← Clinical Trials
COPD Phase 3 Exacerbation Rate NCT04133909

MATINEE: Complete Statistical Analysis of Mepolizumab in COPD

A statistical review of the randomized, double-blind, placebo-controlled phase 3 MATINEE trial (Mepolizumab as Add-on Treatment IN Participants With COPD Characterized by Frequent Exacerbations and Eosinophil Level), with a focus on how an annualized exacerbation rate is modelled with a negative binomial regression and how the supporting time-to-event and responder analyses should be read.

Sponsor: GlaxoSmithKline  ·  Start: 2019-10-30  ·  Primary completion: 2024-08-08  ·  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

MATINEE was a randomized, double-blind, parallel-group phase 3 trial comparing mepolizumab 100 mg with placebo as add-on treatment in people with chronic obstructive pulmonary disease (COPD) characterized by frequent exacerbations and an eosinophil-based entry criterion. Its primary endpoint was the annualized rate of moderate or severe exacerbations up to Week 104.

806
Enrolled
2 parallel arms
0.79
Primary rate ratio
95% CI 0.66–0.94
0.011
Primary p-value
Superiority test
0.77
HR, first exacerbation
95% CI 0.64–0.93
FeatureMATINEE
PhasePhase 3
ConditionPulmonary disease, chronic obstructive (COPD)
DesignRandomized, parallel-group, double-masked, placebo-controlled; primary purpose treatment
Arms2: Mepolizumab 100 mg (biological) and placebo (drug)
Enrollment806
Primary endpointAnnualized rate of moderate or severe exacerbations, up to Week 104
Primary analysisCovariate-adjusted negative binomial model; rate ratio mepolizumab 100 mg / placebo
Analysis populationModified intent-to-treat (mITT)
DatesStart 2019-10-30; primary completion 2024-08-08
ClinicalTrials.govNCT04133909
SponsorGlaxoSmithKline (industry)

2. Clinical Question

The central question was whether adding mepolizumab 100 mg to existing treatment reduces the rate of moderate or severe COPD exacerbations compared with placebo, in a population selected for a history of frequent exacerbations and an eosinophil-level criterion. Because exacerbations are recurrent events, the question is framed as a comparison of event rates over time rather than of the proportion of participants who ever have an event.

Population

Participants with COPD characterized by frequent exacerbations and an eosinophil-level criterion, as described in the registry title. The registry record does not list the eligibility thresholds among its results.

Intervention

Mepolizumab 100 mg, a biological, given as add-on treatment.

Comparator

Placebo, under double masking, given as add-on treatment.

Primary question

Is the annualized rate of moderate or severe exacerbations up to Week 104 lower with mepolizumab 100 mg than with placebo?

3. Trial Design

01
Enroll806 participants
02
RandomizeMepolizumab 100 mg or placebo
03
Double-blindAdd-on treatment period
04
Follow-upExacerbations up to Week 104
05
AnalyzemITT, by randomized arm
ARM A · 403 at risk in safety summary

Mepolizumab 100 mg

  • Mepolizumab 100 mg (biological)
  • Added to existing COPD treatment
  • Double-masked
ARM B · 401 at risk in safety summary

Placebo

  • Placebo (drug)
  • Added to existing COPD treatment
  • Double-masked

Double masking matters particularly for this endpoint. A moderate exacerbation is defined by a treatment decision (oral or systemic corticosteroids and/or antibiotics), so knowledge of the assigned arm could plausibly influence whether an episode is treated and therefore counted. Blinding protects the comparison from this kind of ascertainment bias.

4. Analysis Population and Covariate Adjustment

All efficacy analyses posted to the registry used the modified intent-to-treat (mITT) population: all randomized participants who received at least one dose of trial medication, analyzed by randomized treatment. For the patient-reported responder endpoints, only participants with baseline data available were analyzed. The analysis descriptions for those endpoints also refer to an mITT2 population, which the registry does not define in its posted analysis summaries.

Analysis populationDefinition / role
mITTAll randomized participants who received at least one dose of trial medication, analyzed by randomized treatment; used for the primary and secondary efficacy analyses.
mITT with baseline dataSubset of the mITT population with available baseline scores; used for the CAT, SGRQ and E-RS: COPD responder analyses.
Safety summaryParticipants at risk: 403 (mepolizumab 100 mg) and 401 (placebo).

The exacerbation and time-to-event models adjusted for the same set of baseline covariates:

Geographic region
Categorical covariate capturing regional differences in practice and exacerbation management.
Prior-year exacerbations
Moderate/severe exacerbations in the previous year, coded ≤2, 3, ≥4 and entered as ordinal.
Baseline % predicted FEV1
Continuous measure of airflow limitation at baseline.
Smoking status
Current vs former smoker.

Prior exacerbation history is usually the strongest single predictor of future exacerbations, so adjusting for it reduces unexplained between-participant variation and sharpens the treatment estimate. In a randomized trial, covariate adjustment is not primarily about correcting imbalance; it is about precision. The responder models adjusted for treatment group, smoking status, geographic region and the baseline score of the relevant questionnaire.

5. Endpoints

Primary endpoint

EndpointRegistry definitionTime frame
Annualized Rate of Moderate or Severe ExacerbationsModerate exacerbations: clinically significant exacerbations that require treatment with oral or systemic corticosteroids and/or antibiotics. Severe exacerbations: clinically significant exacerbations that require in-patient hospitalization (≥24 hours) or result in death.Up to Week 104

Secondary endpoints with posted statistical analyses

EndpointTime frameUnit / effect measure
Time to First Moderate or Severe ExacerbationAssessed at Weeks 8, 16, 24, 32, 40, 48, 52, 56, 64, 72, 80, 88, 96 and 104Percentage of participants; hazard ratio
CAT responders (≥2 point reduction from baseline) at Week 52Baseline and Week 52Percentage of participants; odds ratio
SGRQ total score responders (≥4 point reduction from baseline) at Week 52Baseline and Week 52Percentage of participants; odds ratio
E-RS: COPD responders (≥2 point reduction from baseline)Baseline and 4 weeks prior to Week 52Percentage of participants; odds ratio
Annualized Rate of Exacerbations Requiring ED Visit and/or HospitalizationUp to Week 104Exacerbations per year; rate ratio

These endpoints use three distinct data types: counts of recurrent events (analyzed as rates), a time to first event (analyzed with a hazard model), and binary responder status derived from continuous questionnaire scores (analyzed with logistic regression). Each effect measure answers a different question and cannot be directly compared in magnitude with the others.

6. Primary Result: Annualized Rate of Moderate or Severe Exacerbations

Rate ratio, mepolizumab 100 mg / placebo

0.79

95% CI: 0.66–0.94 (two-sided)   ·   P = 0.011

Negative binomial model · mITT population · superiority hypothesis · up to Week 104

ItemPosted value
OutcomeAnnualized rate of moderate or severe exacerbations (exacerbations per year)
ComparisonMepolizumab 100 mg vs placebo
MethodNegative binomial model with covariates of treatment group, geographic region, prior-year exacerbations (≤2, 3, ≥4 as ordinal), baseline % predicted FEV1 and smoking status; offset log(time on- and off-treatment)
Effect measureRate ratio (mepolizumab 100 mg / placebo)
Estimate (95% CI)0.79 (0.66–0.94)
P-value0.011

The registry's statistical analysis reports the ratio and its test rather than the arm-level adjusted annualized rates, so the absolute number of exacerbations avoided per participant-year cannot be read off this comparison alone.

Clinical Biostats interpretation

What the estimate means. A rate ratio of 0.79 means that, after adjustment for the model covariates, the estimated annualized rate of moderate or severe exacerbations in the mepolizumab 100 mg arm was 79% of the rate in the placebo arm, i.e. a 21% lower estimated rate. It is a ratio of population-average event rates, estimated on the log scale and exponentiated.

What it does not mean. It does not mean that 21% of participants were spared an exacerbation, nor that each participant had 21% fewer exacerbations. Exacerbation counts are highly variable between individuals; the ratio summarizes the arm-level rate, not an individual response. It is also a relative measure: the same ratio corresponds to a larger absolute reduction in a population with a higher baseline rate.

Precision. The 95% confidence interval of 0.66–0.94 excludes 1, and its range spans estimated rate reductions of roughly 6% to 34%. The data are therefore compatible with anything from a modest to a moderately large relative reduction; the point estimate is the single best guess, not a precise value.

The p-value. P = 0.011 indicates that data this far from a rate ratio of 1 would be unusual if mepolizumab had no effect on the exacerbation rate. It says nothing about how large the effect is; a smaller p-value could arise from a bigger trial with the same rate ratio. The confidence interval, not the p-value, carries the information about magnitude.

Cautions. The negative binomial model accounts for overdispersion (participants differing in their underlying exacerbation propensity), which a Poisson model would ignore and which would make its intervals too narrow. The offset uses time on- and off-treatment, so follow-up after treatment discontinuation contributes to the analysis; the estimate therefore reflects the effect of being assigned to mepolizumab rather than the effect under perfect adherence. The mITT population excludes randomized participants who received no dose, a small departure from a strict intention-to-treat analysis.

7. Secondary Endpoint Results

Time to first moderate or severe exacerbation

Hazard ratio, mepolizumab 100 mg vs placebo

0.77

95% CI: 0.64–0.93 (two-sided)   ·   P = 0.009

Cox proportional hazards model · mITT population · superiority hypothesis

The Cox model adjusted for treatment group, geographic region, prior-year exacerbations (≤2, 3, ≥4 as ordinal), baseline % predicted FEV1 and smoking status. A hazard ratio of 0.77 corresponds to a 23% lower estimated instantaneous rate of a first moderate or severe exacerbation in the mepolizumab arm. This endpoint discards all events after the first, so it answers a narrower question than the primary rate analysis: how long participants remained free of any exacerbation. The two estimates pointing in the same direction with similar magnitude is reassuring about consistency, but they are not independent confirmations; they are derived from overlapping events in the same participants.

Reading the hazard ratio

The single hazard ratio summarizes the relative hazard across the whole follow-up and is most interpretable if the hazards are roughly proportional over time. The registry does not report a test of proportionality. The interval 0.64–0.93 excludes 1, and P = 0.009 again reflects compatibility with the null rather than size of effect.

Exacerbations requiring an emergency department visit and/or hospitalization

Rate ratio, mepolizumab 100 mg / placebo

0.65

95% CI: 0.43–0.96 (two-sided)   ·   P = 0.032

Negative binomial model · mITT population · up to Week 104

The point estimate corresponds to a 35% lower estimated rate of exacerbations requiring ED care and/or hospitalization. The confidence interval is noticeably wider than for the primary endpoint (0.43–0.96 versus 0.66–0.94), which is expected: these more severe events are rarer, and fewer events mean less information and less precision. The upper limit lies close to 1, so the data remain compatible with a small effect. The registry notes that estimates were based on weighting each level of the class variables by their observed proportions, which produces marginal (population-averaged) rates rather than rates for a single reference covariate profile.

Patient-reported responder endpoints at Week 52

Responder definitionOdds ratio, mepolizumab vs placebo95% CIP-value
CAT, ≥2 point reduction from baseline0.810.60–1.090.161
SGRQ total score, ≥4 point reduction from baseline1.170.87–1.570.291
E-RS: COPD, ≥2 point reduction from baseline0.820.60–1.120.209

All three odds ratios have confidence intervals that include 1, and none of the p-values is below conventional thresholds. The point estimates do not point in a consistent direction: the SGRQ odds ratio is above 1, while the CAT and E-RS: COPD odds ratios are below 1. The appropriate reading is that these analyses did not show a difference in the odds of achieving the responder thresholds; it is not evidence that mepolizumab has no effect on symptoms or health status, because the intervals are wide enough to include modest effects in either direction.

Posted ratio estimates with 95% CIs (log scale, 0.4 to 2; vertical line = 1)
Mod/severe exac. rate (RR)
0.79 (0.66–0.94)
Time to first exac. (HR)
0.77 (0.64–0.93)
ED/hospital exac. rate (RR)
0.65 (0.43–0.96)
CAT responder (OR)
0.81 (0.60–1.09)
SGRQ responder (OR)
1.17 (0.87–1.57)
E-RS: COPD responder (OR)
0.82 (0.60–1.12)

The display places different effect measures on one ratio axis purely to compare precision and direction relative to 1. Rate ratios, hazard ratios and odds ratios are not interchangeable, and their magnitudes should not be compared with each other directly.

Multiplicity: the registry does not describe a hierarchical testing procedure or other multiplicity adjustment for the secondary endpoints. When several secondary endpoints are each tested at a nominal level, the chance that at least one reaches nominal significance by chance alone increases. The secondary p-values, including P = 0.032 for ED visits/hospitalizations, are best treated as supportive rather than confirmatory unless the protocol's testing strategy says otherwise.

8. Safety: Serious Adverse Events

ArmParticipants with ≥1 serious adverse eventParticipants at risk
Mepolizumab 100 mg101403
Placebo115401

The registry reports 101 of 403 participants with a serious adverse event in the mepolizumab 100 mg arm and 115 of 401 in the placebo arm. No formal statistical comparison of serious adverse events is posted, which is typical: safety tables are descriptive, and trials are rarely powered to test differences in specific adverse events. In COPD, serious adverse events include hospitalized exacerbations, so the serious adverse event count partly overlaps with the efficacy outcome and should not be read as a separate, independent signal.

9. Statistical Methodology

Negative binomial regression for exacerbation rates

The primary and ED/hospitalization endpoints are counts of events accumulated over different lengths of follow-up. A negative binomial model treats each participant's count as arising from a Poisson process whose rate varies between participants according to a gamma distribution. This extra-Poisson variation (overdispersion) is characteristic of exacerbation data: some participants have many events and many have none.

Conceptual form
log E[Yi] = log(ti) + β0 + β1·Treatmenti + γ′Xi,    Var(Yi) = μi + k·μi2

Here Yi is the number of exacerbations, ti the follow-up time (on- and off-treatment) entering as an offset, Xi the baseline covariates and k the dispersion parameter. The rate ratio is exp(β1), reported as 0.79 for the primary endpoint.

Why the offset matters

Participants are followed for different lengths of time, for example because of withdrawal. Including log(follow-up time) as an offset converts the model from one for counts to one for rates, so a participant followed for one year is compared fairly with one followed for two. Because the offset includes off-treatment time, events after stopping study drug are still counted, consistent with a treatment-policy view of the effect.

Cox proportional hazards model

For time to first moderate or severe exacerbation, the Cox model estimates the ratio of hazards between arms without specifying the shape of the baseline hazard, adjusting for the same covariates as the rate model. Participants without an exacerbation are censored at the end of their follow-up and contribute information up to that point.

Interpretation of the hazard ratio
h(t | X) = h0(t) · exp(β1·Treatment + γ′X)  →  HR = exp(β1)

An HR below 1 indicates a lower estimated instantaneous rate of a first exacerbation among participants still exacerbation-free. It is not a relative risk, and it is not the proportion of participants who avoid an exacerbation.

Logistic regression for responder endpoints

For CAT, SGRQ and E-RS: COPD, each participant was classified as a responder or not using a prespecified minimum change from baseline. Logistic regression then estimated the odds ratio of being a responder, adjusted for treatment group, smoking status, geographic region and the baseline score. Adjusting for the baseline score is important because participants with worse baseline scores have more room to improve by a fixed number of points.

Conceptual form
logit P(responderi) = α + β1·Treatmenti + δ·Baselinei + γ′Zi  →  OR = exp(β1)

Dichotomizing a continuous score loses information, which is one reason responder analyses often have wide confidence intervals.

Modified intent-to-treat analysis

Participants were analyzed according to randomized treatment, preserving the comparability created by randomization. The modification, requiring at least one dose, is common in double-blind trials, where dosing status is unlikely to depend on the assigned arm; it usually excludes few participants and has little effect on the comparison.

10. Statistical Methods Explained

Why was a negative binomial model used instead of a t-test or a Poisson model?

Exacerbations are counts, not continuous measurements, and follow-up time differs between participants, so comparing mean counts with a t-test would ignore exposure and the skewed distribution. A Poisson model handles counts and exposure but assumes the variance equals the mean. Exacerbation counts are overdispersed, and a Poisson model would understate uncertainty, producing confidence intervals that are too narrow and p-values that are too small. The negative binomial model adds a dispersion parameter to fix this.

What does a rate ratio of 0.79 mean in practical terms?

It means the adjusted annualized rate of moderate or severe exacerbations was estimated to be 21% lower with mepolizumab 100 mg than with placebo. To translate that into exacerbations avoided per year, one needs the placebo-arm rate: the same relative reduction yields more avoided events in a population with a higher underlying rate.

Why do the primary rate ratio (0.79) and the time-to-first-event HR (0.77) differ?

They measure different things. The rate ratio uses every exacerbation over follow-up; the hazard ratio uses only the first. A treatment that delays the first event but affects later events differently could produce different values. Here the two are close, suggesting that the relative effect is broadly similar whether measured by first events or by all events.

Why was the ED/hospitalization rate ratio (0.65) smaller but less precise?

A smaller ratio indicates a larger estimated relative reduction for the more severe events. But these events are less frequent, and the precision of a rate ratio depends mainly on the number of events. With fewer events, the 95% CI widens to 0.43–0.96, so the data are compatible with anything from a large reduction to a small one.

Why do the responder odds ratios not show a difference when exacerbations do?

Symptom and quality-of-life questionnaires, dichotomized at a threshold change, are a different outcome with higher measurement variability and less statistical information than a count of clinical events over two years. The CAT (0.81, 95% CI 0.60–1.09), SGRQ (1.17, 0.87–1.57) and E-RS: COPD (0.82, 0.60–1.12) intervals all include 1. That is an absence of evidence for a difference in responder odds, not evidence of absence.

Is a p-value of 0.011 stronger evidence of benefit than a p-value of 0.032?

Only in the narrow sense of being less compatible with the null hypothesis. Neither p-value measures the size of the effect. The ED/hospitalization endpoint (P = 0.032) actually has the larger estimated relative effect but a less precise estimate. Effect size should be judged from the estimate and its confidence interval.

11. Limitations

12. Why This Trial Matters Statistically

MATINEE is a useful teaching case for recurrent-event endpoints. It shows how a single clinical question, whether a treatment reduces exacerbations, can be approached with three different models, each targeting a different estimand, and how the results of those models should be read side by side rather than as interchangeable.

ConceptHow it appears in MATINEE
Negative binomial regressionPrimary analysis of the annualized exacerbation rate, allowing for overdispersion
Offset for exposure timelog(time on- and off-treatment) converts counts to rates and follows a treatment-policy approach
Rate ratio0.79 for moderate/severe and 0.65 for ED/hospitalization exacerbations
Cox model and hazard ratioTime to first moderate or severe exacerbation, HR 0.77
Logistic regression and odds ratioCAT, SGRQ and E-RS: COPD responder analyses
Covariate adjustmentRegion, prior exacerbations, baseline FEV1, smoking status and baseline scores
Confidence intervalsPrecision differs by event frequency: narrower for all exacerbations, wider for severe ones
Modified intent-to-treatAnalysis by randomized arm among those receiving at least one dose
MultiplicitySeveral secondary endpoints tested, with nominal p-values requiring cautious reading

13. Related Tutorials

Learn more about the methods used in this trial:

14. Related Calculators

15. Sources

Continue exploring Clinical Biostats

Connect the endpoints and methods in this trial to deeper statistical tutorials and to calculators for count, time-to-event and binary outcomes.

16. Record Summary

MATINEE compared mepolizumab 100 mg with placebo in 806 enrolled participants with COPD characterized by frequent exacerbations and an eosinophil criterion. The primary negative binomial analysis estimated a rate ratio of 0.79 (95% CI 0.66–0.94; P = 0.011) for moderate or severe exacerbations up to Week 104, supported by a hazard ratio of 0.77 (0.64–0.93; P = 0.009) for time to first exacerbation and a rate ratio of 0.65 (0.43–0.96; P = 0.032) for exacerbations requiring ED care and/or hospitalization. Responder analyses for CAT, SGRQ and E-RS: COPD produced odds ratios with confidence intervals including 1. The most useful reading combines relative effects, their uncertainty intervals, the estimand each model targets and the design features, blinding, covariate adjustment and a treatment-policy offset, that shape causal interpretation.

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