This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
HYPERION was a randomized, double-blind, phase 3, parallel-group trial comparing sotatercept plus background pulmonary arterial hypertension (PAH) therapy with placebo plus background PAH therapy in participants with newly diagnosed intermediate- and high-risk pulmonary arterial hypertension.
| Feature | HYPERION |
|---|---|
| Trial name | HYPERION |
| Brief title | Study of Sotatercept in Newly Diagnosed Intermediate- and High-Risk PAH Participants (MK-7962-005/A011-13) |
| Phase | Phase 3 |
| Condition | Pulmonary Arterial Hypertension |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 321 |
| Primary endpoint | Median Time to Clinical Worsening |
| Primary endpoint type | Time-to-event |
| Hypothesis type | Superiority |
| Status | Terminated |
The ClinicalTrials.gov record states that the study started on March 18, 2022 and reached primary completion on April 3, 2025. The lead sponsor was Acceleron Pharma, Inc., a wholly-owned subsidiary of Merck & Co., Inc., Rahway, NJ USA, and the sponsor type was industry.
2. Clinical Question
The central statistical question was whether participants receiving sotatercept plus background PAH therapy experienced a different time to clinical worsening than participants receiving placebo plus background PAH therapy.
Population
Newly diagnosed intermediate- and high-risk participants with pulmonary arterial hypertension.
Intervention
Sotatercept plus background pulmonary arterial hypertension therapy.
Comparator
Placebo plus background pulmonary arterial hypertension therapy.
Primary question
Does sotatercept plus background PAH therapy reduce the hazard of clinical worsening compared with placebo plus background PAH therapy?
This is a time-to-event question rather than simply a comparison of whether an event occurred. The distinction matters because participants can be followed for different lengths of time, and participants who have not yet experienced the event contribute censored follow-up information rather than being treated as if they had no risk.
3. Trial Design
Sotatercept plus background PAH therapy
- Sotatercept
- Background pulmonary arterial hypertension therapy
- Included in the randomized comparison of clinical worsening and secondary outcomes
Placebo plus background PAH therapy
- Placebo
- Background pulmonary arterial hypertension therapy
- Comparator for the randomized efficacy analyses
4. Primary Endpoint
| Endpoint | Registry definition / time frame | Analysis |
|---|---|---|
| Median Time to Clinical Worsening |
Time frame: Up to ~35 months. Time to clinical worsening (TTCW) is defined as time from randomization to the first confirmed morbidity event or death. Clinical worsening events are defined as all-cause death, non-planned PAH-related hospitalization of ≥ 24 hours in duration, atrial septostomy, lung transplant, and/or deterioration in performance in 6-minute walk test from baseline combined with one of the following conditions: worsening of WHO FC from baseline, signs/symptoms of increased right heart failure, addition of a background PAH therapy or change in the background PAH therapy delivery route to parenteral. All events were adjudicated by a blinded, independent committee of clinical experts. The median TTCW is presented. |
Log-rank test; Cox regression model with Efron's method of tie handling; treatment as a covariate; stratification by the randomization stratification factors. |
The registry definition identifies clinical worsening as a composite time-to-event endpoint. Importantly, the analysis is based on the first confirmed morbidity event or death. This means the primary statistical estimand concerns time until the first qualifying event, rather than the total number of events experienced by each participant.
5. Primary Results: Time to Clinical Worsening
The primary analysis compared sotatercept plus background PAH therapy with placebo plus background PAH therapy among all participants who received at least one dose of study intervention. The reported method was a log-rank test, with the analysis text specifying stratification by the randomization stratification factors.
Hazard ratio for clinical worsening or death
95% CI: 0.14–0.41 · P < 0.0001
Superiority hypothesis; two-sided 95% confidence interval.
| Primary analysis feature | Reported value |
|---|---|
| Outcome | Median Time to Clinical Worsening |
| Analysis population | All participants who received at least one dose of study intervention |
| Comparison | Sotatercept plus background PAH therapy vs placebo plus background PAH therapy |
| Method | Log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.24 |
| 95% CI | 0.14–0.41 |
| P-value | <0.0001 |
| Hypothesis | Superiority |
A hazard ratio of 0.24 means that, under the reported time-to-event model, the estimated instantaneous rate of the primary clinical-worsening event was about 24% as high in the sotatercept-plus-background-therapy group as in the placebo-plus-background-therapy group. Expressed as a simple relative-hazard interpretation, this corresponds to an estimated 76% lower hazard because 1 − 0.24 = 0.76.
The HR does not mean that 76% of participants avoided clinical worsening, nor does it mean that every participant experienced exactly a 76% reduction in individual risk. It is a relative time-to-event measure derived from the analysis model.
The 95% confidence interval of 0.14–0.41 describes statistical uncertainty around the estimated hazard ratio. It does not describe the range of treatment effects that individual participants experienced. The entire interval is below 1, which is consistent with a lower estimated hazard in the sotatercept group under the specified analysis.
The reported P < 0.0001 addresses the statistical evidence against the null hypothesis used for the superiority analysis. It does not measure the size of the treatment effect; the hazard ratio and its confidence interval provide the effect estimate and its precision.
The analysis notes are also important: the p-value was described as one-sided and based on a log-rank test stratified by the randomization stratification factors, while the reported confidence interval was two-sided at 95%. The Cox regression model used Efron's method for tied event times and included treatment as a covariate with stratification by the randomization stratification factors.
Finally, the HR interpretation depends on the time-to-event modeling framework. A single hazard ratio summarizes a relative event-rate comparison under the model; it is not itself a direct description of the absolute probability of clinical worsening at a particular month.
Why the primary analysis is a survival-analysis problem
The primary endpoint is not adequately represented by a simple proportion because participants can experience clinical worsening at different times and some participants may not experience the event during the available observation period. Kaplan-Meier methods describe the event-time distribution, the log-rank test compares the distributions between randomized groups, and the Cox model provides a hazard-ratio summary.
The corresponding relative reduction in estimated hazard is 1 − 0.24 = 76%. This is a transformation of the reported hazard ratio, not a separate trial result.
6. Secondary Endpoint Results
The registry extract contains seven secondary statistical analyses. These cover binary multicomponent and risk-score endpoints, continuous changes in NT-proBNP and 6MWD, and overall survival.
6.1 Multicomponent Improvement Endpoint
The percentage of participants achieving the multicomponent improvement endpoint of 6-minute walk distance (6MWD), N-terminal Prohormone B-type Natriuretic Peptide (NT-ProBNP), and WHO FC was assessed at baseline and Week 24.
Percentage-point difference
95% CI: 5.1–23.9 · P = 0.003
Cochran-Mantel-Haenszel analysis stratified by the randomization stratification factors.
The analysis population consisted of all participants who received at least one dose of study intervention and had baseline and Week 24 data for 6MWD, NT-ProBNP and WHO FC. The reported effect measure was a percentage-point difference, with the confidence interval based on the Miettinen and Nurminen method stratified by the randomization stratification factors.
A percentage-point difference of 14.5 is an absolute difference between the proportions meeting the multicomponent improvement definition in the two randomized groups. It is not a relative percentage increase and does not mean that an individual participant's probability changed by exactly 14.5 percentage points.
The 95% CI of 5.1–23.9 quantifies uncertainty around the estimated between-group percentage-point difference. The reported P = 0.003 addresses the hypothesis test; it does not indicate that the effect size is 0.003 or that the result is clinically important solely because of its p-value.
6.2 REVEAL Lite 2 Risk Score
Maintained or achieved a low REVEAL Lite 2 risk score
95% CI: 0.8–23.5 · P = 0.036
Cochran-Mantel-Haenszel analysis stratified by the randomization stratification factors.
The endpoint was measured at baseline and Week 24. The analysis population included participants who received at least one dose and had data used to calculate the REVEAL Lite 2 risk score at both time points. The reported analysis used the Cochran-Mantel-Haenszel method, with the Miettinen and Nurminen method used for the confidence interval.
The estimate of 12.3 percentage points represents an absolute between-group difference in the percentage of participants who maintained or achieved a low REVEAL Lite 2 risk score. The confidence interval, 0.8–23.5, indicates appreciable uncertainty in the size of that difference.
The P = 0.036 result is a hypothesis-test result. It should not be interpreted as a 3.6% probability that the null hypothesis is true, nor as a measure of the magnitude of the observed difference.
6.3 Simplified French Risk Score
Maintained or achieved a low Simplified French Risk Score
95% CI: −3.5–13.6 · P = 0.246
Cochran-Mantel-Haenszel analysis stratified by the randomization stratification factors.
The analysis included participants who received at least one dose and had data used to determine the Simplified French Risk Score at baseline and Week 24.
The estimated difference was 5.0 percentage points. The 95% CI extends from −3.5 to 13.6, so the interval includes zero. In descriptive terms, the reported data are compatible with a range of between-group differences that includes no difference.
The reported P = 0.246 is not evidence that the two groups are identical. A non-significant hypothesis test does not prove equality; it indicates that the specified analysis did not provide sufficiently strong statistical evidence against its null hypothesis.
6.4 Median Change From Baseline in NT-proBNP
Hodges-Lehmann location shift
95% CI: −434.84 to −182.37 · P = 0.001
Aligned rank stratified Wilcoxon analysis stratified by the randomization stratification factors.
The endpoint was median change from baseline in NT-proBNP levels at Week 24. The analysis population included all participants who received at least one dose and had baseline and Week 24 NT-proBNP data. The reported method was Wilcoxon (Mann-Whitney), with a Hodges-Lehmann location-shift estimate and confidence interval.
The reported Hodges-Lehmann location shift of −308.60 pg/mL summarizes the between-group difference in the distribution of change using the nonparametric analysis framework. The negative direction indicates a lower change in the sotatercept-plus-background-therapy group relative to the comparator under the reported contrast.
The 95% CI of −434.84 to −182.37 provides the corresponding uncertainty interval for the location-shift estimate. Because the interval remains below zero, the reported statistical comparison is directionally separated from zero.
The P = 0.001 is evidence from the specified nonparametric hypothesis test; it is not a direct measure of how large the NT-proBNP difference is.
6.5 WHO Functional Class
Improved WHO FC or maintained WHO FC II at 24 weeks
95% CI: 4.7–26.3 · P = 0.005
Cochran-Mantel-Haenszel analysis stratified by the randomization stratification factors.
The analysis population consisted of participants who received at least one dose and had WHO FC data at baseline and Week 24. The registry reports the Miettinen and Nurminen method stratified by the randomization factors for the percentage-point comparison.
The estimate of 15.6 percentage points represents an absolute difference in the percentage of participants who improved in WHO FC or maintained WHO FC II at 24 weeks from baseline. The 95% CI of 4.7–26.3 describes uncertainty around that absolute difference.
The P = 0.005 is the reported superiority-test result. It should not be converted into a probability that the treatment works, and it does not quantify clinical importance.
6.6 Median Change From Baseline in 6-Minute Walk Distance
Hodges-Lehmann location shift
95% CI: 4.65–38.24 · P = 0.013
Aligned rank stratified Wilcoxon analysis using the randomization stratification factors.
The endpoint measured median change from baseline in six-minute walk distance at Week 24. The analysis included participants who received at least one dose and had 6MWD data at baseline and Week 24.
The reported Hodges-Lehmann location shift of 21.40 meters summarizes the between-group difference in the distribution of change in 6MWD under the reported nonparametric method. The 95% CI of 4.65–38.24 indicates uncertainty around that estimate while remaining above zero.
The P = 0.013 is evidence from the specified hypothesis test. It does not mean that the probability of observing the effect is 98.7%, nor does it establish the clinical importance of a 21.40-meter location shift by itself.
6.7 Overall Survival
Hazard ratio for overall survival
95% CI: 0.31–2.81 · P = 0.453
Reported as a superiority analysis with stratification by the randomization stratification factors.
The overall-survival analysis included all participants who received at least one dose of study intervention and used a time frame of up to ~35 months. The ClinicalTrials.gov record reports a hazard ratio as the effect measure. Its analysis notes describe a one-sided p-value based on a stratified log-rank test and a Cox regression model with Efron's method of tie handling.
An HR of 0.94 is close to 1, so the estimated relative hazard difference between the randomized groups was comparatively small under this model. However, the confidence interval is wide: 0.31–2.81. This interval includes 1 and therefore encompasses both a potentially lower and a potentially higher hazard under the model.
The reported P = 0.453 does not establish equality of overall survival. It indicates that the specified superiority analysis did not provide strong statistical evidence against its null hypothesis.
This result also illustrates why an HR should always be considered together with its confidence interval. A point estimate near 1 with a very wide interval does not imply that the true effect is known to be near 1.
7. Secondary Results Summary
| Endpoint | Effect measure | Estimate | 95% CI | P-value | Method |
|---|---|---|---|---|---|
| Multicomponent improvement: 6MWD, NT-ProBNP, WHO FC | Percentage point difference | 14.5 | 5.1–23.9 | 0.003 | Cochran-Mantel-Haenszel |
| Maintained or achieved low REVEAL Lite 2 risk score | Percentage point difference | 12.3 | 0.8–23.5 | 0.036 | Cochran-Mantel-Haenszel |
| Maintained or achieved low Simplified French Risk Score | Percentage point difference | 5.0 | −3.5–13.6 | 0.246 | Cochran-Mantel-Haenszel |
| Median change in NT-proBNP | Hodges-Lehmann location shift | −308.60 pg/mL | −434.84 to −182.37 | 0.001 | Wilcoxon / Mann-Whitney |
| Improved WHO FC or maintained WHO FC II | Percentage point difference | 15.6 | 4.7–26.3 | 0.005 | Cochran-Mantel-Haenszel |
| Median change in 6MWD | Hodges-Lehmann location shift | 21.40 meters | 4.65–38.24 | 0.013 | Wilcoxon / Mann-Whitney |
| Overall survival | Hazard ratio | 0.94 | 0.31–2.81 | 0.453 | Wilcoxon / Mann-Whitney as reported; analysis notes specify stratified log-rank and Cox regression |
The secondary endpoints use different estimands and therefore should not be collapsed into a single summary number. Percentage-point differences describe absolute differences in binary outcomes, Hodges-Lehmann location shifts summarize distributional differences for the nonparametric continuous analyses, and the hazard ratio describes a relative time-to-event comparison.
8. Statistical Methodology
Log-rank testing for the primary time-to-event endpoint
The primary clinical-worsening analysis used a log-rank test. The registry-reported analysis notes specify that the log-rank analysis was stratified by the randomization stratification factors. Stratification allows the comparison to account for the factors used in randomization rather than treating all participants as belonging to one completely homogeneous risk set.
Cox regression for the hazard ratio
The primary analysis notes state that the hazard ratio was based on a Cox regression model with Efron's method of tie handling. Treatment was included as a covariate and the model was stratified by the randomization stratification factors.
For the primary analysis, HR = 0.24. The corresponding simple relative-hazard interpretation is a 76% lower estimated hazard, calculated as 1 − 0.24.
Cochran-Mantel-Haenszel analysis
The binary secondary endpoints were analyzed using the Cochran-Mantel-Haenszel test, stratified by the randomization stratification factors. This approach provides a way to compare categorical outcomes while accounting for the specified strata.
For the percentage-point estimates, the registry analysis notes specify the Miettinen and Nurminen method stratified by the randomization stratification factors. This is relevant because the confidence interval for an absolute difference is not simply obtained by treating the two observed percentages as unrelated quantities.
Wilcoxon / Mann-Whitney analysis
The analyses of median change from baseline in NT-proBNP and 6MWD used the Wilcoxon (Mann-Whitney) framework. The reported effect measure was the Hodges-Lehmann location shift. The analysis notes specify an aligned rank stratified Wilcoxon approach using the randomization stratification factors.
One-sided and two-sided inference
The primary clinical-worsening analysis is unusual enough to deserve explicit attention: the registry-reported analysis notes describe a one-sided p-value from the stratified log-rank test, while the reported confidence interval is two-sided at 95%. The same one-sided p-value convention is described in the overall-survival analysis notes.
A one-sided hypothesis test and a two-sided confidence interval are not interchangeable descriptions of the same inferential quantity. The p-value addresses the prespecified direction of the hypothesis test; the confidence interval provides a two-sided interval around the estimated effect.
9. Stratification and Covariate Adjustment
The ClinicalTrials.gov record repeatedly identify randomization stratification factors as part of the analysis. The exact factors themselves are not included in the ClinicalTrials.gov record, so they are not specified here.
Why stratify?
Stratified analysis preserves the distinction between participants assigned through different randomization strata and can improve alignment between the analysis and the randomization scheme.
What was adjusted?
The registry analysis text states that treatment was a covariate in the Cox model, with the model stratified by the randomization stratification factors.
This distinction matters statistically. A stratified Cox model does not necessarily estimate a separate treatment coefficient for every stratum. Instead, the baseline hazard can differ across strata while the treatment effect is summarized through the model's treatment coefficient.
10. Censoring and Time-to-Event Interpretation
Clinical worsening and overall survival are time-to-event endpoints. Participants who do not experience the relevant event during their observed follow-up do not simply disappear from the analysis. Their available follow-up contributes information until the point at which they are censored.
Kaplan-Meier estimation uses the observed event and censoring times to estimate the event-free survival function. The registry-reported HYPERION analysis data report log-rank and Cox methods, but do not provide Kaplan-Meier estimates or a survival curve for reproduction on this page.
For a clinical-worsening composite, the event definition also matters. The analysis concerns the first confirmed morbidity event or death. A participant can therefore contribute time to the risk set until the first qualifying event, censoring, or the end of available follow-up.
11. Statistical Methods Explained
Why was a log-rank test used?
The primary endpoint is time to clinical worsening or death, so the relevant information includes both whether an event occurred and when it occurred. The log-rank test is designed to compare time-to-event distributions between groups while incorporating censored observations.
What does an HR of 0.24 mean?
It means that the estimated instantaneous event rate under the reported model was 0.24 times that of the comparator. A convenient descriptive transformation is a 76% lower estimated hazard. It does not mean a 76% reduction in every participant's individual probability of an event.
Why does the confidence interval matter?
The point estimate alone can give a false sense of precision. The primary 95% CI is 0.14–0.41, showing the range of hazard-ratio values compatible with the statistical uncertainty represented by that interval. The interval is about the estimated treatment effect, not individual patient outcomes.
Why use a Cochran-Mantel-Haenszel test?
Several secondary endpoints are binary. The Cochran-Mantel-Haenszel framework permits comparison of the treatment groups while stratifying on the randomization stratification factors. This makes the binary analysis consistent with the trial's stratified design.
What is a Hodges-Lehmann location shift?
For the NT-proBNP and 6MWD analyses, the reported effect measure is a Hodges-Lehmann location shift rather than a conventional arithmetic mean difference. It summarizes a distributional difference within the nonparametric Wilcoxon framework and is accompanied by a confidence interval.
Why can a non-significant p-value not prove equality?
For the Simplified French Risk Score, the estimate was 5.0 percentage points, with a 95% CI of −3.5–13.6 and P = 0.246. The result does not establish that the groups are identical. Rather, the confidence interval includes zero and the specified hypothesis test did not provide strong statistical evidence against its null hypothesis.
12. Understanding the Primary Hazard Ratio
The primary HR of 0.24 is a relative time-to-event estimate comparing sotatercept plus background PAH therapy with placebo plus background PAH therapy. Under the reported Cox-model framework, the estimated instantaneous rate of clinical worsening or death was substantially lower in the sotatercept group.
The HR does not provide the percentage of participants who avoided clinical worsening, the absolute difference in event-free survival at a particular time, or the treatment effect experienced by any particular participant.
The interval of 0.14–0.41 communicates the precision of the estimated hazard ratio. It is considerably more informative than the point estimate alone because it shows how much statistical uncertainty surrounds the estimate.
The reported P < 0.0001 addresses the hypothesis test. A p-value is not an effect-size metric and should not be interpreted as the probability that the treatment has a particular effect size.
13. Overall Survival: A Different Time-to-Event Question
Overall survival was a secondary endpoint with a time frame of up to ~35 months. The reported hazard ratio was 0.94, with a 95% CI of 0.31–2.81 and P = 0.453.
| Feature | Overall survival analysis |
|---|---|
| Endpoint role | Secondary |
| Time frame | Up to ~35 months |
| Analysis population | All participants who received at least one dose of study intervention |
| Effect measure | Hazard ratio |
| Estimate | 0.94 |
| 95% CI | 0.31–2.81 |
| P-value | 0.453 |
| Hypothesis | Superiority |
Statistically, the contrast between the primary endpoint and overall survival is instructive. The primary endpoint concerns a composite of clinical-worsening events or death, while overall survival considers death itself. These endpoints therefore measure different clinical processes and can produce different estimates even within the same randomized comparison.
The wide OS confidence interval is particularly important. The interval extends below and above 1, so the ClinicalTrials.gov record does not establish a precise direction or magnitude of an overall-survival effect. This is a good example of why the confidence interval should be read alongside the point estimate and p-value.
14. Multiplicity and the Secondary Endpoints
The trial has one registered primary endpoint and multiple posted secondary analyses. The ClinicalTrials.gov record identifies the secondary analyses as superiority hypotheses, but do not provide an alpha-allocation or multiplicity-adjustment procedure for the collection of secondary endpoints.
| Endpoint role | Number in the ClinicalTrials.gov record | Statistical interpretation |
|---|---|---|
| Primary endpoint | 1 | Primary confirmatory comparison reported in the registry analysis field |
| Secondary analyses | 7 | Additional hypothesis tests and estimates; the ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure across these analyses |
| Statistical analyses posted | 8 | One primary analysis plus seven secondary analyses in the registry-reported extract |
This does not erase the effect estimates or their confidence intervals. It simply distinguishes the descriptive and inferential information provided by each analysis from a stronger claim that every secondary endpoint was independently confirmatory.
15. Early Termination and Clinical Equipoise
The registry caveat states that the study was stopped early due to loss of clinical equipoise based on robust evidence of clinical benefit of sotatercept demonstrated in previous studies.
What this means statistically
The available observation period and information accumulation were affected by the decision to stop the study early. Early stopping is therefore part of the context for interpreting the reported estimates.
What it does not mean
The registry statement does not provide a separate numerical adjustment to the reported hazard ratio, confidence interval, or p-value in the ClinicalTrials.gov record.
The stated reason for termination also distinguishes the trial's registry status from a simple administrative completion. The ClinicalTrials.gov record explicitly attributes the decision to loss of clinical equipoise based on evidence from previous studies.
16. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm using affected participants over the corresponding at-risk denominator.
| Safety measure | Sotatercept | Placebo |
|---|---|---|
| Serious adverse events, affected / at risk | 39 / 160 | 45 / 160 |
The ClinicalTrials.gov record is counts of affected participants and participants at risk. They should not be confused with the efficacy analysis population, which was defined as all participants who received at least one dose of study intervention. The ClinicalTrials.gov record does not provide a formal statistical comparison or confidence interval for serious adverse events, so this page does not construct one.
17. Analysis Populations
| Analysis | Population specified in the ClinicalTrials.gov record |
|---|---|
| Primary clinical worsening | All participants who received at least one dose of study intervention |
| Multicomponent improvement endpoint | All participants who received at least one dose and had baseline and Week 24 data for 6MWD, NT-ProBNP and WHO FC |
| REVEAL Lite 2 risk score | All participants who received at least one dose and had data used to calculate the score at baseline and Week 24 |
| Simplified French Risk Score | All participants who received at least one dose and had data used to determine the score at baseline and Week 24 |
| NT-proBNP | All participants who received at least one dose and had baseline and Week 24 NT-proBNP data |
| WHO FC | All participants who received at least one dose and had WHO FC data at baseline and Week 24 |
| 6MWD | All participants who received at least one dose and had baseline and Week 24 6MWD data |
| Overall survival | All participants who received at least one dose of study intervention |
The variation in analysis populations is important. A participant must have the required measurements to contribute to a particular Week 24 secondary endpoint. Consequently, the population for a secondary analysis can differ from the broader treated population even though all analyses originate from the same randomized trial.
18. What the Different Effect Measures Mean
Hazard ratio
Used for the primary time-to-clinical-worsening endpoint and overall survival. It summarizes a relative time-to-event comparison under the reported Cox model.
Percentage-point difference
Used for binary secondary endpoints. It expresses the absolute difference between the percentages of participants meeting the endpoint.
Hodges-Lehmann location shift
Used for the nonparametric analyses of NT-proBNP and 6MWD. It summarizes a distributional location difference within the Wilcoxon framework.
Confidence interval
Provides an uncertainty interval around the corresponding effect estimate. Its interpretation depends on the effect measure and statistical method used.
These measures should not be compared as though they were interchangeable scales. An HR of 0.24, a percentage-point difference of 14.5, and a Hodges-Lehmann shift of −308.60 pg/mL answer different statistical questions.
19. Limitations
- Early termination: the study was stopped early due to loss of clinical equipoise based on robust evidence of clinical benefit of sotatercept demonstrated in previous studies. This is an important part of the context for interpreting the available follow-up and estimates.
- Composite primary endpoint: clinical worsening combines several possible events, including death and specified morbidity events. A treatment effect on the composite does not by itself establish that every component is affected to the same degree.
- Analysis population: the primary analysis used all participants who received at least one dose rather than an explicitly stated all-randomized ITT population in the ClinicalTrials.gov record.
- Secondary endpoint populations: Week 24 analyses required availability of the relevant baseline and Week 24 measurements, so their populations are more restricted than the full enrolled population.
- Multiplicity: the ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure across the seven reported secondary statistical analyses.
- Hazard-ratio interpretation: an HR is a model-based relative time-to-event measure and should not be interpreted as an absolute risk difference or a direct probability of benefit for an individual.
- Wide OS uncertainty: the reported OS HR of 0.94 has a 95% CI of 0.31–2.81, indicating substantial uncertainty around the estimated relative hazard.
- Safety comparison: the registry-reported serious-adverse-event data provide affected/at-risk counts but no formal comparative confidence interval or p-value.
- Registry detail: the registry-reported endpoint definition for clinical worsening ends with a colon after describing deterioration in 6-minute walk test from baseline combined with one of the following conditions. No additional conditions are included in the ClinicalTrials.gov record, so none are added here.
20. Why This Trial Matters Statistically
HYPERION is a useful statistical teaching case because the trial combines a composite time-to-event primary endpoint with stratified survival analysis, a Cox hazard-ratio model, multiple binary secondary outcomes, nonparametric analyses of continuous changes, and an overall-survival endpoint.
| Concept | How it appears in HYPERION |
|---|---|
| Randomization | Randomized allocation to two parallel treatment groups |
| Double blinding | Double-masked trial design |
| Time-to-event endpoint | Time to clinical worsening or death |
| Log-rank test | Primary comparison of time to clinical worsening |
| Hazard ratio | Primary effect measure; also used for overall survival |
| Cox regression | Model underlying the reported primary hazard ratio |
| Stratified analysis | Used across the reported survival and categorical analyses |
| Cochran-Mantel-Haenszel test | Binary secondary endpoints |
| Wilcoxon / Mann-Whitney | NT-proBNP, 6MWD, and the reported overall-survival method field |
| Hodges-Lehmann estimate | Effect measure for the NT-proBNP and 6MWD nonparametric analyses |
| Confidence intervals | Reported for the primary and secondary effect estimates |
| One-sided testing | Specified in the primary and overall-survival analysis notes |
| Multiplicity | Multiple secondary analyses require distinction between individual tests and an overall error-control strategy |
| Early termination | Study stopped due to loss of clinical equipoise according to the registry caveat |
21. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The primary randomized comparison produced a hazard ratio of 0.24 with a 95% CI of 0.14–0.41 and a reported P < 0.0001 under the specified stratified log-rank/Cox analysis framework.
Clinical interpretation
The primary endpoint was defined around time to clinical worsening or death. The secondary analyses additionally assessed functional, biomarker, risk-score, and overall-survival outcomes. Each endpoint measures a different aspect of the clinical course.
A statistically strong result on one endpoint should therefore not be treated as a substitute for examining the other endpoints. The primary clinical-worsening endpoint, overall survival, functional measures, biomarkers, risk scores, and safety outcomes have different meanings and different statistical estimands.
22. Longitudinal Trial History
Study start
The HYPERION study began according to the ClinicalTrials.gov record.
Randomized phase 3 evaluation
The trial enrolled 321 participants and compared sotatercept plus background PAH therapy with placebo plus background PAH therapy in a randomized, double-blind, parallel design.
Early termination
The study was stopped early due to loss of clinical equipoise based on robust evidence of clinical benefit of sotatercept demonstrated in previous studies.
Primary completion
The ClinicalTrials.gov record lists April 3, 2025 as the primary completion date.
23. A Closer Look at the Primary Endpoint
The phrase time to clinical worsening contains several statistical ideas that are easy to overlook.
Time matters
Two participants can both experience clinical worsening but contribute very different amounts of event-free follow-up before the event. A time-to-event method retains that timing information.
Censoring matters
A participant who has not experienced clinical worsening by the end of observed follow-up is not equivalent to a participant who will never experience clinical worsening. Survival methods treat the available follow-up differently from an observed event.
The composite definition matters
The primary endpoint combines morbidity events and death. A hazard ratio for the composite therefore summarizes the treatment comparison for the first qualifying component of the composite, not a separate effect estimate for every component.
The analysis population matters
The registry-reported primary analysis population was all participants who received at least one dose of study intervention. This should be stated alongside the HR rather than assuming that every trial's time-to-event estimate automatically represents an all-randomized population.
24. Reading the Secondary Results as a Set
The secondary analyses provide a useful example of why clinical-trial interpretation should be multidimensional.
| Domain | Reported result | Statistical perspective |
|---|---|---|
| Multicomponent improvement | 14.5 percentage-point difference; 95% CI 5.1–23.9; P = 0.003 | Absolute binary-outcome contrast |
| REVEAL Lite 2 | 12.3 percentage-point difference; 95% CI 0.8–23.5; P = 0.036 | Absolute binary-outcome contrast |
| Simplified French Risk Score | 5.0 percentage-point difference; 95% CI −3.5–13.6; P = 0.246 | Interval includes zero |
| NT-proBNP | −308.60 pg/mL; 95% CI −434.84 to −182.37; P = 0.001 | Nonparametric location-shift estimate |
| WHO FC | 15.6 percentage-point difference; 95% CI 4.7–26.3; P = 0.005 | Absolute binary-outcome contrast |
| 6MWD | 21.40 meters; 95% CI 4.65–38.24; P = 0.013 | Nonparametric location-shift estimate |
| Overall survival | HR 0.94; 95% CI 0.31–2.81; P = 0.453 | Time-to-event relative effect with wide uncertainty |
The results are not contradictory merely because they use different scales or have different p-values. Each analysis answers a different question. A statistically significant change in a biomarker, for example, is not the same estimand as a hazard ratio for clinical worsening or overall survival.
25. Related Tutorials
Learn more about the methods used in this trial:
26. Related Statistical Calculators
27. Sources
- ClinicalTrials.gov: HYPERION, NCT04811092.
- PubMed: PubMed record for PMID 41025556.
Continue through the Clinical Biostats statistical tutorials
Use the trial's endpoints and methods as a practical route into survival analysis, stratified testing, nonparametric methods, confidence intervals, and clinical-trial interpretation.
28. Record Summary
HYPERION provides a detailed example of a randomized phase 3 trial in which the primary question is fundamentally a time-to-event problem. The registry analysis reports a hazard ratio of 0.24 for time to clinical worsening, with a 95% confidence interval of 0.14–0.41 and P < 0.0001. The analysis used a stratified log-rank test and a Cox regression model with Efron's method of tie handling.
The secondary analyses demonstrate several other statistical frameworks: Cochran-Mantel-Haenszel testing for binary endpoints, Miettinen and Nurminen confidence intervals for percentage-point differences, and aligned rank stratified Wilcoxon analyses with Hodges-Lehmann location-shift estimates for NT-proBNP and 6MWD. Overall survival was analyzed as a secondary time-to-event endpoint and produced an HR of 0.94 with a 95% CI of 0.31–2.81 and P = 0.453.
The most important statistical lesson is that these results should be interpreted according to their estimand, analysis population, confidence interval, hypothesis-test framework, and endpoint definition. The primary clinical-worsening result, secondary functional and biomarker results, overall survival, and serious-adverse-event counts describe different dimensions of the trial and should not be reduced to a single statistic.