← Clinical Trials
COPD Phase 3 52-Week Study NCT01782326

FLAME: Complete Statistical Analysis of QVA149 in COPD

An independent statistical analysis of the randomized phase 3 FLAME trial comparing QVA149 with a long acting B2 agonist and inhaled corticosteroid for COPD exacerbations, with emphasis on exacerbation rates, time-to-event methods, non-inferiority, and subsequent superiority testing.

FLAME  ·  Phase 3  ·  Completed  ·  Enrollment 3362
Scope of this record

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

FLAME was a randomized, parallel-group, triple-masked phase 3 study in chronic obstructive pulmonary disease (COPD), comparing QVA149 with a long acting B2 agonist (LABA) and inhaled corticosteroid (ICS). The registry reports a total enrollment of 3362 patients and a 52-week primary endpoint focused on the rate of COPD exacerbations.

3362
Enrollment
Randomized phase 3 study
2
Arms
Parallel allocation
0.89
Primary rate ratio
95% CI 0.83–0.96
0.84
First exacerbation HR
95% CI 0.78–0.91
FeatureFLAME
Trial nameFLAME
Brief titleQVA vs. Salmeterol/Fluticasone, 52-week Exacerbation Study, FLAME (EFfect of Indacaterol Glycopyronium Vs Fluticasone Salmeterol on COPD Exacerbations)
PhasePhase 3
StatusCompleted
Therapeutic areaPulmonology
ConditionChronic Obstructive Pulmonary Disease (COPD)
AllocationRandomized
Design modelParallel
MaskingTriple
Primary purposeTreatment
Enrollment3362
Lead sponsorNovartis Pharmaceuticals
Sponsor typeIndustry
Start2013-07
Primary completion2015-09
ClinicalTrials.govNCT01782326

2. Clinical Question

The central statistical question was whether QVA149 could be shown to be non-inferior to a long acting B2 agonist and inhaled corticosteroid combination for the rate of COPD exacerbations over 52 weeks, and, if non-inferiority was established, whether the same analysis could support a claim of superiority for reducing exacerbation rate.

Population

Patients with chronic obstructive pulmonary disease enrolled in a phase 3 randomized treatment study. The ClinicalTrials.gov record does not provide a complete baseline-characteristic table.

Intervention

QVA149, identified in the registry as a drug intervention.

Comparator

Long acting B2 agonist (LABA) and inhaled corticosteroid (ICS), identified in the registry as a drug intervention.

Primary question

Is the COPD exacerbation rate with QVA149 no more than the prespecified non-inferiority margin above the comparator rate, and, conditional on non-inferiority, is QVA149 superior?

3. Trial Design

01
Randomize3362 enrolled
02
QVA149One randomized treatment arm
03
LABA + ICSComparator arm
04
52 weeksPrimary exacerbation endpoint
05
AnalysisRate and time-to-event methods
Allocation
Randomized, with a parallel-group design.
Masking
Triple-masked according to the registry.
Study phase
Phase 3, with treatment as the primary purpose.
Primary follow-up
52 weeks for the registered primary endpoint.
ARM 1 · QVA149

QVA149

  • Drug intervention
  • Compared with a long acting B2 agonist and inhaled corticosteroid
  • Primary outcome assessed over 52 weeks
ARM 2 · LABA + ICS

Long acting B2 agonist and inhaled corticosteroid

  • Drug intervention
  • Comparator for QVA149
  • Primary outcome assessed over 52 weeks

4. Endpoints

The registry reports one primary endpoint and multiple secondary endpoints with formal statistical analyses. The primary endpoint is an event-rate measure rather than a conventional binary endpoint, while several secondary endpoints use time-to-event methods.

EndpointTime frameAnalysis / effect measure
Rate of COPD Exacerbations 52 weeks Generalized linear model assuming a negative binomial distribution; rate ratio
Time to First COPD Exacerbation. 52 weeks Cox proportional-hazards model; hazard ratio
Rate of Moderate to Severe COPD Exacerbations. 52 weeks Generalized linear model; rate ratio
Time to First Moderate to Severe COPD Exacerbation. 52 weeks. Cox proportional-hazards model; hazard ratio
Time to First Moderate to Severe COPD Exacerbations Requiring Treatment With Systemic Corticosteroids 52 weeks Cox proportional-hazards model; hazard ratio
Time to First Moderate to Severe COPD Exacerbations Requiring Treatment With Antibiotics 52 weeks Cox proportional-hazards model; hazard ratio
Time to First Moderate to Severe COPD Exacerbations Requiring Hospitalization 52 weeks Cox proportional-hazards model; hazard ratio
Time to First Moderate to Severe COPD Exacerbations Requiring Re-hospitalization Within 30 Days 52 weeks Cox proportional-hazards model; hazard ratio

Primary endpoint definition

The registry definition specifies that COPD exacerbations starting between first dose and one day after last treatment are included. COPD exacerbations that occurred within 7 days of each other are collapsed as one event. Estimates are from a generalized linear model assuming a negative binomial distribution, with terms for treatment, baseline total symptom score, and baseline COPD exacerbation history.

Endpoint structure matters. The primary outcome is a rate of exacerbations over a defined period. That makes the rate ratio the natural effect measure in the posted analysis. It is not the same estimand as the hazard ratio for time to first exacerbation.

5. Statistical Methodology

Generalized linear model with a negative binomial distribution

The primary exacerbation-rate analysis used a generalized linear model assuming a negative binomial distribution. This is appropriate to the structure of recurrent event counts when variability can exceed what a simple Poisson model would accommodate.

Conceptual rate comparison
Rate Ratio = estimated exacerbation rate under QVA149 ÷ estimated exacerbation rate under LABA + ICS

A rate ratio below 1 indicates a lower estimated exacerbation rate in the QVA149 group relative to the comparator. The reported model also incorporated baseline total symptom score and baseline COPD exacerbation history.

Per-protocol analysis for the primary endpoint

The posted primary analysis was conducted in the per-protocol set (PPS). The registry defines the PPS as all patients in the full analysis set without any major protocol deviations, with only PPS patients having non-missing values for all terms in the negative binomial model included in the analysis.

This is particularly important because the trial was designed around a non-inferiority question. A per-protocol analysis can provide information about treatment performance under adherence to the protocol, while the exact relationship between the PPS and other analysis populations must be kept explicit rather than treating all analyses as interchangeable.

Cox proportional-hazards model

The secondary time-to-event analyses used Cox proportional-hazards models. These models estimate a relative hazard for experiencing the first specified event, allowing patients who have not experienced the event during available follow-up to contribute information through their censoring time.

Conceptual hazard-ratio interpretation
HR < 1  →  lower estimated instantaneous event rate in the QVA149 group

The hazard ratio describes a relative model-based event rate over time. It is not a rate ratio for recurrent exacerbations and is not an absolute risk difference.

Full analysis set and missing values

The secondary analyses identify the full analysis set (FAS) as including randomized patients who received at least one dose of study drug and had no major GCP violations. The posted analyses also reference non-missing values for all model terms and identify multiple imputation / missing data as an analysis concept.

Population distinction: the primary exacerbation-rate analysis is explicitly identified as a PPS analysis, whereas the posted secondary analyses identify the FAS. These are different analysis populations and their estimates should not be silently combined into a single analysis population.

Non-inferiority followed by superiority

The primary endpoint was evaluated under a two-stage inferential logic. The study was designed to have >95% power to rule out a 1.15-fold increase in the rate of exacerbations for QVA149 versus salmeterol/fluticasone. The registry states that if the upper limit of the confidence interval was <1.15, non-inferiority could be claimed. If non-inferiority was demonstrated, superiority could be claimed if the upper limit of the same confidence interval was <1.

Non-inferiority threshold

The relevant boundary was 1.15. The question was whether the observed uncertainty interval excluded a QVA149 exacerbation rate more than 1.15 times the comparator rate.

Superiority threshold

After non-inferiority, the same two-sided confidence interval was assessed against 1. A rate ratio below 1 with an upper confidence limit below 1 supports superiority under the posted analysis rule.

6. Results: Primary Endpoint

The registry posts two formal analyses for the primary endpoint, reflecting the prespecified non-inferiority and superiority framework. Both use the same estimated rate ratio and 95% two-sided confidence interval.

Rate of COPD Exacerbations — Non-inferiority Analysis

Rate ratio for COPD exacerbations

0.89

95% CI: 0.83–0.96   ·   Two-sided CI

Non-inferiority margin: 1.15-fold increase

The upper confidence limit of 0.96 is below the prespecified non-inferiority boundary of 1.15. According to the registry's stated analysis rule, this satisfies the criterion for non-inferiority of QVA149 compared with the LABA/ICS comparator.

Clinical Biostats interpretation

The estimated rate ratio of 0.89 means that the modeled COPD exacerbation rate for QVA149 was estimated at 89% of the corresponding comparator rate over the 52-week analysis period. Put another way, the point estimate corresponds to a 11% lower estimated exacerbation rate.

That interpretation is about the estimated rate, not about 11% of patients avoiding an exacerbation and not about an 11% absolute reduction in individual patient risk.

The 95% confidence interval of 0.83–0.96 describes uncertainty around the estimated rate ratio under the statistical model. It does not describe the range of individual patient responses.

No p-value is posted for this non-inferiority analysis. For non-inferiority, the prespecified margin and the confidence interval are central to the decision rule; simply asking whether a conventional null-hypothesis p-value is below a threshold is not the same as applying the stated non-inferiority criterion.

The analysis is based on the PPS and includes only patients with non-missing values for all terms in the negative binomial model. That analysis-population restriction is therefore part of the interpretation of the estimate.

Rate of COPD Exacerbations — Superiority Analysis

Rate ratio for COPD exacerbations

0.89

95% CI: 0.83–0.96   ·   P = 0.003

Superiority analysis after non-inferiority

The upper confidence limit of 0.96 is also below 1, and the posted superiority analysis reports P = 0.003. Under the registry's stated sequence, superiority of QVA149 compared with the LABA/ICS comparator for reducing exacerbation rate could therefore be claimed after non-inferiority was demonstrated.

Clinical Biostats interpretation

The estimate remains a rate ratio of 0.89; the superiority analysis does not turn it into a different effect measure. The point estimate still represents the modeled relative exacerbation rate between the two treatment groups.

The confidence interval of 0.83–0.96 is entirely below 1, indicating that the values supported by the interval under the model are below equal rates. It is also entirely below the non-inferiority margin of 1.15.

The p-value of 0.003 is evidence against the null hypothesis used for the posted superiority comparison. It does not measure the size of the treatment effect. The effect size is described by the rate ratio, while the confidence interval describes its precision.

The result should be understood in the context of the prespecified non-inferiority-then-superiority framework rather than as an isolated p-value. The primary analysis is also explicitly a PPS analysis with model-term completeness requirements.

Primary analysisEstimate95% CIP-valueHypothesis
Rate of COPD Exacerbations 0.89 0.83–0.96 Not posted Non-inferiority
Rate of COPD Exacerbations 0.89 0.83–0.96 0.003 Superiority

7. Results: Secondary Endpoints

The registry posts seven secondary statistical analyses. Five are time-to-event analyses involving the Cox proportional-hazards model, while one is a recurrent exacerbation-rate analysis using a generalized linear model and one additional time-to-event analysis evaluates first exacerbation.

Time to First COPD Exacerbation

Hazard ratio

0.84

95% CI: 0.78–0.91   ·   P < 0.001

52 weeks · Cox proportional-hazards model

Clinical Biostats interpretation

The hazard ratio of 0.84 corresponds to an estimated instantaneous hazard of first COPD exacerbation that was approximately 16% lower with QVA149 than with the comparator, under the fitted Cox model.

This is a time-to-first-event estimand. It should not be interpreted as a 16% reduction in the number of exacerbations, because a patient can contribute more than one exacerbation to a recurrent-event rate endpoint while the time-to-first analysis stops at the first event.

The 95% CI of 0.78–0.91 quantifies uncertainty around the hazard-ratio estimate. The p-value of <0.001 addresses the statistical comparison and does not quantify effect magnitude.

As with other Cox analyses, the hazard ratio is model-based. Its interpretation relies on the proportional-hazards framework being a reasonable description of the relative hazards over the relevant follow-up.

Rate of Moderate to Severe COPD Exacerbations

Rate ratio

0.83

95% CI: 0.75–0.91   ·   P < 0.001

52 weeks · Generalized linear model

Clinical Biostats interpretation

The rate ratio of 0.83 means that the estimated rate of moderate to severe COPD exacerbations was 83% of the comparator rate under the posted model, corresponding to a 17% lower estimated rate.

The 95% CI of 0.75–0.91 provides the uncertainty interval for this relative rate estimate. It remains below 1 throughout the interval.

The p-value of <0.001 indicates strong statistical evidence against the null comparison used for this superiority analysis. It does not mean that the probability of the observed treatment effect being due to chance is <0.001, nor does it describe the clinical magnitude of the effect.

Time to First Moderate to Severe COPD Exacerbation

Hazard ratio

0.78

95% CI: 0.70–0.86   ·   P < 0.001

52 weeks · Cox proportional-hazards model

Clinical Biostats interpretation

The estimated hazard ratio of 0.78 corresponds to a 22% lower estimated instantaneous hazard of a first moderate to severe COPD exacerbation with QVA149 under the Cox model.

The 95% CI of 0.70–0.86 indicates relatively tight statistical uncertainty around the point estimate. It does not imply that every patient experienced the same relative reduction.

The p-value of <0.001 is a measure of evidence against the statistical null hypothesis, not a measure of effect size. The hazard ratio and confidence interval provide the magnitude and precision information.

Time to First Moderate to Severe COPD Exacerbation Requiring Systemic Corticosteroids

Hazard ratio

0.90

95% CI: 0.74–1.08   ·   P = 0.256

52 weeks · Cox proportional-hazards model

Clinical Biostats interpretation

The point estimate of 0.90 corresponds to an estimated 10% lower instantaneous hazard of the specified event with QVA149 under the Cox model.

However, the 95% CI of 0.74–1.08 includes 1. This means the interval includes values compatible with either a lower or higher hazard under the model. The p-value of 0.256 does not provide conventional statistical evidence against a null hazard ratio of 1 for this superiority comparison.

This result illustrates why the point estimate alone is insufficient. A hazard ratio below 1 does not automatically establish superiority; the uncertainty interval and prespecified inferential framework also matter.

Time to First Moderate to Severe COPD Exacerbation Requiring Antibiotics

Hazard ratio

0.81

95% CI: 0.69–0.95   ·   P = 0.008

52 weeks · Cox proportional-hazards model

Clinical Biostats interpretation

The hazard ratio of 0.81 corresponds to a 19% lower estimated instantaneous hazard of the specified first event with QVA149 under the fitted Cox model.

The 95% CI of 0.69–0.95 remains below 1, while the p-value of 0.008 provides evidence against the null comparison used in this superiority analysis.

The confidence interval is important because it communicates the precision of the estimate; the p-value alone cannot tell the reader whether the estimated effect is large or small.

Time to First Moderate to Severe COPD Exacerbation Requiring Hospitalization

Hazard ratio

0.81

95% CI: 0.66–1.00   ·   P = 0.046

52 weeks · Cox proportional-hazards model

Clinical Biostats interpretation

The hazard ratio of 0.81 corresponds to a 19% lower estimated instantaneous hazard of the specified first hospitalization-related event with QVA149 under the Cox model.

The upper confidence limit is 1.00. Thus, the interval reaches the null value rather than lying strictly below it. The posted p-value is 0.046, which is below 0.05, but this should not be converted into a claim that the effect is large or that statistical significance establishes a particular clinical magnitude.

The relatively broad interval also illustrates the additional uncertainty that can arise for a more specific time-to-event outcome compared with the broader exacerbation endpoint.

Time to First Moderate to Severe COPD Exacerbation Requiring Re-hospitalization Within 30 Days

Hazard ratio

0.89

95% CI: 0.38–2.10   ·   P = 0.790

52 weeks · Cox proportional-hazards model

Clinical Biostats interpretation

The point estimate of 0.89 corresponds to an estimated 11% lower instantaneous hazard under the Cox model. But the 95% CI of 0.38–2.10 is very wide and includes both substantially lower and substantially higher hazards.

The p-value of 0.790 does not provide evidence against the null comparison. More importantly, the wide confidence interval shows that the point estimate is imprecise for this specific endpoint.

This is a useful statistical distinction: a non-significant result does not prove that the two hazards are identical. The confidence interval shows the range of effect sizes that remain compatible with the observed analysis under its assumptions.

Secondary endpointEffect95% CIP-value
Time to First COPD Exacerbation.HR 0.840.78–0.91<0.001
Rate of Moderate to Severe COPD Exacerbations.Rate ratio 0.830.75–0.91<0.001
Time to First Moderate to Severe COPD Exacerbation.HR 0.780.70–0.86<0.001
Time to first moderate to severe exacerbation requiring systemic corticosteroidsHR 0.900.74–1.080.256
Time to first moderate to severe exacerbation requiring antibioticsHR 0.810.69–0.950.008
Time to first moderate to severe exacerbation requiring hospitalizationHR 0.810.66–1.000.046
Time to first moderate to severe exacerbation requiring re-hospitalization within 30 daysHR 0.890.38–2.100.790
Educational note: the registry provides summary hazard-ratio results but does not provide sufficient underlying event and censoring information in the ClinicalTrials.gov record to reconstruct valid Kaplan-Meier curves. A Kaplan-Meier graphic should not be fabricated from summary hazard ratios alone.

8. Safety Results

The ClinicalTrials.gov record reports serious adverse events by randomized arm as affected patients divided by patients at risk. These are the safety figures available for this page.

Safety measureQVA149LABA + ICS
Serious adverse events308 / 1678334 / 1680

QVA149

Serious adverse events affected 308 of 1678 patients at risk in the ClinicalTrials.gov record.

LABA + ICS

Serious adverse events affected 334 of 1680 patients at risk in the ClinicalTrials.gov record.

The serious-adverse-event figures are descriptive counts with denominators as reported in the registry by the registry. The available trial data do not provide a formal statistical comparison, confidence interval, or p-value for this safety measure, so none is added here.

9. Statistical Methods Explained

Why was a negative binomial generalized linear model used?

The primary endpoint counts COPD exacerbations over a 52-week period. A negative binomial model is a generalized linear modeling approach suited to count-rate outcomes when the variability in counts can be greater than the simple Poisson assumption permits. The registry explicitly identifies this distributional assumption for the primary analysis.

What does a rate ratio of 0.89 mean?

A rate ratio of 0.89 means that the estimated exacerbation rate under QVA149 was 0.89 times the comparator rate in the posted model. The point estimate therefore corresponds to a 11% lower estimated rate. It does not mean that 11% of patients benefited or that each patient had exactly an 11% reduction in risk.

Why is the non-inferiority margin important?

The study was designed to rule out a 1.15-fold increase in the exacerbation rate. That margin defines what degree of worsening would still be considered compatible with non-inferiority under the trial's statistical framework. Because the upper 95% confidence limit was 0.96, it was below 1.15.

Why could superiority be tested after non-inferiority?

The registry explicitly states a sequential rule: first establish non-inferiority using the 1.15 boundary, then assess superiority using the same confidence interval against 1. The posted analysis has an upper limit of 0.96, which is below 1, and reports P = 0.003 for superiority.

Why are both rate ratios and hazard ratios reported?

They answer different questions. The rate ratio compares the modeled rate of exacerbations over the study period, whereas the hazard ratio compares the instantaneous rate of experiencing a specified first event over time. A recurrent exacerbation-rate endpoint can count multiple events; a time-to-first endpoint focuses on when the first event occurs.

Why does the analysis population matter?

The primary endpoint was analyzed in the per-protocol set, whereas the posted secondary analyses identify the full analysis set. Non-inferiority trials often pay particular attention to protocol adherence because substantial deviations can complicate interpretation of whether the tested treatment strategy was actually delivered as intended.

What does a p-value not tell us?

A p-value describes the evidence against a specified null hypothesis under the statistical model and testing framework. It does not measure treatment-effect magnitude, does not give the probability that the null hypothesis is true, and does not describe the range of plausible effects. Those questions are addressed more directly by the effect estimate and confidence interval.

10. Non-Inferiority Logic in FLAME

The primary endpoint provides a useful example of why non-inferiority trials require a different reading strategy from a conventional superiority trial.

StepRegistry criterionFLAME result
1. Define the marginRule out a 1.15-fold increase in exacerbation ratePrespecified margin = 1.15
2. Estimate treatment effectRate ratio with 95% two-sided CI0.89
3. Examine upper confidence limitUpper limit must be <1.15 for non-inferiority0.96 < 1.15
4. Test superiority if non-inferiority is demonstratedUpper limit of same CI must be <10.96 < 1
5. Review superiority evidenceFormal superiority analysisP = 0.003
Why this matters

The non-inferiority conclusion is not based on the point estimate being below 1 alone. The crucial comparison is between the upper confidence limit and the prespecified margin. A point estimate of 0.89 would not, by itself, establish non-inferiority if the uncertainty interval extended beyond 1.15.

Here, the entire posted 95% confidence interval, 0.83–0.96, lies below both the non-inferiority boundary of 1.15 and the equality boundary of 1.

11. Missing Data and Analysis Populations

The registry data explicitly identify missing-data considerations in the posted analyses. For the primary negative-binomial analysis, only PPS patients with non-missing values for all terms in the model are included. The secondary analyses similarly specify inclusion of patients with non-missing values for all relevant model terms and identify multiple imputation / missing data as an analysis concept.

Primary analysis

The PPS included patients in the FAS without major protocol deviations, with the primary model restricted to patients having non-missing values for all model terms.

Secondary analyses

The posted secondary analyses identify the FAS and reference intention-to-treat analysis and multiple imputation / missing data concepts.

What the ClinicalTrials.gov record does not establish: the registry excerpt does not provide enough detail to reconstruct a complete missing-data algorithm, imputation model, number of imputations, or sensitivity-analysis framework. Those details should not be inferred from the mere presence of the "multiple imputation / missing data" concept.

12. Cox Proportional-Hazards Interpretation

Several FLAME secondary endpoints are time-to-event outcomes. The Cox model expresses the treatment comparison as a hazard ratio, making it possible to use the timing of first events rather than reducing follow-up to a simple yes/no endpoint.

Hazard ratio examples from the posted analyses
First COPD exacerbation: HR 0.84   |   First moderate/severe exacerbation: HR 0.78

Both estimates are below 1, but they refer to different event definitions and therefore should not be treated as interchangeable estimates of one endpoint.

A Cox hazard ratio also depends on the proportional-hazards framework. If the relative hazard changes materially over time, a single HR can compress a more complicated time-varying pattern into one summary measure. The ClinicalTrials.gov record does not report a formal test or diagnostic for the proportional-hazards assumption, so no such conclusion is added here.

13. Primary vs Secondary Estimands

QuestionEffect measureWhat it captures
How frequently did COPD exacerbations occur?Rate ratioRelative modeled exacerbation rate over 52 weeks
How quickly did the first COPD exacerbation occur?Hazard ratioRelative instantaneous hazard of first exacerbation
How frequently did moderate/severe exacerbations occur?Rate ratioRelative modeled rate of moderate/severe exacerbations
How quickly did a first moderate/severe exacerbation occur?Hazard ratioRelative instantaneous hazard of first moderate/severe exacerbation
How quickly did specific treated or serious exacerbation events occur?Hazard ratioRelative instantaneous hazard for the specified first event

This distinction is central to interpreting the page. A lower rate ratio and a lower hazard ratio can point in the same descriptive direction while still representing different statistical estimands. The values should therefore be reported with their exact endpoint definitions.

14. Limitations

15. Why This Trial Matters Statistically

FLAME is a useful teaching case because the primary analysis combines a recurrent-event rate endpoint with a prespecified non-inferiority margin and a subsequent superiority assessment. The secondary analyses then show how closely related clinical questions can require different statistical estimands.

ConceptHow it appears in FLAME
RandomizationRandomized phase 3 parallel-group design
Triple maskingRegistry identifies the trial as triple-masked
Rate ratioPrimary COPD exacerbation rate and moderate/severe exacerbation-rate analyses
Negative binomial modelPrimary rate analysis assumes a negative binomial distribution
Non-inferiorityUpper confidence limit compared with a 1.15-fold margin
SuperiorityAfter non-inferiority, the same CI is evaluated against 1
Per-protocol analysisPrimary endpoint analyzed in the PPS
Full analysis setPosted secondary analyses identify the FAS
Missing dataModel inclusion requires non-missing values; multiple imputation / missing data is identified as an analysis concept
Cox modelUsed for multiple time-to-first-exacerbation endpoints
Hazard ratioPrimary effect measure for the posted time-to-event analyses
Confidence intervals95% two-sided intervals are reported for all posted effect estimates
P-valuesPosted for the superiority analyses of the primary and secondary endpoints

16. Statistical Interpretation of the Main Findings

Rate ratio

The primary rate ratio of 0.89 summarizes the modeled relative rate of COPD exacerbations. The confidence interval of 0.83–0.96 provides the precision of that estimate. It does not translate directly into an absolute number of exacerbations prevented for an individual patient.

Non-inferiority

The trial's prespecified non-inferiority criterion was based on excluding a 1.15-fold increase. Because the upper confidence limit was 0.96, the posted result satisfies that criterion. This is a margin-based inference rather than a conventional superiority p-value interpretation.

Superiority

After non-inferiority, the upper limit of the same confidence interval was below 1, and the posted superiority analysis reported P = 0.003. The p-value supplies evidence for the statistical comparison; the rate ratio and confidence interval describe the magnitude and precision of the effect.

Time-to-event consistency

The posted secondary time-to-first-exacerbation analyses include HR estimates of 0.84 for first COPD exacerbation and 0.78 for first moderate to severe COPD exacerbation. These are different estimands from the recurrent exacerbation rate ratio and should be interpreted on their own endpoint definitions.

17. Results Across the Posted Endpoints

EndpointMeasureEstimate95% CIP-value
Rate of COPD ExacerbationsRate ratio0.890.83–0.960.003 for superiority analysis
Time to First COPD Exacerbation.HR0.840.78–0.91<0.001
Rate of Moderate to Severe COPD Exacerbations.Rate ratio0.830.75–0.91<0.001
Time to First Moderate to Severe COPD Exacerbation.HR0.780.70–0.86<0.001
Time to First Moderate to Severe COPD Exacerbations Requiring Treatment With Systemic CorticosteroidsHR0.900.74–1.080.256
Time to First Moderate to Severe COPD Exacerbations Requiring Treatment With AntibioticsHR0.810.69–0.950.008
Time to First Moderate to Severe COPD Exacerbations Requiring HospitalizationHR0.810.66–1.000.046
Time to First Moderate to Severe COPD Exacerbations Requiring Re-hospitalization Within 30 DaysHR0.890.38–2.100.790

The table demonstrates an important statistical principle: the results do not all have identical precision or identical endpoint definitions. In particular, the confidence interval for re-hospitalization within 30 days is substantially wider than the interval for first COPD exacerbation. The appropriate interpretation therefore considers the estimate, endpoint, confidence interval, and analysis population together.

18. Related Tutorials

Learn more about the methods used in this trial:

19. Related Statistical Calculators

20. Sources

Continue through the Clinical Biostats statistical tutorials

Use the trial's endpoints and methods as a practical pathway into survival analysis, generalized linear models, non-inferiority design, confidence intervals, and clinical-trial methodology.

21. Record Summary

FLAME provides a compact example of several important clinical-trial statistical principles. Its primary endpoint was a 52-week rate of COPD exacerbations, analyzed with a negative-binomial generalized linear model in the per-protocol set. The posted rate ratio was 0.89 with a 95% CI of 0.83–0.96. Because the upper confidence limit was below the prespecified 1.15 non-inferiority margin, the registry's stated criterion for non-inferiority was met; the same upper limit was also below 1, and the posted superiority analysis reported P = 0.003.

The secondary analyses demonstrate why endpoint definition matters. Recurrent exacerbation rates were analyzed with a rate ratio, while time-to-first events were analyzed with Cox proportional-hazards models. The posted hazard ratios ranged from 0.78 to 0.90 for several secondary endpoints, while the re-hospitalization endpoint had a much wider 95% CI of 0.38–2.10. These results should be read with their exact endpoint definitions and analysis populations rather than treated as interchangeable measures of one underlying effect.

The statistical story is therefore broader than the headline p-value. Understanding FLAME requires attention to the estimand being analyzed, the non-inferiority margin, the confidence interval, the distinction between rate ratios and hazard ratios, the difference between the PPS and FAS, and the assumptions behind the negative-binomial and Cox models.

Clinical Biostats methodology: This page separates registry-reported numerical results from statistical interpretation. Numbers are reported only from the registry-reported FLAME trial data, while methodological explanations describe how the posted analyses should be understood without adding unreported trial results.