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
DREAM was a four-arm, parallel-group, randomized phase 2 trial designed to characterize how the rate of clinically significant asthma exacerbations changed across three intravenous doses of mepolizumab (75 mg, 250 mg and 750 mg) relative to placebo over 52 weeks.
| Feature | DREAM |
|---|---|
| Full title | Dose Ranging Efficacy And Safety With Mepolizumab in Severe Asthma |
| Phase | Phase 2 |
| Condition | Asthma |
| Design | Randomized, parallel-group, quadruple-masked, placebo-controlled |
| Arms | 4: mepolizumab 75 mg IV, 250 mg IV, 750 mg IV, and placebo (saline) IV |
| Enrollment | 621 |
| Primary endpoint | Number of clinically significant exacerbations of asthma per year (Week 0 to Week 52 or early withdrawal) |
| Primary analysis | Negative binomial regression; rate ratio for each dose versus placebo |
| Primary purpose | Treatment |
| Dates | Start November 2009; primary completion March 23, 2012 |
| ClinicalTrials.gov | NCT01000506 |
| Sponsor | GlaxoSmithKline (industry) |
2. Clinical Question
The central question in DREAM was whether intravenous mepolizumab, at any of three doses spanning a tenfold range, reduced the frequency of clinically significant asthma exacerbations compared with placebo in patients with severe asthma, and how that effect varied with dose. Because the trial was a phase 2 dose-ranging study, the dose comparison was as important to its purpose as the comparison with placebo.
Population
Patients with severe asthma. The registry lists the condition as asthma; some participants were on maintenance oral corticosteroids at baseline, which is reflected in the endpoint definition and the model covariates.
Intervention
Mepolizumab given intravenously at 75 mg, 250 mg or 750 mg, each as a separate randomized arm.
Comparator
Placebo (saline) given intravenously, under quadruple masking.
Primary question
Does each mepolizumab dose lower the annual rate of clinically significant exacerbations relative to placebo over 52 weeks, tested as superiority?
3. Trial Design
Mepolizumab 75 mg IV
- Lowest dose studied
- Intravenous administration
- Compared with placebo for the primary endpoint
Mepolizumab 250 mg IV
- Intermediate dose
- Intravenous administration
- Compared with placebo for the primary endpoint
Mepolizumab 750 mg IV
- Highest dose studied
- Intravenous administration
- Compared with placebo for the primary endpoint
Placebo (saline) IV
- Shared control group for all three dose comparisons
- Intravenous saline to maintain masking
- Reference group in every rate ratio
Why a dose-ranging design?
A phase 2 dose-ranging trial aims to locate the part of the dose–response curve where most of the achievable benefit is obtained. Including a tenfold spread (75 mg to 750 mg) against a common placebo lets investigators ask two different questions with one randomized dataset: is there an effect at all, and does more drug produce more effect. The second question matters for choosing a dose to carry forward, since a higher dose that adds no efficacy only adds exposure.
4. Randomization, Masking, and Analysis Population
Allocation was randomized to four parallel arms. Masking was quadruple, meaning participants, care providers, investigators and outcome assessors were all blinded to assignment. For an endpoint such as an exacerbation, where the decision to prescribe systemic corticosteroids or to admit a patient involves clinical judgment, masking of those making the decision protects the event count from being influenced by knowledge of treatment.
| Analysis population | Definition / role |
|---|---|
| Intent-to-Treat (ITT) | All participants who were randomized and who received at least one dose of study medication; used for the primary efficacy analyses. |
| Safety (at risk) | Placebo 155; mepolizumab 75 mg 153; 250 mg 152; 750 mg 156; used for serious adverse event reporting. |
The registry's ITT definition requires at least one dose of study medication. This is sometimes called a modified ITT population. Excluding randomized participants who never received any study drug is usually a small departure from strict ITT in a masked trial, because the decision not to dose cannot depend on an unknown assignment, but it is a departure worth naming.
5. Primary Endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Number of Clinically Significant Exacerbations of Asthma Per Year | Worsening of asthma which required use of oral/systemic corticosteroids (for participants on maintenance OCS, an exacerbation requiring OCS is defined as use of oral/systemic corticosteroids at least double the existing maintenance dose for at least 3 days) and/or hospitalization and/or emergency department (ED) visit. Frequency over the 52-week treatment period is expressed as exacerbation rate per year. | From randomization (Week 0) to Week 52 or early withdrawal (EW) |
Three features of this definition shape the analysis. First, the endpoint is a count of recurrent events, not a yes/no outcome: a participant can contribute zero, one or several exacerbations. Second, the definition is composite, triggered by systemic corticosteroids, hospitalization or an ED visit, so a single rate combines events of different severity. Third, follow-up ends at Week 52 or early withdrawal, so participants are observed for different lengths of time and the analysis must account for exposure.
6. Primary Results: Exacerbation Rate Ratios
Each posted analysis compares one mepolizumab dose with placebo using a negative binomial regression model, in the ITT population, under a superiority hypothesis, with two-sided 95% confidence intervals. The effect measure is the rate ratio: the number of exacerbations per year in the mepolizumab arm divided by the number per year in the placebo arm.
| Comparison | Rate ratio | 95% CI (two-sided) | P-value | Estimated rate reduction |
|---|---|---|---|---|
| Mepolizumab 75 mg IV vs placebo | 0.52 | 0.39–0.69 | <0.001 | 48% lower |
| Mepolizumab 250 mg IV vs placebo | 0.61 | 0.46–0.81 | <0.001 | 39% lower |
| Mepolizumab 750 mg IV vs placebo | 0.48 | 0.36–0.64 | <0.001 | 52% lower |
Mepolizumab 75 mg IV versus placebo
Rate ratio for clinically significant exacerbations
95% CI: 0.39–0.69 · P < 0.001
Negative binomial regression · ITT population · superiority
What it means. A rate ratio of 0.52 means that, after adjustment for the model covariates, the estimated annual exacerbation rate in the 75 mg arm was 52% of the placebo rate, a 48% lower estimated rate.
What it does not mean. It does not mean that 48% of participants avoided exacerbations, nor that each participant's own exacerbation frequency was halved. The ratio compares group-level event rates; exacerbations cluster in some individuals, and a rate ratio says nothing directly about how many people had none.
Precision. The 95% CI of 0.39–0.69 excludes 1.00 by a clear margin. Its upper limit implies that, even at the less favorable end of the range compatible with the data, the estimated rate reduction is roughly 31%. Ratio intervals are symmetric on the log scale, which is why the interval extends further below 0.52 than above it in relative terms.
The p-value. P < 0.001 indicates that data this extreme would be unlikely if the true rate ratio were 1.00. It does not quantify how large the effect is; the size is carried by the estimate and interval.
Cautions. This is one of three comparisons against the same placebo group, so the three results are correlated rather than independent confirmations. The registry does not describe how multiplicity across doses was handled, although with P < 0.001 in each comparison a conventional adjustment would not change the conclusion that each dose differed from placebo.
Mepolizumab 250 mg IV versus placebo
Rate ratio for clinically significant exacerbations
95% CI: 0.46–0.81 · P < 0.001
Negative binomial regression · ITT population · superiority
What it means. A rate ratio of 0.61 corresponds to a 39% lower estimated annual exacerbation rate with 250 mg than with placebo, conditional on the covariates in the model.
What it does not mean. The point estimate is numerically closer to 1.00 than those for 75 mg or 750 mg, but this does not show that 250 mg is less effective than the lowest dose. The three estimates come from separate arms of about 150 participants each, and between-arm differences of this size are well within ordinary sampling variation.
Precision. The 95% CI of 0.46–0.81 overlaps substantially with the intervals for the other two doses. Its upper limit still lies below 1.00, so the data are compatible with reductions ranging from modest to large, but not with no effect.
The p-value. The same reported threshold (P < 0.001) applies here as for the other doses. Identical p-value thresholds across comparisons do not mean identical effects; the p-value reflects both effect size and precision.
Cautions. Differences between doses would require a direct dose-versus-dose comparison or a dose–response model. The posted analyses compare each dose only with placebo.
Mepolizumab 750 mg IV versus placebo
Rate ratio for clinically significant exacerbations
95% CI: 0.36–0.64 · P < 0.001
Negative binomial regression · ITT population · superiority
What it means. A rate ratio of 0.48 means the estimated annual exacerbation rate with 750 mg was 48% of the placebo rate, a 52% lower estimated rate after covariate adjustment.
What it does not mean. Although this is the smallest point estimate, it is only slightly below the 0.52 observed at a dose ten times lower. The result does not establish that the highest dose is superior to the lowest one.
Precision. The 95% CI of 0.36–0.64 is similar in width to the 75 mg interval, as expected from arms of similar size and a shared control group. The two intervals overlap almost entirely.
The p-value. P < 0.001 supports rejecting a rate ratio of 1.00; it provides no information on whether 750 mg adds benefit beyond lower doses.
Cautions. The model uses the logarithm of time on treatment as an offset, so participants who withdrew early contribute their observed exposure rather than a full year. If early withdrawal was related to asthma control and differed between arms, the rate estimates could be affected; the registry does not describe sensitivity analyses for this.
Reading the three doses together
The pattern across doses is flat rather than rising: 0.52 at 75 mg, 0.61 at 250 mg and 0.48 at 750 mg, with overlapping confidence intervals. The most natural statistical reading is that all three doses reduced exacerbations to a broadly similar extent and that the trial did not detect a clear gradient across a tenfold dose range. That is itself an informative dose-ranging result: it suggests the lowest studied dose already sits near the plateau of the dose–response curve for this endpoint. The non-monotonic ordering of point estimates (250 mg appearing weakest) is best treated as noise rather than biology, absent a formal dose–response analysis.
7. Safety: Serious Adverse Events
The ClinicalTrials.gov record reports the number of participants with at least one serious adverse event (SAE) in each arm.
| Arm | Participants with SAE | Participants at risk |
|---|---|---|
| Placebo IV | 25 | 155 |
| Mepolizumab 75 mg IV | 20 | 153 |
| Mepolizumab 250 mg IV | 24 | 152 |
| Mepolizumab 750 mg IV | 19 | 156 |
SAE counts were similar across the four arms, with none of the mepolizumab arms showing more participants with SAEs than placebo, and no rising pattern with dose. No formal statistical comparison of SAEs is posted, and none would be expected: safety comparisons in a trial of this size are descriptive, and the absence of a numerical difference is not evidence of equivalence. In a severe asthma population, some SAEs will themselves be asthma-related hospitalizations, which overlap with the efficacy endpoint; the SAE table does not separate these.
8. Statistical Methodology
Negative binomial regression for exacerbation counts
The number of exacerbations per participant was analyzed with a negative binomial regression model. Count data are often first modeled with a Poisson distribution, which assumes the variance equals the mean. Exacerbations do not behave this way: many patients have none while a minority have several, producing variance larger than the mean (overdispersion). The negative binomial model adds a dispersion parameter that allows for this extra between-patient variability. Ignoring overdispersion would make standard errors too small and confidence intervals too narrow.
where Yi is the number of exacerbations for participant i and ti is time on treatment. The rate ratio for each dose versus placebo is exp(βdose).
The offset for time on treatment
The model includes the logarithm of time on treatment as an offset. An offset is a term with its coefficient fixed at 1. It converts the model from one for raw counts into one for event rates, so a participant followed for six months who had one exacerbation is treated as having a rate comparable to a participant followed for a year with two. This is how the endpoint can be expressed as exacerbations per year even though follow-up ended at Week 52 or early withdrawal.
Covariate adjustment
The registry lists the covariates as treatment group, baseline maintenance OCS therapy (OCS vs no OCS), region, exacerbations in the year prior to the study (as an ordinal variable) and baseline percent predicted FEV1. These are strong predictors of future exacerbations. In a randomized trial, adjusting for prognostic baseline variables does not correct bias (randomization handles that) but improves precision by explaining outcome variation that would otherwise sit in the error term. Treating prior-year exacerbations as ordinal preserves their ordering without assuming a linear effect per event.
Rate ratio as the effect measure
A rate ratio below 1 indicates fewer exacerbations per unit of follow-up time in the mepolizumab arm. The percentage reduction is 1 minus the rate ratio. It is a relative measure; the absolute number of exacerbations avoided per patient-year depends on the placebo rate.
Intention-to-treat analysis
The primary efficacy analyses used the ITT population as defined in the registry (randomized participants who received at least one dose). Analyzing participants according to their randomized arm preserves the comparability created by randomization, even if some later discontinued treatment.
9. Multiplicity and the Shared Placebo Group
DREAM produced three primary-endpoint comparisons, one per dose, all against the same placebo arm. Two statistical consequences follow.
Multiple comparisons
Testing three doses against placebo increases the chance that at least one comparison is significant by chance alone. Designs such as Dunnett-type or hierarchical testing are commonly used to control this. The registry does not describe the multiplicity strategy for DREAM.
Correlated estimates
Because every rate ratio shares the same denominator (the placebo rate), the three estimates move together. If the placebo arm happened to have a high exacerbation rate by chance, all three ratios would look more favorable at once.
In DREAM each comparison reported P < 0.001, which is well below the thresholds that common adjustments for three comparisons would impose. The multiplicity question therefore does not alter the conclusion that each dose differed from placebo, but the shared-control correlation is a reason to read the three results as one body of evidence rather than three independent replications.
10. Statistical Methods Explained
Why was negative binomial regression used instead of a Poisson model?
Asthma exacerbations are overdispersed: most participants have few or none, while some have many. A Poisson model assumes the variance equals the mean and would understate uncertainty. The negative binomial model estimates an extra dispersion parameter, giving more honest standard errors and confidence intervals for the rate ratios.
What does a rate ratio of 0.52 mean in practical terms?
It means the estimated exacerbation rate with mepolizumab 75 mg was 52% of the placebo rate, a 48% lower rate. If a placebo group averaged a given number of exacerbations per year, the model estimates about half that number in the 75 mg group. The absolute number avoided depends on how frequent exacerbations were in the first place.
Why include log time on treatment as an offset?
Participants who withdrew early had less time to experience exacerbations. Without an offset, shorter follow-up would look like fewer events. The offset converts counts into rates per unit time, so each participant contributes information in proportion to their observed exposure.
Why adjust for baseline OCS use, region, prior exacerbations and FEV1?
These factors predict how often a patient will exacerbate regardless of treatment. Including them in the model reduces unexplained variability, which narrows confidence intervals and increases power. It does not change what is being estimated in a randomized comparison; it estimates it more precisely.
Does the smaller effect at 250 mg mean it is a worse dose?
No. The 250 mg rate ratio (0.61, 95% CI 0.46–0.81) overlaps heavily with the 75 mg (0.39–0.69) and 750 mg (0.36–0.64) intervals. The posted analyses compare each dose with placebo, not doses with each other, and the ordering of point estimates is consistent with random variation around a similar underlying effect.
Why do the confidence intervals look lopsided around the estimate?
Rate ratios are estimated on the log scale, where the interval is symmetric, and then exponentiated. On the original scale this produces an interval that extends further below the estimate than above it in absolute terms. For 0.48 (0.36–0.64), the lower limit is 0.12 below the estimate and the upper limit 0.16 above, reflecting this transformation.
11. Limitations
- Composite endpoint: the exacerbation definition combines systemic corticosteroid use, hospitalization and ED visits. A reduction in the rate may be driven mainly by the most common, least severe component; the posted rate ratios do not separate them.
- Relative measure only: the rate ratios describe proportional change. Clinical relevance also depends on the baseline exacerbation rate, which determines the absolute number of events avoided.
- Multiplicity not described: three dose comparisons against one placebo arm were reported without a registry description of how type I error was controlled across them.
- No formal dose–response analysis posted: conclusions about the shape of the dose–response relationship rest on inspection of three overlapping intervals rather than a dedicated model or test.
- Modified ITT population: the analysis population required at least one dose of study medication, a minor departure from analysing everyone randomized.
- Early withdrawal: the offset handles differing follow-up but assumes that the exacerbation rate while observed is representative. If withdrawal was related to asthma control, rates may be biased; the registry does not describe missing-data sensitivity analyses.
- Phase 2 scale: with roughly 150 participants per arm, the trial was sized for efficacy against placebo, not for detecting uncommon adverse events or modest differences between doses.
- Intravenous dosing: the results apply to the intravenous doses studied; they do not by themselves describe other routes or doses.
12. Why This Trial Matters Statistically
DREAM is a useful teaching case for the analysis of recurrent-event count data and for the logic of dose-ranging trials.
| Concept | How it appears in DREAM |
|---|---|
| Count outcomes | Exacerbations per participant over up to 52 weeks |
| Overdispersion | Motivates negative binomial rather than Poisson regression |
| Offsets and exposure | Log time on treatment converts counts to annual rates when follow-up varies |
| Rate ratio | Primary effect measure for each dose versus placebo |
| Covariate adjustment | Baseline OCS, region, prior exacerbations and FEV1 improve precision |
| Confidence intervals | Log-scale intervals that are asymmetric on the ratio scale; overlap across doses |
| P-values | All three comparisons P < 0.001, yet effect sizes differ; the p-value does not rank doses |
| Intention-to-treat | Modified ITT: randomized participants who received at least one dose |
| Multiple comparisons | Three doses against a shared placebo control |
| Dose ranging | A flat dose–response across a tenfold range informs dose selection |
13. Related Tutorials
Learn more about the methods used in this trial:
14. Related Calculators
15. Sources
- ClinicalTrials.gov: NCT01000506: Dose Ranging Efficacy And Safety With Mepolizumab in Severe Asthma (DREAM).
- Publications linked to the registry record (PubMed):
Explore the methods behind the results
Follow the statistical ideas in this trial, from count models and rate ratios to confidence intervals, through tutorials and calculators.
16. Record Summary
DREAM randomized 621 participants with severe asthma to one of three intravenous mepolizumab doses or placebo and counted clinically significant exacerbations over 52 weeks. Negative binomial regression with an exposure offset and prognostic covariates gave rate ratios of 0.52 (75 mg), 0.61 (250 mg) and 0.48 (750 mg) versus placebo, each with P < 0.001 and overlapping 95% confidence intervals. Serious adverse event counts were similar across arms. Statistically, the trial shows a consistent reduction in exacerbation rate at every dose studied with no clear dose gradient, and it illustrates how count models, offsets, relative rate measures and interval-based comparison across doses fit together in a dose-ranging design.