VTE4Randomized trialsSeriousgNot seriousNot seriousSerioushNone1,959/3,737 (52.4%)1,778/3,737 (47.6%)OR: 3.69 (0.71–19.14)16 more per 1,000 (from 2 fewer to 100 more)i⊕⊕○○ Lowg,hCRITICAL

GRADE evidence profiles for the overall quality of the evidence assessment.

Abbreviations: VAT, vascular access thrombosis; CI, confidence interval; OR, odds ratio; AE, adverse event; VTE, venous thromboembolism.

3.4 Secondary outcomes

Data for the composite endpoints of any VTE were available from four trials (Figure 3B). The pooled estimates suggested increased odds with roxadustat but lacked precision (OR: 3.69; 95% CI: 0.71–19.14) and were accompanied by low heterogeneity (I2 = 9.7%, τ2 = 0.3104) (Figure 3B). The 95% CI was wide (0.29–46.82) and included unity, reflecting sparse events and substantial uncertainty regarding effect magnitudes. Although the fixed effects sensitivity analysis yielded a statistically significant estimate (OR: 4.22; 95% CI: 1.84–9.67; Supplementary Figure S2B), this result was heavily driven by a single large trial (∼70% weight). Consequently, given the sparse event counts, the primary inference was anchored upon the more conservative random effects model. The leave-one-out estimates of the ORs ranged from 2.44 to 7.84 (Supplementary Figure S1B), and the certainty of evidence was rated as low (Table 2).

In the pooled analysis of 16 trials, roxadustat was associated with significantly higher odds of all-cause mortality than the controls (OR: 1.14; 95% CI: 1.06–1.22), with negligible between-study heterogeneity (I2 = 0.0%, τ2 = 0) (Figure 3C). The 95% CI (1.00–1.29) indicates that the effects in comparable future trials may likely range from the null effect to a moderate increase in mortality risk. These results remained robust in the leave-one-out analyses (OR: 1.11–1.16; Supplementary Figure S1C) and in fixed effects modeling (OR: 1.14; 95