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
GOLDEN-2 was a randomized, parallel, quadruple-masked phase 2 study of EP-101 in subjects with moderate to severe COPD. The registry reports 275 enrolled subjects across five treatment groups and a primary endpoint of change from baseline in morning trough FEV1 measured at baseline and Day 29.
| Feature | GOLDEN-2 |
|---|---|
| Trial name | GOLDEN-2 |
| ClinicalTrials.gov identifier | NCT01706536 |
| Phase | Phase 2 |
| Condition | COPD |
| Design | Randomized, parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 275 |
| Arms | 5 |
| Interventions | Placebo; EP-101 12.5 mcg; EP-101 25 mcg; EP-101 50 mcg; EP-101 100 mcg |
| Status | Completed |
| Lead sponsor | Sunovion Respiratory Development Inc. |
| Start | 2012-10 |
| Primary completion | 2013-04 |
2. Clinical Question
The central statistical question was whether EP-101, evaluated at 12.5 mcg, 25 mcg, 50 mcg, and 100 mcg, produced a different change from baseline in morning trough FEV1 than placebo over the registered assessment period.
Population
Subjects with moderate to severe chronic obstructive pulmonary disease.
Intervention
EP-101 at 12.5 mcg, 25 mcg, 50 mcg, or 100 mcg.
Comparator
Placebo.
Primary question
Does treatment assignment produce a difference in change from baseline in morning trough FEV1 at Day 29?
3. Trial Design
Placebo
- Placebo drug
- 57 subjects at risk for the reported serious adverse-event count
EP-101 12.5 mcg
- EP-101 drug
- 55 subjects at risk for the reported serious adverse-event count
EP-101 25 mcg
- EP-101 drug
- 54 subjects at risk for the reported serious adverse-event count
EP-101 50 mcg
- EP-101 drug
- 57 subjects at risk for the reported serious adverse-event count
EP-101 100 mcg
- EP-101 drug
- 59 subjects at risk for the reported serious adverse-event count
4. Registered Primary Endpoint
| Endpoint | Registered definition | Time frame | Endpoint type |
|---|---|---|---|
| Change From Baseline in Morning Trough Forced Expiratory Volume in 1 Second (FEV1) | Spirometry measurements were conducted in accordance with the current ATS/ERS 2005 guidelines. Trough FEV1 was defined as the mean of the spirometry values collected at 23 hours 30 minutes and 24 hours after the in-clinic morning dose, approximately 12 hours after the previous evening dose. | Baseline and Day 29 | Other / unclear |
The endpoint is a continuous pulmonary-function measurement expressed in liters. The clinically important statistical feature is that the analysis concerns change from baseline, rather than simply comparing Day 29 FEV1 values without regard to baseline measurements.
5. Analysis Population
The reported formal analyses used an intention-to-treat population. The registry defines the ITT population as all subjects who were randomized to treatment and received at least 1 dose of double-blind study medication.
This definition anchors the reported efficacy analyses to randomized treatment assignment while requiring at least one dose of double-blind study medication for inclusion in the stated analysis population.
Using an ITT population is important because the principal comparison remains connected to the randomized design. It also prevents the analysis from being restricted only to subjects who completed every planned assessment, which can introduce selection if completion is related to treatment response or tolerability.
6. Statistical Methodology
Linear regression
The registry identifies linear regression as the statistical method for the reported primary analyses. The outcome is continuous: change from baseline in morning trough FEV1, measured in liters.
For a continuous endpoint, a linear model estimates differences in mean outcome between treatment groups while accounting for the structure specified by the analysis. The registry reports the resulting treatment effect as a least squares mean (SE), with the effect measure identified as a mean difference.
The treatment coefficient represents the estimated difference in mean change associated with the treatment comparison represented in the analysis. The exact covariate specification beyond the registry's linear-regression classification is not provided in the ClinicalTrials.gov record.
Least squares means
A least squares mean is a model-based estimate of the mean outcome associated with a treatment group. Unlike a simple arithmetic mean, it can reflect adjustment for variables included in the underlying regression model. Because the registry supplies the result as a least squares mean with a standard error and normalizes the effect to a mean difference, the reported estimates should be read as model-based treatment differences.
Confidence intervals
Each of the four reported primary analyses includes a two-sided 95% confidence interval. The interval quantifies statistical uncertainty around the estimated treatment effect under the analysis model and sampling framework.
Superiority testing
The registry classifies the hypothesis type as superiority. Conceptually, the question is whether the observed treatment difference is inconsistent with a null difference under the prespecified statistical framework. This differs from a non-inferiority analysis, where the interpretation depends on a prespecified non-inferiority margin.
Sample-size rationale
The registry analysis notes that a sample size of 45 subjects per treatment arm provides approximately 80% power to detect a 0.12 L treatment difference in mean change in trough FEV1 between an active arm and placebo at a significance level of 0.05 using a 2-group t-test, assuming a standard deviation for change in trough FEV1 of 200. Assuming a 20% discontinuation rate, approximately 55 subjects were enrolled into each treatment arm.
7. Primary Results: Morning Trough FEV1
The registry posts four statistical analyses for the registered primary endpoint. Three are explicitly labeled as comparisons of placebo with an EP-101 dose. The fourth is listed for EP-101 100 mcg without a comparator label in the ClinicalTrials.gov record. All four use least squares mean (SE), are classified as linear regression analyses, and report two-sided 95% confidence intervals.
EP-101 12.5 mcg vs Placebo
Least squares mean difference
95% CI: 0.0369–0.1966 · P = 0.0043
Endpoint: change from baseline in morning trough FEV1, measured at baseline and Day 29.
The reported estimate of 0.1168 L represents a model-based mean difference in change from baseline in morning trough FEV1 for the reported comparison of EP-101 12.5 mcg versus placebo. In the context of the registry's effect-measure classification, the positive estimate indicates a greater estimated change in the EP-101 group than the placebo group for this analysis.
The estimate does not mean that every treated subject experienced a 0.1168 L increase, nor does it describe the response of an individual patient. It is a group-level model estimate.
The 95% confidence interval, 0.0369 to 0.1966 L, describes the statistical precision of the estimated mean difference under the analysis framework. It is not a range containing 95% of individual patient responses.
The p-value of 0.0043 addresses evidence against the relevant null hypothesis under the statistical testing framework. It is not a measure of the magnitude or clinical importance of the effect. The estimate and its confidence interval are needed to understand effect size and precision.
Because the registry identifies a linear-regression analysis rather than a time-to-event model, proportional-hazards assumptions are not relevant to this particular endpoint. The multiplicity implications of testing multiple EP-101 dose comparisons should nevertheless be considered when interpreting the collection of primary analyses; the ClinicalTrials.gov record does not provide an adjustment procedure.
EP-101 25 mcg vs Placebo
Least squares mean difference
95% CI: 0.0479–0.2089 · P = 0.0019
Endpoint: change from baseline in morning trough FEV1, measured at baseline and Day 29.
The reported estimate of 0.1284 L is the model-based mean difference in change from baseline in morning trough FEV1 for EP-101 25 mcg versus placebo.
The positive estimate indicates a greater estimated change in the EP-101 25 mcg group for this comparison. It does not mean that every patient experienced that amount of change, and it should not be interpreted as an individual-level treatment effect.
The 95% confidence interval of 0.0479–0.2089 L gives the statistical uncertainty around the estimated group difference. A narrower interval would indicate greater precision; the width here reflects the amount of uncertainty remaining around the point estimate.
The p-value of 0.0019 describes the strength of evidence against the null hypothesis under the specified analysis. It does not tell us that the treatment effect is "0.0019 large," nor does a smaller p-value automatically imply a larger or more clinically meaningful effect.
As with the other dose comparisons, interpretation of this result as one member of a set of multiple treatment comparisons depends on how multiplicity was handled. No multiplicity-adjustment procedure is reported in the ClinicalTrials.gov record.
EP-101 50 mcg vs Placebo
Least squares mean difference
95% CI: 0.0667–0.2257 · P = 0.0004
Endpoint: change from baseline in morning trough FEV1, measured at baseline and Day 29.
The estimated mean difference for EP-101 50 mcg versus placebo is 0.1462 L. This is a comparison of average model-estimated change from baseline, not a guarantee that individual patients experienced the same magnitude of change.
The 95% confidence interval of 0.0667–0.2257 L describes uncertainty around the estimated treatment difference. It provides a more informative description of precision than the point estimate alone.
The p-value of 0.0004 indicates strong statistical evidence against the relevant null hypothesis under the reported testing framework. It does not quantify the clinical value of the treatment, the probability that a particular patient will respond, or the probability that the estimated effect is exactly correct.
The result comes from a linear-regression analysis of a continuous endpoint, so the interpretation does not require a proportional-hazards assumption. The ClinicalTrials.gov record also does not establish whether a formal dose-response test or multiplicity adjustment was used across the four dose-related analyses.
EP-101 100 mcg
Reported least squares mean difference
95% CI: 0.0992–0.2548 · P < 0.0001
Endpoint: change from baseline in morning trough FEV1, measured at baseline and Day 29.
The registry reports an estimate of 0.1770 L for the EP-101 100 mcg analysis, with a 95% confidence interval of 0.0992–0.2548 L and a p-value of <0.0001.
The ClinicalTrials.gov record lists the group as EP-101 100 mcg but does not state a comparator in its "Groups Compared" field. Accordingly, this page does not infer a comparator or reconstruct a comparison that is not explicitly present in the ClinicalTrials.gov record.
The estimate is a group-level least squares mean result, not an individual response. The confidence interval describes uncertainty around the reported estimate, while the p-value describes statistical evidence under the relevant hypothesis-testing framework rather than effect magnitude.
Because four primary statistical analyses are posted for the same registered endpoint, readers should distinguish the individual nominal p-values from any broader claim about familywise error control. The ClinicalTrials.gov record does not identify a multiplicity-adjustment method.
| Primary analysis | Groups compared | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| EP-101 12.5 mcg | Placebo vs EP-101 12.5 mcg | 0.1168 L | 0.0369–0.1966 | 0.0043 |
| EP-101 25 mcg | Placebo vs EP-101 25 mcg | 0.1284 L | 0.0479–0.2089 | 0.0019 |
| EP-101 50 mcg | Placebo vs EP-101 50 mcg | 0.1462 L | 0.0667–0.2257 | 0.0004 |
| EP-101 100 mcg | EP-101 100 mcg | 0.1770 L | 0.0992–0.2548 | <0.0001 |
8. Reading the Pattern Across Dose Analyses
The four reported point estimates are 0.1168, 0.1284, 0.1462, and 0.1770 L for the 12.5 mcg, 25 mcg, 50 mcg, and 100 mcg analyses, respectively. These estimates increase across the listed EP-101 doses.
The same distinction applies to the p-values. The sequence of increasingly small p-values may accompany increasingly large estimates, but p-values are influenced by both effect size and statistical precision. They are not a scale for ranking treatment effects.
9. Serious Adverse Events
The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected subjects divided by subjects at risk.
| Treatment arm | Affected | At risk | Reported format |
|---|---|---|---|
| Placebo | 1 | 57 | 1/57 |
| EP-101 12.5 mcg | 2 | 55 | 2/55 |
| EP-101 25 mcg | 2 | 54 | 2/54 |
| EP-101 50 mcg | 0 | 57 | 0/57 |
| EP-101 100 mcg | 1 | 59 | 1/59 |
These figures describe the number affected and the corresponding number at risk for the serious-adverse-event measure reported by the registry. They should not be combined with the efficacy estimates into a single statistical quantity. Efficacy and safety answer different questions and use different denominators and interpretations.
Why the denominator matters
A count such as 2 affected subjects has a different interpretation depending on whether the underlying group contains 10, 50, or 500 subjects. The registry therefore reports the affected count together with the number at risk.
Small event counts
The reported serious-adverse-event counts are small. Small counts can produce substantial statistical uncertainty, making formal between-arm safety comparisons difficult without the underlying event-level data.
10. Statistical Methods Explained
Why was linear regression used?
Morning trough FEV1 is a continuous measurement expressed in liters, and the registered endpoint is its change from baseline. Linear regression is a natural framework for estimating mean differences in a continuous outcome between treatment groups. The registry explicitly identifies linear regression as the statistical method.
What does a mean difference of 0.1462 L mean?
For the reported EP-101 50 mcg versus placebo analysis, 0.1462 L is the estimated difference in mean change from baseline in morning trough FEV1. It is a population-level estimate, not the amount by which every individual patient's FEV1 changed.
Why report a least squares mean rather than only a raw mean?
A least squares mean is generated from the fitted statistical model. It can therefore represent a model-adjusted treatment-group mean when covariates or other model terms are included. The registry reports the result as a least squares mean with standard error and identifies the effect measure as a mean difference.
What does the 95% confidence interval tell us?
The confidence interval describes the precision of the estimated treatment difference under the specified statistical framework. For the 50 mcg analysis, the interval is 0.0667–0.2257 L. It should not be interpreted as the range of individual patient responses or as a probability distribution for the true effect.
Why doesn't a smaller p-value mean a larger treatment effect?
A p-value depends on the observed effect relative to its statistical uncertainty. A large effect with substantial variability can produce a less impressive p-value than a smaller effect estimated very precisely. The treatment estimate and confidence interval should therefore be read alongside the p-value.
What does intention-to-treat mean here?
The registry definition includes all randomized subjects who received at least 1 dose of double-blind study medication. The approach keeps the efficacy analysis linked to randomized treatment assignment rather than redefining treatment groups according to later behavior or observed response.
Does the sequence of estimates prove a dose-response relationship?
No. The estimates increase from 0.1168 L to 0.1770 L across the listed doses, which is a descriptive pattern. Establishing a formal dose-response relationship requires an explicitly specified model and hypothesis test rather than inspection of the point estimates alone.
11. Multiplicity and Multiple Dose Comparisons
Four statistical analyses are posted for the same registered primary endpoint, corresponding to the four EP-101 dose levels in the ClinicalTrials.gov record. This creates an important statistical issue: multiple hypothesis tests can increase the probability of obtaining at least one statistically significant result by chance when every test is evaluated at the same nominal significance level without adjustment.
| Question | What the ClinicalTrials.gov record establishes |
|---|---|
| Number of posted primary analyses | 4 |
| Hypothesis type | Superiority |
| Confidence intervals | 95%, two-sided |
| Multiplicity adjustment reported? | Not provided in the ClinicalTrials.gov record |
| Formal dose-response test reported? | Not provided in the ClinicalTrials.gov record |
This does not invalidate the individual estimates. It means that the inferential meaning of the collection of four p-values depends on the prespecified testing strategy. A nominal p-value should not automatically be treated as if it represented familywise-error-controlled evidence across all dose comparisons.
12. Randomization and Blinding as Statistical Protection
Randomization
Random allocation is the foundation for a causal comparison because treatment assignment is not determined by investigators' or subjects' expectations of response.
Quadruple masking
The registry identifies the study as quadruple masked. Masking can reduce the influence of treatment knowledge on trial behavior, assessment, and reporting.
Parallel design
Each randomized treatment group is evaluated as a parallel group rather than having participants serve as their own crossover comparison.
Continuous endpoint
FEV1 provides a quantitative outcome, allowing treatment differences to be summarized directly on the liter scale.
These design elements matter because statistical analysis cannot fully repair systematic bias introduced during trial conduct. Randomization and masking are therefore part of the statistical architecture of the study, not merely administrative features.
13. Understanding the Primary Endpoint
Morning trough FEV1 was defined using two spirometry measurements collected at specific times after the in-clinic morning dose. The trough value was the mean of the measurements collected at 23 hours 30 minutes and 24 hours after the in-clinic morning dose, approximately 12 hours after the previous evening dose.
The primary endpoint then evaluates change from baseline in this defined trough FEV1 measure between baseline and Day 29.
This construction is statistically useful because it attempts to characterize pulmonary function at a defined point in the dosing interval rather than using an unspecified spirometry measurement. Averaging the two specified measurements also creates a prespecified endpoint rather than selecting a single favorable observation after seeing the data.
14. What the Results Do — and Do Not — Establish
The registry reports positive mean-difference estimates for all four posted primary analyses: 0.1168, 0.1284, 0.1462, and 0.1770 L. Each is accompanied by a two-sided 95% confidence interval and a p-value.
The estimates do not describe the response of every patient, do not establish an individual patient's expected change, and do not by themselves establish a clinically meaningful threshold. The statistical results also do not establish a formal dose-response relationship unless such a test was prespecified and conducted.
The p-values do not measure effect size, probability of treatment benefit for an individual patient, or clinical importance. They describe statistical evidence under the relevant null hypothesis and analysis framework.
The confidence intervals provide the uncertainty surrounding each estimate. They should be examined together with the point estimate because an estimate without a measure of precision provides an incomplete statistical description.
15. Planned and Reported Statistical Framework
The ClinicalTrials.gov record indicates that the study was completed, results were posted, and four formal statistical analyses were posted for the registered primary endpoint. The analysis framework can therefore be summarized as a randomized superiority comparison of a continuous pulmonary-function outcome using linear regression and an ITT population as defined by the registry.
| Component | Reported information |
|---|---|
| Design | Randomized, parallel |
| Masking | Quadruple |
| Primary endpoint | Change from baseline in morning trough FEV1 |
| Assessment | Baseline and Day 29 |
| Outcome unit | Liters |
| Analysis population | ITT as defined by the registry |
| Statistical method | Linear regression |
| Reported effect | Least squares mean (SE); normalized as mean difference |
| Confidence interval | 95%, two-sided |
| Hypothesis type | Superiority |
16. Trial Timeline
Study start
The registry lists October 2012 as the trial start.
Primary completion
The registry lists April 2013 as the primary completion date.
Results posted
The trial is listed as completed, with 10 outcome measures and four statistical analyses posted.
17. Limitations
- Limited statistical detail in the registry extract: the ClinicalTrials.gov record identifies linear regression and the reported least squares mean results but does not provide the complete regression specification.
- Comparator ambiguity for the 100 mcg record: the ClinicalTrials.gov record lists EP-101 100 mcg in the groups-compared field without explicitly naming a comparator. This page does not infer one.
- Multiplicity: four statistical analyses are posted for the same primary endpoint, but the ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure.
- Dose-response inference: the increasing point estimates are descriptive and should not be treated as proof of a formal dose-response relationship.
- Individual response: group-level mean differences do not describe the distribution of individual patient responses.
- Safety information: the registry-reported serious-adverse-event data contain affected/at-risk counts by arm but do not provide a broader safety dataset for this analysis.
- Generalizability: the trial population is defined in the registry as subjects with moderate to severe COPD; the ClinicalTrials.gov record does not provide additional baseline characteristics with which to assess applicability to other populations.
- Missing-data details: the ClinicalTrials.gov record does not identify the imputation or missing-data strategy used for the primary endpoint.
- Covariate specification: the ClinicalTrials.gov record classifies the analysis as linear regression but do not provide the complete list of model covariates.
18. Why This Trial Matters Statistically
GOLDEN-2 is a useful teaching example because its primary endpoint is a continuous physiologic measure rather than a survival endpoint. The statistical story therefore shifts from hazard ratios and Kaplan-Meier curves to treatment-group mean differences, regression modeling, confidence intervals, and the interpretation of repeated dose comparisons.
| Concept | How it appears in GOLDEN-2 |
|---|---|
| Randomization | Subjects were randomized in a parallel design. |
| Blinding | The registry identifies the study as quadruple masked. |
| Intention-to-treat analysis | The formal analyses used the registry-defined ITT population. |
| Continuous outcome | Change from baseline in morning trough FEV1 was measured in liters. |
| Linear regression | The statistical method for the posted analyses. |
| Least squares mean | The reported effect format for each primary analysis. |
| Mean difference | The effect measure. |
| Confidence interval | Each posted estimate has a two-sided 95% CI. |
| P-value | Each posted analysis includes a p-value. |
| Superiority testing | The registry classifies the hypothesis type as superiority. |
| Multiplicity | Four formal analyses are posted for the same primary endpoint. |
| Sample-size planning | The registry supplies assumptions for power, treatment difference, variability, significance level, and discontinuation. |
19. A Deeper Statistical Reading of the Four Estimates
The reported estimates can be read on three separate dimensions: magnitude, precision, and statistical evidence.
Magnitude
The point estimate describes the estimated mean difference in change from baseline. For example, 0.1462 L is larger than 0.1284 L as a point estimate, but that alone does not establish that the corresponding underlying treatment effects differ statistically.
Precision
The confidence interval describes uncertainty around each point estimate. The interval should be considered when deciding how precisely the treatment difference has been estimated.
Statistical evidence
The p-value quantifies evidence against the null hypothesis under the statistical testing framework. It is not a substitute for the effect estimate or confidence interval.
Multiplicity
Because several dose analyses are posted, interpretation of the collection of p-values depends on the prespecified multiplicity strategy, which is not reported in the ClinicalTrials.gov record.
This separation is especially important in dose-ranging trials. A dose can have a numerically larger estimate without the difference between two doses being statistically established. Conversely, a statistically detectable difference does not automatically establish clinical importance.
20. Sample Size and Power
The registry's analysis notes provide a specific design-stage sample-size rationale. A sample size of 45 subjects per treatment arm was stated to provide approximately 80% power to detect a 0.12 L treatment difference in mean change in trough FEV1 between an active arm and placebo at a significance level of 0.05, using a 2-group t-test and assuming a standard deviation for change in trough FEV1 of 200.
| Design assumption | Registry value |
|---|---|
| Subjects per treatment arm for power calculation | 45 |
| Approximate power | 80% |
| Treatment difference to detect | 0.12 L |
| Significance level | 0.05 |
| Assumed standard deviation | 200 |
| Power calculation test | 2-group t-test |
| Assumed discontinuation rate | 20% |
| Approximate enrollment per arm after discontinuation allowance | 55 subjects |
The distinction between the planning calculation and the final regression analysis is worth emphasizing. The registry describes the sample-size calculation using a 2-group t-test, while the posted analyses are identified as linear regression. These are not contradictory: a two-group comparison of means can be represented within the broader linear-model framework. The design calculation and final analysis, however, answer different statistical questions.
21. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The registry reports four linear-regression analyses of change from baseline in morning trough FEV1, each with a least squares mean estimate, a two-sided 95% confidence interval, and a p-value. The reported estimates are positive for the four listed analyses.
Clinical interpretation
The ClinicalTrials.gov record establishes the direction and magnitude of the reported statistical estimates but do not provide a clinical-response threshold with which to judge whether a particular change is clinically meaningful for an individual subject.
This distinction is central to responsible interpretation. Statistical significance addresses compatibility with a null hypothesis; clinical importance asks whether the magnitude and consequences of the observed difference matter in practice. The two questions should not be treated as interchangeable.
22. Important Questions for Interpreting GOLDEN-2
As point estimates, 0.1770 L is numerically larger than 0.1168 L. The ClinicalTrials.gov record does not provide a formal statistical test comparing those two dose effects directly, so the numerical difference should not be converted into a claim of statistically established superiority of one dose over another.
Each reported interval lies above zero: 0.0369–0.1966, 0.0479–0.2089, 0.0667–0.2257, and 0.0992–0.2548. That pattern is consistent with the corresponding positive point estimates and reported p-values under the posted analyses. It still does not answer whether the magnitude is clinically meaningful.
No. A p-value is not the probability that the treatment works and is not the probability that the null hypothesis is true. It is a measure of statistical evidence against the null hypothesis under the specified testing framework.
No. Proportional hazards is an assumption associated with Cox-type time-to-event models. The GOLDEN-2 primary endpoint is continuous and the registry reports linear regression, so proportional-hazards assumptions are not part of this primary analysis.
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Statistical Calculators
25. Sources
- ClinicalTrials.gov: NCT01706536 — GOLDEN-2.
- Linked publication: PubMed record for PMID 29202767.
Continue through Clinical Biostats
Use the related tutorials and statistical calculators to explore the concepts underlying randomized trials, continuous endpoints, regression, confidence intervals, and hypothesis testing.
26. Record Summary
GOLDEN-2 provides a compact example of how statistical reasoning changes when the primary outcome is a continuous physiologic measure rather than a time-to-event endpoint. The study used randomization, quadruple masking, a parallel design, an ITT analysis population, and linear regression for change from baseline in morning trough FEV1 at baseline and Day 29. Four formal primary analyses were posted, with least squares mean estimates of 0.1168, 0.1284, 0.1462, and 0.1770 L and corresponding two-sided 95% confidence intervals and p-values.
The most important statistical lesson is to keep several concepts separate: the point estimate describes the magnitude of the estimated group difference; the confidence interval describes its precision; the p-value describes statistical evidence against a null hypothesis; and the trial design determines how those quantities should be interpreted. The increasing estimates across the four listed doses are descriptive evidence of a pattern, but the ClinicalTrials.gov record does not establish a formal dose-response test or a multiplicity-adjusted inference across the four analyses.