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
DB2113361 was a randomized, quadruple-masked, parallel-group phase 3 trial evaluating GSK573719/GW642444 125/25 mcg and its individual components in patients with chronic obstructive pulmonary disease. The primary endpoint was change from baseline in trough FEV1 on Day 169 (Week 24).
| Feature | DB2113361 |
|---|---|
| Phase | Phase 3 |
| Condition | Chronic obstructive pulmonary disease |
| Design | Randomized, quadruple-masked, parallel-group trial |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 1493 |
| Arms | 4 |
| Primary endpoint | Change from baseline in trough FEV1 on Day 169 (Week 24) |
| Primary endpoint type | Binary in the registry profile; the posted statistical analyses identify the analyzed outcome as continuous and report least squares mean differences in liters |
| Hypothesis type | Superiority |
| Primary statistical method | Mixed-effects model / repeated measures model |
| Study status | Completed |
| Start | March 1, 2011 |
| Primary completion | March 1, 2012 |
| Lead sponsor | GlaxoSmithKline |
2. Clinical Question
The primary statistical question was whether the randomized treatment groups differed in change from baseline in trough FEV1 at Day 169 (Week 24). The registry reports five primary- endpoint comparisons, each evaluated using a mixed-model repeated-measures approach.
Population
Participants enrolled in a phase 3 randomized trial of treatment for chronic obstructive pulmonary disease.
Combination intervention
GSK573719/GW642444 125/25 mcg.
Component interventions
GSK573719 125 mcg and GW642444 25 mcg.
Primary question
How did each active treatment comparison differ in adjusted mean change from baseline in trough FEV1 at Day 169?
3. Trial Design
GSK573719/GW642444 125/25 mcg
- GSK573719/GW642444 125/25 mcg
- Reported in the statistical analyses as UMEC/VI 125/25 µg
GSK573719 125 mcg
- GSK573719 125 mcg
- Reported in the statistical analyses as UMEC 125 µg
GW642444 25 mcg
- GW642444 25 mcg
- Reported in the statistical analyses as VI 25 µg
Placebo only
- Placebo only
- Used as the comparator in three of the posted primary analyses
The design was parallel, so participants were assigned to one randomized treatment arm rather than being described in the registry as moving between treatment sequences. The study was quadruple-masked, and its primary purpose was treatment.
4. Primary Endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Change From Baseline (BL) in Trough FEV1 on Day 169 (Week 24) | FEV1 is the maximal amount of air that can be forcefully exhaled in one second. Trough FEV1 measurements were taken electronically by spirometry on Days 2, 28, 56, 84, 112, 168, and 169. Baseline is the mean of assessments made 30 minutes pre-dose and 5 minutes pre-dose on Treatment Day 1. Trough FEV1 is the mean of the FEV1 values obtained at 23 and 24 hours after the previous morning's dosing. Change from baseline is trough FEV1 at the visit minus baseline. | Baseline and Day 169 |
The endpoint is therefore not simply the raw FEV1 value at Week 24. It is the change from a prespecified baseline, with trough measurements designed to represent lung function near the end of the dosing interval.
For Day 169, the registry defines trough FEV1 using the mean of FEV1 values obtained 23 and 24 hours after the morning dosing on Day 168.
5. How the FEV1 Endpoint Was Measured
The registry describes a longitudinal measurement schedule rather than a single post-baseline assessment. Trough FEV1 was measured on Days 2, 28, 56, 84, 112, 168, and 169. This repeated structure is central to understanding why a mixed-effects repeated-measures model was used.
| Assessment | Role in the registered endpoint |
|---|---|
| Baseline | Mean of assessments 30 minutes pre-dose and 5 minutes pre-dose on Treatment Day 1 |
| Day 2 | Post-baseline trough FEV1 assessment |
| Day 28 | Post-baseline trough FEV1 assessment |
| Day 56 | Post-baseline trough FEV1 assessment |
| Day 84 | Post-baseline trough FEV1 assessment |
| Day 112 | Post-baseline trough FEV1 assessment |
| Day 168 | Post-baseline trough FEV1 assessment |
| Day 169 | Primary endpoint assessment |
This repeated-measures structure provides more information than a simple comparison of two isolated means. A longitudinal model can use the pattern of observations over time while estimating the treatment contrast at the prespecified Day 169 endpoint.
6. Statistical Methodology
Mixed-effects repeated-measures model
The posted analyses use a restricted maximum likelihood (REML)-based repeated-measures approach, described in the registry as mixed models analysis. The model includes treatment, baseline, smoking status, center group, day, and day-by-baseline and day-by-treatment interactions.
The model is fitted across repeated FEV1 observations, with the treatment effect at the specified endpoint summarized as a least squares mean difference.
Covariate adjustment
The analysis adjusts for baseline, smoking status, and center group, while also allowing the relationship with baseline and treatment to vary across day through interaction terms. This is more informative than an unadjusted comparison because the reported treatment contrast is an adjusted model-based estimate.
Least squares mean difference
The registry reports the effect measure as a least squares mean difference. For each comparison, the difference is oriented according to the named treatment contrast in the analysis record.
How to read a positive estimate
A positive value means the first-named treatment in the comparison had a greater adjusted mean change from baseline in trough FEV1 than the second-named treatment.
The unit is liters because the outcome measure is change in FEV1.
Intent-to-Treat population
The primary analyses use the Intent-to-Treat (ITT) Population, defined in the registry as all participants randomized to treatment who received at least one dose of randomized study drug. For a particular presented time point, the registry notes that participants represent those with data available at that time point; participants in the ITT population without missing covariate information and with at least one post-baseline measurement are included in the analysis.
7. Primary Results
The registry posts five statistical analyses for the primary endpoint. All five use the same broad repeated-measures methodology, and all five report a two-sided 95% confidence interval and a superiority hypothesis.
| Comparison | LS mean difference | 95% CI | P-value |
|---|---|---|---|
| UMEC 125 µg vs Placebo | 0.160 L | 0.122 to 0.198 | <0.001 |
| VI 25 µg vs Placebo | 0.124 L | 0.086 to 0.162 | <0.001 |
| UMEC/VI 125/25 µg vs Placebo | 0.238 L | 0.200 to 0.276 | <0.001 |
| UMEC/VI 125/25 µg vs UMEC 125 µg | 0.079 L | 0.046 to 0.112 | <0.001 |
| UMEC/VI 125/25 µg vs VI 25 µg | 0.114 L | 0.081 to 0.148 | <0.001 |
UMEC 125 µg vs Placebo
Least squares mean difference in change from baseline
95% CI: 0.122–0.198 · P < 0.001
Contrast: UMEC 125 µg minus Placebo
The estimated adjusted difference in change from baseline in trough FEV1 was 0.160 liters, with UMEC 125 µg having the larger adjusted change. The 95% confidence interval extends from 0.122 to 0.198 liters, describing the statistical uncertainty around the estimated mean contrast under the fitted repeated-measures model.
The estimate is a mean difference, not a statement that every participant experienced a 0.160-liter improvement. It also does not describe an individual patient's expected change. Because the model is longitudinal and adjusted for prespecified covariates, the estimate is a model-based least squares mean difference rather than a simple subtraction of two observed arithmetic means.
The P-value of <0.001 addresses the statistical evidence against the null hypothesis of no difference under the specified model; it does not measure the size or clinical importance of the effect. The confidence interval is therefore essential for understanding precision.
VI 25 µg vs Placebo
Least squares mean difference in change from baseline
95% CI: 0.086–0.162 · P < 0.001
Contrast: VI 25 µg minus Placebo
The estimated adjusted difference was 0.124 liters in favor of VI 25 µg relative to placebo. The 95% CI of 0.086 to 0.162 liters quantifies uncertainty around that estimated mean difference.
Again, the effect measure is not a ratio and should not be interpreted as a percentage increase in FEV1. It is expressed directly in liters of change from baseline. The P-value indicates the strength of evidence against a zero difference under the fitted model, not the magnitude of the treatment effect.
Because the analysis is based on repeated measurements, its interpretation depends on the longitudinal model and its assumptions, including the adequacy of the model for the observed covariance and mean structure. The registry identifies the approach as a REML-based repeated-measures model.
UMEC/VI 125/25 µg vs Placebo
Least squares mean difference in change from baseline
95% CI: 0.200–0.276 · P < 0.001
Contrast: UMEC/VI 125/25 µg minus Placebo
The combination had an estimated adjusted difference of 0.238 liters in change from baseline compared with placebo. The corresponding 95% CI, 0.200 to 0.276 liters, indicates the precision of this model-based estimate.
This comparison is the largest of the three posted active-versus-placebo estimates for the primary endpoint. That descriptive observation does not by itself establish the relative contribution of each component; the two direct combination-versus-component analyses provide additional information about that question.
The P-value of <0.001 is evidence against a zero mean difference under the prespecified superiority framework. It should not be interpreted as the probability that the null hypothesis is true or as a measure of how large the treatment effect is.
UMEC/VI 125/25 µg vs UMEC 125 µg
Least squares mean difference in change from baseline
95% CI: 0.046–0.112 · P < 0.001
Contrast: UMEC/VI 125/25 µg minus UMEC 125 µg
The combination was estimated to have a 0.079-liter greater adjusted change from baseline in trough FEV1 than UMEC 125 µg alone. The 95% CI ranges from 0.046 to 0.112 liters.
This direct comparison is important statistically because it addresses a different question from the combination-versus-placebo comparison. A combination can have a larger effect than placebo partly because each active component contributes to the response. Comparing the combination directly with UMEC isolates the additional adjusted difference associated with adding the second component within this randomized four-arm framework.
The confidence interval excludes zero, while the P-value is <0.001. Neither statistic alone describes the clinical importance of the magnitude; that requires interpretation of the effect size in the context of the endpoint and population.
UMEC/VI 125/25 µg vs VI 25 µg
Least squares mean difference in change from baseline
95% CI: 0.081–0.148 · P < 0.001
Contrast: UMEC/VI 125/25 µg minus VI 25 µg
The combination had an estimated adjusted difference of 0.114 liters in change from baseline compared with VI 25 µg alone. The 95% CI extends from 0.081 to 0.148 liters, providing a quantitative measure of uncertainty around the estimated incremental difference.
The direct comparison complements the combination-versus-placebo analysis. Together, the two component comparisons describe the incremental adjusted differences observed when the combination is compared separately with each active component.
As with the other analyses, the P-value of <0.001 is evidence against a zero difference under the specified model. It is not a measure of effect size, and the confidence interval should be considered alongside the point estimate.
8. Comparing the Five Statistical Contrasts
The figure is a visual representation of the posted point estimates. It should not be interpreted as a replacement for the confidence intervals or the model specification. In particular, the estimates come from separate model contrasts and the graph does not display their uncertainty.
| Contrast | Estimate | 95% CI | Interpretive question |
|---|---|---|---|
| UMEC vs Placebo | 0.160 L | 0.122–0.198 | Difference associated with UMEC relative to placebo |
| VI vs Placebo | 0.124 L | 0.086–0.162 | Difference associated with VI relative to placebo |
| UMEC/VI vs Placebo | 0.238 L | 0.200–0.276 | Difference associated with the combination relative to placebo |
| UMEC/VI vs UMEC | 0.079 L | 0.046–0.112 | Incremental difference of the combination relative to UMEC |
| UMEC/VI vs VI | 0.114 L | 0.081–0.148 | Incremental difference of the combination relative to VI |
9. What the Least Squares Mean Difference Means
A least squares mean difference of 0.238 liters for UMEC/VI 125/25 µg versus placebo means that the model-estimated adjusted mean change from baseline in trough FEV1 was 0.238 liters greater for the combination than for placebo at the specified primary endpoint.
The word adjusted matters. The estimate is generated from a model incorporating treatment, baseline, smoking status, center group, day, and the specified interactions. It is therefore not simply the difference between two raw sample means at Day 169.
The estimate also does not imply that every participant experienced the same difference. Individual responses can vary substantially around an adjusted group mean.
The 95% CI describes uncertainty around the estimated treatment contrast. For the combination-versus-placebo comparison, the interval is 0.200 to 0.276 liters. A narrower interval generally reflects greater statistical precision than a wider interval, all else equal.
The confidence interval is not a range containing 95% of individual participant responses. It concerns uncertainty in the estimated population-level treatment contrast under the statistical framework used for the analysis.
All five posted primary analyses have P < 0.001. This establishes strong statistical evidence against a zero treatment contrast under each specified analysis, but the P-value itself does not tell the reader whether the estimated difference is 0.079, 0.114, 0.160, or 0.238 liters.
For that reason, effect estimates and confidence intervals should be reported alongside P-values rather than replacing them with a significance statement.
10. Statistical Methods Explained
Why was a mixed-effects repeated-measures model used?
The primary endpoint was derived from a series of FEV1 assessments collected across multiple days. A repeated-measures model is designed for this longitudinal structure. Rather than reducing every participant's information to one isolated post-baseline observation, the model analyzes the repeated outcome measurements together and estimates the treatment contrast at the prespecified endpoint.
Why use restricted maximum likelihood?
The registry describes the approach as restricted maximum likelihood (REML)-based. REML is a likelihood-based estimation method commonly used for variance and covariance components in mixed models. In a longitudinal analysis, those components describe how repeated measurements are related and help define the statistical model used to estimate treatment effects.
What is a least squares mean?
A least squares mean is a model-adjusted mean predicted by the fitted statistical model. It can be thought of as the expected outcome for a treatment group after accounting for the covariates and model terms included in the analysis. The reported difference between two least squares means therefore represents an adjusted treatment contrast rather than a simple unadjusted arithmetic difference.
Why was baseline included as a covariate?
Baseline FEV1 is directly related to the outcome being studied because the primary endpoint is defined as change from baseline. Including baseline in the model can improve precision by accounting for baseline differences in lung function and can provide an adjusted comparison of treatment groups.
Why include day-by-treatment and day-by-baseline interactions?
The repeated assessments occur at multiple times. A day-by-treatment interaction permits the treatment effect to vary across the assessment schedule rather than forcing a single treatment difference to apply identically at every time point. A day-by-baseline interaction similarly permits the relationship between baseline FEV1 and the outcome to vary with day.
What does the ITT population mean here?
The registry defines the ITT population as all randomized participants who received at least one dose of randomized study drug. The analysis includes participants without missing covariate information who have at least one post-baseline measurement. The important statistical principle is that randomized treatment assignment remains the basis of the efficacy comparison rather than switching participants to another randomized group because of later events.
How should the five P-values be interpreted?
Each posted comparison has a P-value of <0.001. Each P-value addresses its own treatment contrast under the corresponding model. A P-value does not quantify the magnitude of the effect, the probability that the treatment works, or the probability that the null hypothesis is true. The point estimate and confidence interval provide the information needed to understand the magnitude and precision of each contrast.
11. Multiplicity and the Five Posted Comparisons
The registry posts five primary analyses for the same registered endpoint. They comprise three comparisons against placebo and two direct comparisons of the combination with an active component.
| Comparison family | Number posted | Statistical purpose |
|---|---|---|
| Active treatment vs placebo | 3 | Compare each active regimen with placebo |
| Combination vs active component | 2 | Estimate the incremental adjusted difference for the combination |
Multiple formal comparisons create a multiplicity question because each additional hypothesis test can contribute to the overall probability of at least one false-positive finding if tests are treated independently at the same nominal threshold. The registry data identify the hypothesis type as superiority and provide the individual P-values, but do not provide a multiplicity-adjustment procedure in the information summarized here.
12. Randomization and Masking
The study was randomized, parallel, and quadruple-masked. Randomization is important because it establishes the treatment groups before outcome data are observed, supporting comparison of outcomes according to randomized assignment.
Randomization
Participants were allocated randomly to one of four treatment arms. This provides the fundamental design basis for the between-group efficacy comparisons.
Quadruple masking
The registry classifies the study as quadruple-masked, reducing opportunities for knowledge of treatment assignment to influence trial conduct or assessment.
Parallel design
Each participant was assigned to one of the four parallel treatment arms rather than the trial being described as a crossover design.
Superiority framework
The posted analyses use superiority hypotheses, so the statistical question is whether the treatment contrast differs from zero rather than whether it remains within a prespecified non-inferiority margin.
13. Safety
The registry reports serious adverse events by treatment arm. These data are presented as affected participants divided by the participants at risk in each arm.
| Treatment arm | Serious adverse events | Affected / at risk |
|---|---|---|
| Placebo | Serious adverse events | 17 / 275 |
| UMEC 125 µg | Serious adverse events | 22 / 407 |
| VI 25 µg | Serious adverse events | 20 / 404 |
| UMEC/VI 125/25 µg | Serious adverse events | 23 / 403 |
These counts describe the number affected and the corresponding number at risk in each arm. They should not be treated as an efficacy endpoint or combined with the FEV1 analysis into a single treatment-effect statistic.
The registry presents serious adverse events as affected/at-risk counts. The denominator is therefore essential to understanding the scale of the reported safety finding.
14. Analysis Population and Missing Measurements
The primary analysis population is the ITT population, but the registry specifies an important qualification for the longitudinal model: participants without missing covariate information and with at least one post-baseline measurement are included in the analysis. At a particular presented time point, the registry describes participants as those with data available at that time point.
This distinction matters because a repeated-measures analysis does not require every participant to have a complete set of measurements at every scheduled visit in order to contribute information. However, the statistical interpretation depends on assumptions about the missing-data process and the adequacy of the model for the observed longitudinal data.
15. Why This Is a Longitudinal Analysis Rather Than a Simple Week-24 Comparison
A simple analysis could compare change from baseline at Day 169 between two treatment groups. The registered analysis goes further by modeling repeated trough FEV1 measurements across the study.
| Approach | What it would use | Statistical consequence |
|---|---|---|
| Single-time-point comparison | Day 169 outcome only | Uses only the endpoint observation |
| Registered repeated-measures approach | Multiple post-baseline observations across days | Models the longitudinal outcome pattern and estimates the prespecified Day 169 contrast |
The repeated-measures approach is especially relevant because the endpoint was measured on several scheduled days. It allows the statistical analysis to account for the fact that observations from the same participant are related rather than treating all observations as independent.
16. Understanding the Direction of the Contrasts
The registry explicitly defines the direction of each least squares mean difference. This is important because reversing the treatment order would reverse the sign of the estimate while leaving the underlying magnitude of the contrast unchanged.
| Reported contrast | Definition of estimate |
|---|---|
| UMEC vs Placebo | UMEC 125 µg minus Placebo |
| VI vs Placebo | VI 25 µg minus Placebo |
| UMEC/VI vs Placebo | UMEC/VI 125/25 µg minus Placebo |
| UMEC/VI vs UMEC | UMEC/VI 125/25 µg minus UMEC 125 µg |
| UMEC/VI vs VI | UMEC/VI 125/25 µg minus VI 25 µg |
Because all five estimates are positive, the direction of every reported contrast is toward a larger adjusted change from baseline in the first-named treatment group.
17. Confidence Intervals Across the Primary Analyses
| Comparison | Estimate | Lower 95% CI | Upper 95% CI | Interval width |
|---|---|---|---|---|
| UMEC vs Placebo | 0.160 | 0.122 | 0.198 | 0.076 |
| VI vs Placebo | 0.124 | 0.086 | 0.162 | 0.076 |
| UMEC/VI vs Placebo | 0.238 | 0.200 | 0.276 | 0.076 |
| UMEC/VI vs UMEC | 0.079 | 0.046 | 0.112 | 0.066 |
| UMEC/VI vs VI | 0.114 | 0.081 | 0.148 | 0.067 |
The widths above are simple differences between the posted upper and lower confidence limits. They provide a descriptive view of the precision of the reported contrasts. The intervals should not be interpreted as ranges of individual treatment responses.
18. What the Results Do Not Establish
- Individual response: a group-level least squares mean difference does not imply that every participant experienced the same change.
- Percentage change: the reported effect measure is a difference in liters, not a percentage or ratio.
- Probability of benefit: a P-value below 0.001 does not mean that there is a corresponding probability that an individual patient will benefit.
- Clinical importance from P-values alone: statistical significance does not by itself establish whether a numerical difference is clinically meaningful.
- Independence of repeated observations: the longitudinal measurements are not treated as independent observations from unrelated participants; the mixed-model framework is designed for repeated measurements within participants.
- Multiplicity adjustment: the summarized registry information does not identify a multiplicity-adjustment method for the five posted primary comparisons.
- Individual-level prediction: these group-level estimates should not be used as individualized predictions of FEV1 response.
19. Limitations
- Multiplicity: five primary statistical contrasts are posted for the registered endpoint. The registry information summarized here does not identify an adjustment procedure across these comparisons.
- Model dependence: the reported least squares mean differences depend on the specified repeated-measures model, covariates, interactions, and covariance structure.
- Missing observations: the registry describes the analysis population and availability of post-baseline measurements but does not provide a separate imputation method in the summarized information.
- Endpoint interpretation: FEV1 is a continuous physiological measure, so its statistical interpretation as a mean difference differs from interpretation of binary or time-to-event endpoints.
- Generalizability: the registry information summarized here establishes the study condition and design but does not provide a detailed baseline-characteristics table for assessing how closely the enrolled population represents other COPD populations.
- Safety interpretation: the reported serious-adverse-event counts are descriptive arm-level data and are not accompanied here by a formal comparative safety model.
- Clinical interpretation: statistical evidence for a difference does not by itself determine the clinical importance of that difference.
20. Why This Trial Matters Statistically
DB2113361 is a useful teaching example because its primary endpoint demonstrates how a modern randomized trial can combine repeated physiological measurements, covariate adjustment, and multiple prespecified treatment contrasts in a single statistical framework.
| Concept | How it appears in DB2113361 |
|---|---|
| Randomization | Four-arm randomized parallel-group design |
| Blinding | Quadruple-masked study |
| ITT analysis | Primary analyses based on the registry-defined ITT population |
| Repeated measures | Trough FEV1 measured across multiple scheduled days |
| Mixed-effects model | REML-based repeated-measures analysis |
| Covariate adjustment | Treatment, baseline, smoking status, center group, day, and specified interactions |
| Least squares mean | Adjusted treatment contrast reported in liters |
| Confidence interval | Two-sided 95% intervals around every posted primary estimate |
| P-value | All five posted comparisons report P < 0.001 |
| Superiority testing | All five analyses are identified as superiority hypotheses |
| Multiplicity | Five posted primary statistical comparisons |
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: DB2113361, NCT01313637.
- PubMed: PMID 24385182.
Continue through Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical-trial analyses organized around the methods used in randomized clinical research.
24. Record Summary
DB2113361 provides a clear example of longitudinal clinical-trial analysis in which the primary outcome is a repeated continuous physiological measurement. The registry-defined endpoint is change from baseline in trough FEV1 on Day 169, with trough FEV1 measured repeatedly during the study. The posted analyses use a REML-based repeated-measures model incorporating treatment, baseline, smoking status, center group, day, and day-by-baseline and day-by-treatment interactions.
Five primary contrasts are reported. The least squares mean differences range from 0.079 to 0.238 liters, with two-sided 95% confidence intervals that exclude zero and P-values of <0.001 for every posted comparison. The statistical interpretation therefore depends on more than the P-values: the direction and magnitude of each contrast, its confidence interval, the longitudinal model, the ITT analysis population, and the multiplicity created by the five posted comparisons all contribute to the correct reading of the results.