Roxadustat in the Treatment of Anemia in End Stage Renal Disease (ESRD) Patients on Stable Dialysis: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Roxadustat in the Treatment of Anemia in End Stage Renal Disease (ESRD) Patients on Stable Dialysis (NCT02278341), a randomized phase 3 study. The primary outcome is Change From Baseline (BL) to the Average Hemoglobin (Hb) in Weeks 28-36 Without Rescue Therapy [EU (EMA)].

2. Clinical Question

3. Trial Design

The registry describes a phase 3, randomized, open-label, active-controlled study with a parallel-group design. The trial compared roxadustat with erythropoiesis-stimulating agent treatment in participants with anemia and end stage renal disease on stable dialysis. The active comparator included epoetin alfa and darbepoetin alfa. The study is marked completed, and the registry lists 838 participants as enrolled.

Participants assigned to roxadustat received it three times a week. The starting dose was 100 mg, 150 mg or 200 mg, according to the average weekly dose of epoetin or darbepoetin alfa before randomization. Participants in the comparator group received epoetin alfa or darbepoetin alfa on the schedules listed in the registry. Both treatment descriptions give a minimum treatment duration of 52 weeks and a maximum of 104 weeks. Comparator dosage was adjusted according to a prespecified rule, but its full details are not reported in the registry record.

The study was open-label, meaning treatment assignment was not masked from participants or study personnel according to the registry design description. Random allocation can help balance measured and unmeasured characteristics at the start of a trial, while an active comparator makes the question a comparison with established anemia treatments rather than with no treatment. Open-label treatment can affect aspects of care or assessment, so the endpoint definition and analysis procedures matter when interpreting the result.

The registry lists iron among the interventions, but the available descriptions do not establish how iron was used or whether its use differed by group. The planned sample-size justification, allocation ratio, randomization procedure, interim analyses and detailed statistical plan are not reported in the registry record. Enrollment is not the same as the number analyzed for a particular endpoint: the primary endpoint reports separate analysis counts for the two groups, and the secondary outcomes also have their own analysis counts.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Before assignment: Participants were randomized in a 1:1 ratio to roxadustat or ESA (epoetin alfa or darbepoetin alfa). Randomization was stratified by 5 factors: previous ESA treatment, region, history of cardiovascular, cerebrovascular or thromboembolic diseases, average weekly ESA dose 4 weeks prior to randomization and the screening hemaglobin (Hb) value.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: The analysis population was the Per Protocol Set (PPS) which consisted of all Full Analysis Set (FAS) participants who did not meet any of exclusion criteria from the PPS. The FAS consisted of all randomized participants who received at least one dose of study drug and had at least one post-dose valid Hb assessment.

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

Population: The analysis population was the All Randomized, which consisted of participants who were randomized into the study and had the available data.

6. Primary Endpoint

The registry lists the primary outcome as Change From Baseline (BL) to the Average Hemoglobin (Hb) in Weeks 28-36 Without Rescue Therapy [EU (EMA)], measured over Baseline and weeks 28 to 36.

The primary endpoint measures change in hemoglobin from baseline to the average hemoglobin during weeks 28 to 36, without rescue therapy. Hemoglobin is the laboratory measure named in the registry; the endpoint is expressed in g/dL. The baseline value was defined as the mean of four central laboratory hemoglobin values, using the four latest values obtained before or on the date of the first study-drug intake. The endpoint therefore compares a specified baseline measure with a later period average, rather than relying on a single later measurement.

Hemoglobin is relevant to the trial because the study concerns maintenance treatment of anemia. Using an average over a defined period can represent the treatment-period level while reducing reliance on one measurement. The “without rescue therapy” qualification also makes the estimate conditional on the registry’s handling of rescue treatment: after rescue began, observations were set to missing for six weeks from its start date. When an endpoint value was unavailable, the record describes an imputation approach that carried forward the mean of available values from Day 1 to the minimum of the End of Efficacy Emergent Period.

The primary analysis used the Per Protocol Set. The registry defines this as Full Analysis Set participants who did not meet any of the Per Protocol Set exclusion criteria, but does not give the complete Full Analysis Set definition in the available description. These population and missing-data rules are part of what the endpoint estimate represents. They should be considered alongside the numerical difference and its confidence interval, rather than treating the endpoint as a simple average for every randomized participant.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The primary result is a model-based comparison of change in hemoglobin. The registry identifies the arm values as least-squares means and the comparison as a least-squares mean difference from a mixed models analysis. A least-squares mean is an estimate derived from a statistical model, rather than necessarily the raw arithmetic average of observed values. A mixed model can account for repeated measurements or other modeled structure, but the exact model specification, covariates and covariance assumptions are not reported in the registry record. The posted estimate should therefore be interpreted as the sponsor-reported model result, not as a fully reproducible analysis from the summary alone.

A confidence interval gives a range of values compatible with the data and the analysis assumptions at the stated confidence level. For the primary comparison, the registry reports a 95% confidence interval from 0.132 to 0.339 g/dL around the least-squares mean difference of 0.235 g/dL. A confidence interval is not a guarantee that the true effect lies inside its limits, nor does it describe the probability that a particular treatment effect is correct. Its width conveys statistical precision: narrower intervals indicate greater precision than wider intervals in the context of the same measure and analysis.

The registry labels the primary comparison as noninferiority. In a noninferiority design, the objective is generally to assess whether the estimate is not worse than a prespecified margin. This requires knowing the margin, the direction of the comparison, and the analysis rules. The margin is not reported in the registry record, so the posted confidence interval cannot be judged against that decision threshold from the registry information alone. In particular, a p-value for a difference from zero does not substitute for a noninferiority assessment.

For binary outcomes, a percentage is the observed proportion meeting the outcome within the stated analysis group. A risk difference expresses the contrast between proportions in the groups, while a risk ratio and an odds ratio express other relative contrasts. The registry also reports sponsor analyses of differences of percentages with confidence intervals. These quantities answer related but distinct questions and should not be treated as interchangeable. The secondary hemoglobin-response analyses include a recalculated unadjusted comparison and a sponsor-reported analysis; their uncertainty and design context matter more than any isolated point estimate.

Unadjusted recalculations from summary statistics can be useful as checks, but they do not reproduce analyses that use participant-level data, covariate adjustment, imputation or a specified model. For example, a simple comparison of means does not account for the mixed-model structure used for the primary outcome. Any recalculation is consequently an approximation based on the posted summaries. The primary result relies on the reported mixed-model analysis; a summary-statistic calculation should not be used to override it.

The hospitalization outcome is a count summarized as a mean and standard deviation in each group. Count outcomes often have distributions that are skewed or have more variability than a simple model assumes. Over-dispersion is the general statistical situation in which count variability exceeds what a basic count model expects; it can make standard errors too small if ignored. This is a general caution, not a claim about the data or model used in this trial. The registry does not report a sponsor analysis for the hospitalization outcome, and the simple unadjusted comparison does not establish whether a count model or other adjustment would be appropriate.

Planning illustrations based on an observed effect are not observed power and do not describe the original trial’s design justification. They show what group sizes might be needed under stated assumptions if a future study were planned around a fraction of the observed effect. The stated assumptions are a two-sided significance level of 0.05, equal allocation between two groups and a normal approximation. These are illustrations only; the trial’s own sample-size assumptions and justification are not reported in the registry record. Such planning estimates are especially uncertain when the observed effect used as their starting point is itself imprecise.

Missing-data methods are also central to interpretation. The primary endpoint description sets measurements to missing for six weeks after rescue therapy starts and describes carrying forward a mean of available values when no hemoglobin value is available. Imputation fills in missing information according to an assumption; it does not recover an unobserved value with certainty. Different assumptions about why measurements are missing can lead to different estimates. The registry does not report enough information to assess sensitivity to alternative missing-data assumptions, nor does it give full missing-data details for all secondary outcomes.

10. Analyses Reported by the Sponsor

The sponsor-reported primary analysis is a mixed models analysis of the least-squares mean difference. It gives a difference of 0.235 g/dL, with a 95% confidence interval from 0.132 to 0.339 and p < 0.001. This is the analysis to use when describing the sponsor’s primary result. The registry’s arm-level least-squares means and intervals provide context for that comparison, but they are not a substitute for the fitted model.

No separate simple unadjusted recalculation is reported for the primary endpoint. The registry’s summary does not provide the participant-level data or all model details needed to reproduce the mixed-model estimate. Accordingly, there is no recalculated primary estimate against which to claim agreement or disagreement. The registry does not flag a discrepancy between a recalculation and the sponsor analysis. This does not remove the limitations that come from incomplete reporting of model specification, analysis populations and noninferiority criteria.

For the secondary hemoglobin-response outcome, the sponsor reports a Miettinen and Nurminen difference of percentages of 2.3, with a 95% confidence interval from -2.9 to 7.6 and p < 0.05, designated as noninferiority. The unadjusted recalculation is a distinct summary-statistic check, not a replacement for the sponsor’s method. The sponsor’s stated analysis and the recalculation may differ in estimand, method or rounding. The noninferiority margin is not reported in the registry record, so neither result alone permits independent verification of the noninferiority criterion.

11. Multiplicity, Missing Data and Analysis Populations

The record includes many outcome measures and analyses, but it does not report a multiplicity strategy. When investigators test many outcomes or make multiple comparisons, the chance of at least one small p-value due to random variation can rise. Statistical approaches such as prespecified testing sequences or adjustments can address that issue, but the registry does not state whether any such approach was used. This missing information matters most when interpreting secondary results as a group; individual unadjusted p-values should not be treated as if they were the sole planned test.

The primary endpoint description does provide some information about missing measurements. Baseline hemoglobin was based on the mean of four central laboratory values, using the four latest values before or on the day of first study drug intake. For participants without a hemoglobin value in weeks 28 to 36, imputation rules were applied. For the analysis without rescue therapy, values after rescue began were set to missing for six weeks from the rescue start date. If no hemoglobin value was available, an imputation technique carried forward the mean of available values from Day 1 through the minimum of the End of Efficacy Emergent Period. The registry does not report the full algorithm or sensitivity analyses that would show how results change under alternative assumptions.

The primary analysis used the Per Protocol Set, which the record describes as Full Analysis Set participants who did not meet any of the Per Protocol Set exclusion criteria. The description of the Full Analysis Set is cut off, and the full criteria for the Per Protocol Set are not reported in the registry record. The primary analysis counts are 354 and 381; the fact that these counts are not the full enrollment total is consistent with an endpoint-specific analysis population, but the reasons for exclusion or missingness by group are not posted in the record. The posted secondary endpoints have their own analysis counts, and their population definitions and missing-data handling are not comprehensively described.

Details of interim analyses, the complete statistical analysis plan, the noninferiority margin, the sample-size justification, and the handling of multiplicity are not reported in the registry record. These gaps limit independent assessment of prespecification, sensitivity to missing data and the strength of evidence across the full set of outcomes.

12. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 5% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: arteriovenous fistula thrombosis (7.0%), pneumonia (5.0%), atrial fibrillation (2.9%), acute myocardial infarction (2.6%), peritonitis (2.4%).

The registry reports 210 participants with serious adverse events among 414 at risk in the roxadustat group and 189 among 420 at risk in the erythropoiesis-stimulating agent group. It also reports other adverse events for 210 participants among 414 at risk and 209 among 420 at risk. Deaths are reported for 78 participants in the roxadustat group and 59 in the comparator group. These are descriptive counts; they do not by themselves estimate a causal difference in risk between treatments.

The most frequently listed serious adverse event terms include arteriovenous fistula thrombosis at 7.0%, pneumonia at 5.0%, atrial fibrillation at 2.9%, acute myocardial infarction at 2.6%, peritonitis at 2.4% and hypertension at 2.4%. The listed percentages are not sufficient on their own to establish whether an event was caused by treatment, how its frequency compared between groups, or how it relates to time exposed. The summary does not give uncertainty intervals or exposure-adjusted comparisons for these terms.

The treatment period could range from at least 52 weeks up to a maximum of 104 weeks. Counts may be influenced by how long participants were observed, and the registry summaries do not provide the information needed to adjust for that difference. Seriousness, attribution, event timing and the full event definitions also matter. These figures should therefore be read as a record of reported events during the study, not as proof that either treatment caused or prevented a particular event.

13. What the Result Does, and Does Not, Mean

The primary analysis estimates a difference between treatment groups in mean hemoglobin change over a defined period, under the registry’s analysis population, model and missing-data rules. The reported difference is 0.235 g/dL, with a 95% confidence interval from 0.132 to 0.339. The interval describes statistical uncertainty around this estimate under the model’s assumptions. It does not state the probability that the treatment effect is any one value, and it does not determine whether the magnitude matters to patients or clinicians.

The registry calls the primary comparison a noninferiority analysis. Establishing noninferiority requires comparison with a prespecified margin and interpretation under the planned analysis rules. Because the margin and full analysis plan are not reported in the registry record, the numerical estimate alone does not allow an independent assessment of whether the formal noninferiority criterion was met. A p-value for a difference from zero answers a different question from a test against a noninferiority margin.

The result also does not establish how the treatments compare for every secondary outcome, for safety, or for participants outside the analyzed population. The primary endpoint is hemoglobin change without rescue therapy, not a comprehensive measure of all effects of treatment. The secondary outcomes cover distinct measures and are subject to multiplicity and reporting limitations. Adverse-event counts are descriptive and do not, by themselves, establish causation or comparative risk.

Finally, an observed trial estimate is not a guarantee of the result in every setting or for every participant. Interpretation depends on the study population, treatment schedules, follow-up, analysis definitions and assumptions. The registry gives the trial’s broad design and selected endpoint methods, but several details needed for a complete evaluation—among them the noninferiority margin, full model specification, full analysis population definitions and complete missing-data sensitivity analyses—are not reported in the registry record.

14. Why This Trial Matters Statistically

This trial is statistically informative because it compares a treatment with active anemia therapies in a randomized design, rather than relying only on before-and-after measurements. Randomization supports a group comparison by reducing the risk that baseline differences systematically determine treatment assignment. The active-controlled design also makes the intended contrast explicit: roxadustat versus epoetin alfa or darbepoetin alfa under the trial’s treatment and analysis procedures.

The primary endpoint uses a period average of hemoglobin and a specified baseline definition, while the main analysis is a mixed-model least-squares mean comparison. This highlights why endpoint definition and analysis method matter: estimates depend not only on the measured laboratory values but also on which participants are included, how repeated measurements are modeled and how missing or rescue-related values are handled. The reported interval conveys uncertainty that a point estimate alone would hide.

The trial also illustrates a key distinction in noninferiority research. A result that differs statistically from zero is not the same as a result that satisfies a noninferiority margin. The registry labels the analysis as noninferiority, yet the margin is not reported in the registry record. That omission limits independent interpretation of the trial’s central claim even though a model-based estimate and confidence interval are available.

Lastly, the many secondary outcomes and analyses demonstrate why a hierarchy of endpoints and transparent multiplicity methods are important. Broad measurement can be valuable, but separate unadjusted results need cautious interpretation. The trial’s posted figures can inform statistical understanding of estimates, confidence intervals, analysis populations and reporting gaps without turning those figures into treatment advice.

15. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

Several trial-specific limitations constrain interpretation. The study was open-label, so treatment assignment was not masked. The registry does not describe whether any endpoint assessment steps were blinded or how potential effects of unmasked treatment were mitigated. The primary outcome is a laboratory measure, but the record does not give the full operational detail for all assessment procedures beyond the stated baseline and follow-up definitions.

The most important interpretive gap is that the primary analysis is designated as noninferiority but the noninferiority margin is not reported in the registry record. Without that margin and the complete decision framework, the formal noninferiority conclusion cannot be independently assessed from the posted result. The record also does not provide the complete statistical model specification, including its covariates and assumptions, nor the complete analysis plan.

Information about participants is limited in the available trial description. The registry states the condition and stable-dialysis setting but does not provide further demographic or clinical characteristics in the registry record. The primary population is described only partly: the Full Analysis Set definition is truncated, and the full Per Protocol Set criteria are not available. The reported primary analysis counts are therefore difficult to reconcile with the enrollment total in detail.

Missing-data handling is described for the primary endpoint, including rescue-related missingness and an imputation approach, but the full algorithm and sensitivity analyses are not reported. The record does not give comparable detail for all secondary endpoints. In addition, the breadth of outcomes and analyses, alongside no reported multiplicity strategy, makes isolated secondary p-values less definitive.

Safety summaries are descriptive, without exposure-adjusted rates, uncertainty intervals or the attribution details necessary for a comparative causal interpretation. The broad permitted treatment duration also means counts alone do not indicate how much observation time underlies each group. Finally, the sample-size justification, interim analyses and full role of iron are not reported in the registry record.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Roxadustat in the Treatment of Anemia in End Stage Renal Disease (ESRD) Patients on Stable Dialysis — NCT02278341.
• Registry record last updated 2024-11-27; status last verified 2024-11.
• Results first posted 2020-08-26.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

19. Record Summary

The registry record describes NCT02278341 as a completed phase 3 randomized, open-label, active-controlled, parallel-group study of roxadustat for anemia in participants with end stage renal disease on stable dialysis. It lists 838 enrolled participants and compares roxadustat with erythropoiesis-stimulating agent treatment using epoetin alfa or darbepoetin alfa.

The primary endpoint is change from baseline to average hemoglobin during weeks 28 to 36 without rescue therapy. The sponsor-reported mixed-model least-squares mean difference is 0.235 g/dL, with a 95% confidence interval from 0.132 to 0.339 and p < 0.001. The registry designates the analysis as noninferiority, but does not report the margin needed to independently assess that criterion.

The record also includes numerous secondary outcomes and descriptive adverse-event summaries. Interpretation is limited by incomplete reporting of the analysis plan, multiplicity strategy, model details, full analysis-population definitions and safety comparisons.