← Clinical Trials
Asthma Phase 3 Two Primary Endpoints NCT02414854

LIBERTY ASTHMA QUEST: Complete Statistical Analysis of Dupilumab in Persistent Asthma

An independent statistical review of the randomized, triple-masked, placebo-controlled phase 3 LIBERTY ASTHMA QUEST trial, which compared dupilumab 200 mg and 300 mg every two weeks with matched placebo in patients with persistent asthma, using negative binomial regression for severe exacerbation rates and mixed-effect models with repeated measures for lung function.

Sponsor: Sanofi  ·  Start: April 2015  ·  Primary completion: July 2017  ·  Status: Completed
About this page

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

LIBERTY ASTHMA QUEST was a randomized, triple-masked, parallel-group phase 3 trial evaluating whether dupilumab, given every two weeks at one of two doses, reduced the rate of severe asthma exacerbations over 52 weeks and improved pre-bronchodilator FEV1 at Week 12 compared with matched placebo, in patients with persistent asthma receiving inhaled corticosteroid therapy.

1902
Enrolled
4 randomized arms
0.54
Exacerbation rate ratio
300 mg vs placebo; 95% CI 0.43–0.68
0.523
Exacerbation rate ratio
200 mg vs placebo; 95% CI 0.413–0.662
0.13 L
FEV1 difference, Wk 12
300 mg vs placebo; 95% CI 0.08–0.18
FeatureLIBERTY ASTHMA QUEST
Official titleEvaluation of Dupilumab in Patients With Persistent Asthma (Liberty Asthma Quest)
PhasePhase 3
ConditionAsthma
DesignRandomized, parallel-group, placebo-controlled, triple-masked
Arms4: dupilumab 200 mg q2w, placebo (for dupilumab 200 mg) q2w, dupilumab 300 mg q2w, placebo (for dupilumab 300 mg) q2w
Enrollment1902
Primary endpointsAnnualized rate of severe exacerbation events over 52 weeks; absolute change from baseline in pre-bronchodilator FEV1 at Week 12 (both in the ITT population)
Primary analysis methodsNegative binomial regression (exacerbations); mixed-effect model with repeated measures, MMRM (FEV1)
Error controlHierarchical (fixed-sequence) testing at the 0.05 level across the primary and key secondary outcome measures
DatesStart 2015-04-27; primary completion 2017-07-29
ClinicalTrials.govNCT02414854
SponsorSanofi (industry)

2. Clinical Question

The trial asked whether adding dupilumab to ongoing inhaled corticosteroid therapy reduces the frequency of severe asthma exacerbations over one year and improves airflow (pre-bronchodilator FEV1) within 12 weeks, relative to placebo. Because two dose levels were tested, each against its own matched placebo, the trial effectively contains two parallel randomized comparisons that share a common analysis model.

Population

Patients with persistent asthma receiving inhaled corticosteroid (ICS) therapy, with albuterol/salbutamol or levalbuterol/levosalbutamol available as reliever medication.

Intervention

Dupilumab 200 mg every two weeks (q2w) or dupilumab 300 mg q2w, added to ICS therapy.

Comparator

Placebo q2w, matched separately to each dupilumab dose (placebo for 200 mg and placebo for 300 mg).

Primary question

Does each dupilumab regimen lower the annualized rate of severe exacerbations over 52 weeks and increase pre-bronchodilator FEV1 at Week 12 compared with its matched placebo?

3. Trial Design

01
Randomize1902 enrolled into 4 arms
02
BackgroundICS therapy continued; reliever available
03
Treat q2wDupilumab 200 / 300 mg or matched placebo
04
Week 12Pre-bronchodilator FEV1 primary assessment
05
Week 52End of exacerbation counting period
200 MG COMPARISON · SAFETY AT RISK 631 vs 313

Dupilumab 200 mg q2w vs matched placebo

  • Dupilumab 200 mg every two weeks
  • Placebo (for dupilumab 200 mg) every two weeks
  • Inhaled corticosteroid therapy in both arms
  • Albuterol/salbutamol or levalbuterol/levosalbutamol as reliever
300 MG COMPARISON · SAFETY AT RISK 632 vs 321

Dupilumab 300 mg q2w vs matched placebo

  • Dupilumab 300 mg every two weeks
  • Placebo (for dupilumab 300 mg) every two weeks
  • Inhaled corticosteroid therapy in both arms
  • Albuterol/salbutamol or levalbuterol/levosalbutamol as reliever

The numbers of participants at risk in the safety data are roughly twice as large in each dupilumab arm as in its matched placebo arm, consistent with unequal allocation within each dose pair. The registry fields summarized here do not state the allocation ratio explicitly, so the pattern is best read as descriptive.

Why two placebo arms? Each placebo arm was defined as a placebo "for" a specific dupilumab dose, a common way of preserving masking when active doses differ. The consequence for analysis is that every efficacy comparison is dose-specific: 200 mg is compared with its own placebo, and 300 mg with its own placebo. The analysis models nevertheless included all four treatment groups, so the covariate effects and dispersion parameters are estimated from the whole trial.
Allocation
Randomized
Masking
Triple masking
Design model
Parallel assignment, four arms
Primary purpose
Treatment

4. Endpoints

Primary endpoints

Endpoint (registry wording)DefinitionTime frame
Annualized Rate of Severe Exacerbation Events During The 52-Week Treatment Period: Intent-to-Treat (ITT) PopulationA severe exacerbation was defined as a deterioration of asthma requiring: use of systemic corticosteroids for ≥3 days; or hospitalization or emergency room visit because of asthma, requiring systemic corticosteroids. Annualized event rate was the total number of exacerbations that occurred during the treatment period divided by the total number of participant-years treated.Baseline to Week 52
Absolute Change From Baseline in Pre-Bronchodilator Forced Expiratory Volume in 1 Second (FEV1) at Week 12: ITT PopulationFEV1 was the volume of air exhaled in the first second of a forced expiration as measured by spirometer.Baseline, Week 12

Key secondary endpoints with posted analyses

Endpoint (registry wording, shortened)PopulationTime frame
Percent change from baseline in pre-bronchodilator FEV1 at Week 12ITTBaseline, Week 12
Annualized rate of severe exacerbation events during the 52-week treatment periodITT with baseline eosinophils ≥0.15 Giga/LBaseline to Week 52
Absolute change from baseline in pre-bronchodilator FEV1 at Week 12ITT with baseline eosinophils ≥0.15 Giga/LBaseline, Week 12
Annualized rate of severe exacerbation events during the 52-week treatment periodITT with baseline eosinophils ≥0.3 Giga/LBaseline to Week 52
Absolute change from baseline in pre-bronchodilator FEV1 at Week 12ITT with baseline eosinophils ≥0.3 Giga/LBaseline, Week 12
Annualized rate of severe exacerbation events during the 52-week treatment periodITT with baseline eosinophils <0.3 Giga/LBaseline to Week 52

The two primary endpoints are statistically very different objects. The exacerbation endpoint is a count of recurrent events accumulated over a year, divided by exposure time; it is skewed, contains many zeros and is sensitive to how long each participant was observed. The FEV1 endpoint is a continuous measurement taken repeatedly at visits up to Week 12 and summarized as a change from baseline. Each requires its own model, which is why the trial used negative binomial regression for one and MMRM for the other.

5. Analysis Populations

Analysis populationRegistry descriptionUsed for
Intent-to-treat (ITT)All randomized population analyzed according to the treatment group allocated by randomizationBoth primary endpoints; percent change in FEV1
ITT with available dataFor FEV1 analyses, "Number analyzed" signifies number of participants with available data for the specified categoryMMRM analyses of FEV1 (n=954 for the 300 mg vs placebo percent-change analysis)
ITT with baseline eosinophils ≥0.15 Giga/LITT participants in this baseline eosinophil categorySecondary exacerbation and FEV1 analyses
ITT with baseline eosinophils ≥0.3 Giga/LITT participants in this baseline eosinophil categorySecondary exacerbation and FEV1 analyses
ITT with baseline eosinophils <0.3 Giga/LITT participants in this baseline eosinophil categorySecondary exacerbation analysis

The eosinophil populations are defined by a baseline measurement, so they are subsets of randomized participants in which the treatment comparison remains randomized. They are, however, not mutually exclusive: everyone with eosinophils ≥0.3 Giga/L is also in the ≥0.15 Giga/L population, and the ≥0.3 and <0.3 populations partition the ITT population. These overlaps matter when reading the results as a set.

6. Primary Results: Severe Exacerbations

Both dupilumab regimens were compared with their matched placebo using a negative binomial model. The registry labels the effect measure "relative risk"; for an event-rate endpoint of this kind it is a rate ratio: the ratio of the model-estimated annualized exacerbation rates.

Dupilumab 300 mg q2w vs placebo (test 1 of the hierarchy)

Rate ratio for severe exacerbations

0.54

95% CI: 0.43–0.68   ·   P < 0.0001 (two-sided CI)

Unit: exacerbations per participant-year  ·  Population: ITT  ·  Hypothesis: superiority

Clinical Biostats interpretation

What it means. A rate ratio of 0.54 means that, after adjustment for the model covariates and for differences in observation time, the estimated annualized rate of severe exacerbations with dupilumab 300 mg was 54% of the rate with its matched placebo, a 46% lower estimated rate.

What it does not mean. It does not mean that 46% fewer participants had an exacerbation, or that each individual's risk fell by 46%. A rate ratio compares average event frequencies across groups; a participant with several events contributes more to the rate than one with none. It also says nothing directly about the absolute number of exacerbations avoided, which depends on the placebo-group rate.

Precision. The 95% confidence interval of 0.43 to 0.68 is entirely below 1, and even its upper limit corresponds to a 32% lower estimated rate. The interval describes sampling uncertainty in the population-average ratio under the negative binomial model, not the spread of responses among individuals.

The p-value. P < 0.0001 indicates that data this extreme would be very unlikely if the true rate ratio were 1. It does not measure how large the effect is; that information is in the estimate and its interval.

Cautions. This was the first test in the fixed sequence, so it was performed at the full 0.05 level. Rate ratios from a negative binomial model depend on the assumed variance structure; if overdispersion is poorly captured, the interval may be too narrow or too wide. Participants who stopped early contribute their observed follow-up via the offset, which assumes that their exacerbation rate while observed is representative of the rate they would have had over the full period.

Dupilumab 200 mg q2w vs placebo (test 3 of the hierarchy)

Rate ratio for severe exacerbations

0.523

95% CI: 0.413–0.662   ·   P < 0.0001 (two-sided CI)

Unit: exacerbations per participant-year  ·  Population: ITT  ·  Hypothesis: superiority

Clinical Biostats interpretation

What it means. The estimated annualized severe exacerbation rate with dupilumab 200 mg was 52.3% of the rate with its matched placebo, a 47.7% lower estimated rate, under the same adjusted negative binomial model.

What it does not mean. The point estimate for 200 mg (0.523) is numerically slightly lower than for 300 mg (0.54), but this does not show that the lower dose is more effective. The two estimates come from different placebo comparators, and their confidence intervals overlap almost completely. No direct 200 mg versus 300 mg comparison is reported in the registry analyses.

Precision. The interval 0.413 to 0.662 is similar in width to the 300 mg interval, as expected from comparisons of similar size.

The p-value and multiplicity. This test was eligible for confirmatory interpretation only because tests 1 and 2 were both statistically significant. Under the hierarchical procedure the whole sequence shares one 0.05 error budget, so P < 0.0001 here is a confirmatory result, not a nominal one. As before, the p-value reflects the strength of evidence against no difference, not the size of the difference.

Severe exacerbation rate ratio, dupilumab vs matched placebo (track spans 0 to 1.0, where 1.0 = no difference)
300 mg q2w, ITT
0.54
200 mg q2w, ITT
0.523

7. Primary Results: Pre-Bronchodilator FEV1 at Week 12

The FEV1 endpoint was analysed with an MMRM model, and the effect measure is the least squares (LS) mean difference in absolute change from baseline, in liters, between dupilumab and matched placebo at Week 12.

Dupilumab 300 mg q2w vs placebo (test 2 of the hierarchy)

LS mean difference in change from baseline in FEV1

0.13 L

95% CI: 0.08–0.18 L   ·   P < 0.0001 (two-sided CI)

Population: ITT with available data  ·  Hypothesis: superiority

Clinical Biostats interpretation

What it means. After adjustment for baseline FEV1 and the other covariates, the model-estimated mean improvement in pre-bronchodilator FEV1 from baseline to Week 12 was 0.13 L greater with dupilumab 300 mg than with its matched placebo.

What it does not mean. This is a difference between group means in the change from baseline. It is not the average improvement within the dupilumab arm (placebo participants may also have changed), and it does not imply that every treated participant gained 0.13 L. It is also a Week 12 estimate; it does not by itself describe lung function at Week 52.

Precision. The 95% CI of 0.08 to 0.18 L excludes zero and indicates that the data are compatible with differences between 0.08 L and 0.18 L. Whether the lower end is clinically meaningful is a clinical judgment, not a statistical one.

The p-value. P < 0.0001 says the difference is unlikely to be zero; it does not grade its clinical importance.

Cautions. The analysis used participants with available data and an MMRM framework, which handles intermittent or dropout-related missing visits validly if data are missing at random given the observed data and covariates. If participants with poorer lung function were more likely to miss visits for reasons not captured by the model, the estimate could be biased. This was test 2 in the sequence, performed after test 1 was significant.

Dupilumab 200 mg q2w vs placebo (test 4 of the hierarchy)

LS mean difference in change from baseline in FEV1

0.14 L

95% CI: 0.08–0.19 L   ·   P < 0.0001 (two-sided CI)

Population: ITT with available data  ·  Hypothesis: superiority

Clinical Biostats interpretation

What it means. The estimated adjusted mean improvement in pre-bronchodilator FEV1 at Week 12 was 0.14 L greater with dupilumab 200 mg than with its matched placebo.

What it does not mean. The similarity of 0.14 L and 0.13 L across doses is not evidence of equivalence between doses; the trial did not test doses against each other, and each estimate has its own placebo reference.

Precision. The interval of 0.08 to 0.19 L is nearly identical to that for 300 mg, reflecting similar sample sizes and variability.

The p-value and multiplicity. As test 4, this comparison could be claimed as confirmatory only because tests 1 to 3 had all succeeded. The p-value is again a statement about compatibility with no difference, not about magnitude. The same missing-at-random caution applies as for the 300 mg comparison.

Primary endpointComparisonMethodEstimate (95% CI)P-valueTest no.
Severe exacerbation rate, 52 weeks300 mg vs placeboNegative binomialRate ratio 0.54 (0.43–0.68)<0.00011
Pre-BD FEV1 change, Week 12300 mg vs placeboMMRMLS mean diff 0.13 L (0.08–0.18)<0.00012
Severe exacerbation rate, 52 weeks200 mg vs placeboNegative binomialRate ratio 0.523 (0.413–0.662)<0.00013
Pre-BD FEV1 change, Week 12200 mg vs placeboMMRMLS mean diff 0.14 L (0.08–0.19)<0.00014

8. Secondary Results

The posted secondary analyses compare dupilumab 300 mg q2w with its matched placebo. They continue the same hierarchical sequence, so each result carries confirmatory weight only if every earlier test in the sequence was significant.

Test no.EndpointPopulationEstimate (95% CI)P-value
5Percent change in pre-BD FEV1, Week 12 (MMRM, n=954)ITTLS mean diff 9.41 (5.74–13.07) percentage points<0.0001
6Severe exacerbation rate, 52 weeks (negative binomial)Eosinophils ≥0.15 Giga/LRate ratio 0.402 (0.307–0.526)<0.0001
7Absolute change in pre-BD FEV1, Week 12 (MMRM)Eosinophils ≥0.15 Giga/LLS mean diff 0.15 L (0.09–0.21)<0.0001
8Severe exacerbation rate, 52 weeks (negative binomial)Eosinophils ≥0.3 Giga/LRate ratio 0.326 (0.234–0.454)<0.0001
9Absolute change in pre-BD FEV1, Week 12 (MMRM)Eosinophils ≥0.3 Giga/LLS mean diff 0.24 L (0.16–0.32)<0.0001
10Severe exacerbation rate, 52 weeks (negative binomial)Eosinophils <0.3 Giga/LRate ratio 0.834 (0.608–1.144)0.2599
Severe exacerbation rate ratio, 300 mg vs matched placebo, by population (track spans 0 to 1.0)
ITT
0.54
Eos ≥0.15 Giga/L
0.402
Eos ≥0.3 Giga/L
0.326
Eos <0.3 Giga/L
0.834

Reading the eosinophil gradient

The rate ratios form a clear ordering: 0.326 in the ≥0.3 Giga/L population (a 67.4% lower estimated rate), 0.402 in the ≥0.15 Giga/L population (59.8% lower), 0.54 overall (46% lower) and 0.834 in the <0.3 Giga/L population (16.6% lower). FEV1 differences follow the same direction, from 0.13 L overall to 0.15 L at ≥0.15 Giga/L and 0.24 L at ≥0.3 Giga/L.

In the <0.3 Giga/L population the interval of 0.608 to 1.144 includes 1, and P = 0.2599. This does not show that dupilumab has no effect in that group: the interval is compatible with a moderate reduction as well as a small increase. It shows that the data in this subset do not establish a difference from placebo.

Gradient is not the same as a demonstrated interaction. A significant result in one eosinophil group and a nonsignificant result in another does not by itself prove that the treatment effect differs between the groups. That claim requires a treatment-by-eosinophil interaction test or a comparable prespecified analysis, which is not among the analyses posted to the registry. The consistent ordering across nested populations and both endpoints is nonetheless more persuasive than a single subgroup contrast would be.
Where the hierarchy stopped. Test 10 (exacerbations in the <0.3 Giga/L population) was not statistically significant at the 0.05 level. Under a fixed-sequence procedure, testing is performed only if all previous outcome measures were significant, so any outcome measures placed after test 10 in the prespecified order cannot be claimed as confirmatory, whatever their nominal p-values.

Percent change in FEV1

The percent-change analysis (test 5) estimated a 9.41 percentage-point greater improvement with 300 mg than placebo (95% CI 5.74 to 13.07). Percent change and absolute change carry similar information, but percent change scales each participant's improvement by their own baseline, which gives more weight to improvements in participants with lower baseline FEV1. Because the model also adjusts for baseline FEV1, the two analyses answer closely related questions and would be expected to agree in direction, as they do here.

9. Safety: Serious Adverse Events

The registry reports serious adverse events as participants affected out of participants at risk in each arm. No formal between-arm statistical comparison of serious adverse events is posted.

ArmParticipants with serious adverse events / at risk
Placebo (for dupilumab 200 mg) q2w26 / 313
Dupilumab 200 mg q2w51 / 631
Placebo (for dupilumab 300 mg) q2w28 / 321
Dupilumab 300 mg q2w56 / 632

Because the dupilumab arms have about twice as many participants at risk as their placebo arms, raw counts should not be compared directly; the counts must be read relative to their denominators. Serious adverse events are a heterogeneous category of events of different types and causes, so an overall proportion is a screening summary, not an estimate of a specific drug-related risk. A risk ratio with a confidence interval would be the natural summary for a formal comparison of proportions, but interpreting such a comparison would need to account for the many safety categories typically examined without multiplicity control.

10. Statistical Methodology

Negative binomial regression for exacerbation rates

The exacerbation analysis used the total number of events with onset from randomization up to Week 52 or the last contact date, whichever came earlier, as the response. The model included the four treatment groups, age, region, baseline eosinophil strata, baseline ICS dose level and the number of severe exacerbation events within 1 year prior to the study as covariates, with the log-transformed standardized observation duration as an offset.

Conceptual form
log E[Yi] = log(ti) + β0 + βtrt + γ·covariatesi,   Var(Yi) = μi + k·μi2

Here Yi is the number of severe exacerbations, ti the standardized observation time, and k the dispersion parameter. The rate ratio is exp(βtrt) for a given dupilumab arm relative to its matched placebo.

The offset converts counts into rates: a participant observed for half a year with one exacerbation is treated differently from one observed for a full year with one exacerbation. The dispersion term k allows the variance to exceed the mean, which is typical of exacerbation data because a minority of participants experience several events while many have none. A Poisson model would ignore this extra variation and produce intervals that are too narrow.

Mixed-effect model with repeated measures (MMRM) for FEV1

The FEV1 analysis used the change from baseline in FEV1 values up to Week 12 as the response, with treatment, age, sex, baseline height, region, baseline eosinophil strata, baseline ICS dose level, visit, treatment-by-visit interaction, baseline FEV1 and baseline-by-visit interaction as covariates.

Conceptual form
ΔFEV1ij = μ + trti + visitj + (trt×visit)ij + β·FEV10i + (FEV10×visit)ij + covariatesi + εij

The residuals εij for the same participant are allowed to be correlated across visits. The treatment effect at Week 12 is the difference in LS means for the treatment-by-visit combination at that visit.

The treatment-by-visit interaction lets the treatment effect differ at each visit, so the Week 12 estimate is not forced to equal an average over visits. Height, age and sex are natural covariates for FEV1 because they are major determinants of lung volume; adjusting for them and for baseline FEV1 reduces residual variability and narrows the confidence interval without introducing bias in a randomized comparison.

Intention-to-treat principle

The primary analyses were performed in the ITT population, including all randomized participants analysed according to the treatment group allocated by randomization. This preserves the comparability created by randomization. In the FEV1 analyses, participants contribute to the MMRM according to the visits at which they have data, so the practical population is ITT participants with available data.

Missing data

The two endpoints handle incomplete follow-up differently. In the exacerbation model, events are counted up to Week 52 or the last contact date, and the offset adjusts for shorter observation. In the FEV1 model, MMRM uses all observed visits and yields valid estimates under a missing-at-random assumption. Neither approach protects against missingness that depends on unobserved outcomes, and sensitivity analyses for alternative missing-data assumptions are not among the analyses posted to the registry.

11. Multiplicity and the Hierarchical Testing Procedure

The registry states that a hierarchical testing procedure was used to control the type I error rate at the 0.05 level. The procedure included the 2 primary outcome measures and the first 13 secondary outcome measures, and considered 2 pair-wise comparisons: dupilumab 200 mg q2w versus its placebo and dupilumab 300 mg q2w versus its placebo.

OrderComparisonEndpoint / populationResult
1300 mg vs placeboSevere exacerbation rate, ITTSignificant; continue
2300 mg vs placeboFEV1 absolute change, ITTSignificant; continue
3200 mg vs placeboSevere exacerbation rate, ITTSignificant; continue
4200 mg vs placeboFEV1 absolute change, ITTSignificant; continue
5300 mg vs placeboFEV1 percent change, ITTSignificant; continue
6300 mg vs placeboSevere exacerbation rate, eosinophils ≥0.15 Giga/LSignificant; continue
7300 mg vs placeboFEV1 absolute change, eosinophils ≥0.15 Giga/LSignificant; continue
8300 mg vs placeboSevere exacerbation rate, eosinophils ≥0.3 Giga/LSignificant; continue
9300 mg vs placeboFEV1 absolute change, eosinophils ≥0.3 Giga/LSignificant; continue
10300 mg vs placeboSevere exacerbation rate, eosinophils <0.3 Giga/LNot significant (P = 0.2599); sequence stops

Why a fixed sequence works

Each hypothesis is tested at the full 0.05 level, but only if every earlier hypothesis was rejected. Because a false positive can only enter the chain through the first false null hypothesis encountered, the familywise error rate stays at 0.05 without splitting alpha.

The price of the ordering

Power is concentrated on early hypotheses. A failure anywhere ends confirmatory testing for everything below it, even for outcomes that would have been highly significant on their own.

Why 300 mg came first

Placing the 300 mg comparisons at positions 1 and 2 gave the higher dose the first claim on alpha. The 200 mg comparisons were reached only after both 300 mg primary tests succeeded.

Subgroups inside the hierarchy

Unusually, the eosinophil populations were placed inside the confirmatory sequence, so tests 6 to 9 are controlled confirmatory findings rather than exploratory subgroup results.

12. 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. Comparing the proportion of participants with at least one exacerbation would discard information from participants with multiple events and ignore differences in follow-up time. The negative binomial model uses every event, accounts for variable observation time through the offset, and allows for the overdispersion that is characteristic of exacerbation counts.

The registry calls 0.54 a "relative risk". Is that the right term?

In this context it is a ratio of event rates (exacerbations per participant-year), usually called a rate ratio. It is interpreted like a relative risk, as a multiplicative comparison with 1 meaning no difference, but it applies to event frequency rather than to the probability of having any event. A 0.54 rate ratio does not translate into 46% fewer participants experiencing an exacerbation.

Why use MMRM for a Week 12 endpoint rather than a simple analysis of covariance at Week 12?

MMRM uses FEV1 measurements from all visits up to Week 12 and models their correlation within participants. A participant missing the Week 12 visit still contributes information through earlier visits, and under a missing-at-random assumption the Week 12 estimate remains valid. An ANCOVA at Week 12 alone would drop such participants or require imputation.

What does an LS mean difference of 0.13 L represent?

It is the difference between the model-adjusted mean changes from baseline in the two groups at Week 12, estimated at common covariate values. "Least squares" refers to the model-based averaging, which removes chance imbalances in covariates such as baseline FEV1 and height. It is an average group-level contrast, not an individual response.

Why is test 3 (200 mg exacerbations) confirmatory even though it came after two other tests?

In a fixed-sequence procedure, alpha is not divided among hypotheses. Each test uses the full 0.05 level, conditional on all preceding tests having been significant. Since tests 1 and 2 succeeded, test 3 was performed at 0.05 with the familywise error rate still controlled.

Does the nonsignificant result in eosinophils <0.3 Giga/L mean dupilumab does not work in that group?

No. The rate ratio of 0.834 with interval 0.608 to 1.144 is compatible both with a meaningful reduction and with no reduction. Absence of statistical significance is not evidence of absence of effect. The result also halted the hierarchy, which limits what can be claimed for any later outcome measures in the prespecified order.

Why were all four treatment groups in the same model?

Fitting one model to all arms estimates covariate effects and the dispersion (or residual covariance) from the full trial, which improves precision. The dose-specific contrasts, such as 300 mg versus its placebo, are then extracted from that common model.

13. Limitations

14. Why This Trial Matters Statistically

LIBERTY ASTHMA QUEST is a strong teaching case because it combines two very different primary endpoint types, two dose-specific comparisons and a long fixed-sequence hierarchy that includes biomarker-defined populations, and because that hierarchy actually stopped partway through.

ConceptHow it appears in LIBERTY ASTHMA QUEST
Count data and overdispersionSevere exacerbations modelled by negative binomial regression
Offsets and exposure timeLog-transformed standardized observation duration converts counts to annualized rates
Rate ratio0.54 (300 mg) and 0.523 (200 mg) versus matched placebo
Repeated measuresMMRM for FEV1 change up to Week 12 with treatment-by-visit interaction
Covariate adjustmentBaseline FEV1, height, age, sex, region, eosinophil strata, ICS dose, prior exacerbations
Intention-to-treatAll randomized participants analysed as allocated
MultiplicityHierarchical testing at 0.05 across primary and key secondary outcome measures for two dose comparisons
Biomarker-defined populationsEosinophil ≥0.15, ≥0.3 and <0.3 Giga/L analyses within the hierarchy
Non-significance and error controlTest 10 (P = 0.2599) stops the confirmatory sequence
Multi-arm placebo designSeparate matched placebo for each dose within a single analysis model

15. Related Tutorials

Learn more about the methods used in this trial:

16. Related Calculators

17. Sources

Explore more trial statistics

Follow the methods used in this trial into deeper tutorials, then try them out with an appropriate calculator.

18. Record Summary

LIBERTY ASTHMA QUEST compared two dupilumab doses with matched placebo in 1902 enrolled participants with persistent asthma. Both doses produced severe exacerbation rate ratios below 1 (0.54 and 0.523) and Week 12 FEV1 improvements over placebo (0.13 L and 0.14 L), all significant within a hierarchical procedure controlling the type I error rate at 0.05. Secondary analyses for 300 mg showed larger estimated effects in higher-eosinophil populations and a nonsignificant exacerbation result below 0.3 Giga/L, where the testing sequence stopped. The most useful reading combines rate ratios, adjusted mean differences, their confidence intervals and the structure of the testing hierarchy.

Clinical Biostats methodology: A trial-results page should not merely repeat the registry entry. The goal is to reconstruct the statistical story of the trial in a standardized format while clearly separating reported evidence from educational interpretation.