This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Reported estimates are separated from the interpretation that follows them, so readers can see what the registry states and what the statistics do and do not imply.
1. Trial at a Glance
BOREAS was a randomized, quadruple-masked, parallel-group phase 3 trial asking whether dupilumab, added to inhaled maintenance therapy, reduces the annualized rate of moderate or severe COPD exacerbations compared with placebo in patients with moderate-to-severe COPD and type 2 inflammation.
| Feature | BOREAS |
|---|---|
| Registry ID | NCT03930732 |
| Phase | Phase 3 |
| Condition | Chronic obstructive pulmonary disease (moderate-to-severe, with type 2 inflammation) |
| Design | Randomized, parallel-group, quadruple-masked, placebo-controlled; 2 arms |
| Arms | Dupilumab 300 mg every 2 weeks (q2w) vs placebo |
| Background therapy listed | Inhaled corticosteroid, inhaled long-acting beta agonist, inhaled long-acting muscarinic antagonist |
| Primary endpoint | Annualized rate of moderate or severe COPD exacerbations over the 52-week treatment period |
| Primary analysis | Negative binomial model, ITT population; rate difference derived by the delta method |
| Enrollment | 939 |
| Dates | Start 2019-04-15; primary completion 2023-02-08 |
| Status | Completed; results posted |
| Lead sponsor | Sanofi (industry) |
2. Clinical Question
The trial addressed whether blocking a type 2 inflammatory pathway with dupilumab reduces the frequency of clinically important COPD exacerbations in a population selected for type 2 inflammation, when given on top of inhaled maintenance treatment. Lung function, health-related quality of life and respiratory symptoms were examined as secondary questions within a prespecified testing hierarchy.
Population
Participants with moderate-to-severe COPD with type 2 inflammation, receiving inhaled maintenance therapy.
Intervention
Dupilumab (SAR231893) 300 mg every 2 weeks, alongside inhaled corticosteroid, long-acting beta agonist and long-acting muscarinic antagonist therapy.
Comparator
Matching placebo, alongside the same inhaled maintenance therapy.
Primary question
Is the annualized rate of moderate or severe COPD exacerbations over 52 weeks lower with dupilumab than with placebo (superiority)?
3. Trial Design
Dupilumab 300 mg q2w
- Dupilumab SAR231893 300 mg every 2 weeks
- Inhaled corticosteroid
- Inhaled long-acting beta agonist
- Inhaled long-acting muscarinic antagonist
- 52-week treatment period
Placebo
- Matching placebo
- Inhaled corticosteroid
- Inhaled long-acting beta agonist
- Inhaled long-acting muscarinic antagonist
- 52-week treatment period
Covariates and analysis populations
The registry does not describe the randomization strata directly, but the analysis models consistently adjust for region (pooled country), ICS dose and smoking status at screening, which is the typical pattern when these factors were used to stratify randomization. The exacerbation models additionally adjust for baseline disease severity and the number of moderate or severe exacerbations in the year before the study.
| Analysis population | Definition / role |
|---|---|
| Intent-to-treat (ITT) | Randomized population analyzed according to the treatment group allocated by randomization; used for the primary exacerbation analysis and the SGRQ responder analysis. |
| ITT with available data | ITT participants with data available; used for the MMRM analyses of FEV1, SGRQ total score and E-RS total score. |
| FeNO ≥20 ppb subgroup | ITT participants with baseline fractional exhaled nitric oxide ≥20 parts per billion; used for the prespecified subgroup endpoints within the hierarchy. |
4. Endpoints
Primary endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Annualized rate of moderate or severe COPD exacerbations over the 52-week treatment period | Moderate exacerbations: acute exacerbation of COPD (AECOPD) requiring systemic corticosteroids (such as intramuscular, intravenous or oral) and/or antibiotics. Severe exacerbations: AECOPD requiring hospitalization, or observation for >24 hours in an emergency department/urgent care facility, or resulting in death. Events had to be separated by at least 14 days to be counted separately. The annualized event rate was the total number of exacerbations during the treatment period divided by the total number of participant-years treated. Events were adjudicated by an independent third party. | Baseline (Day 1) to Week 52 |
Secondary endpoints with posted analyses
| Endpoint | Time frame | Unit | Method |
|---|---|---|---|
| Change from baseline in pre-bronchodilator (BD) FEV1 at Week 12 | Baseline (Day 1) to Week 12 | liters | MMRM |
| Change from baseline in pre-BD FEV1 at Week 52 | Baseline (Day 1) to Week 52 | liters | MMRM |
| Change from baseline in pre-BD FEV1 at Week 12, baseline FeNO ≥20 ppb subgroup | Baseline (Day 1) to Week 12 | liters | MMRM |
| Change from baseline in pre-BD FEV1 at Week 52, baseline FeNO ≥20 ppb subgroup | Baseline (Day 1) to Week 52 | liters | MMRM |
| Change from baseline in St. George's Respiratory Questionnaire (SGRQ) total score at Week 52 | Baseline (Day 1) to Week 52 | score on a scale | MMRM |
| Percentage of participants with SGRQ improvement ≥4 points at Week 52 | Baseline (Day 1) to Week 52 | percentage of participants | Logistic regression |
| Change from baseline in E-RS: COPD RS-Total score at Week 52 | Baseline (Day 1) to Week 52 | score on a scale | MMRM |
| Annualized rate of moderate or severe COPD exacerbations over 52 weeks, baseline FeNO ≥20 ppb subgroup | Baseline (Day 1) to Week 52 | exacerbation per participant-year | Negative binomial model |
5. Primary Result: Annualized Exacerbation Rate
The primary analysis compared the annualized rate of moderate or severe COPD exacerbations between dupilumab 300 mg q2w and placebo in the ITT population. The registry reports the treatment effect as a difference in annualized rates (labelled a risk difference), expressed in exacerbations per participant-year.
Difference in annualized exacerbation rate (dupilumab − placebo)
exacerbations per participant-year
95% CI (two-sided): −0.508 to −0.140 · P = 0.0005
Negative binomial model with log treatment duration as offset; difference derived by the delta method; superiority hypothesis
| Component | Reported |
|---|---|
| Groups compared | Placebo vs Dupilumab 300 mg q2w |
| Effect measure | Difference in annualized rate (registry label: Risk Difference) |
| Estimate | −0.324 exacerbations per participant-year |
| 95% CI | −0.508 to −0.140 |
| P-value | 0.0005 |
| Response variable | Total number of events during the 52-week treatment period |
| Covariates | Treatment group, region (pooled country), ICS dose, smoking status at screening, baseline disease severity, number of moderate or severe exacerbations within one year before the study |
| Offset | Log-transformed treatment duration |
| Significance level | Two-sided 0.049, first step of a hierarchical testing procedure |
The group-specific annualized rates and the rate ratio are not included among the posted statistical analyses, so the effect is described here on the absolute (difference) scale that the registry reports.
What the estimate means. Under the fitted negative binomial model, the adjusted annualized rate of moderate or severe exacerbations was an estimated 0.324 events per participant-year lower with dupilumab than with placebo. Put differently, across 100 participant-years of treatment, the model estimates roughly 32 fewer moderate or severe exacerbations with dupilumab.
What it does not mean. It does not mean that 32.4% of patients avoided an exacerbation, nor that each patient had 0.324 fewer events. Exacerbation counts are highly uneven across patients: many have none and a few have several, which is exactly why a negative binomial rather than a Poisson model was used. The difference is a population-average contrast in rates, not a per-patient guarantee. It is also an absolute difference, so it cannot be read as a percentage reduction without the placebo rate, which is not part of the posted analysis.
Precision. The 95% CI from −0.508 to −0.140 excludes zero and indicates that the data are compatible with reductions ranging from modest (about 14 events per 100 participant-years) to larger (about 51 events per 100 participant-years). Because the interval was derived with the delta method, it relies on a normal approximation of a nonlinear function of model parameters, which is generally adequate at this sample size.
The p-value. P = 0.0005 is below the two-sided 0.049 threshold, so the superiority hypothesis was rejected and the hierarchy was allowed to proceed. The p-value says how surprising the data would be if there were no difference in rates; it does not measure how large or clinically important the reduction is. That information comes from the estimate and its interval.
Cautions. The estimate is covariate-adjusted, so it is conditional on the model's covariates and on the offset correctly accounting for differing treatment durations. Patients who discontinue early contribute fewer participant-years; the offset handles unequal exposure but assumes the event rate during observed follow-up is representative. The ITT framing preserves randomization, but the estimand (for example, how events after treatment discontinuation were handled) is not detailed in the registry.
6. Secondary Endpoint Results
All secondary analyses were tested within the hierarchical procedure, each performed only if the preceding endpoint was statistically significant at the two-sided 0.049 level. The table lists them in the order the registry reports them.
| Endpoint | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Pre-BD FEV1, Week 12 (L) | LS mean difference | 0.083 | 0.042 to 0.125 | <0.0001 |
| Pre-BD FEV1, Week 52 (L) | LS mean difference | 0.083 | 0.038 to 0.128 | 0.0003 |
| Pre-BD FEV1, Week 12, FeNO ≥20 ppb (L) | LS mean difference | 0.124 | 0.045 to 0.203 | 0.0022 |
| Pre-BD FEV1, Week 52, FeNO ≥20 ppb (L) | LS mean difference | 0.127 | 0.042 to 0.212 | 0.0034 |
| SGRQ total score, Week 52 | LS mean difference | −3.363 | −5.459 to −1.266 | 0.0017 |
| SGRQ improvement ≥4 points, Week 52 | Odds ratio | 1.439 | 1.096 to 1.890 | 0.0089 |
| E-RS: COPD RS-Total score, Week 52 | LS mean difference | −1.137 | −1.823 to −0.450 | 0.0012 |
| Annualized exacerbation rate, FeNO ≥20 ppb (per participant-year) | Rate difference (registry: Risk Difference) | −0.418 | −0.728 to −0.109 | 0.0052 |
Lung function: pre-bronchodilator FEV1
The adjusted mean improvement in pre-BD FEV1 relative to placebo was 0.083 L at both Week 12 (95% CI 0.042 to 0.125) and Week 52 (95% CI 0.038 to 0.128). The identical point estimates at the two time points suggest the lung-function difference, once established, was maintained over the treatment year. The slightly wider interval at Week 52 is consistent with more missing data and greater variability at the later visit.
In the FeNO ≥20 ppb subgroup the estimated differences were larger, 0.124 L at Week 12 and 0.127 L at Week 52, but with notably wider intervals (0.045 to 0.203 and 0.042 to 0.212) because the subgroup is smaller. These larger point estimates are compatible with a stronger effect in participants with higher FeNO, but a subgroup estimate alone is not a formal test of effect modification; that would require a treatment-by-FeNO interaction test, which is not among the posted analyses.
Health-related quality of life: SGRQ
On the SGRQ total score, where lower scores indicate better health status, the adjusted mean change favored dupilumab by −3.363 points (95% CI −5.459 to −1.266). The responder analysis looked at the same instrument differently: the odds of achieving an improvement of at least 4 points at Week 52 were 1.439 times higher with dupilumab (95% CI 1.096 to 1.890), that is, an estimated 43.9% higher odds of response.
Respiratory symptoms: E-RS: COPD
The E-RS: COPD RS-Total score, where lower values indicate fewer respiratory symptoms, showed an adjusted mean difference of −1.137 (95% CI −1.823 to −0.450) at Week 52.
Exacerbations in the FeNO ≥20 ppb subgroup
The exacerbation rate difference in the FeNO ≥20 ppb subgroup was −0.418 per participant-year (95% CI −0.728 to −0.109), using the same negative binomial model and delta-method derivation as the primary analysis. The point estimate is larger in magnitude than the overall −0.324, but the confidence intervals overlap substantially. On an absolute scale, a larger difference can also arise simply because the underlying exacerbation rate is higher in a subgroup, without any change in the relative effect.
7. Safety: Serious Adverse Events
The registry reports serious adverse events by arm as the number of participants affected over the number at risk.
| Arm | Participants with serious adverse events | Participants at risk |
|---|---|---|
| Dupilumab 300 mg q2w | 65 | 469 |
| Placebo | 74 | 470 |
The number of participants with serious adverse events was numerically lower in the dupilumab arm (65 of 469) than in the placebo arm (74 of 470). No formal statistical comparison of safety is posted, which is typical: safety outcomes in efficacy trials are usually summarized descriptively rather than tested, because the trial is not powered for them and many categories are examined at once. Serious adverse events in a COPD population include exacerbations requiring hospitalization, so the safety and efficacy endpoints are not fully independent here.
8. Statistical Methodology
Negative binomial regression for exacerbation counts
Exacerbations are counts of recurrent events accumulated over variable follow-up time. A Poisson model assumes the variance equals the mean, which rarely holds for exacerbations: some patients are exacerbation-prone and others are not. The negative binomial model adds a dispersion parameter that allows the variance to exceed the mean, giving more honest standard errors.
where Yi is the number of exacerbations, ti is treatment duration (entered as a log offset so the model estimates a rate per unit time), and k is the dispersion parameter. The model naturally yields a rate ratio (exp β1); the registry instead reports a difference in adjusted annualized rates, obtained by transforming the model estimates and computing its standard error with the delta method.
Mixed-effect model for repeated measures (MMRM)
FEV1, SGRQ total score and E-RS total score were measured repeatedly over time. The MMRM uses all available post-baseline visits in one model, with visit, treatment-by-visit interaction, the baseline value and a baseline-by-visit interaction, plus the adjustment covariates (for FEV1 also age, sex and height). The treatment effect at a specific week is the least-squares (LS) mean difference at that visit.
With an unstructured or flexible within-patient covariance Σ, participants with some missing visits still contribute information, and the estimate is valid under the missing-at-random assumption: missingness may depend on observed earlier values but not on the unobserved value itself.
Logistic regression for the SGRQ responder endpoint
The proportion of participants with an SGRQ improvement of at least 4 points was analyzed with logistic regression adjusted for treatment group, region (pooled country), ICS dose, smoking status at screening and baseline SGRQ total score. The model estimates an adjusted odds ratio. The registry analyzes this endpoint in the full ITT population, which usually implies that participants without Week 52 data were classified by a prespecified rule (often as non-responders); the exact rule is not described in the registry.
Covariate adjustment
Every analysis adjusts for baseline prognostic factors. In randomized trials, adjustment is not needed to remove bias, but it reduces residual variance (for continuous outcomes) and yields more precise, and in nonlinear models conditional rather than marginal, treatment effect estimates. Adjustment for the baseline value of the outcome (FEV1, SGRQ, E-RS) and for prior exacerbation history is particularly powerful because these are strong predictors of the outcome.
Intention-to-treat principle
The primary analysis used the ITT population, analyzing participants according to the arm allocated by randomization. The MMRM analyses restrict to ITT participants with data available, which is a modified ITT in practice; the credibility of these results then depends on the missing-at-random assumption underpinning the MMRM.
9. Multiplicity and the Hierarchical Testing Procedure
BOREAS tested one primary and several secondary endpoints. Without adjustment, testing many endpoints at 0.05 each would inflate the probability of at least one false-positive claim. The registry describes a hierarchical (fixed-sequence) testing procedure at a two-sided significance level of 0.049: endpoints are tested sequentially in the order reported, and testing continues only if the previous endpoint was statistically significant at the 0.049 level.
How fixed-sequence testing controls error
Each hypothesis is tested at the full 0.049 level, but only after all hypotheses earlier in the sequence have been rejected. Once one fails, all subsequent tests are non-confirmatory. This controls the familywise type I error at 0.049 without splitting alpha.
Why 0.049 rather than 0.05
A significance level slightly below 0.05 usually indicates that a small amount of alpha was set aside for another purpose, such as an interim look. The registry does not describe the reason, so this should be treated as a design detail rather than a documented interim analysis.
| Order as reported | Endpoint | P-value | Below 0.049? |
|---|---|---|---|
| 1 | Annualized exacerbation rate (primary) | 0.0005 | Yes |
| 2 | Pre-BD FEV1, Week 12 | <0.0001 | Yes |
| 3 | Pre-BD FEV1, Week 52 | 0.0003 | Yes |
| 4 | Pre-BD FEV1, Week 12, FeNO ≥20 ppb | 0.0022 | Yes |
| 5 | Pre-BD FEV1, Week 52, FeNO ≥20 ppb | 0.0034 | Yes |
| 6 | SGRQ total score, Week 52 | 0.0017 | Yes |
| 7 | SGRQ improvement ≥4 points, Week 52 | 0.0089 | Yes |
| 8 | E-RS: COPD RS-Total score, Week 52 | 0.0012 | Yes |
| 9 | Annualized exacerbation rate, FeNO ≥20 ppb | 0.0052 | Yes |
Every posted p-value in the sequence falls below 0.049, so under the fixed-sequence logic each of these hypotheses can be regarded as confirmed with familywise error control. The ordering matters: had an early endpoint failed, later endpoints, however small their p-values, could not have been claimed as confirmatory.
10. Statistical Methods Explained
Why was a negative binomial model used instead of comparing proportions?
The endpoint is a rate of recurrent events, not a yes/no outcome. Counting only whether a patient had any exacerbation would discard information from patients with multiple events and would ignore differences in follow-up time. The negative binomial model uses every event, scales by exposure through the log treatment-duration offset, and accommodates the extra-Poisson variability typical of exacerbation data.
What does a rate difference of −0.324 per participant-year mean, and why was it derived with the delta method?
It is the adjusted dupilumab rate minus the adjusted placebo rate: about 0.324 fewer moderate or severe exacerbations for every participant-year of treatment. The negative binomial model works on the log scale, so the difference of two rates is a nonlinear function of the model coefficients. The delta method approximates its variance using a first-order Taylor expansion, which gives the standard error for the reported confidence interval.
Why was MMRM used for FEV1, SGRQ and E-RS?
These outcomes were measured at several visits. MMRM models all visits jointly, uses partial data from participants who miss later visits, and estimates the treatment effect at the target week (Week 12 or Week 52) without imputing missing values. It assumes data are missing at random given the observed data and covariates.
What does an odds ratio of 1.439 for SGRQ response mean?
The odds of achieving an SGRQ improvement of at least 4 points were an estimated 1.439 times as high with dupilumab as with placebo, after adjustment. That is a 43.9% increase in odds, not in probability. When response is common, an odds ratio overstates the corresponding relative risk, so it should not be read as "43.9% more patients responded."
Why do the FEV1 results at Weeks 12 and 52 share the same estimate but have different intervals?
Both LS mean differences are 0.083 L, but the Week 52 interval (0.038 to 0.128) is wider than the Week 12 interval (0.042 to 0.125). Later visits typically have more missing data and more between-patient variability, which increases the standard error even when the point estimate is unchanged.
Does a significant result in the FeNO ≥20 ppb subgroup prove dupilumab works better in that subgroup?
No. It shows that the effect was statistically demonstrated in that subgroup under hierarchical error control. Claiming that the effect is larger there than in participants with lower FeNO would require a formal interaction test comparing the subgroups, which is not among the posted analyses.
11. Limitations
- Absolute scale only for the primary effect: the posted primary analysis reports a rate difference; without the group-specific rates and a rate ratio, the relative size of the reduction cannot be assessed from the registry alone.
- Missing-data assumptions: MMRM analyses include only participants with available data and rely on the missing-at-random assumption; sensitivity analyses under missing-not-at-random scenarios are not described in the registry.
- Responder definition: the SGRQ responder analysis dichotomizes a continuous score at 4 points, which loses information, and the rule used for participants without Week 52 data is not stated.
- Subgroup interpretation: FeNO ≥20 ppb analyses have wider intervals and do not establish effect modification without an interaction test.
- Conditional estimates: covariate-adjusted odds ratios and negative binomial estimates are conditional on the covariates and may differ numerically from unadjusted or marginal estimates.
- Selected population: the trial enrolled patients with moderate-to-severe COPD with type 2 inflammation on inhaled maintenance therapy; results apply to that population and not to COPD in general.
- Safety summary: the registry summary of serious adverse events is descriptive and not powered to detect differences in specific or rare events.
- Limited descriptive detail: the posted analyses do not include baseline characteristics, group-specific summary values or the precise estimand strategy, which limits independent assessment of these aspects.
12. Why This Trial Matters Statistically
BOREAS is a useful teaching case because it combines the three most common analysis families in respiratory trials, count models, repeated-measures models and binary-outcome regression, inside a single, clearly described multiplicity framework.
| Concept | How it appears in BOREAS |
|---|---|
| Negative binomial regression | Annualized exacerbation rates with log exposure offset, overall and in the FeNO ≥20 ppb subgroup |
| Rate / risk difference | Absolute rate difference of −0.324 per participant-year derived by the delta method |
| MMRM | Pre-BD FEV1, SGRQ total score and E-RS total score over time |
| Logistic regression and odds ratio | SGRQ ≥4-point responder analysis, OR 1.439 |
| Covariate adjustment | Region, ICS dose, smoking status and baseline values in every model |
| Intention-to-treat | Primary analysis by randomized allocation |
| Multiplicity adjustment | Fixed-sequence hierarchy at two-sided 0.049 |
| Confidence intervals and p-values | Distinguishing magnitude and precision from statistical significance across nine tested endpoints |
| Prespecified subgroups | FeNO ≥20 ppb endpoints embedded in the error-controlled hierarchy |
13. Related Tutorials
Learn more about the methods used in this trial:
14. Related Calculators
15. Sources
- ClinicalTrials.gov: NCT03930732 — Pivotal Study to Assess the Efficacy, Safety and Tolerability of Dupilumab in Patients With Moderate-to-severe COPD With Type 2 Inflammation (BOREAS).
- PubMed: PMID 42337097.
- PubMed: PMID 42153327.
- PubMed: PMID 41794122.
- PubMed: PMID 40651490.
- PubMed: PMID 40024335.
Continue learning with Clinical Biostats
Explore the tutorials and calculators behind the methods used in this trial, or browse other trial analyses.
16. Record Summary
BOREAS compared dupilumab 300 mg every 2 weeks with placebo in 939 participants with moderate-to-severe COPD and type 2 inflammation. The primary negative binomial analysis estimated 0.324 fewer moderate or severe exacerbations per participant-year with dupilumab (95% CI −0.508 to −0.140; P = 0.0005), and all eight secondary endpoints in the reported hierarchy, covering lung function, quality of life, symptoms and the FeNO ≥20 ppb subgroup, met the two-sided 0.049 threshold in sequence. The most useful interpretation combines the absolute rate difference, the precision of each interval, the fixed-sequence error control that makes the secondary claims confirmatory, and the modelling assumptions (overdispersion, missing at random, covariate conditioning) on which each estimate rests.