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
AZALEA was a phase 4, randomized, parallel-group, quadruple-masked trial in asthma. The registry records 199 participants, two study arms, and a primary endpoint based on a diary card summary symptom score assessed 10 days after randomisation.
| Feature | AZALEA |
|---|---|
| Trial acronym | AZALEA |
| ClinicalTrials.gov identifier | NCT01444469 |
| Trial title | AZithromycin Against pLacebo in Exacerbations of Asthma |
| Condition | Asthma |
| Phase | Phase 4 |
| Status | Completed |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 199 |
| Number of arms | 2 |
| Lead sponsor | Imperial College London |
| Sponsor type | Other |
| Start date | 2011-09 |
| Primary completion date | 2014-06 |
2. Clinical Question
The registry identifies asthma as the condition under study and describes the trial as evaluating treatment. The intervention arms are Zithromax and placebo. The primary statistical question can therefore be framed as whether the randomized treatment groups differ in the diary card summary symptom score assessed 10 days after randomisation.
Population
Participants enrolled in the phase 4 trial for asthma.
Intervention
Zithromax (drug), as recorded in the intervention field.
Comparator
Placebo (drug), as recorded in the intervention field.
Primary question
How does the diary card summary symptom score differ between randomized Zithromax and placebo groups 10 days after randomisation?
The registry-defined symptom endpoint is clinically interpretable because it summarizes symptoms rather than relying on a single symptom observation. The description specifies wheezing, breathlessness and coughing as the symptoms assessed at 10 days after randomisation.
3. Trial Design
Zithromax
- Intervention type: drug
- Randomized treatment arm
Placebo
- Intervention type: drug
- Randomized comparator arm
Why the parallel randomized structure matters
Randomization is the central design feature that permits a between-group comparison without requiring investigators to decide which participants receive which intervention. In a parallel trial, the randomized groups remain the fundamental units of comparison. The statistical analysis therefore asks whether the distribution of the prespecified outcome differs between the two groups after randomisation.
The quadruple-masked designation is also important. Masking can reduce the opportunity for expectations or knowledge of treatment assignment to influence aspects of trial conduct and outcome assessment. The registry identifies the masking level but does not provide, in the information reported here, a breakdown of which four groups of people were masked.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Diary card summary symptom score | Symptoms include wheezing, breathlessness and coughing assessed at 10 days after randomisation. | 10 days after randomisation |
What the primary endpoint measures
The primary endpoint is a summary symptom score derived from diary-card assessment. The registry specifically identifies wheezing, breathlessness and coughing as the symptoms included in the assessment.
The time frame is part of the endpoint definition. A symptom score assessed 10 days after randomisation is not interchangeable with an unspecified later assessment, an overall follow-up average, or a time-to-event endpoint. Keeping the measurement window fixed is important because symptom burden can change over time, and treatment-group comparisons depend on the prespecified timing of assessment.
5. Statistical Methodology
No formal statistical analyses are posted in the ClinicalTrials.gov record. The registry nevertheless provides enough information to identify the statistical structure of the primary analysis: a randomized two-arm comparison of a symptom score measured at a defined time point.
Primary analysis framework
For a continuous summary symptom score measured once after randomisation, a conventional analysis would compare the outcome distribution between the randomized Zithromax and placebo groups. The exact method should depend on the scale of the score, its distribution, the amount and mechanism of missing data, and any prespecified analysis plan.
A negative difference would indicate a lower mean score in the Zithromax group if lower scores represent fewer symptoms. The direction of clinical improvement cannot be assumed without the scoring convention for the diary card.
ANCOVA as a possible analysis
If a baseline symptom score were available and prespecified for adjustment, analysis of covariance (ANCOVA) could compare the 10-day score between treatment groups while accounting for baseline symptom severity. This can improve precision when baseline measurements are strongly related to the follow-up outcome.
However, the registry information available here does not specify that baseline symptom score adjustment was used. It therefore should be regarded as a standard statistical option for this type of endpoint, not as a documented feature of the AZALEA analysis.
Two-sample comparison without baseline adjustment
If the endpoint were analyzed without covariate adjustment and the score were approximately continuous and reasonably well behaved, a two-sample comparison of treatment-group means could be used. The standard error of the estimated difference would then determine the confidence interval and test statistic.
The corresponding confidence interval would combine the estimated difference with its standard error. This is a generic representation of the statistical logic, not a claim about the analysis actually used in the registry.
Nonparametric analysis
If the diary score were ordinal, markedly skewed, or otherwise poorly suited to a mean-based model, a rank-based comparison could be considered. Such a method answers a somewhat different question from a mean difference, so the estimand should match the interpretation intended for the symptom score.
Regression-based analysis
A regression model can provide a treatment-group contrast while allowing prespecified covariates to be incorporated. For a continuous outcome, a linear model is a natural framework when its assumptions are reasonable. For an ordinal outcome, an ordinal regression model may be more appropriate. The registry does not identify which modeling framework was used.
Confidence intervals
A confidence interval is particularly useful for a symptom-score endpoint because it communicates the precision of the estimated treatment difference rather than reducing the analysis to a binary significant/not-significant conclusion. The interval should be interpreted together with the direction and clinical scale of the symptom score.
6. Planned Analysis
The registry records the diary card summary symptom score as the primary endpoint, assessed 10 days after randomisation. No statistical analysis is posted in the ClinicalTrials.gov record.
What is measured
A summary symptom score incorporating wheezing, breathlessness and coughing at 10 days after randomisation.
What the comparison concerns
The randomized Zithromax group is compared with the randomized placebo group.
Typical statistical approach
A two-group continuous or ordinal outcome analysis would ordinarily be selected according to the scoring scale and distribution.
Results status
Results have not been posted on ClinicalTrials.gov.
How the primary endpoint would ordinarily be analyzed
The first statistical question is whether the symptom score is best treated as a continuous measure or as an ordered categorical measure. That decision depends on the actual scoring system. If the score is approximately continuous, the treatment effect could be expressed as a mean difference, adjusted or unadjusted according to the prespecified model. If the scale is ordinal, an ordinal model or rank-based method may better preserve the information contained in the score.
The second question is whether baseline symptom information should enter the model. If baseline measurements exist and were specified for adjustment, an ANCOVA-style model can increase statistical efficiency by explaining some of the variation in the 10-day score. The adjusted treatment coefficient would then represent the expected between-group difference after accounting for the prespecified baseline covariate.
The third question concerns missing observations. A 10-day symptom endpoint requires an observation at or corresponding to the prespecified assessment window. Participants without an evaluable score create missing-outcome issues. The appropriate handling depends on why observations are missing and on the prespecified statistical analysis plan.
These choices matter because different estimators answer different questions. A mean difference, median difference, odds ratio from an ordinal model, and rank-based effect are not interchangeable descriptions of the same treatment effect.
7. Statistical Methods Explained
What is a diary card summary symptom score?
A diary card summary symptom score is an outcome constructed from repeated or structured symptom assessments recorded by participants. In AZALEA, the registry identifies wheezing, breathlessness and coughing as the symptoms assessed at 10 days after randomisation. The summary score converts those symptom observations into a single endpoint for the primary comparison.
Why compare the two randomized groups?
The purpose of randomization is to create treatment groups whose assignment is determined independently of subsequent outcomes. The primary statistical comparison therefore focuses on the difference between the randomized Zithromax and placebo groups rather than selecting participants after the outcome is observed.
Why might ANCOVA be useful?
ANCOVA can adjust a follow-up outcome for a prespecified baseline measurement. If participants begin the study with different levels of symptoms, baseline adjustment can account for some of that variation and can make the treatment-group comparison more precise. The important point is that the baseline variable and model should be specified before examining treatment effects.
When would a rank-based method be useful?
A rank-based method can be useful when a symptom score is ordinal or substantially non-normal. Rather than relying directly on the numerical distances between scores, the analysis uses their ordering. This can make fewer assumptions about the shape of the outcome distribution, although it also changes the interpretation of the treatment effect.
Why does the 10-day time frame matter?
The treatment comparison is tied to a specific assessment window. If symptom severity changes rapidly during an asthma exacerbation, measurements at different times can represent different clinical states. A treatment difference at 10 days should therefore be interpreted as a treatment difference at the registry-defined time point, not as a general statement about symptoms at every subsequent time.
Why is missing data an important issue?
If some participants do not contribute a symptom score at the primary assessment, the analysis must distinguish between the treatment effect and the mechanism producing the missing observations. Complete-case analysis, multiple imputation, mixed models, and other approaches make different assumptions. The registry information here does not specify a missing-data method.
8. Interpretation of the Primary Endpoint
The diary card summary symptom score is designed to provide a quantitative comparison of wheezing, breathlessness and coughing between the two randomized treatment groups at 10 days after randomisation. A treatment-group difference in this score would therefore describe the relative symptom burden at the specified assessment time.
A difference in a symptom score does not automatically establish an effect on every clinical outcome associated with asthma. It is a specific endpoint measured at a specific time. It should not be interpreted as a measure of long-term disease control, survival, or another outcome unless those outcomes are separately defined and analyzed.
For many symptom scales, lower scores represent fewer or less severe symptoms. But the registry information available here does not provide the numerical scoring convention. Consequently, the direction of a favorable treatment difference should be determined from the actual endpoint definition and analysis plan rather than assumed from the endpoint name alone.
If a formal comparison were reported, the p-value would address evidence against a specified null hypothesis under the statistical model. It would not describe the magnitude of the symptom difference. A complete interpretation would therefore consider the estimated treatment difference, its confidence interval, the scale of the symptom score, and the clinical meaning of the observed change.
9. Randomization and Causal Interpretation
Randomization is the foundation for the causal comparison in a parallel-group trial. Because participants are assigned to Zithromax or placebo through the randomized allocation process, the treatment groups can be compared as randomized rather than relying on observational differences between people who happened to receive different treatments.
That does not mean every post-randomisation difference is automatically attributable to treatment. Adherence, missing outcome assessments, deviations from the intended intervention, and other post-randomisation events can affect the observed data. The statistical analysis must therefore preserve the advantages of randomization while handling outcome data according to the prespecified analysis strategy.
The quadruple-masked design provides an additional layer of protection against knowledge of treatment assignment affecting trial conduct or assessment. The registry records the masking level but does not identify the four masked groups in the information available here.
10. Endpoint Scale and Estimand
A useful statistical analysis begins by defining exactly what quantity is being estimated. For AZALEA, the registry defines a symptom score at 10 days after randomisation. The treatment effect could therefore be expressed in several mathematically different ways depending on the underlying scoring system and prespecified analysis.
| Potential estimand | What it would describe | When it could be appropriate |
|---|---|---|
| Mean difference | Difference in average symptom scores between randomized groups at 10 days. | When the score can reasonably be treated as continuous and a mean is clinically interpretable. |
| Adjusted mean difference | Difference in model-adjusted 10-day symptom scores after accounting for prespecified baseline information. | When baseline adjustment is specified and a regression/ANCOVA framework is appropriate. |
| Ordinal treatment effect | Relative shift in the distribution across ordered symptom-score categories. | When the endpoint is ordinal and category ordering is central to interpretation. |
| Rank-based treatment comparison | Evidence that one treatment tends to produce higher or lower scores than the other. | When distributional assumptions for a mean-based analysis are inappropriate. |
These are alternative statistical representations rather than multiple versions of the same estimate. The appropriate estimand should be selected in advance and should correspond to the clinical question the trial is intended to answer.
11. Missing Data and Follow-Up
The primary endpoint is assessed at a defined point 10 days after randomisation. In any trial with a fixed-time symptom endpoint, an important practical issue is whether all randomized participants contribute an evaluable score at that assessment.
If outcome data are missing, the reason for missingness becomes relevant. For example, missingness related to participant withdrawal, inability to complete a diary, worsening symptoms, improvement, or administrative factors can have different implications for the statistical analysis.
A complete-case analysis uses only participants with observed endpoint values. That approach is simple but can be biased when missingness is related to treatment or outcome. Multiple imputation can incorporate information from observed variables under assumptions about the missing-data mechanism. Model-based repeated-measures approaches can also be used when repeated symptom observations are available.
The registry information available here does not specify a missing-data or imputation strategy. The appropriate method would therefore need to be taken from the trial's prespecified statistical analysis documentation if such documentation is available.
12. Multiplicity and Statistical Testing
The registry identifies one primary endpoint: the diary card summary symptom score at 10 days after randomisation. The information reported here does not identify a multiplicity strategy involving multiple primary endpoints, multiple interim looks, or a hierarchy of confirmatory hypotheses.
Multiplicity becomes important whenever several statistical tests contribute to a single confirmatory conclusion. If investigators test many endpoints or many treatment contrasts independently, the probability of obtaining at least one small p-value by chance can exceed the nominal type I error rate.
For a single primary endpoint, the central issue is instead the prespecified treatment comparison, its estimand, its confidence interval, and the assumptions supporting the chosen analysis. Secondary or exploratory analyses, if specified elsewhere, would need to be interpreted separately from the primary endpoint.
13. Interim Analysis and Sequential Testing
The registry information available here does not identify an interim statistical analysis, alpha-spending procedure, group-sequential boundary, or early stopping rule. Such features should not be assumed merely because the trial was randomized or because it was a phase 4 study.
If a trial does conduct repeated interim efficacy analyses, the statistical threshold must generally account for repeated looks at the accumulating data. Otherwise, repeatedly examining the data can increase the overall probability of a false-positive finding.
For AZALEA, the documented information establishes the primary endpoint and study design but does not establish an interim-testing framework.
14. Non-Inferiority, Factorial Design, and Bayesian Methods
| Design topic | AZALEA interpretation |
|---|---|
| Non-inferiority margin | No non-inferiority margin is identified in the registry information available here. |
| Factorial design | The registry describes a parallel design with two arms rather than a factorial design. |
| Bayesian analysis | No Bayesian statistical method is identified in the registry information available here. |
| Interim analysis | No interim statistical analysis is identified in the registry information available here. |
These distinctions matter because each design framework changes the interpretation of the statistical evidence. A non-inferiority trial, for example, requires comparison with a prespecified margin rather than simply testing whether a treatment difference is statistically different from zero. A Bayesian analysis would instead incorporate a prior distribution and produce a posterior distribution for the treatment effect. Neither framework is identified here.
15. Why the Endpoint Is a Useful Statistical Teaching Example
The AZALEA primary endpoint illustrates a common clinical-trial situation: a treatment comparison is based on a patient-level symptom outcome collected at a defined follow-up time. Unlike a time-to-event endpoint, the diary card summary symptom score is assessed at a specified time rather than describing the time until an event occurs.
| Statistical concept | How it applies to AZALEA |
|---|---|
| Randomization | Creates the treatment groups for the primary between-group comparison. |
| Parallel design | Keeps the two randomized intervention groups as separate comparison groups. |
| Masked treatment assignment | Quadruple masking can reduce the influence of treatment knowledge on trial conduct and assessment. |
| Continuous or ordinal outcome | The appropriate model depends on the actual structure of the diary card summary score. |
| Baseline adjustment | ANCOVA may improve precision if an appropriate baseline measurement is available and prespecified. |
| Confidence interval | Quantifies uncertainty around the estimated treatment contrast. |
| Missing data | Missing 10-day symptom scores require an analysis strategy rather than automatic assumptions about their values. |
| Estimand | Defines exactly which treatment-group contrast is being estimated at the 10-day assessment. |
| Clinical interpretation | The numerical treatment difference must be interpreted using the score's actual scale and clinical meaning. |
16. A Statistical Walkthrough of the Primary Comparison
Consider the analysis as a sequence of statistical decisions rather than as a single test.
Specify the estimand
Define the treatment-group contrast in the diary card summary symptom score at 10 days after randomisation.
Understand the score
Determine whether the score is continuous, ordinal, bounded, skewed, or otherwise constrained in a way that affects model choice.
Select the statistical framework
Choose an appropriate two-group comparison, regression model, or rank-based method according to the prespecified analysis plan and endpoint properties.
Estimate the treatment effect
Report the treatment contrast together with an appropriate confidence interval rather than relying solely on a hypothesis-test p-value.
Connect the number to symptoms
Interpret the magnitude and direction of the estimated difference using the actual diary-card scoring scale and clinical context.
17. What a Complete Statistical Result Would Contain
For a primary symptom-score endpoint, a statistically complete results presentation would normally make several elements visible at the same time. The treatment-group sample sizes contributing to the analysis would establish the analysis population. The observed summary statistics would describe the outcome distribution. The treatment contrast would quantify the between-group difference. A confidence interval would communicate precision. A hypothesis-test result, if part of the prespecified analysis, would address evidence against the null hypothesis.
Descriptive statistics
Group-specific summaries of the 10-day diary card symptom score establish the observed outcome distribution.
Effect estimate
The treatment contrast states how the randomized groups differ on the prespecified symptom-score scale.
Confidence interval
The interval communicates statistical precision around the estimated treatment effect.
Hypothesis test
If prespecified, the p-value evaluates the compatibility of the observed data with the null hypothesis under the selected model.
These components answer different questions. Descriptive statistics show what was observed. The effect estimate describes the treatment contrast. The confidence interval addresses uncertainty. The p-value addresses the hypothesis test. None of these elements alone provides the entire statistical interpretation.
18. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The central statistical comparison concerns the diary card summary symptom score at 10 days after randomisation between the randomized Zithromax and placebo groups.
Clinical interpretation
The clinical meaning of any observed difference depends on the scoring scale, the magnitude of the difference, and what constitutes a meaningful change in wheezing, breathlessness and coughing.
This distinction is important. Statistical significance and clinical importance are not synonyms. A sufficiently precise estimate can detect a small difference, while a clinically substantial difference may be estimated imprecisely in a smaller study. The magnitude of the effect and its confidence interval therefore belong alongside any p-value.
19. Important Limitations
- Endpoint scale: the registry information available here identifies the diary card summary symptom score and the symptoms included but does not provide the numerical scoring scale.
- Statistical analysis: no formal statistical analyses are posted in the ClinicalTrials.gov record.
- Missing data: the available registry information does not specify how missing 10-day symptom scores were handled.
- Baseline adjustment: the available registry information does not establish whether baseline symptom measurements were included in the primary model.
- Distributional assumptions: the appropriate statistical model depends on the actual distribution and measurement properties of the symptom score.
- Clinical meaning: the magnitude of a score difference cannot be interpreted clinically without knowing the score's scale and the meaning of its units.
- Generalizability: the trial concerns participants enrolled in a specific randomized phase 4 study of asthma, and the registry information available here does not provide a detailed baseline population description.
20. Why This Trial Matters Statistically
AZALEA is a useful statistical teaching case because its primary endpoint is neither a mortality endpoint nor a conventional time-to-event outcome. Instead, the trial asks a treatment question using a patient symptom score measured at a fixed follow-up time.
That structure highlights a fundamental principle of clinical-trial statistics: The statistical method should follow the endpoint. Kaplan-Meier estimation and Cox regression are natural for time-to-event outcomes, but a fixed-time symptom score calls for a different analytical framework.
The trial also illustrates why the endpoint definition should be read before selecting a statistical test. A continuous symptom score, an ordinal symptom category, and a binary responder definition can all describe symptoms while requiring different estimands and models.
| Question | Statistical issue |
|---|---|
| What was randomized? | Treatment assignment in a two-arm parallel design. |
| What was measured? | Diary card summary symptom score. |
| When was it measured? | 10 days after randomisation. |
| Which symptoms contributed? | Wheezing, breathlessness and coughing. |
| What is being compared? | The symptom-score outcome between randomized Zithromax and placebo groups. |
| What model should be used? | Determined by the score's measurement properties and the prespecified analysis plan. |
| What determines precision? | Outcome variability, effective sample size, and the statistical model. |
| What determines clinical meaning? | The score's scale, direction, magnitude, and clinically meaningful change. |
21. Trial Timeline
Trial start
The registry records September 2011 as the start date for AZALEA.
Randomized parallel trial
The study was conducted as a randomized, quadruple-masked, parallel-group phase 4 treatment trial with two arms: Zithromax and placebo.
Primary completion
The registry records June 2014 as the primary completion date.
Completed
The ClinicalTrials.gov record identifies the study status as completed.
22. Sources
- ClinicalTrials.gov: AZALEA — NCT01444469.
- PubMed: PubMed record.
Continue through Clinical Biostats
Explore statistical methods, clinical-trial analyses, and quantitative tools for understanding how trial endpoints are designed and analyzed.
23. Record Summary
AZALEA is a randomized, parallel-group, quadruple-masked phase 4 trial in asthma with 199 enrolled participants and two intervention arms: Zithromax and placebo. Its registered primary endpoint is the diary card summary symptom score assessed 10 days after randomisation, with wheezing, breathlessness and coughing identified in the endpoint description.
From a statistical perspective, the central issue is how to compare a fixed-time symptom score between randomized treatment groups. The appropriate analysis depends on the score's measurement scale, distribution, baseline information, missing-data pattern, and prespecified estimand. A mean-based comparison, ANCOVA, ordinal model, or rank-based method could each be appropriate under different endpoint specifications, but the registry information available here does not identify which formal statistical analysis was used.