This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. View NCT00853658 on ClinicalTrials.gov.
1. Trial at a Glance
ATMOSPHERE was a randomized, parallel, triple-masked phase 3 trial in chronic heart failure. The study enrolled 7064 participants and evaluated three treatment groups involving aliskiren and enalapril. The registered primary endpoint was a time-to-event composite of cardiovascular death or heart failure hospitalization.
| Feature | ATMOSPHERE |
|---|---|
| Trial name | ATMOSPHERE |
| Brief title | Efficacy and Safety of Aliskiren and Aliskiren/Enalapril Combination on Morbidity-mortality in Patients With Chronic Heart Failure |
| Phase | Phase 3 |
| Condition | Chronic Heart Failure |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Triple |
| Primary purpose | Treatment |
| Enrollment | 7064 |
| Primary endpoint | First occurrence of the composite endpoint defined as either cardiovascular death or heart failure hospitalization |
| Primary endpoint time frame | Up to End of Study (78 months) |
| Primary analysis | Cox proportional-hazards model |
| Primary effect measure | Hazard ratio |
| Results posted | Yes |
| Statistical analyses posted | 4 |
| Lead sponsor | Novartis Pharmaceuticals |
2. Clinical Question
The ATMOSPHERE trial addressed whether treatment with aliskiren alone or the combination of aliskiren and enalapril affected the time to the first occurrence of a composite endpoint consisting of cardiovascular death or heart failure hospitalization in patients with chronic heart failure.
Population
Patients with chronic heart failure enrolled in the phase 3 ATMOSPHERE study.
Interventions
Aliskiren and the combination of aliskiren and enalapril were the active intervention strategies represented in the posted analyses.
Comparator
Enalapril served as the comparator in the posted primary analyses.
Primary question
How do the aliskiren/enalapril combination and aliskiren alone compare with enalapril for the first occurrence of cardiovascular death or heart failure hospitalization?
3. Trial Design
Combination
- Aliskiren and enalapril combination
- Compared directly with enalapril in the posted primary analyses
Aliskiren
- Aliskiren
- Compared directly with enalapril in the posted primary analyses
Enalapril
- Enalapril
- Comparator for both posted treatment comparisons
From a statistical perspective, the three-arm parallel structure is important because the primary endpoint can support more than one prespecified comparison. The posted analyses show two different inferential questions involving aliskiren versus enalapril—one framed as non-inferiority or equivalence and another as superiority—as well as a superiority comparison of the combination with enalapril.
4. Endpoints
| Endpoint | Registry definition | Time frame | Endpoint type |
|---|---|---|---|
| Primary composite endpoint | Number of participants that had first occurrence of the composite endpoint, which is defined as either cardiovascular (CV) death or heart failure (HF) hospitalization due to HF. | Up to End of Study (78 months) | Time-to-event |
The endpoint is a first-event composite. Statistically, each participant is followed until the first qualifying component event, another prespecified censoring point, or the end of available follow-up. The registry's posted formal analyses treat the endpoint as a time-to-event outcome rather than simply comparing the proportion of participants who eventually experienced an event.
5. Analysis Population
The posted primary analyses used the Full Analysis Set (FAS). The registry defines this as all randomized patients with the exception of mis-randomized patients that took no study drug and patients from sites with major GCP violations.
| Analysis population | Registry definition | Role |
|---|---|---|
| Full Analysis Set | All randomized patients with exception of mis-randomized patients that took no study drug and patients from the sites with major GCP violations. | Primary efficacy analyses |
This distinction is important because the analysis population is not simply described as every person initially enrolled. The posted Cox analyses were based on the registry-defined FAS, which begins from randomized participants but excludes the specified categories.
6. Statistical Methodology
Time-to-event analysis
The primary endpoint was analyzed as a time-to-event outcome. This is appropriate because the endpoint concerns the first occurrence of cardiovascular death or heart failure hospitalization and the follow-up extends to the end of the study.
A time-to-event analysis uses both whether an event occurred and when it occurred. Participants who do not experience the endpoint during the period in which they are observed contribute information through their censoring time rather than being treated as if their eventual event status were known.
Cox proportional-hazards model
The posted statistical method was regression using a Cox model, normalized here as a Cox proportional-hazards model. The effect measure was the hazard ratio.
For a treatment indicator, the exponentiated treatment coefficient, exp(β), corresponds to the estimated hazard ratio between the treatment groups under the model.
Hazard ratio
The hazard ratio compares the estimated instantaneous event rate between two groups within the Cox modeling framework. An HR below 1 corresponds to a lower estimated hazard in the numerator treatment group relative to its comparator; an HR above 1 corresponds to a higher estimated hazard.
HR < 1 → lower estimated hazard in the treatment group
HR > 1 → higher estimated hazard in the treatment group
The hazard ratio is not an absolute risk difference, probability of hospitalization, probability of death, or percentage of patients who benefit.
Intention-to-treat principle
The analysis text associated with the posted primary analyses identifies intention-to-treat analysis as an analysis concept. In a randomized trial, the intention-to-treat principle preserves comparison according to randomized assignment rather than redefining treatment groups based on subsequent treatment behavior.
For ATMOSPHERE, the registry-defined FAS is the population used for the posted primary analyses. The precise FAS exclusions should therefore be kept in view when interpreting the phrase "intention-to-treat" rather than assuming that the posted analysis necessarily includes every enrolled participant without exception.
Non-inferiority and superiority are different questions
The posted analyses contain both superiority hypotheses and a non-inferiority or equivalence hypothesis. These should not be treated as interchangeable tests.
For superiority, the question is whether the data provide evidence that the treatment effect differs from the null value in the direction specified by the hypothesis. For a hazard ratio, the conventional null value is 1.
For non-inferiority, the question is instead whether the treatment can be ruled out as being worse than a prespecified amount. The registry specifically reports a non-inferiority margin of 1.104 for the aliskiren-versus-enalapril comparison.
7. Primary Results: Aliskiren/Enalapril Combination vs Enalapril
The first posted primary analysis compared the aliskiren/enalapril combination with enalapril for the first occurrence of the cardiovascular death or heart failure hospitalization composite endpoint through the end of study.
Combination vs Enalapril
95% CI: 0.85–1.03 · P = 0.1724
Hypothesis type: Superiority
| Element | Posted result |
|---|---|
| Endpoint | First occurrence of cardiovascular death or heart failure hospitalization |
| Time frame | Up to End of Study (78 months) |
| Groups compared | Combination Aliskiren / Enalapril vs Enalapril |
| Analysis population | Full Analysis Set |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.93 |
| 95% CI | 0.85–1.03 |
| P-value | 0.1724 |
| Hypothesis | Superiority |
An HR of 0.93 means that the fitted Cox model estimated a lower instantaneous rate of first cardiovascular death or heart failure hospitalization in the combination group than in the enalapril group, with the estimated hazard approximately 7% lower on a relative hazard scale.
It does not mean that 7% fewer participants experienced the endpoint, nor does it mean that each participant had exactly a 7% lower probability of an event. A hazard ratio is a model-based relative measure for the time-to-event process.
The 95% CI of 0.85–1.03 expresses uncertainty around the estimated hazard ratio. Because the interval includes 1, the posted confidence interval does not exclude equal hazards under the usual two-sided confidence-interval interpretation.
The p-value of 0.1724 addresses the statistical evidence against the superiority null hypothesis; it is not a measure of effect size. The HR and its confidence interval describe the estimated magnitude and precision, while the p-value addresses a hypothesis-testing question.
As with any Cox analysis, interpretation also depends on the proportional-hazards model being a reasonable description of the relative hazards over follow-up. The posted registry data do not provide enough information here to independently assess that assumption.
8. Primary Results: Aliskiren vs Enalapril — Non-Inferiority Analysis
A second primary analysis compared aliskiren with enalapril using a non-inferiority or equivalence hypothesis. The registry reports a prespecified non-inferiority margin of 1.104.
Aliskiren vs Enalapril · Non-Inferiority
95% CI: 0.90–1.10 · P = 0.0368
Prespecified non-inferiority margin: 1.104
| Element | Posted result |
|---|---|
| Endpoint | First occurrence of cardiovascular death or heart failure hospitalization |
| Time frame | Up to End of Study (78 months) |
| Groups compared | Aliskiren vs Enalapril |
| Analysis population | Full Analysis Set |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.99 |
| 95% CI | 0.90–1.10 |
| P-value | 0.0368 |
| Hypothesis | Non-inferiority or equivalence |
| Non-inferiority margin | 1.104 |
An HR of 0.99 is very close to 1. On the relative hazard scale, the estimated hazard of first cardiovascular death or heart failure hospitalization was approximately the same in the two groups, with the point estimate corresponding to an estimated 1% lower hazard for aliskiren.
The important feature of this analysis is the non-inferiority margin, not simply whether a conventional p-value crosses 0.05. The upper confidence limit was 1.10, while the prespecified non-inferiority margin was 1.104. Thus, the reported confidence interval remains below the stated margin.
This is the logic of non-inferiority testing: the analysis asks whether sufficiently large harm can be ruled out relative to a clinically prespecified threshold. The margin is therefore part of the definition of the statistical question.
The p-value of 0.0368 should not be interpreted as saying that the treatment effect is "3.68%" or that there is a 96.32% probability that the treatment is non-inferior. A p-value does not provide either interpretation.
The 95% CI of 0.90–1.10 also shows that the data are compatible with effects ranging from a lower estimated hazard to a modestly higher estimated hazard, while remaining below the specified non-inferiority boundary.
9. Primary Results: Aliskiren vs Enalapril — Non-Diabetic Patients
The registry also posts a primary-endpoint Cox analysis for non-diabetic patients, comparing aliskiren with enalapril. This analysis is labeled as a superiority analysis in the registry.
Aliskiren vs Enalapril · Non-Diabetic Patients
95% CI: 0.85–1.07 · P = 0.4579
Hypothesis type: Superiority
| Element | Posted result |
|---|---|
| Endpoint | First occurrence of cardiovascular death or heart failure hospitalization |
| Time frame | Up to End of Study (78 months) |
| Groups compared | Aliskiren vs Enalapril |
| Analysis population | Full Analysis Set |
| Analysis subset | Non-Diabetic patients |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.96 |
| 95% CI | 0.85–1.07 |
| P-value | 0.4579 |
| Hypothesis | Superiority |
An HR of 0.96 corresponds to an estimated 4% lower hazard in the aliskiren group relative to enalapril within the fitted model.
The 95% CI of 0.85–1.07 includes 1, indicating that the posted interval does not exclude equal hazards under the conventional two-sided interpretation.
The p-value of 0.4579 is a test statistic for the superiority hypothesis; it does not quantify the size or clinical importance of the estimated treatment effect. The estimate and confidence interval provide the more direct description of magnitude and precision.
This analysis is specifically identified as involving non-diabetic patients. It should therefore not automatically be generalized to the full randomized population or interpreted as a formal comparison of treatment effects between diabetic and non-diabetic patients.
10. Primary Results: Aliskiren vs Enalapril — Superiority Analysis
The registry separately posts a superiority analysis of aliskiren versus enalapril for the same primary composite endpoint and time frame.
Aliskiren vs Enalapril · Superiority
95% CI: 0.90–1.10 · P = 0.9118
Hypothesis type: Superiority
| Element | Posted result |
|---|---|
| Endpoint | First occurrence of cardiovascular death or heart failure hospitalization |
| Time frame | Up to End of Study (78 months) |
| Groups compared | Aliskiren vs Enalapril |
| Analysis population | Full Analysis Set |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.99 |
| 95% CI | 0.90–1.10 |
| P-value | 0.9118 |
| Hypothesis | Superiority |
The point estimate of 0.99 indicates an estimated hazard very close to the comparator's hazard, corresponding to approximately a 1% lower estimated hazard for aliskiren.
The 95% CI of 0.90–1.10 includes the null value of 1. This means the interval is compatible with both a modestly lower and a modestly higher relative hazard under the model.
The p-value of 0.9118 is not an effect-size measure. A large p-value does not establish that the two treatments are identical; it indicates that the posted superiority test does not provide strong evidence against its null hypothesis.
This analysis should also be distinguished from the separate non-inferiority analysis. The same HR and confidence interval can be viewed through different inferential frameworks depending on the prespecified hypothesis and margin.
11. Reading the Four Posted Primary Analyses Together
The four posted analyses are most clearly understood by keeping their comparison, population, and hypothesis labels separate rather than collapsing them into a single result.
| Comparison / analysis | HR | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Combination vs Enalapril | 0.93 | 0.85–1.03 | 0.1724 | Superiority |
| Aliskiren vs Enalapril | 0.99 | 0.85–1.07 | 0.4579 | Superiority; non-diabetic patients |
| Aliskiren vs Enalapril | 0.99 | 0.90–1.10 | 0.0368 | Non-inferiority or equivalence |
| Aliskiren vs Enalapril | 0.99 | 0.90–1.10 | 0.9118 | Superiority |
The combination analysis and the aliskiren analyses answer different treatment-comparison questions. The non-inferiority analysis also cannot be interpreted using exactly the same decision rule as a superiority analysis because its relevant benchmark is the prespecified margin of 1.104 rather than the null value alone.
12. Non-Inferiority: A Closer Statistical Look
The ATMOSPHERE registry data provide a particularly useful example of why non-inferiority analysis requires explicit attention to the margin.
95% CI = 0.90–1.10
Non-inferiority margin = 1.104
The upper confidence limit of 1.10 is below the prespecified margin of 1.104.
Why 1 is not the only relevant number
For a superiority analysis of a hazard ratio, 1 is the usual null value. A non-inferiority analysis asks a different question: whether the treatment effect could be worse than a prespecified acceptable limit.
Here, the registry gives that limit as 1.104. The point estimate of 0.99 is close to 1, but the statistical logic of non-inferiority does not require the confidence interval to exclude 1. Instead, the relevant question is whether the confidence interval extends beyond the non-inferiority margin.
Why the p-value should not replace the margin
A conventional superiority p-value answers a null-hypothesis question. It does not tell us whether a treatment satisfies a non-inferiority requirement. The posted non-inferiority result therefore needs to be read with its stated margin and confidence interval.
13. Confidence Intervals and Precision
Confidence intervals are particularly useful in ATMOSPHERE because the posted estimates are close to the null value. The intervals communicate how much uncertainty surrounds each estimated hazard ratio.
Combination vs Enalapril
HR 0.93 with a 95% CI of 0.85–1.03. The interval spans the null value of 1.
Aliskiren vs Enalapril
HR 0.99 with a 95% CI of 0.90–1.10. The interval also spans 1 while remaining below the stated non-inferiority margin of 1.104.
The width of a confidence interval is influenced by the amount of information available for the estimate. The ClinicalTrials.gov record does not provide the event counts needed to reconstruct the underlying information content of the Cox analyses, so the confidence intervals are best treated as the reported measures of statistical precision rather than as quantities to be reverse-engineered.
14. Safety Results
The registry reports serious adverse events by treatment arm as the number affected divided by the number at risk.
| Treatment arm | Serious adverse events | At risk |
|---|---|---|
| Combination of Aliskiren and Enalapril | 1466 | 2347 |
| Aliskiren | 1504 | 2348 |
| Enalapril | 1501 | 2345 |
These figures are safety counts, not the primary efficacy endpoint. The denominators also differ slightly across the three arms, so raw affected-participant counts should not be interpreted as directly comparable risks without accounting for their respective at-risk populations.
The ClinicalTrials.gov record does not provide a formal statistical comparison of serious adverse events, so the appropriate interpretation here is descriptive rather than a claim of statistical equivalence or difference.
15. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The primary endpoint is a time-to-event endpoint: the first occurrence of cardiovascular death or heart failure hospitalization. A Cox model uses both event occurrence and event timing and can accommodate censored observations. It produces a hazard ratio that summarizes the relative event hazard between treatment groups under the model.
What does an HR of 0.99 mean?
An HR of 0.99 means that the estimated instantaneous event hazard in the treatment group is approximately 99% of that in the comparator group within the fitted model. It does not mean that 99% of patients avoided the endpoint, nor does it imply an absolute risk reduction of 1 percentage point.
Why is the non-inferiority margin important?
The non-inferiority margin defines how much worse the treatment could be while still satisfying the prespecified non-inferiority criterion. In ATMOSPHERE, the posted margin was 1.104. The reported upper confidence limit of 1.10 lies below that margin.
Why doesn't the p-value measure the treatment effect?
A p-value measures the compatibility of the observed data with a specified null hypothesis under the statistical model. It is not a percentage chance that the hypothesis is true and it does not quantify the magnitude of treatment benefit or harm. The HR supplies the point estimate, while the confidence interval describes its uncertainty.
What does intention-to-treat mean here?
The posted analysis text identifies intention-to-treat analysis as a statistical concept, while the analysis population is specifically defined as the Full Analysis Set. The practical lesson is that the exact analysis-population definition matters: one should use the registry's stated FAS definition rather than assuming an undifferentiated "all enrolled" population.
Why does the composite endpoint require care?
The primary endpoint combines cardiovascular death and heart failure hospitalization. A single HR therefore summarizes time to the first occurrence of either component. It cannot by itself tell us whether the treatment effect was the same for cardiovascular death and heart failure hospitalization separately.
Why should the non-diabetic analysis not be treated as a separate randomized trial?
The posted non-diabetic analysis is a subset analysis of the randomized study. Its HR of 0.96 and 95% CI of 0.85–1.07 describe that specified subgroup. A subgroup estimate does not automatically establish that the treatment effect differs from the treatment effect in another subgroup; a formal interaction analysis would be needed for that question.
16. What the Hazard Ratio Does — and Does Not — Mean
The combination-versus-enalapril HR of 0.93 means that the Cox model estimated a lower instantaneous rate of first cardiovascular death or heart failure hospitalization for the combination, corresponding to approximately a 7% lower estimated hazard.
The aliskiren-versus-enalapril HR of 0.99 corresponds to an estimated hazard approximately 1% lower for aliskiren. Neither HR should be translated directly into an absolute probability or an individual patient's treatment effect.
The confidence interval gives a range of values representing uncertainty around the estimated hazard ratio under the statistical model and sampling framework. For the combination comparison, the 95% CI is 0.85–1.03. For the principal aliskiren comparison, it is 0.90–1.10.
These intervals should not be interpreted as the range of effects that individual patients experience.
Two groups can have similar eventual event proportions but different timing of events, or different proportions with different patterns of follow-up. A time-to-event model incorporates the timing information and censoring structure instead of reducing the endpoint to a simple yes/no outcome.
17. Analysis Population, Censoring, and Model Assumptions
Several statistical concepts are easy to overlook when reading a hazard ratio in isolation.
Analysis population
The primary analyses used the registry-defined Full Analysis Set, including the stated exclusions for mis-randomization without study drug and sites with major GCP violations.
Censoring
Time-to-event analysis allows participants without an observed first event during their available follow-up to contribute information until their censoring point.
Proportional hazards
The Cox model is described as a proportional-hazards model. The ClinicalTrials.gov record does not provide diagnostics for independently assessing that assumption.
Composite endpoint
The hazard ratio summarizes time to first occurrence of either cardiovascular death or heart failure hospitalization rather than either component separately.
These issues matter because a hazard ratio is not a purely descriptive statistic. It is an estimate generated by a particular model applied to a particular analysis population and follow-up structure.
18. Multiplicity and Multiple Primary Analyses
The ClinicalTrials.gov record reports 4 statistical analyses for the registered primary endpoint. These analyses include superiority hypotheses, a non-inferiority or equivalence hypothesis, and a separately identified non-diabetic analysis.
| Feature | What the registry data show | Statistical implication |
|---|---|---|
| Primary endpoint | One registered composite endpoint | The same endpoint can be examined under different prespecified hypotheses. |
| Combination comparison | Combination vs enalapril, superiority | Tests whether the combination provides evidence of superiority under the posted analysis. |
| Aliskiren non-inferiority comparison | Aliskiren vs enalapril, non-inferiority/equivalence | Uses the stated non-inferiority margin of 1.104. |
| Aliskiren superiority comparison | Aliskiren vs enalapril, superiority | Uses a superiority hypothesis rather than the non-inferiority margin. |
| Non-diabetic analysis | Aliskiren vs enalapril, non-diabetic patients | Describes a specified subgroup analysis. |
The ClinicalTrials.gov record does not report an alpha-allocation or multiplicity-adjustment procedure for these four posted analyses. Accordingly, this page does not assign an overall familywise-error interpretation to the collection of p-values beyond the individual hypothesis labels reported by the registry.
19. Interim Analysis, Crossover, and Bayesian Methods
The ClinicalTrials.gov record does not report an interim-analysis procedure, alpha-spending approach, crossover design, or Bayesian analysis method for ATMOSPHERE. These topics are therefore not used to characterize the statistical design of this page.
The important design information that is documented is the randomized, parallel, triple-masked structure, the three treatment arms, the FAS analysis population, the Cox model, and the stated non-inferiority margin.
20. Limitations
- Registry-level detail: the ClinicalTrials.gov record identifies the primary endpoint, analysis population, model, effect estimates, confidence intervals, p-values, and selected design features, but do not provide the full statistical analysis plan.
- Composite endpoint: the reported primary HR concerns first occurrence of either cardiovascular death or heart failure hospitalization. It does not separately quantify the treatment effect for each component.
- Hazard-ratio assumptions: Cox proportional-hazards estimates depend on the suitability of the proportional-hazards framework. The ClinicalTrials.gov record does not provide diagnostics for that assumption.
- Non-inferiority interpretation: the posted non-inferiority result depends on the prespecified margin of 1.104 and should not be interpreted using a superiority p-value alone.
- Analysis-population exclusions: the primary FAS is not simply every enrolled participant because the registry specifies exclusions for mis-randomized patients who took no study drug and patients from sites with major GCP violations.
- Subgroup interpretation: the non-diabetic analysis is a subgroup analysis and should not automatically be interpreted as evidence of treatment-effect modification.
- Multiplicity: the ClinicalTrials.gov record identifies four posted primary analyses but do not provide the complete multiplicity-control strategy. The p-values therefore should be read in the context of their individual posted hypotheses.
- Safety comparisons: serious adverse-event counts are provided descriptively by arm, but the ClinicalTrials.gov record does not report a formal statistical comparison of these safety outcomes.
- Limited endpoint reporting: although three outcome measures are posted in the registry, the ClinicalTrials.gov record provides formal estimates only for the registered primary endpoint. No unsupported secondary endpoint results are added here.
21. Why This Trial Matters Statistically
ATMOSPHERE is a useful teaching case because it places several core clinical-trial concepts around one time-to-event endpoint. The same randomized trial includes a three-arm parallel structure, triple masking, Cox regression, hazard ratios, confidence intervals, superiority testing, a prespecified non-inferiority margin, and a subgroup analysis.
| Concept | How it appears in ATMOSPHERE |
|---|---|
| Randomization | The study uses randomized allocation across three parallel treatment arms. |
| Blinding | The registry describes the study as triple masked. |
| Time-to-event endpoint | The primary endpoint is time to first cardiovascular death or heart failure hospitalization. |
| Cox model | Primary analyses use a Cox proportional-hazards model. |
| Hazard ratio | HR is the posted effect measure for the primary analyses. |
| Confidence interval | 95% two-sided confidence intervals quantify uncertainty around the HR estimates. |
| Superiority | Several posted analyses use a superiority hypothesis. |
| Non-inferiority | The aliskiren-versus-enalapril analysis uses a prespecified margin of 1.104. |
| Intention-to-treat concept | The analysis text identifies intention-to-treat analysis; the formal population is the registry-defined FAS. |
| Subgroup analysis | A separate posted analysis is identified for non-diabetic patients. |
| Safety analysis | Serious adverse-event counts are reported separately by treatment arm. |
The most important statistical lesson is that the same numerical estimate can answer different questions depending on the hypothesis. An HR of 0.99 can appear in both a non-inferiority analysis and a superiority analysis, yet the inferential benchmark is different. Reading the estimate without reading the hypothesis would therefore lose an essential part of the statistical story.
22. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
23. Related Statistical Calculators
24. Sources
- ClinicalTrials.gov: ATMOSPHERE, NCT00853658.
- PubMed: PMID 38904251.
- PubMed: PMID 37366061.
- PubMed: PMID 34101002.
- PubMed: PMID 33089502.
- PubMed: PMID 32720404.
Continue through the Clinical Biostats statistical pathway
Explore the statistical concepts behind randomized trials, time-to-event endpoints, hazard ratios, confidence intervals, and non-inferiority analysis.
25. Record Summary
ATMOSPHERE provides a compact example of how a randomized phase 3 trial can generate several distinct statistical questions from the same primary time-to-event endpoint. The study enrolled 7064 participants in a randomized, parallel, triple-masked design with three treatment arms. The registered primary endpoint was the first occurrence of cardiovascular death or heart failure hospitalization through the end of study at 78 months.
The posted analyses used a Cox proportional-hazards model and reported hazard ratios with two-sided 95% confidence intervals. The combination-versus-enalapril superiority analysis produced an HR of 0.93 (95% CI 0.85–1.03; P = 0.1724). The principal aliskiren-versus-enalapril comparison produced an HR of 0.99 (95% CI 0.90–1.10), with one analysis framed around non-inferiority using the prespecified margin of 1.104 and another framed around superiority.
The statistical lesson is broader than any individual p-value. Correct interpretation requires keeping the endpoint definition, analysis population, hazard-ratio scale, confidence interval, and hypothesis type together. In particular, the non-inferiority analysis illustrates why a prespecified margin can be more important to the inferential question than a conventional superiority threshold.