BREEZE: Complete Statistical Analysis of Treprostinil Inhalation Powder in Pulmonary Arterial Hypertension

1. Trial at a Glance

BREEZE was a completed phase 1, open-label, single-group clinical study evaluating Treprostinil Inhalation Powder (TreT) in subjects with pulmonary arterial hypertension who were currently using Tyvaso. The study enrolled 51 subjects and used no randomized comparator arm.

2. Clinical Question

The clinical question in BREEZE is fundamentally different from that of a randomized superiority trial. The study evaluated Treprostinil Inhalation Powder in subjects with pulmonary arterial hypertension who were already using Tyvaso. Its primary purpose was treatment, and the design was single-group and open-label.

This design determines what can be learned statistically. A within-subject change can describe how measurements changed after treatment exposure, but it does not provide the same causal contrast as a randomized treatment-versus-control comparison. Without a concurrent randomized comparator, a change from baseline cannot by itself establish how much of the observed change was caused by TreT rather than by time, repeated measurement, background treatment, regression to the mean, or other factors.

3. Trial Design

The study's statistical structure is therefore primarily longitudinal and within-subject. Baseline measurements provide the reference point, and follow-up measurements after exposure to TreT describe change over time. This is a natural framework for endpoints such as 6-Minute Walk Distance and patient-reported symptom or impact scores, because each subject can serve as their own baseline reference.

4. Treatment and Dose Groups

The safety information is reported in treatment-phase and optional-extension groupings corresponding to the transition from Tyvaso to TreT. The registry identifies three treatment-phase dose groups and three optional-extension groupings.

The denominator is important when interpreting these safety summaries. The treatment-phase groups and optional-extension groups should not automatically be treated as independent randomized arms. The registry describes the overall study as single-group, while the safety display separates subjects according to dose and study phase. Consequently, the affected/at-risk counts are best read as phase- and dose-specific safety summaries rather than as evidence from a randomized dose-comparison experiment.

5. Endpoints

These endpoints represent three related but statistically distinct types of information. 6MWD is an objective functional measure expressed on a continuous scale. PAH-SYMPACT is a patient-reported outcome with domain scores ranging from 0 to 4, where higher values indicate more severe symptoms or impacts. The PQ-ITD is a structured preference and satisfaction questionnaire with categorical response options.

That distinction matters because the most appropriate statistical summary depends on the measurement scale. A continuous change score can be summarized using a mean and standard deviation or a median and interquartile range, while ordinal questionnaire responses are more naturally summarized using response distributions, proportions, or other methods that respect their categorical structure.

6. Statistical Methodology

The registry identifies the endpoints and their assessment time points, but no formal statistical analyses are posted in the ClinicalTrials.gov record. The appropriate statistical framework can therefore be described based on the structure of each endpoint without attributing an unreported analysis to the study.

Change from baseline for 6MWD

The 6MWD endpoint is defined as the change from baseline to Week 3. For each subject, the basic change score is:

For a single-group study, a natural descriptive analysis is the distribution of these subject-level changes. The mean change summarizes the average numerical change, while the median change describes the middle of the distribution and is less sensitive to unusually large or small observations.

If a formal inferential analysis were prespecified for a continuous within-subject change endpoint, a one-sample method could compare the mean change with zero, provided its assumptions were appropriate. A nonparametric alternative could be considered when the distribution of changes is strongly non-normal. The choice between these approaches should be determined from the prespecified statistical analysis plan rather than selected after viewing the results.

PAH-SYMPACT domain scores

PAH-SYMPACT domain scores are constructed from item responses and range from 0 to 4. The registry explicitly states that higher scores indicate more severe symptoms or impacts. Thus, unlike 6MWD, a decrease in a domain score represents movement toward fewer or less severe reported symptoms or impacts.

The baseline-to-Week-3 and baseline-to-Week-11 endpoints can therefore be represented as subject-level changes in each applicable domain. The primary interpretive task is to preserve the direction of the scale: an apparently negative numerical change is not inherently unfavorable when the measured construct is symptom severity.

Because the endpoint is patient-reported and contains multiple domains, analysis should also distinguish the statistical unit from the questionnaire item. A subject contributes observations across domains, but the domains should not automatically be treated as independent observations. Correlation among repeated patient-reported measurements is an important consideration when formal modeling is undertaken.

PQ-ITD satisfaction and preference

The PQ-ITD contains 12 statements concerning inhaled device satisfaction and provides five response options: strongly disagree, disagree, neutral, agree, and strongly agree. These are ordered categorical responses rather than measurements on a conventional continuous scale.

A straightforward statistical presentation would show the distribution of responses for each statement. For example, the proportion selecting each of the five response categories can provide more information than collapsing all responses into a single average score. If a composite score were defined in a prespecified analysis plan, its construction and interpretation would need to respect the questionnaire's measurement properties.

Safety summaries

Safety is naturally summarized using counts of subjects experiencing adverse events, with the number affected presented alongside the number at risk. For serious adverse events, the registry provides affected/at-risk counts separately for the treatment phase and optional extension and by TreT dose grouping.

Because the study is single-group and the reported safety groupings arise from treatment phase and dose rather than randomization, these counts describe observed safety experience. They do not establish a comparative risk ratio against a concurrent control group.

7. Planned Analysis

6MWD: change from baseline to Week 3

The registry defines this endpoint as the change in 6MWD from baseline to Week 3 of treatment with TreT. For an endpoint of this type in a single-group study, the principal analysis would ordinarily begin with subject-level changes and descriptive statistics for those changes. A formal one-sample analysis could test whether the mean or another prespecified measure of change differs from zero, depending on the analysis plan and distributional assumptions.

The important distinction is between describing a change and estimating a comparative treatment effect. In this study design, a baseline-to-Week-3 change is a within-subject quantity. It does not have the same interpretation as a difference in mean change between TreT and a randomized control group.

PQ-ITD: satisfaction and preference

The PQ-ITD endpoint is based on 12 statements and five ordered response categories. An appropriate analysis would ordinarily preserve the five-category response structure and summarize the distribution of responses for each statement. If a prespecified composite or summary measure were used, its definition would determine the appropriate inferential method.

The questionnaire is also evaluating satisfaction and preference rather than a direct physiologic measurement. Statistical interpretation should therefore focus on the pattern of reported responses rather than treating the response categories as though they were equally spaced numerical measurements.

PAH-SYMPACT: baseline to Week 3

The Week-3 PAH-SYMPACT endpoint is a change score derived from the questionnaire's symptom and impact domains. For this type of endpoint, the analysis would ordinarily summarize the baseline score, Week-3 score, and subject-level change for each domain. Because higher scores represent more severe symptoms or impacts, the direction of change must be interpreted accordingly.

PAH-SYMPACT: baseline to Week 11

The Week-11 endpoint applies to subjects participating in the OEP. This is an important analysis-population distinction. The Week-11 endpoint is not defined for every enrolled subject simply because 51 subjects were enrolled; it is specifically restricted to subjects participating in the OEP.

For a longitudinal endpoint assessed at both baseline and Week 11, a natural descriptive analysis would report the distribution of subject-level changes. If a formal repeated-measures model were prespecified, it could account for the correlation among measurements within the same subject, but the registry does not identify such a model.

8. How to Interpret the Single-Group Design

The most important statistical feature of BREEZE is not a particular test statistic. It is the absence of a randomized concurrent comparator. That feature changes the meaning of almost every efficacy-related result.

This does not make a single-group analysis statistically uninformative. It simply means the estimand is different. The central quantity is generally a within-subject change rather than a between-group treatment effect. That distinction should remain visible whenever the results are interpreted.

For example, suppose a subject's 6MWD increased between baseline and Week 3. The increase is real as a descriptive change if measured correctly. But the statistical question "Did the subject improve?" is not equivalent to the causal question "Did TreT cause the improvement?" The second question requires a counterfactual: what would the same subject's Week-3 6MWD have been under an alternative treatment or under no change in treatment? A single-group design does not observe that counterfactual directly.

9. Statistical Methods Explained

Why analyze change from baseline?

A change score puts each subject's follow-up measurement on a common individual reference point. This can be especially useful when baseline measurements differ across subjects. Rather than comparing raw Week-3 values alone, the analysis asks how much each subject moved relative to their own starting measurement.

Why does the single-group design matter?

With one treatment group, there is no randomized between-group contrast. A mean change, median change, or other summary describes the observed trajectory of the treated subjects. It does not directly estimate the difference that would have been observed between TreT and a concurrent randomized comparator.

How should a PAH-SYMPACT score be interpreted?

The registry defines domain scores from 0 to 4 and states that higher scores indicate more severe symptoms or impacts. Therefore, a lower follow-up score indicates less severe reported symptoms or impacts relative to a higher score. The sign of a change must always be interpreted in the context of what the scale measures.

Why should PQ-ITD responses not automatically be averaged?

The five PQ-ITD response options are ordered categories: strongly disagree, disagree, neutral, agree, and strongly agree. The ordering contains information, but the numerical distance between adjacent categories is not necessarily equivalent. Reporting the response distribution preserves information that can be lost when ordinal responses are converted into a simple average.

Why are Week-3 and Week-11 PAH-SYMPACT analyses different?

The Week-3 endpoint is defined for the treatment period, whereas the Week-11 endpoint is specifically for subjects participating in the OEP. The Week-11 population therefore has a different scope. Analyses at the two time points should not automatically be interpreted as though they were based on exactly the same set of subjects.

What does 0/2 for serious adverse events mean?

It means that 0 subjects were affected among 2 subjects at risk in that particular treatment-phase dose grouping. It is an observed count, not evidence that the true probability of a serious adverse event is zero. With only 2 subjects at risk, the estimate is intrinsically very uncertain.

Why should the optional-extension safety counts be separated from the treatment-phase counts?

The registry reports the treatment-phase and optional-extension groupings separately. The study is single-group, and the extension groupings reflect a later phase of exposure rather than an independent randomized cohort. Combining the counts without regard to study phase could obscure differences in exposure period and population composition.

10. Safety Results

The ClinicalTrials.gov record reports serious adverse events by dose grouping and study phase. These are the principal safety counts available for the registered record.

The affected/at-risk format is useful because the numerator alone does not provide the context needed to interpret an event count. One serious adverse event among 27 subjects has a different statistical meaning from one event among 2 subjects. Small denominators produce substantial uncertainty even when the observed count is simple to report.

The optional-extension data also illustrate why exposure time and study phase matter in safety analysis. The registry separates the extension from the treatment phase rather than presenting one undifferentiated safety count. That distinction should be retained when describing the observed events.

11. Understanding Uncertainty in Small Safety Groups

Small denominators are particularly important in the BREEZE safety data. The treatment-phase 32 mcg grouping has 2 subjects at risk, while the treatment-phase 48 mcg and 64 mcg groupings have 27 and 22 subjects at risk, respectively. The optional-extension groupings have 2, 26, and 21 subjects at risk.

A zero count in a small group does not demonstrate absence of risk. If no serious adverse event is observed among a small number of subjects, the data are compatible with a wide range of underlying event probabilities. Conversely, a relatively large observed fraction in a small denominator can be highly unstable as an estimate of a broader population rate.

This is a general lesson in clinical-trial statistics: observed frequency is not the same thing as estimated underlying risk. Confidence intervals are particularly valuable for sparse event data because they communicate how much uncertainty surrounds an observed proportion.

The same principle applies to dose-specific comparisons. Even if one dose grouping has a higher observed event proportion than another, the difference may reflect random variation, differences in exposure, differences in the subjects represented in each grouping, or the small number of observations. A formal comparison would require an appropriate analysis plan and sufficiently informative data.

12. Patient-Reported Outcomes: Why Direction Matters

The PAH-SYMPACT questionnaire provides an instructive example of why statistical interpretation must begin with the measurement scale rather than with the sign of a numerical result.

This is more than a presentation issue. It prevents a common statistical error: treating every negative change as unfavorable and every positive change as favorable. The direction of a numerical effect has no inherent meaning until the underlying variable is understood.

For PAH-SYMPACT, the registry explicitly defines higher scores as more severe symptoms or impacts. Therefore, the clinically meaningful direction is toward lower domain scores. A statistical analysis should preserve that interpretation in tables, confidence intervals, and any graphical display.

The PQ-ITD requires another layer of care because it consists of 12 statements rather than a single continuously measured outcome. Agreement may indicate greater satisfaction for some statements, while the interpretation of a particular response depends on what the statement says. A response distribution is therefore often more transparent than reducing the questionnaire to one number without a prespecified scoring rule.

13. Longitudinal Interpretation

BREEZE contains assessments at multiple time points, including baseline, Week 3, and Week 11 for subjects participating in the OEP. Longitudinal data create opportunities for more informative analysis than a simple comparison of group averages, because repeated observations from the same subject are correlated.

A repeated-measures model can be useful when the scientific question concerns the trajectory over time rather than a single follow-up measurement. Depending on the endpoint and analysis plan, possibilities include mixed-effects models for continuous outcomes or models designed for ordinal questionnaire responses. But the registry does not specify such a model, so it should not be represented as the analysis actually used in BREEZE.

The distinction between observed follow-up and longitudinal modeling is important. Simply having multiple observations does not require a repeated-measures model. The analysis method should follow the endpoint definition, the intended estimand, the measurement schedule, and the prespecified statistical plan.

14. Missing Data and Analysis Population

The registered Week-11 PAH-SYMPACT endpoint is explicitly limited to subjects participating in the OEP. That wording creates an analysis-population boundary that is different from the overall enrollment of 51 subjects.

For a baseline-to-follow-up change endpoint, missing Week-3 or Week-11 measurements create an incomplete change score. The statistical consequences depend on why measurements are missing and on the analysis method used. A complete-case analysis uses only subjects with the required measurements, whereas longitudinal models or multiple-imputation approaches can make different assumptions about incomplete observations.

The choice is not merely technical. If subjects with missing follow-up measurements differ systematically from those with observed measurements, a complete-case summary may not represent the full enrolled population. Conversely, an imputation model introduces assumptions of its own. The registry does not specify a missing-data or imputation method in the information available here, so no particular approach should be treated as the registered method.

15. Statistical Interpretation of a Within-Subject Change

16. Statistical Interpretation of the PQ-ITD

The PQ-ITD endpoint is fundamentally categorical. Each of its 12 statements has five possible responses: strongly disagree, disagree, neutral, agree, and strongly agree.

For an ordinal endpoint, the most transparent statistical display is often a response table showing the number or proportion of subjects selecting each category for each statement. This preserves the shape of the response distribution and avoids assuming that the distance from strongly disagree to disagree is numerically identical to the distance from agree to strongly agree.

If a formal ordinal regression model were prespecified, it could use the ordered structure of the responses directly. Such a model would require an explicit definition of the response variable, the analysis population, and any assumptions governing the relationship among response categories. Those details are not identified in the registry information available here.

Preference is also conceptually different from satisfaction. A subject can report favorable satisfaction characteristics without necessarily expressing the same degree of preference across all device-related statements. Keeping the individual questionnaire statements visible can therefore make the statistical interpretation more faithful to the endpoint.

17. What BREEZE Does Statistically — and Does Not — Establish

This distinction is central to responsible interpretation of phase 1 and single-group clinical research. A study can provide useful information about tolerability, patient experience, and changes over time without being designed to provide the same type of confirmatory causal evidence as a randomized controlled trial.

18. Important Limitations and Interpretation Issues

• No randomized comparator: BREEZE is registered as a single-group study with one arm. Baseline-to-follow-up changes therefore do not provide a randomized estimate of comparative treatment effect.
• Open-label design: masking is listed as none. Patient-reported outcomes such as satisfaction, preference, symptoms, and impacts can be influenced by knowledge of treatment exposure.
• Single-group efficacy interpretation: an observed change after switching from Tyvaso to TreT cannot by itself distinguish treatment effects from temporal or other influences.
• Ordinal questionnaire responses: the five PQ-ITD response categories are ordered categorical data and should not automatically be treated as equally spaced numerical measurements.
• Patient-reported outcome direction: PAH-SYMPACT uses a 0-to-4 scale in which higher scores indicate more severe symptoms or impacts. The direction of a change must therefore be interpreted accordingly.
• Week-11 population: the Week-11 PAH-SYMPACT endpoint is specifically for subjects participating in the OEP, so its population is not synonymous with overall enrollment.
• Small safety denominators: some dose-phase safety groupings contain only 2 subjects at risk, making their observed event frequencies highly uncertain.
• Phase-specific safety: treatment-phase and optional-extension counts should not be combined without considering their different study periods and populations.
• Unreported formal analyses: the registry provides no posted statistical analysis objects for the listed endpoints, so the exact inferential procedures and model specifications cannot be attributed to the registry.
• Missing-data considerations: the interpretation of longitudinal changes depends on which subjects have evaluable follow-up and on the assumptions used to address incomplete observations.

19. Why This Trial Matters Statistically

BREEZE is a useful teaching example because it demonstrates a type of clinical-trial question that is sometimes misunderstood when the statistical framework developed for randomized trials is applied mechanically to a single-group study.

The most important lesson is that design determines estimand. In a randomized two-arm trial, the principal question may be a treatment-versus-control contrast. In BREEZE, the registered endpoints are centered on changes observed after treatment exposure in a single group. The statistical analysis must therefore preserve the distinction between a descriptive longitudinal estimate and a comparative causal effect.

20. Trial Timeline

21. Clinical Interpretation vs Statistical Interpretation

The two perspectives should remain connected but distinct. A statistical change estimate describes the data numerically. Clinical interpretation asks what that change means in the context of pulmonary arterial hypertension and the patient experience.

22. A Practical Statistical Reading of the BREEZE Endpoints

Step 1: Identify the estimand

For 6MWD and PAH-SYMPACT, the registered estimand is a change from baseline at a specified follow-up time. This is fundamentally a within-subject estimand. For PQ-ITD, the estimand concerns subject satisfaction and preference responses after treatment with TreT.

Step 2: Identify the measurement scale

6MWD is a continuous functional measure. PAH-SYMPACT domain scores occupy a bounded 0-to-4 scale. PQ-ITD responses are ordinal categories. The statistical method should follow that measurement structure.

Step 3: Identify the analysis population

The overall enrollment is 51 subjects, but not every endpoint necessarily uses every enrolled subject. The Week-11 PAH-SYMPACT endpoint specifically concerns subjects participating in the OEP. Safety summaries likewise use phase- and dose-specific denominators.

Step 4: Describe the observed data

For continuous changes, useful summaries include the central tendency and variability of individual changes. For ordinal responses, the distribution across response categories is central. For safety, affected and at-risk counts should remain paired.

Step 5: Quantify uncertainty

Confidence intervals can quantify uncertainty around a prespecified population parameter. Their interpretation depends on the estimator and model used. In small groups, uncertainty can be substantial even when the observed count appears straightforward.

Step 6: Respect the study design

Because BREEZE is single-group and open-label, statistical results should be interpreted as observations associated with treatment exposure and follow-up rather than as randomized comparative treatment effects.

23. Sources

• ClinicalTrials.gov: BREEZE, NCT03950739.
• PubMed: PubMed record.
• PubMed: PubMed record.

24. Record Summary

BREEZE is a completed phase 1, single-group, open-label study of Treprostinil Inhalation Powder in subjects with pulmonary arterial hypertension currently using Tyvaso. The registry reports an enrollment of 51 and defines primary endpoints around change in 6-Minute Walk Distance from baseline to Week 3, subject satisfaction and preference for inhaled treprostinil devices after 3 weeks, PAH-SYMPACT change from baseline to Week 3, and PAH-SYMPACT change from baseline to Week 11 for subjects participating in the OEP.

Statistically, the defining feature is the single-group design. The central efficacy quantities are within-subject changes rather than randomized between-group contrasts. That makes baseline-to-follow-up analysis appropriate for describing observed trajectories while limiting causal interpretation. The PAH-SYMPACT endpoints additionally require careful attention to scale direction, because higher domain scores indicate more severe symptoms or impacts. The PQ-ITD endpoint requires attention to the ordered categorical structure of its five response options across 12 statements.

The safety record provides affected/at-risk counts for serious adverse events across treatment-phase and optional-extension dose groupings. These counts are useful descriptions of observed safety experience, but their interpretation is constrained by the absence of a randomized comparator and, in some groupings, very small denominators.