This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical results on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
SONAR was a randomized, parallel-group, phase 3 trial in diabetic nephropathy comparing atrasentan with placebo. The registered primary endpoint was a time-to-event composite renal endpoint, analyzed in the Intent-to-Treat (ITT) Responder Set consisting of participants who achieved at least 30% reduction in their albumin to creatinine ratio (UACR) during the Enrichment Period.
| Feature | SONAR |
|---|---|
| Trial name | SONAR |
| Brief title | Study Of Diabetic Nephropathy With Atrasentan |
| Phase | Phase 3 |
| Condition | Diabetic Nephropathy |
| Design | Randomized, parallel |
| Masking | Quadruple |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 5107 |
| Interventions | Atrasentan; placebo |
| Primary endpoint type | Time-to-event |
| ClinicalTrials.gov | NCT01858532 |
| Trial status | Terminated |
2. Clinical Question
The central statistical question was whether atrasentan was associated with a different time to the first occurrence of a component of the composite renal endpoint than placebo among randomized participants who met the prespecified responder criterion during the Enrichment Period.
Population
Participants with diabetic nephropathy enrolled in the phase 3 SONAR trial. The primary efficacy analysis was conducted in the ITT Responder Set: participants who achieved at least 30% reduction in UACR during the Enrichment Period.
Intervention
Atrasentan during the trial's randomized treatment comparison.
Comparator
Placebo.
Primary question
How did time to the first occurrence of a component of the composite renal endpoint compare between randomized atrasentan and placebo responders?
3. Trial Design
Atrasentan
- Atrasentan was the active intervention.
- Primary comparisons were made against placebo.
- The primary analysis used the ITT Responder Set as randomized.
Placebo
- Placebo was the randomized comparator.
- The primary comparison was performed against randomized atrasentan.
- The same ITT Responder Set framework was used for the primary analysis.
The enrichment feature is statistically important. The primary analysis was not simply a comparison of every enrolled participant: it was conducted among participants who achieved at least 30% reduction in UACR during the Enrichment Period, while retaining their randomized treatment assignment for the subsequent comparison. This creates a clearly defined analysis population and preserves the randomized comparison within that responder set.
4. Endpoints
Primary Endpoint
| Endpoint | Definition / time frame | Analysis |
|---|---|---|
| Time to the First Occurrence of a Component of the Composite Renal Endpoint in the Intent-to-Treat (ITT) Responder Set (as Randomized) | From randomization to individual end of observation, up to 53 months. The time to the first occurrence of a component of the composite renal endpoint was defined as doubling of serum creatinine, confirmed by a 30-day serum creatinine measurement, or onset of end stage renal disease: eGFR less than 15 ml/min/1.73 m2 confirmed by a 90-day eGFR measurement, receiving chronic dialysis, renal transplantation, or renal death. | Cox proportional-hazards model and stratified log-rank test |
The registry's registry-reported endpoint definition ends after the listed renal-death component. The analysis on this page uses only those components and timing details contained in the ClinicalTrials.gov record.
Secondary Endpoints
| Secondary endpoint | Time frame | Analysis method |
|---|---|---|
| Time to a 50% Estimated Glomerular Filtration Rate Reduction in the ITT Responder Set (as Randomized) | From randomization to individual end of observation, up to 53 months | Cox proportional-hazards model; stratified log-rank test |
| Time to Cardio-renal Composite Endpoint in the ITT Responder Set (as Randomized) | From randomization to individual end of observation, up to 53 months | Cox proportional-hazards model; stratified log-rank test |
| Time to First Occurrence of a Component of Composite Renal Endpoint for All Randomized Participants (Pooled) | From randomization to individual end of observation, up to 53 months | Cox proportional-hazards model |
| Time to the Cardiovascular Composite Endpoint in the ITT Responder Set (as Randomized) | From randomization to individual end of observation, up to 53 months | Cox proportional-hazards model; stratified log-rank test |
5. Analysis Population and Estimand
The primary efficacy analysis used the Intent-to-Treat (ITT) Responder Set. The registry defines these participants as those who achieved at least 30% reduction in their UACR during the Enrichment Period. They were then compared according to randomized treatment assignment: ITT Responder Atrasentan versus ITT Responder Placebo.
The treatment comparison remains based on the randomized assignment within this responder set. This is different from an as-treated comparison, which would classify participants according to treatment actually received.
This distinction matters because the responder criterion was determined during the Enrichment Period. The resulting primary analysis therefore answers a narrower question than a comparison of all 5107 enrolled participants: it estimates the randomized treatment difference among the prespecified population that met the UACR-response criterion.
6. Results: Primary Renal Endpoint
ClinicalTrials.gov reports two formal primary analyses for the same registered time-to-event endpoint. One used a Cox proportional-hazards regression model with prespecified covariate adjustment and reported a hazard ratio with a two-sided 95% confidence interval. The other used a stratified log-rank test and reported a P-value.
Cox Proportional-Hazards Analysis
Hazard ratio for the composite renal endpoint
95% CI: 0.488–0.878 · P = 0.005
Analysis population: ITT Responder Set, as randomized.
| Measure | Reported result |
|---|---|
| Effect measure | Hazard Ratio (HR) |
| Estimate | 0.654 |
| 95% confidence interval | 0.488–0.878 |
| Confidence interval | Two-sided |
| P-value | =0.005 |
| Model | Cox proportional-hazards model |
| Adjustment | Prespecified covariates were adjusted for in the Cox regression model. |
An HR of 0.654 means that, under the Cox model, the estimated instantaneous rate of experiencing the first qualifying component of the composite renal endpoint was 34.6% lower for randomized atrasentan responders relative to randomized placebo responders over the analyzed follow-up. This is a relative hazard measure, not a statement that 34.6% of patients avoided the endpoint.
The 95% CI of 0.488–0.878 describes the precision of the estimated hazard ratio under the model. It excludes 1, so the reported interval is compatible with a lower hazard for atrasentan than placebo under the specified model. It does not establish that every patient benefits, nor does it provide an absolute risk reduction.
The reported P = 0.005 addresses the statistical evidence against the null hypothesis represented by the analysis; it is not a measure of the size or clinical importance of the treatment effect. The P-value should therefore be interpreted alongside the hazard ratio and confidence interval.
Because this is a Cox analysis, the hazard-ratio interpretation depends on the model and its proportional-hazards framework. The ClinicalTrials.gov record does not report a separate assessment of the proportional-hazards assumption, so the HR should not be interpreted as a guaranteed constant risk ratio at every time point.
Stratified Log-Rank Analysis
Stratified log-rank treatment comparison
Primary endpoint · ITT Responder Set · Atrasentan vs placebo
The second primary analysis used a stratified log-rank test for the treatment comparison and reported P = 0.029. Unlike the Cox analysis, this result does not itself provide a treatment-effect estimate such as a hazard ratio or a confidence interval.
The stratified log-rank P-value provides a hypothesis-test result for the difference between the treatment-group time-to-event experience under the specified stratified analysis. It does not quantify the magnitude of that difference. The HR of 0.654 and its 95% CI come from the separate Cox model.
A P-value of 0.029 should not be read as a 2.9% probability that the treatment effect is real, nor as a 97.1% probability that atrasentan is beneficial. It measures compatibility with the null hypothesis under the statistical test, given the analysis framework.
The combination of the log-rank result and the Cox estimate is informative because the two methods address related aspects of the same time-to-event comparison: the stratified log-rank test supplies the formal comparison, while the Cox model supplies an estimated relative effect and its precision.
As with any time-to-event analysis, censoring and the assumptions underlying the statistical methods affect interpretation. The ClinicalTrials.gov record does not provide enough information to independently reconstruct the censoring process or a Kaplan-Meier curve.
7. Secondary Endpoint Results
50% eGFR Reduction
Time to a 50% eGFR reduction
95% CI: 0.573–1.060 · P = 0.112
The Cox model estimated a hazard ratio of 0.779 for time to a 50% estimated glomerular filtration rate reduction in the ITT Responder Set. The two-sided 95% CI was 0.573–1.060, with P = 0.112.
The registry also reports a stratified log-rank analysis for this endpoint with P = 0.289.
The HR of 0.779 corresponds to a model-estimated 22.1% lower instantaneous event rate for atrasentan relative to placebo. The 95% CI includes 1, extending from 0.573 to 1.060, so the estimate is imprecise enough that the data are compatible with effects on either side of the conventional HR = 1 reference under this model.
The P-value of 0.112 is not a measure of effect size. The HR and confidence interval provide the information about estimated magnitude and precision; the P-value addresses the hypothesis test.
Cardio-renal Composite Endpoint
Time to cardio-renal composite endpoint
95% CI: 0.642–0.999 · P = 0.049
For the cardio-renal composite endpoint in the ITT Responder Set, the prespecified-covariate-adjusted Cox model produced an HR of 0.801 with a two-sided 95% CI of 0.642–0.999 and P = 0.049.
The corresponding stratified log-rank analysis reported P = 0.089.
An HR of 0.801 represents an estimated 19.9% lower instantaneous event rate under the Cox model for atrasentan versus placebo in the responder set. The upper confidence-limit value of 0.999 is very close to 1, illustrating why the point estimate alone should not be treated as a complete description of the uncertainty.
The Cox-model P-value is 0.049, while the separate stratified log-rank P-value is 0.089. These are results from different statistical procedures, so they should not be substituted for one another or treated as if they were two independent treatment effects. The registry does not provide additional information here that would justify a broader multiplicity interpretation.
Composite Renal Endpoint in All Randomized Participants
All randomized participants, pooled
95% CI: 0.58–0.89 · P = 0.002
In the pooled ITT population of participants randomized to atrasentan or placebo during the Double-Blind Treatment Period, the Cox model estimated an HR of 0.72, with a 95% CI of 0.58–0.89 and P = 0.002.
An HR of 0.72 corresponds to a 28% lower estimated instantaneous event rate for atrasentan versus placebo under the Cox model. This analysis differs importantly from the primary responder-set analysis because its population is described as all randomized participants rather than only participants who achieved the UACR-response criterion during enrichment.
Because the analysis populations differ, the HR 0.72 should not be presented as interchangeable with the primary HR 0.654. They answer related but distinct statistical questions.
Cardiovascular Composite Endpoint
Time to cardiovascular composite endpoint
95% CI: 0.643–1.215 · P = 0.447
The Cox model for the cardiovascular composite endpoint in the ITT Responder Set produced an HR of 0.884, with a two-sided 95% CI of 0.643–1.215 and P = 0.447.
The corresponding stratified log-rank analysis reported P = 0.446.
The HR of 0.884 corresponds to a 11.6% lower estimated instantaneous event rate under the Cox model, but the 95% CI extends from 0.643 to 1.215 and therefore includes the null value of 1. The ClinicalTrials.gov record therefore do not provide a precise estimate of the direction or magnitude of the cardiovascular composite effect.
The closely corresponding Cox and stratified log-rank P-values, 0.447 and 0.446, are results from different analyses of the same endpoint. Neither P-value is an effect-size measure.
8. Secondary Endpoint Summary
| Endpoint | Population | Effect / test | 95% CI | P-value |
|---|---|---|---|---|
| 50% eGFR reduction | ITT Responder Set | HR 0.779 | 0.573–1.060 | 0.112 |
| Cardio-renal composite | ITT Responder Set | HR 0.801 | 0.642–0.999 | 0.049 |
| Composite renal endpoint, all randomized participants | All randomized participants | HR 0.72 | 0.58–0.89 | 0.002 |
| Cardiovascular composite | ITT Responder Set | HR 0.884 | 0.643–1.215 | 0.447 |
| Endpoint | Stratified log-rank P-value |
|---|---|
| Primary composite renal endpoint | =0.029 |
| 50% eGFR reduction | =0.289 |
| Cardio-renal composite | 0.089 |
| Cardiovascular composite | 0.446 |
The registry reports nine statistical analyses in total, including two formal primary-endpoint analyses and seven secondary analyses. The ClinicalTrials.gov record therefore support a broad statistical description of the registered time-to-event results, but they do not support treating every reported P-value as an independently confirmatory test.
9. Safety
The ClinicalTrials.gov record reports serious adverse-event counts for three labeled analysis contexts. Because the denominators and labels differ, these figures should be presented exactly as reported rather than converted into percentages or treated as directly comparable treatment-group rates.
| Safety context | Serious adverse events | Affected / at risk |
|---|---|---|
| Enrichment Atrasentan | Serious adverse events | 292/5107 |
| Double-Blind Atrasentan | Serious adverse events | 685/1829 |
| Double-Blind Placebo | Serious adverse events | 626/1830 |
10. Statistical Methodology
Cox Proportional-Hazards Model
The primary Cox analysis used a proportional-hazards regression model with prespecified covariate adjustment to estimate the hazard ratio of atrasentan to placebo and its 95% confidence interval. The same model family was used for several secondary time-to-event endpoints.
For a binary treatment indicator, the treatment hazard ratio is represented by exp(β1). The registry's analysis notes specify that prespecified covariates were adjusted for.
The Cox model is particularly useful for time-to-event outcomes because it uses the ordering and timing of observed events while accommodating right censoring. Its treatment effect is expressed as a hazard ratio rather than a difference in proportions at one fixed time.
Hazard Ratio
An HR below 1 indicates a lower estimated instantaneous event rate for atrasentan relative to placebo under the fitted Cox model.
The hazard ratio is not a relative risk. It does not mean that the same percentage of participants avoided the event, and it does not directly provide an absolute risk difference. For SONAR, the HR should be interpreted in conjunction with its confidence interval and with the definition of the analyzed population.
Stratified Log-Rank Test
The registered treatment comparison for the primary endpoint also used a stratified log-rank test. The ClinicalTrials.gov record identifies the method as a stratified analysis but does not provide the specific stratification variables. Therefore, no particular stratification factors are asserted here.
The stratified log-rank approach compares the time-to-event experience between randomized treatment groups while incorporating the registered stratified-analysis framework. It produces a P-value for the treatment comparison but, by itself, does not produce the hazard-ratio estimate shown in the Cox analysis.
Intention-to-Treat Analysis
The registry labels the primary analysis population as an Intent-to-Treat (ITT) Responder Set and explicitly states that the comparison was made "as Randomized." This is important: after the responder criterion is applied, participants remain classified according to their randomized atrasentan or placebo assignment rather than being reassigned according to subsequent exposure.
The term ITT therefore describes the treatment-assignment principle within the specified responder set. It should not be confused with a claim that all 5107 enrolled participants entered the primary efficacy analysis.
Time-to-Event Analysis and Censoring
The primary endpoint was followed from randomization to each participant's individual end of observation, up to 53 months. Time-to-event methods allow participants who do not experience the endpoint during observed follow-up to contribute information until their observation ends.
The ClinicalTrials.gov record does not provide individual censoring times, numbers censored, or a censoring table. Consequently, this page does not attempt to reconstruct survival probabilities or a Kaplan-Meier curve.
11. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The primary endpoint is explicitly a time-to-event outcome, with follow-up beginning at randomization and continuing until the individual end of observation. A Cox model is suited to this structure because it estimates a relative hazard while allowing participants to have different observed follow-up times and providing a hazard-ratio effect measure.
What does HR 0.654 mean?
Within the specified ITT Responder Set and Cox model, HR 0.654 means the estimated instantaneous rate of the first qualifying composite renal event was 34.6% lower for atrasentan than placebo. It does not mean that 34.6% fewer participants experienced the endpoint, and it is not an absolute risk reduction.
Why is the confidence interval important?
The 95% CI of 0.488–0.878 shows the statistical precision around the reported HR 0.654 under the fitted model. It is substantially more informative than the point estimate alone because it shows a range of effect estimates compatible with the data under the stated confidence procedure. It also indicates that the interval excludes HR = 1.
Why are the Cox P-value and log-rank P-value different?
The Cox regression and stratified log-rank test are different statistical procedures. The Cox analysis estimates a hazard ratio and its confidence interval, while the log-rank analysis supplies a treatment-comparison test. For the primary endpoint, the registry reports P = 0.005 from the Cox analysis and P = 0.029 from the stratified log-rank analysis. These values should not be treated as two estimates of the same quantity.
Why does the responder population matter?
The primary efficacy analysis was restricted to participants who achieved at least 30% reduction in UACR during the Enrichment Period. Therefore, the primary HR describes the randomized treatment comparison within this prespecified responder population. It is not automatically generalizable to every participant represented by the overall enrollment figure of 5107.
Why should the HR not be interpreted as a fixed risk ratio?
A hazard ratio compares instantaneous event rates under a Cox model. It is not the ratio of cumulative event probabilities at a particular time. Its interpretation also relies on the proportional-hazards framework. The ClinicalTrials.gov record does not report a separate proportional-hazards diagnostic, so no stronger claim about the constancy of the HR over time should be made.
12. Reading the P-values Carefully
SONAR illustrates why statistical interpretation should not reduce a clinical trial to a list of P-values. The primary Cox result combines three distinct pieces of information:
- Effect estimate: HR 0.654.
- Precision: 95% CI 0.488–0.878.
- Hypothesis-test result: P = 0.005.
The stratified log-rank analysis provides a separate treatment-comparison P-value of 0.029. The secondary endpoints then provide a mixture of hazard-ratio estimates, confidence intervals, and log-rank P-values.
Effect size
The HR describes the estimated relative event rate under the Cox model.
Precision
The confidence interval shows uncertainty around the estimated hazard ratio.
Evidence against the null
The P-value quantifies the result of a specified hypothesis test; it does not measure treatment benefit.
Clinical meaning
Clinical importance cannot be inferred from a P-value alone and requires consideration of the endpoint, magnitude, uncertainty, and population.
13. Comparing the Analysis Populations
One of the most important statistical distinctions in the registry-reported SONAR results is between the responder-set analyses and the analysis reported for all randomized participants.
| Analysis | Population | Reported HR | 95% CI | P-value |
|---|---|---|---|---|
| Primary composite renal endpoint | ITT Responder Set | 0.654 | 0.488–0.878 | =0.005 |
| Composite renal endpoint | All randomized participants | 0.72 | 0.58–0.89 | 0.002 |
These two estimates should not be interpreted as conflicting versions of one statistic. The first is explicitly restricted to participants who achieved the UACR-response criterion during enrichment; the second is reported for all randomized participants. Different analysis populations can produce different estimates even when they concern closely related endpoints.
The analysis-population distinction is especially important when communicating results because the headline enrollment number, 5107, should not be silently substituted for the population actually used in the primary efficacy analysis.
14. What the Results Do and Do Not Establish
The reported primary Cox analysis provides evidence about the relative hazard of the specified composite renal endpoint in the ITT Responder Set under the prespecified-covariate-adjusted Cox model. The HR is below 1 and its 95% confidence interval excludes 1.
That statistical result does not by itself establish an absolute reduction in the probability of renal events, a reduction in mortality, or a particular patient-level probability of benefit. Those would require different quantities or additional reported analyses.
Similarly, the secondary results should be interpreted endpoint by endpoint. For example, the 50% eGFR-reduction analysis reports HR 0.779 with a 95% CI of 0.573–1.060 and P = 0.112, whereas the cardiovascular composite analysis reports HR 0.884 with a 95% CI of 0.643–1.215 and P = 0.447. Neither result should be reduced to a binary "worked" or "failed" label without considering the estimate and its uncertainty.
15. Limitations
- Responder-set primary analysis: the primary efficacy analysis was restricted to participants achieving at least 30% UACR reduction during the Enrichment Period. The resulting estimate therefore applies specifically to that defined analysis population.
- Incomplete population detail: the ClinicalTrials.gov record does not provide the number of participants in the ITT Responder Set by randomized arm, so the page does not invent those counts.
- No individual time-to-event data: the ClinicalTrials.gov record contains summary estimates but not participant-level event and censoring information. Kaplan-Meier curves and absolute event probabilities therefore cannot be reconstructed reliably.
- Stratification details: the ClinicalTrials.gov record identifies use of a stratified log-rank test but does not specify the individual stratification variables.
- Proportional-hazards diagnostics: the ClinicalTrials.gov record does not report a formal assessment of the proportional-hazards assumption. Hazard-ratio interpretation should therefore remain tied to the Cox model rather than being treated as a universal fixed risk ratio over time.
- Multiplicity: the registry supplies multiple secondary endpoint analyses but does not provide enough information here to establish a complete multiplicity-adjustment hierarchy. Individual secondary P-values should therefore not automatically be interpreted as independent confirmatory tests.
- Safety denominators: serious adverse-event counts are reported under different labeled analysis contexts, so they should not be converted into an assumed common randomized safety comparison.
- External generalization: the primary estimate should not automatically be generalized beyond the population and analysis framework represented by the registered ITT Responder Set.
16. Why This Trial Matters Statistically
SONAR is particularly useful for teaching clinical trial statistics because its primary analysis combines several important ideas in one design: randomization, enrichment, an ITT analysis principle, time-to-event endpoints, Cox regression, stratified log-rank testing, covariate adjustment, hazard ratios, and confidence intervals.
The enrichment framework also creates an important interpretive distinction. A conventional randomized trial can often be summarized as a treatment comparison across the randomized population. Here, the primary endpoint was analyzed in a responder set defined by a reduction in UACR during the Enrichment Period. That means the statistical question is conditional on entry into that prespecified responder population.
The trial also demonstrates why multiple representations of an endpoint can be useful. The Cox model provides an interpretable effect estimate and confidence interval, while the stratified log-rank test provides a complementary formal treatment comparison. Looking at both prevents the analysis from becoming dependent on a single numerical summary.
Finally, the difference between the responder-set HR of 0.654 and the all-randomized HR of 0.72 is a practical reminder that analysis population is part of the result. A hazard ratio cannot be interpreted correctly without knowing which participants generated it.
17. Trial Timeline
Trial start
The registered SONAR trial began on 2013-05-17.
Randomized parallel-group study
The trial used randomized allocation, a parallel design, and quadruple masking to compare atrasentan with placebo in diabetic nephropathy.
Responder-set definition
The primary analysis population was defined by achieving at least 30% reduction in UACR during the Enrichment Period, after which participants were analyzed according to randomized assignment.
Primary completion
The registered primary completion date was 2018-03-29.
18. Bottom-Line Statistical Reading
Primary Cox result
95% CI 0.488–0.878 · P = 0.005
Primary composite renal endpoint in the ITT Responder Set, as randomized.
The principal reported analysis estimates a lower hazard of the first qualifying composite renal endpoint for atrasentan relative to placebo among the ITT Responder Set, with HR 0.654 and a two-sided 95% CI of 0.488–0.878. The corresponding stratified log-rank test reports P = 0.029.
The most important qualification is the analysis population: the primary estimate concerns participants who achieved at least 30% UACR reduction during the Enrichment Period. Secondary analyses include both responder-set endpoints and an analysis of all randomized participants, and their effect estimates should be interpreted in their respective populations rather than pooled into one overall statistic.
Explore the statistical methods behind SONAR
Use the related tutorials and calculators to build a deeper understanding of time-to-event analysis, hazard ratios, confidence intervals, and randomized clinical-trial methodology.