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
SYGMA2 was a randomized, parallel-group, triple-masked phase 3 trial evaluating whether budesonide/formoterol used as needed could provide a severe asthma exacerbation rate that was non-inferior to twice-daily budesonide with as-needed terbutaline, while also evaluating superiority and several measures of lung function, medication use, asthma control, and quality of life.
| Feature | SYGMA2 |
|---|---|
| Phase | Phase 3 |
| Condition | Asthma |
| Design | Randomized, parallel-group, triple-masked |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 4,215 |
| Primary endpoints | Annual severe asthma exacerbation rate for non-inferiority and superiority analyses |
| Primary analysis population | Full analysis set |
| Lead sponsor | AstraZeneca |
| ClinicalTrials.gov | NCT02224157 |
2. Clinical Question
The central statistical question was whether budesonide/formoterol used as needed could achieve a severe asthma exacerbation rate that was sufficiently close to twice-daily budesonide to meet a prespecified non-inferiority criterion. The trial also evaluated whether the as-needed regimen was superior for annual severe asthma exacerbation rate and compared several secondary clinical and physiological outcomes.
Population
Adults and adolescents with asthma enrolled in a phase 3 randomized treatment study.
Intervention
Budesonide/formoterol used as needed, with budesonide placebo twice daily.
Comparator
Budesonide twice daily with terbutaline used as needed.
Primary question
Is the annual severe asthma exacerbation rate with as-needed budesonide/formoterol non-inferior to the comparator, and is it superior?
3. Trial Design
Symbicort “as needed”
- Budesonide/formoterol “as needed”
- Budesonide placebo twice daily
- Evaluated for annual severe asthma exacerbation rate through up to 52 weeks
Budesonide twice daily
- Budesonide twice daily
- Terbutaline “as needed”
- Evaluated over the same randomized treatment period
The trial therefore compares two different medication-use strategies rather than simply comparing two fixed doses of the same treatment. The statistical design is particularly important because the principal outcome is an event rate accumulated over participant follow-up, not merely the proportion of participants with an event.
4. Endpoints
| Endpoint | Time frame | Definition / outcome unit | Primary method |
|---|---|---|---|
| Annual Severe Asthma Exacerbation Rate — Non-inferiority Analysis | Up to 52 weeks | Severe asthma exacerbations over the randomized treatment period; exacerbations per participant year | Negative binomial model |
| Annual Severe Asthma Exacerbation Rate — Superiority Analysis | Up to 52 weeks | Severe asthma exacerbations over the randomized treatment period; exacerbations per participant year | Negative binomial model |
| Number of Participants Experiencing at Least One Severe Asthma Exacerbation | Day 1 up to 52 weeks | Participants | Cox proportional-hazards model |
| Average Change From Baseline in Pre-bronchodilator FEV1 | Study weeks 0, 17, 34, 52 | mL | Mixed-effects model |
| Average Change From Baseline in “as Needed” Use | Week 0 up to 52 weeks | Number of inhalations per day | ANCOVA |
| Change From Baseline in Percent of “as Needed” Free Days | Week 0 up to 52 weeks | Percentage of “as needed” free days | ANCOVA |
| Percentage of Controller Use Days | Week 0 up to 52 weeks | Percentage of days | ANOVA |
| Average Change From Baseline in Asthma Control Questionnaire (5-item Version) — ACQ-5 Score | Study weeks 0, 17, 34, 52 | Score | Mixed-effects model |
| Average Change From Baseline in Asthma Quality of Life Questionnaire Standardised Version — AQLQ(S) Score | Study weeks 0, 17, 34, 52 | AQLQ(S) overall score | Mixed-effects model |
The registry therefore combines several statistical structures: recurrent event-rate data for severe exacerbations, time-to-first-event data, repeated continuous measurements, and covariate-adjusted continuous outcomes.
5. Statistical Methodology
Negative binomial regression for annual severe exacerbation rate
The primary endpoint was analyzed with a negative binomial model. This is appropriate for count data when the analysis concerns the number of exacerbations accumulated over varying amounts of follow-up and when the variability of the counts can exceed that expected under a simple Poisson model.
A rate ratio below 1 indicates a lower estimated exacerbation rate in the Symbicort “as needed” treatment group. The model used the logarithm of follow-up time as an offset variable, allowing the comparison to account for differing amounts of participant follow-up.
The primary model was adjusted for randomized treatment, pre-study treatment, and region. The analysis population was the full analysis set, and the registry identifies intention-to-treat analysis and covariate adjustment as concepts incorporated into the analysis.
Non-inferiority framework
The non-inferiority analysis used a one-sided 95% confidence limit rather than a p-value. The registry specifies that the upper limit of the one-sided 95% confidence limit had to remain below 1.2 for Symbicort “as needed” to be considered non-inferior to Pulmicort twice daily.
Primary non-inferiority result
One-sided 95% upper confidence limit: 1.16
The registry identifies an upper limit below 1.2 as meeting the stated non-inferiority criterion.
Superiority framework
The same annual severe asthma exacerbation rate was also evaluated under a superiority hypothesis. Here, the registry reports a two-sided 95% confidence interval and a p-value. This is a different inferential question from non-inferiority: the superiority analysis asks whether the data support a difference in either direction and, specifically for this treatment comparison, whether the observed rate ratio is statistically distinguishable from 1.
Cox proportional-hazards model
The secondary endpoint of participants experiencing at least one severe asthma exacerbation was analyzed using Cox regression. The model included randomized treatment, pre-study treatment, and region as covariates.
The registry specifically describes a hazard ratio below 1 as indicating that Symbicort “as needed” prolongs the time to first severe exacerbation.
Mixed-effects models for repeated measurements
Pre-bronchodilator FEV1, ACQ-5, and AQLQ(S) were evaluated at multiple study visits. Mixed-effects models account for the fact that repeated observations from the same participant are correlated rather than statistically independent.
For these longitudinal analyses, treatment, pre-study treatment, region, visit, and the treatment-by-visit interaction were incorporated as fixed effects where specified. Participant was included as a random effect, and baseline FEV1, ACQ-5, or AQLQ was included as a covariate for the corresponding outcome.
ANCOVA and ANOVA
ANCOVA was used for average change in “as needed” use and for change from baseline in the percentage of “as needed” free days. The models incorporated treatment, pre-study treatment, and region, with baseline measures used as continuous covariates where specified.
ANOVA was used for percentage of controller-use days, with adjustment for randomized treatment, pre-study treatment, and region.
6. Primary Results
Annual Severe Asthma Exacerbation Rate — Non-inferiority Analysis
Rate ratio
One-sided 95% confidence limit: upper limit 1.16
Analysis population: full analysis set · Negative binomial regression
The estimated rate ratio was 0.97. The registry states that the upper limit of the one-sided 95% confidence limit was 1.16 and that an upper limit below 1.2 indicates that Symbicort “as needed” is non-inferior to Pulmicort twice daily.
What the estimate means: A rate ratio of 0.97 means that the estimated annual severe asthma exacerbation rate in the Symbicort “as needed” group was 0.97 times the estimated rate in the Pulmicort twice-daily group. In relative terms, 0.97 corresponds to an estimated rate that is 3% lower.
What it does not mean: It does not mean that 3% fewer participants experienced an exacerbation. A rate ratio compares event rates over participant time; it is not a risk ratio or a percentage of participants affected.
Why the confidence limit matters: Non-inferiority is determined by the prespecified margin and the confidence limit. Here, the upper one-sided 95% limit of 1.16 remains below the 1.2 margin. That means the upper end of the uncertainty interval still lies within the range defined as acceptable for non-inferiority.
Why there is no p-value: The registry explicitly states that the non-inferiority analysis was based on the confidence interval rather than a p-value. Consequently, no p-value was calculated for this analysis.
Important caution: Non-inferiority is not established merely because an observed rate ratio is close to 1. The confidence interval must be considered relative to the prespecified non-inferiority margin.
Annual Severe Asthma Exacerbation Rate — Superiority Analysis
Rate ratio
Two-sided 95% CI: 0.78–1.20 · P = 0.754
Analysis population: full analysis set · Negative binomial regression
What the estimate means: The estimated rate ratio of 0.97 indicates a 3% lower estimated annual severe exacerbation rate in the Symbicort “as needed” group relative to the Pulmicort twice-daily group.
What it does not mean: The estimate is not an estimate that 3% fewer people had exacerbations. It compares modeled rates of exacerbations per participant year.
Precision: The two-sided 95% confidence interval extends from 0.78 to 1.20. It therefore includes 1, so the data are compatible with both a lower and a higher exacerbation rate for the Symbicort “as needed” group under the statistical model.
P-value versus effect size: The p-value of 0.754 describes the statistical evidence against the superiority null hypothesis under the specified model. It does not measure the magnitude or clinical importance of the rate ratio.
Non-inferiority versus superiority: The two primary analyses should not be conflated. The same estimated rate ratio of 0.97 can satisfy the stated non-inferiority criterion while not providing evidence of superiority. The non-inferiority margin and the superiority null value answer different questions.
7. Secondary Results
Participants Experiencing at Least One Severe Asthma Exacerbation
Hazard ratio for time to first severe exacerbation
95% CI: 0.777–1.174 · P = 0.664
Cox proportional-hazards model; full analysis set
The hazard ratio of 0.955 indicates an estimated instantaneous hazard of first severe asthma exacerbation that was approximately 4.5% lower in the Symbicort “as needed” group. The registry describes a hazard ratio below 1 as indicating a longer time to first severe exacerbation.
The 95% confidence interval of 0.777 to 1.174 includes 1, and the p-value is 0.664. Thus, the reported analysis does not provide evidence of superiority for this time-to-first-event endpoint.
The hazard ratio is a relative time-to-event measure rather than a difference in the percentage of participants who eventually experienced an exacerbation. Its interpretation also depends on the Cox proportional-hazards model assumptions.
Average Change From Baseline in Pre-bronchodilator FEV1
Least-square mean difference
95% CI: −53.7 to −11.4 mL · P = 0.003
Average treatment effect across all treatment visits
The mixed-effects analysis evaluated study weeks 0, 17, 34, and 52. The model included randomized treatment, pre-study treatment, region, visit, and treatment-by-visit as fixed effects, participant as a random effect, and baseline FEV1 as a covariate.
The estimated difference was −32.6 mL, with the registry defining a mean difference greater than 0 as favoring Symbicort “as needed.” The negative estimate therefore favors the Pulmicort twice-daily group for this outcome under the registry's stated direction.
The confidence interval from −53.7 to −11.4 mL does not include 0, and the reported p-value is 0.003. The confidence interval describes uncertainty around the estimated average treatment difference; it does not describe the range of FEV1 changes among individual participants.
Average Change From Baseline in “as Needed” Use
Least-square mean difference
95% CI: 0.00–0.07 · Superiority analysis
ANCOVA; full analysis set
The ANCOVA adjusted for randomized treatment, pre-study treatment, and region, with the number of “as needed” inhalations at baseline included as a continuous covariate. The registry states that a mean difference less than zero indicates a larger mean reduction in the number of “as needed” inhalations in the Symbicort “as needed” group.
The estimate of 0.03 inhalations per day is positive under the registry's treatment-group direction. That means the estimated difference does not favor the Symbicort “as needed” group under the stated interpretation rule. The reported confidence interval begins at 0.00 and extends to 0.07.
This illustrates why the direction of an effect estimate must always be read together with the prespecified coding and interpretation of the treatment groups. The numerical sign alone does not identify which treatment is favored.
Change From Baseline in Percent of “as Needed” Free Days
Least-square mean difference
95% CI: −8.37 to −5.34 · P < 0.001
ANCOVA; full analysis set
The model adjusted for randomized treatment, pre-study treatment, and region and included the percentage of “as needed” free days during run-in as a continuous covariate. The registry states that an estimate greater than 0 means a larger increase in the percentage of “as needed” free days in the Symbicort “as needed” arm.
The estimated treatment difference was −6.85 percentage points, with a 95% confidence interval from −8.37 to −5.34. Because the entire interval is below 0, the estimated difference is in the direction of a larger increase in “as needed” free days for the Pulmicort twice-daily group according to the registry's stated direction.
The p-value of less than 0.001 indicates strong statistical evidence against a zero mean difference under the specified ANCOVA model. It does not quantify the size or practical importance of the difference.
Percentage of Controller Use Days
Least-square mean difference
95% CI: −39.18 to −35.77 · P < 0.001
ANOVA; full analysis set
The ANOVA was adjusted for randomized treatment, pre-study treatment, and region. The registry states that an estimate greater than 0 means a higher percentage of controller-use days in the Symbicort “as needed” group.
The negative estimate of −37.48 percentage points is therefore in the opposite direction from the registry's stated criterion for a higher percentage of controller-use days in the Symbicort “as needed” group. The confidence interval is entirely below 0, and the p-value is less than 0.001.
The result is a modeled mean difference in percentage of days. It should not be interpreted as a hazard ratio, rate ratio, or percentage reduction in participants.
Average Change From Baseline in ACQ-5 Score
Least-square mean difference
95% CI: 0.068–0.150 · P < 0.001
Mixed-effects model; full analysis set
The mixed-effects model included randomized treatment, pre-study treatment, region, visit, and treatment-by-visit as fixed effects, participant as a random effect, and baseline ACQ-5 as a covariate. The analysis covered study weeks 0, 17, 34, and 52.
The estimated difference was 0.109 points. The registry specifies that a mean difference less than 0 favors Symbicort “as needed.” Therefore the positive estimate is not in the direction specified as favoring the as-needed Symbicort regimen.
The 95% confidence interval is entirely above 0, and the p-value is less than 0.001. Statistical evidence for a nonzero difference therefore exists under the specified model, but the direction and practical interpretation require the registry's endpoint coding rather than simply treating a small p-value as evidence of benefit.
Average Change From Baseline in AQLQ(S) Score
Least-square mean difference
95% CI: −0.137 to −0.054 · P < 0.001
Mixed-effects model; full analysis set
The mixed-effects model included randomized treatment, pre-study treatment, region, visit, and treatment-by-visit as fixed effects, participant as a random effect, and baseline AQLQ as a covariate. The registry states that a mean difference greater than 0 favors Symbicort “as needed.”
The estimated difference of −0.096 is below zero, so it is not in the direction specified by the registry as favoring Symbicort “as needed.” The 95% confidence interval from −0.137 to −0.054 excludes zero, and the p-value is less than 0.001.
As with the other continuous outcomes, the estimate is a model-based average treatment difference rather than an individual-level change. The sign must be interpreted using the prespecified treatment contrast and direction of benefit.
8. Results Summary
| Outcome | Method | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Annual severe asthma exacerbation rate — non-inferiority | Negative binomial regression | Rate ratio 0.97 | One-sided upper limit 1.16 | Not calculated |
| Annual severe asthma exacerbation rate — superiority | Negative binomial regression | Rate ratio 0.97 | 0.78–1.20 | 0.754 |
| At least one severe asthma exacerbation | Cox proportional-hazards model | HR 0.955 | 0.777–1.174 | 0.664 |
| Pre-bronchodilator FEV1 | Mixed-effects model | Mean difference −32.6 mL | −53.7 to −11.4 | 0.003 |
| “As needed” use | ANCOVA | Mean difference 0.03 | 0.00–0.07 | Not reported |
| “As needed” free days | ANCOVA | Mean difference −6.85 | −8.37 to −5.34 | < 0.001 |
| Controller use days | ANOVA | Mean difference −37.48 | −39.18 to −35.77 | < 0.001 |
| ACQ-5 score | Mixed-effects model | Mean difference 0.109 | 0.068–0.150 | < 0.001 |
| AQLQ(S) score | Mixed-effects model | Mean difference −0.096 | −0.137 to −0.054 | < 0.001 |
The table illustrates an important feature of clinical-trial interpretation: statistical significance does not by itself determine whether an outcome favors one treatment. The endpoint definition, direction of the treatment contrast, effect measure, confidence interval, and model specification all matter.
9. Safety Results
The registry reports serious adverse events by randomized treatment arm. These data are presented as affected participants over the participants at risk.
| Treatment group | Serious adverse events | At risk |
|---|---|---|
| Placebo twice daily + Symbicort “as needed” | 66 | 2,089 |
| Pulmicort twice daily + terbutaline “as needed” | 73 | 2,087 |
The corresponding affected/at-risk figures are 66/2,089 and 73/2,087. The registry does not provide a formal comparative statistical analysis for these serious-adverse-event counts in the reported analyses listed here.
Safety summaries answer a different question from the primary efficacy analysis. The serious-adverse-event figures describe the number of affected participants relative to the participants at risk in each arm; they do not constitute a rate ratio, hazard ratio, or formal test of treatment difference.
Because the reported serious-adverse-event information is presented as counts over participants at risk, it should not be treated as an incidence rate per participant year without additional follow-up information.
10. Statistical Methods Explained
Why was negative binomial regression used for the primary endpoint?
The primary outcome counts severe asthma exacerbations over the randomized treatment period. A negative binomial model is designed for count outcomes and can accommodate greater variability in counts than a basic Poisson model. The model also incorporates follow-up time through the logarithm of follow-up time as an offset, making a rate-based comparison possible.
Why is the primary non-inferiority analysis based on a confidence limit?
Non-inferiority asks whether the treatment is not unacceptably worse than the comparator by more than a prespecified amount. Here, that amount is represented by the margin of 1.2 for the rate ratio. The relevant question is therefore whether the upper confidence limit remains below 1.2. A p-value against equality at 1 does not directly answer that question.
What does a rate ratio of 0.97 mean?
A rate ratio of 0.97 means that the modeled severe exacerbation rate in the Symbicort “as needed” group is estimated to be 0.97 times the comparator rate. It is a comparison of event rates, not a comparison of the percentages of participants who experience an event.
Why was a Cox model used for at least one severe exacerbation?
The secondary endpoint asks about the time from Day 1 to the first severe asthma exacerbation. Time-to-event methods use information from both participants who experience the event and participants who remain event-free through their available follow-up. The Cox model summarizes the relative instantaneous event rate through a hazard ratio.
Why were mixed-effects models used for FEV1, ACQ-5, and AQLQ(S)?
These outcomes were measured repeatedly at study weeks 0, 17, 34, and 52. Measurements from the same participant are correlated. A mixed-effects model accounts for that repeated-measures structure by incorporating participant-level random effects while modeling treatment, visit, and other fixed effects.
Why does a statistically significant p-value not automatically identify the clinically preferred treatment?
A p-value evaluates evidence against a specified null hypothesis. It does not encode the direction of benefit, the magnitude of an effect, or its clinical importance. For SYGMA2, several continuous outcomes have p-values below 0.001, but their estimates have different signs and the registry specifies different directions for interpreting those signs. The effect estimate and endpoint definition therefore have to be read alongside the p-value.
11. Non-Inferiority and Superiority Are Different Questions
SYGMA2 is particularly useful statistically because the same primary endpoint was evaluated under both non-inferiority and superiority frameworks.
| Feature | Non-inferiority analysis | Superiority analysis |
|---|---|---|
| Question | Is the annual severe exacerbation rate sufficiently close to the comparator to satisfy the non-inferiority criterion? | Is there evidence that the annual severe exacerbation rate differs between groups in favor of the experimental treatment? |
| Effect measure | Rate ratio | Rate ratio |
| Estimate | 0.97 | 0.97 |
| Confidence interval | One-sided 95% upper limit 1.16 | Two-sided 95% CI 0.78–1.20 |
| Decision framework | Upper confidence limit compared with 1.2 | Evidence assessed against equality of rates |
| P-value | Not calculated | 0.754 |
12. Covariate Adjustment
The primary negative binomial analyses were adjusted for randomized treatment, pre-study treatment, and region. The Cox analysis used the same three covariates. Other secondary analyses used baseline values of the relevant outcome where specified.
| Analysis | Adjustment described in the registry |
|---|---|
| Primary exacerbation-rate analyses | Randomized treatment, pre-study treatment, region; log follow-up time as offset |
| Time to first severe exacerbation | Randomized treatment, pre-study treatment, region |
| Pre-bronchodilator FEV1 | Randomized treatment, pre-study treatment, region, visit, treatment-by-visit; baseline FEV1 covariate; participant random effect |
| “As needed” use | Randomized treatment, pre-study treatment, region; baseline number of “as needed” inhalations as continuous covariate |
| “As needed” free days | Randomized treatment, pre-study treatment, region; run-in percentage of “as needed” free days as continuous covariate |
| Controller use days | Randomized treatment, pre-study treatment, region |
| ACQ-5 | Randomized treatment, pre-study treatment, region, visit, treatment-by-visit; baseline ACQ-5 covariate; participant random effect |
| AQLQ(S) | Randomized treatment, pre-study treatment, region, visit, treatment-by-visit; baseline AQLQ covariate; participant random effect |
Covariate adjustment can improve precision when the covariates are prognostic and appropriately specified. It does not turn a randomized trial into an observational analysis: the treatment comparison remains anchored to randomized assignment.
13. Time-to-Event Analysis
The secondary endpoint of participants experiencing at least one severe asthma exacerbation illustrates the difference between an event-rate analysis and a time-to-first-event analysis.
Rate analysis
Counts severe exacerbations and incorporates participant follow-up time. A participant can contribute more than one exacerbation to the event count.
Time-to-first-event analysis
Focuses on the time until the first severe exacerbation. Once that first event occurs, the participant no longer contributes a later first-event time.
These analyses therefore do not answer identical questions. The negative binomial model asks about the annualized burden of severe exacerbations, whereas the Cox model asks about the timing of the first severe exacerbation.
The estimate summarizes the relative hazard of experiencing the first severe exacerbation under the Cox model. It is not equivalent to the primary rate ratio of 0.97.
14. Longitudinal Analysis
Three secondary outcomes were measured repeatedly at study weeks 0, 17, 34, and 52: pre-bronchodilator FEV1, ACQ-5, and AQLQ(S). The registry identifies mixed-model analyses for these repeated outcomes.
| Repeated outcome | Estimate | 95% CI | P-value |
|---|---|---|---|
| Pre-bronchodilator FEV1 | −32.6 mL | −53.7 to −11.4 | 0.003 |
| ACQ-5 | 0.109 | 0.068–0.150 | < 0.001 |
| AQLQ(S) | −0.096 | −0.137 to −0.054 | < 0.001 |
The treatment-by-visit interaction included in the FEV1, ACQ-5, and AQLQ(S) models allows the treatment contrast to vary across visits rather than forcing every visit to have exactly the same treatment difference. The reported effect for these analyses is described as the average treatment effect across the treatment visits.
A repeated-measures analysis contains more information than simply comparing the final measurement at week 52. It can incorporate the trajectory of observations across the prespecified study visits while accounting for correlation among measurements from the same participant.
15. Missing Data and Censoring Considerations
The registry describes full-analysis-set efficacy analyses and repeated-measures and time-to-event methods, but it does not provide additional details in the reported statistical analyses about a specific missing-data imputation strategy.
For the Cox analysis, participants who have not experienced a first severe exacerbation by the end of their available observation contribute information up to their censoring time. For the repeated-measures outcomes, the mixed-effects framework uses the observed longitudinal measurements according to the fitted model rather than requiring every participant to have an observed value at every visit.
These distinctions matter because missing observations and censoring are not simply numerical blanks. Their treatment is part of the statistical model and can affect the precision and interpretation of estimated treatment differences.
16. Multiplicity and Multiple Statistical Questions
The trial contains two primary analyses of the same severe asthma exacerbation endpoint plus multiple secondary outcomes. Each additional hypothesis test creates another opportunity for a nominally small p-value to occur by chance if many tests are considered together.
| Analysis family | Primary statistical role | Reported inferential quantity |
|---|---|---|
| Annual severe asthma exacerbation rate — non-inferiority | Primary | One-sided confidence limit relative to 1.2 |
| Annual severe asthma exacerbation rate — superiority | Primary | Two-sided 95% CI and p-value |
| Time to first severe exacerbation | Secondary | HR, 95% CI, p-value |
| FEV1, medication use, free days, controller use, ACQ-5, AQLQ(S) | Secondary | Model-specific treatment difference, confidence interval and, where reported, p-value |
A secondary endpoint with a small p-value should therefore be interpreted in the context of its endpoint role, the overall analysis plan, and any multiplicity strategy specified for the trial. The reported p-value remains a property of its individual statistical test; it does not automatically establish that the endpoint had the same confirmatory status as the primary non-inferiority analysis.
17. Reading the Primary Rate Ratio Correctly
The primary rate ratio compares the number of severe asthma exacerbations per participant year. A participant with multiple exacerbations can contribute multiple events to this analysis. A risk ratio, by contrast, would compare the probability of experiencing at least one event over a specified period.
The primary rate ratio of 0.97 and the secondary hazard ratio of 0.955 are not interchangeable. The rate ratio summarizes modeled event rates, while the hazard ratio summarizes the relative instantaneous rate of the first event under the Cox model.
The confidence interval describes uncertainty about a population-level treatment-effect estimate under the specified statistical framework. It does not say that individual participants' true treatment effects fall within the reported interval.
18. Limitations
- Different estimands: the primary annual exacerbation rate and the secondary time-to-first-exacerbation endpoint answer related but distinct questions.
- Non-inferiority margin: interpretation of the primary analysis depends on the prespecified margin of 1.2; a rate ratio close to 1 is not, by itself, the criterion for non-inferiority.
- Model dependence: negative binomial and Cox estimates depend on their respective statistical models and assumptions.
- Cox proportional-hazards assumption: the hazard ratio is most naturally interpreted as a relative hazard under the proportional-hazards framework; departures from that assumption can complicate interpretation of a single HR.
- Multiple endpoints: the trial contains multiple secondary outcomes, so individual p-values should be interpreted according to the prespecified inferential hierarchy.
- Repeated measurements: longitudinal outcomes require assumptions concerning the covariance structure and missing observations. The registry does not provide all implementation details in the reported analysis descriptions.
- Direction of continuous outcomes: the registry defines which direction of each treatment contrast favors the as-needed regimen. The sign of a mean difference therefore cannot be interpreted without the endpoint-specific definition.
- Safety interpretation: serious adverse-event counts by arm do not constitute a formal comparative safety analysis by themselves.
19. Why This Trial Matters Statistically
SYGMA2 is a useful teaching example because a single randomized trial contains several distinct statistical estimands and modeling strategies. The primary endpoint is a recurrent event rate, while a secondary endpoint converts the same clinical phenomenon into a time-to-first-event question. Other outcomes are longitudinal continuous measures analyzed using mixed-effects models, while additional continuous outcomes use ANCOVA or ANOVA.
| Concept | How it appears in SYGMA2 |
|---|---|
| Randomization | Parallel randomized comparison of two treatment strategies |
| Blinding | Triple-masked phase 3 design |
| Non-inferiority | Primary annual severe asthma exacerbation rate evaluated against a margin of 1.2 |
| Negative binomial regression | Primary analysis of recurrent severe asthma exacerbation rates |
| Rate ratio | Primary effect measure for annual exacerbation rate |
| Cox proportional-hazards model | Time to first severe asthma exacerbation |
| Hazard ratio | Secondary time-to-event effect measure |
| ANCOVA | Average change in “as needed” use and percentage of “as needed” free days |
| ANOVA | Percentage of controller-use days |
| Mixed-effects model | Repeated FEV1, ACQ-5, and AQLQ(S) measurements |
| Covariate adjustment | Pre-study treatment, region, baseline values, and other prespecified covariates |
| Confidence intervals | Central to both non-inferiority and superiority interpretation |
The trial therefore demonstrates why statistical analysis should be driven by the structure of the endpoint rather than by a single universal test. Counts, rates, time-to-event outcomes, and repeated continuous measurements require different models because they contain different forms of information.
20. Related Tutorials
Learn more about the methods used in this trial:
21. Related Calculators
22. Sources
- ClinicalTrials.gov: NCT02224157.
- PubMed: PMID 33548020.
- PubMed: PMID 29768147.
- PubMed: PMID 28069068.
- PubMed: PMID 27912982.
Continue through the Clinical Biostats statistical library
Connect the trial's endpoint structure to deeper tutorials and statistical calculators for clinical-trial analysis.
23. Record Summary
SYGMA2 provides a compact example of how the statistical method should follow the clinical endpoint. The primary annual severe asthma exacerbation outcome is a recurrent count-rate endpoint analyzed with negative binomial regression and expressed as a rate ratio. Its non-inferiority analysis uses a one-sided confidence limit against a margin of 1.2, while the superiority analysis uses a two-sided confidence interval and p-value. The same clinical phenomenon is also examined as time to first severe exacerbation using a Cox proportional-hazards model. Other outcomes require ANCOVA, ANOVA, or mixed-effects models because they represent adjusted continuous measurements or repeated observations.
The primary non-inferiority estimate was a rate ratio of 0.97 with a one-sided 95% upper confidence limit of 1.16, which the registry identifies as satisfying the stated non-inferiority criterion. The superiority analysis produced the same rate-ratio estimate of 0.97, with a two-sided 95% confidence interval of 0.78 to 1.20 and a p-value of 0.754. Secondary analyses illustrate why rate ratios, hazard ratios, and mean differences should not be treated as interchangeable effect measures.