Effect of Roflumilast on Lung Function in Chronic Obstructive Pulmonary Disease (COPD) Patients Treated With Salmeterol: The EOS Study (BY217/M2-127): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Effect of Roflumilast on Lung Function in Chronic Obstructive Pulmonary Disease (COPD) Patients Treated With Salmeterol: The EOS Study (BY217/M2-127) (NCT00313209), a randomized phase 3 study. The primary outcome is Pre-bronchodilator Forced Expiratory Volume in First Second (FEV1).

2. Clinical Question

The study’s stated aim was to compare roflumilast’s effects on pulmonary function and symptoms when administered with salmeterol, and to collect further safety and tolerability data. The primary endpoint focuses on lung function rather than symptoms or exacerbations. Secondary outcomes include post-bronchodilator FEV1, COPD exacerbation rate, the Transition Dyspnea Index (TDI) Focal Score, and the Shortness of Breath Questionnaire (SOBQ) Total Score. Results for the SOBQ are not reported in the registry record.

3. Trial Design

The EOS Study (BY217/M2-127; NCT00313209) is recorded as a phase 3, randomized, double-blind, parallel-group study. Its stated enrollment is 933. Participants were assigned to roflumilast or placebo, with salmeterol used as underlying medication in both groups. The study title specifies roflumilast 500 μg once daily versus placebo over 24 weeks; the arm descriptions specify salmeterol 50 μg twice daily by inhalation. The brief summary says the study could last up to 28 weeks.

The registry lists the condition as COPD and the sponsor as AstraZeneca. The recorded start is 2006-04, primary completion is 2007-07, and status is completed. The study is described as double-blind, but details about who was masked, how masking was maintained, or whether it was assessed are not reported in the registry record. Specific eligibility criteria, recruitment locations, participant demographics, and the allocation procedure are also not reported in the registry record.

For the primary endpoint, the registry identifies the analysis population as intention to treat (ITT), and clarifies that the number analyzed is the number of participants with available data. The posted outcome counts are 456 for roflumilast and 463 for placebo. The registry does not explain the relationship between these endpoint-specific counts and the stated enrollment, nor does it report a detailed participant-flow account. Endpoint data availability varies: for example, the primary outcome and the secondary outcomes have different analyzed counts. Those differences do not, by themselves, show why data were unavailable.

The registry lists five outcome measures and five analyses, while four secondary outcomes are posted in the record. Details of the trial’s original sample-size justification and its prespecified statistical analysis plan are not reported in the registry record. These omissions limit assessment of whether the posted analyses fully match the planned analyses.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Overall Study

Analysis population for the primary outcome, as defined in the registry: ITT (Intention to Treat) analysis. Number of participants analyzed = number of participants with data available.

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

6. Primary Endpoint

The registry lists the primary outcome as Pre-bronchodilator Forced Expiratory Volume in First Second (FEV1), measured over Change from baseline over 24 weeks of treatment.

The primary endpoint is change from baseline in pre-bronchodilator forced expiratory volume in the first second (FEV1), measured in milliliters, over 24 weeks of treatment. FEV1 is the volume of air a person can forcefully exhale during the first second of an expiratory maneuver. Measuring it before bronchodilator administration gives an assessment under the specified pre-bronchodilator condition. The endpoint is continuous: each participant contributes a measured value, and the group results are expressed as least-squares means.

Change from baseline focuses the comparison on movement in the measure during treatment, rather than only on the value at the end of treatment. In this study, the posted mean change was 39.0 mL in the roflumilast group and -10.0 mL in the placebo group. The registry’s adjusted analysis expresses the between-group result as a mean difference. Such a difference describes the estimated contrast in average change between assigned groups; it does not show the variation in every individual participant’s response.

FEV1 is used as a lung-function measure, which makes it relevant to a study whose stated aim includes pulmonary function. The registry does not report a threshold for a clinically important change, so the numerical estimate alone cannot determine the importance of the difference to patients. Nor does a lung-function endpoint capture all dimensions of COPD, such as symptoms, exacerbations, or safety. Those are addressed separately among the listed secondary outcomes and adverse-event information.

7. Results Posted in the Registry

• Roflumilast: mean 39.00 versus -10.00 on control; difference 49.00 (95% CI 24.05 to 73.95), standardized effect d = 0.25; z-test p = <0.001.

8. Secondary Efficacy Endpoints

The registry also posts results for secondary outcomes. For each outcome below, the endpoint type, the test applied to the posted summary statistics, the result and its interpretation are given in turn. Secondary endpoints are usually not adjusted for multiple testing, so they support rather than confirm the primary finding.

Post-bronchodilator FEV1

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard errors.
• Result:
  • Roflumilast: mean 68.00 versus 8.00 on control; difference 60.00 (95% CI 35.05 to 84.95), standardized effect d = 0.31; z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the mean is statistically higher for Roflumilast than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

COPD Exacerbation Rate (Mild, Moderate or Severe)

• Endpoint type: Event rate: events per unit of follow-up, compared as a rate ratio.
• Test: Poisson rate ratio, with the standard error taken from the posted confidence limits.
• Result:
  • Roflumilast: 1.90 versus 2.40 (exacerbations per patient per year) on control; rate ratio 0.79 (95% CI 0.56–1.13, derived from the posted confidence limits); p ≈ 0.196.
• Interpretation:
  • The 95% confidence interval includes no difference (a ratio of 1), so these data do not show a statistically significant difference between Roflumilast and the control. The interval gives the range of effects that remain compatible with the data.

Transition Dyspnea Index (TDI) Focal Score

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard errors.
• Result:
  • Roflumilast: mean 1.20 versus 1.10 on control; difference 0.10 (95% CI -0.18 to 0.38), standardized effect d = 0.05; z-test p = 0.480.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Roflumilast and the control. The interval gives the range of effects that remain compatible with the data.

The secondary outcomes broaden the assessment beyond pre-bronchodilator lung function, including post-bronchodilator lung function, exacerbations, and dyspnea. Taken together, their pattern is not uniform: some point estimates favor roflumilast, while the dyspnea comparison is close to no difference. The registry’s own analyses provide stronger statistical evidence for the post-bronchodilator measure than for the exacerbation-rate or dyspnea comparisons. This general pattern is compatible with the primary result but does not establish a consistent effect across all aspects of COPD.

Secondary outcomes have a supporting role in interpreting the primary endpoint. They may indicate whether findings extend to other measures, but each outcome has its own measurement, analysis, and uncertainty. An outcome with an estimate in a favorable direction is not automatically evidence of a reliable difference, especially when its confidence interval includes values compatible with little or no difference. Conversely, a non-significant result does not prove that there is no effect.

Multiple outcomes create a multiplicity concern: the more statistical comparisons considered, the greater the opportunity for an apparently notable result to arise by chance. The registry does not report a multiplicity adjustment or a hierarchy governing secondary testing. That limits how confidently the individual secondary p-values can be interpreted as confirmatory evidence. The SOBQ Total Score appears among the listed secondary outcomes, but its results are not reported in the registry record. The absent result cannot be treated as evidence in either direction.

Accordingly, the secondary findings are most useful as context for the primary FEV1 result, not as a set of independent confirmations. Their differing pattern and the missing posted SOBQ result underscore the need to distinguish the prespecified primary question from additional outcome analyses. Without the full protocol and analysis plan, the intended role of each secondary outcome is not reported in the registry record.

Secondary outcomes registered for this trial, for reference (only those above were recalculated):

• Post-bronchodilator FEV1
• COPD Exacerbation Rate (Mild, Moderate or Severe)
• Transition Dyspnea Index (TDI) Focal Score
• Shortness of Breath Questionnaire (SOBQ) Total Score

9. Statistical Interpretation

10. Statistical Methods Explained

The primary outcome is continuous, and the arm-level results are reported as least-squares mean changes with standard errors. A least-squares mean is a model-based estimate of a group mean, often adjusted for other variables in the analysis. Here, the registry identifies ANCOVA as the method for the primary comparison, but does not list the covariates or the full model specification. The reported 49 mL mean difference is therefore the adjusted between-group estimate as recorded, not merely a subtraction of the two rounded arm-level means.

ANCOVA, or analysis of covariance, compares outcome values while accounting for specified covariates. It can improve precision when the covariates are relevant and the model is appropriate. The registry does not report which covariates were included, how baseline measurements were used, or how assumptions were assessed. As a result, readers can describe the reported adjusted estimate and its uncertainty, but cannot independently reconstruct the model from the registry record.

A confidence interval expresses uncertainty around an estimated effect under the assumptions of the analysis. It gives a range of effect values compatible with the observed data and model at the stated confidence level. It is not a range containing a known proportion of individual participants’ responses, and it is not a guarantee that the true effect lies within the particular interval. The primary analysis reports a 95% confidence interval of 27 to 71 mL. Since this interval lies above zero, it is consistent with a positive between-group difference under that analysis. The interval’s width also matters: estimates within the interval differ in magnitude, and the registry does not define which magnitudes would be clinically important.

The p-value describes how unusual data at least as incompatible with a null comparison would be under a specified statistical model, if the null hypothesis and assumptions held. It is not the probability that the null hypothesis is true, nor a measure of the probability that the result is due to chance in a general sense. The primary analysis reports p < 0.001. That small p-value and the interval excluding zero describe statistical evidence for a difference; they do not establish clinical importance or remove uncertainty about model choices and missing data.

The posted recalculation for the primary comparison uses the group summary statistics and is an unadjusted approximation. The arm values are least-squares means, and the reported standard errors are used in the recalculation; the pooled standard deviation is described as approximate. This calculation gives an independent summary-based check, not a reanalysis of participant-level records. It cannot reproduce the ANCOVA without the model details and individual-level data. Because the registry’s own analysis is adjusted and is the sponsor-reported analysis, it remains the principal result for interpretation.

For the exacerbation outcome, the registry reports a Poisson regression analysis and a rate ratio. A rate ratio compares event rates between groups; a value below one indicates a lower estimated rate in the roflumilast group relative to placebo, while the confidence interval communicates uncertainty about that comparison. Event-rate data can be more variable than a simple model assumes, particularly if participants have differing underlying propensities for events. Over-dispersion means that observed variability exceeds the model’s assumed variability; if present and unaddressed, uncertainty estimates may be too small. Whether over-dispersion was assessed or how the Poisson model handled it is not reported in the registry record. The registry’s own rate-ratio analysis, rather than the simple unadjusted recalculation, is the main reported analysis for that outcome.

Standard errors and confidence intervals describe different aspects of uncertainty. The standard error summarizes the estimated sampling variability of a statistic, while a confidence interval places that uncertainty around an estimate on its measurement scale. They are not measures of the spread of individual outcomes. The registry reports standard errors for the continuous primary and secondary outcomes, and confidence intervals for the exacerbation rates. Details such as distributional checks, transformations, and model diagnostics are not reported in the registry record.

The ITT designation indicates the stated analysis population, but the record’s definition adds that the analyzed number is the number with available data. The registry does not describe a missing-data method, imputation, sensitivity analysis, or whether participants were analyzed according to original assignment when outcome data were available. ITT terminology alone therefore does not resolve how missing measurements were handled. The original protocol and full statistical analysis plan are not reported in the registry record, so it is not possible to verify from this record whether all analyses and outcome time points were prespecified.

Finally, the record includes multiple primary and secondary outcomes, but does not report whether statistical testing was adjusted to account for multiplicity. When many outcomes are tested, the chance of at least one apparently notable result can increase. A multiplicity procedure may control this issue, but no such procedure is described in the registry record. This is why secondary findings should be read as supporting or exploratory evidence unless the prespecified testing strategy is known.

11. Confidence Intervals and Precision

The primary adjusted estimate is a 49 mL mean difference, with a 95% confidence interval from 27 to 71 mL. The interval stays on the positive side of zero, so it supports a positive between-group difference in the modeled primary outcome. Its lower and upper limits also show that the estimate is not pinned down to a single value. The registry does not state a threshold for clinical importance, so the interval cannot by itself distinguish clinically important from less important values.

The recalculated unadjusted interval is 24.05 to 73.95 mL around an estimate of 49.0 mL. It is somewhat wider than the interval from the registry’s ANCOVA analysis. The recalculation is based on posted summary statistics and approximate variability, whereas the sponsor-reported result comes from ANCOVA. These intervals should not be treated as competing estimates from identical methods. The registry’s own adjusted analysis is the principal analysis to interpret.

For the secondary outcomes, the registry reports confidence intervals for the sponsor analyses as well as interval estimates for some recalculations. The exacerbation-rate interval includes values on both sides of a rate ratio of one, so the posted adjusted analysis does not rule out either a lower rate or no difference under that model. The dyspnea interval likewise spans zero for the mean difference. These intervals convey uncertainty that a point estimate alone would hide.

Confidence intervals reflect sampling uncertainty under the method used. They do not account automatically for bias from missing data, outcome selection, model misspecification, or multiple testing. The registry does not report enough detail to evaluate all of those sources. Precision should therefore be considered alongside the analysis method, endpoint availability, and the limits of the registry record.

12. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The planning figures refer to participants per arm for the continuous outcomes, not to total events. The assumptions listed for these planning illustrations are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. The effect to detect is set in turn to the stated fraction of the observed effect. For a rate outcome, the record’s general note distinguishes time-to-event calculations, which use total events, from other calculations, which use participants per arm; the posted exacerbation planning figures are participants per arm.

The 80% power column gives the estimated participants per arm under the stated assumptions for a design with an 80% chance of detecting the specified effect if that effect is the true one and the model assumptions hold. The 90% power column expresses the same planning idea with a higher target chance of detection. These are design illustrations, not guarantees about any future study’s results. The trial’s own original sample-size assumptions and justification are not reported in the registry record.

Rows planning for 100%, 75%, and 50% of the observed effect illustrate how the required sample size changes when the effect to detect is made smaller. Smaller effects need more participants for the same target power, because they are harder to distinguish from sampling variation. The listed sample sizes are planning illustrations based on the observed effect, not observed power and not evidence that the completed trial had a particular probability of success. They should not be read as recommendations without considering the intended effect, outcome variability, design, missingness, and analysis plan.

13. How the Sample Size Is Derived

Power is the probability that a study of a given size rejects the null hypothesis when a specific true effect exists. It depends on four things: the effect to detect, the variability of the outcome, the significance level (here a two-sided α of 0.05, so z_1−α/2 = 1.960) and the desired power (z_1−β = 0.842 for 80% and 1.282 for 90%).

Formula

n is the number of participants in each arm, Δ is the difference in means to detect and σ is the common standard deviation.

Worked calculation for this trial

1. Observed difference Δ = 49.000; pooled standard deviation σ = 192.929.
2. n = 2 × (1.96 + 0.842)² × 192.929² / 49.000² = 243.4, rounded up to 244 participants per arm for 80% power.

The smaller-effect rows in the table above repeat the same calculation with the effect multiplied by 0.75 and 0.5 (for ratios, applied to the logarithm of the ratio).

R code

alpha <- 0.05
z_a   <- qnorm(1 - alpha / 2)          # two-sided
delta <- 49                      # observed difference in means
sigma <- 192.929168665957                     # pooled standard deviation
n_per_arm <- function(power, frac = 1) {
  ceiling(round(2 * (z_a + qnorm(power))^2 * sigma^2 / (delta * frac)^2, 6))
}
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: n per arm = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(n_per_arm(0.80, f), big.mark = ","), format(n_per_arm(0.90, f), big.mark = ",")))

# Check against R's built-in function (t-based, so very slightly larger)
cat("power.t.test:", round(power.t.test(delta = delta, sd = sigma, power = 0.80)$n, 1), "\n")

# Simulation: exact two-sample t-test using simulated means and variances
set.seed(313209)
n    <- n_per_arm(0.80)
reps <- 20000
m_diff <- rnorm(reps, delta, sigma * sqrt(2 / n))             # difference in sample means
s2     <- sigma^2 * rchisq(reps, 2 * (n - 1)) / (2 * (n - 1))   # pooled sample variance
tstat  <- m_diff / sqrt(s2 * 2 / n)
cat(sprintf("Simulated power at n = %s per arm: %.3f (%d replicates)\n", format(n, big.mark = ","), mean(abs(tstat) > qt(0.975, 2 * (n - 1))), reps))

Output from running the code

100% of observed effect: n per arm = 244 (80% power), 326 (90% power)
 75% of observed effect: n per arm = 433 (80% power), 580 (90% power)
 50% of observed effect: n per arm = 974 (80% power), 1,304 (90% power)
power.t.test: 244.3 
Simulated power at n = 244 per arm: 0.802 (20000 replicates)

The simulation draws many sets of group means and variances at the planned sample size and counts how often the two-sample t-test rejects the null. It should land close to the target power; at very small required sample sizes it can fall below it, because the normal approximation is rough there.

14. Analyses Reported by the Sponsor

The registry’s own primary analysis is an ANCOVA estimate of a 49 mL mean difference, with a 95% confidence interval from 27 to 71 mL and p < 0.001. This is the result to use as the trial’s reported primary analysis. ANCOVA is an adjusted method, but the registry does not provide its covariates or full model specification.

The simple recalculation from posted summaries gives a mean difference of 49.0 mL, a 95% confidence interval from 24.05 to 73.95 mL, and p < 0.001. It is an unadjusted summary-statistic calculation, not a reproduction of the sponsor’s ANCOVA. The estimated difference is aligned, while the confidence intervals and p-values are not identical. That is unsurprising when methods differ and the recalculation depends on rounded arm summaries and approximate variability.

The record does not flag a discrepancy between the analyses. This does not make the recalculation interchangeable with the sponsor analysis: the sponsor’s model remains the analysis reported by the trial, and the recalculation is best viewed as a transparent check using summary data. For secondary outcomes, the same distinction applies. The registry’s posted regression or ANCOVA result is the trial’s own analysis; simple unadjusted recalculations are supplementary and may differ because of adjustment, model choice, and available information.

15. Multiplicity, Missing Data and Analysis Populations

The registry identifies a primary endpoint and lists secondary outcomes, but it does not report whether the statistical testing plan adjusted for multiple comparisons. Multiplicity matters when several outcomes are tested because each additional test creates another opportunity for a result to appear statistically notable even when there is no underlying effect. Common approaches include controlling a family-wise error rate or applying a prespecified sequence of tests, but the registry record does not identify such an approach. The secondary p-values should therefore be interpreted cautiously and should not be treated as equivalent to a single prespecified primary test.

Missing-data handling is also not reported in enough detail to assess its effect on the findings. The primary outcome population is labeled ITT, with the stated number analyzed defined as the number with data available. The registry provides outcome-specific analyzed counts, but does not explain why data were unavailable, whether missingness differed by group, or whether any imputation or sensitivity analyses were used. The label ITT does not, on its own, establish how missing outcomes were addressed.

The record does not provide the original protocol or full statistical analysis plan, and does not report whether interim analyses were conducted. The sample-size justification for the original trial is not reported in the registry record. Nor are the covariates for the ANCOVA, model diagnostics, or detailed criteria for analysis populations reported. These gaps make it difficult to determine whether the posted outcomes, time points, models, and testing strategy match the prospectively planned approach.

These omissions do not invalidate the reported estimates, but they constrain the strength of conclusions that can be drawn from a registry entry alone. The primary result can be described using the sponsor’s reported analysis, while uncertainty remains about missing-data assumptions, prespecification, and multiplicity. Independent access to the protocol, analysis plan, or participant-level data would be needed to assess those matters more fully.

16. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 5% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: chronic obstructive pulmonary disease (2.4%), myocardial infarction (0.4%), atrial fibrillation (0.4%), aortic aneurysm (0.4%), prostate cancer (0.4%).

The registry reports serious adverse events affecting 36 participants in the roflumilast group, with 466 at risk, and 42 participants in the placebo group, with 467 at risk. It also reports other adverse events affecting 165 participants in the roflumilast group, with 466 at risk, and 143 participants in the placebo group, with 467 at risk. These are group-level counts and at-risk denominators as posted; they should not be confused with the lung-function outcome counts or interpreted as adjusted comparisons.

The registry lists chronic obstructive pulmonary disease at 2.4% among the top serious-event terms, and myocardial infarction, atrial fibrillation, aortic aneurysm, prostate cancer, and cardiac failure at 0.4% each. The record does not provide enough context alongside these terms to infer comparative risks from those percentages. Death counts are not reported in the registry record.

Adverse-event counts are descriptive and do not by themselves establish whether an event was caused by treatment. Interpretation would require details such as event definitions, duration and completeness of follow-up, severity, and exposure time. These details are not reported in the registry record, so the counts should be read as posted safety information rather than a complete assessment of comparative safety or tolerability.

17. What the Result Does, and Does Not, Mean

The registry’s primary result shows a statistically significant adjusted difference in change from baseline in pre-bronchodilator FEV1 between the roflumilast and placebo groups, with salmeterol used in both. The reported estimate is 49 mL, and its 95% confidence interval is 27 to 71 mL. This describes the estimated average contrast for the study’s primary lung-function outcome over 24 weeks of treatment. It does not describe the response of every participant or establish what size of change matters clinically.

The result applies to the comparison studied: adding roflumilast or placebo alongside salmeterol in the enrolled COPD population. It does not directly answer comparisons involving other background treatments, different populations, or different treatment durations. Participant eligibility and detailed characteristics are not reported in the registry record, which further limits assessment of how broadly the result may apply.

The primary endpoint is one dimension of COPD and should not be treated as a complete measure of symptoms, exacerbations, quality of life, or safety. Secondary outcomes provide additional information, but their results are mixed in strength and are subject to uncertainty and unreported multiplicity procedures. The registry does not report SOBQ results. These limitations mean the FEV1 result cannot stand in for every outcome that matters to patients or clinicians.

Statistical significance indicates evidence against a null comparison under the specified analysis; it does not show that a treatment is beneficial overall, nor does it establish causation for adverse events. A confidence interval describes uncertainty under the model, not all uncertainties arising from missing data, analysis choices, or generalizability. The registry does not report a clinical-importance threshold or enough information to assess all such limitations. The appropriate interpretation is therefore focused: the posted primary analysis estimates a positive between-group difference in FEV1 change, while the clinical meaning and broader implications require evidence beyond that estimate alone.

18. Why This Trial Matters Statistically

This trial is statistically informative because it separates a model-based primary comparison from a simple calculation using posted summary statistics. The sponsor-reported ANCOVA result is the trial’s own adjusted estimate; the unadjusted recalculation offers a transparent check but cannot recreate the model. The close central estimates alongside differing interval limits illustrate why analysis method and available detail matter when interpreting summaries.

The outcome is continuous, making the estimate’s scale and uncertainty directly interpretable in milliliters. The primary confidence interval remains above zero, while secondary outcomes show a less uniform pattern. This contrast is useful: statistical conclusions should be tied to the specified endpoint rather than generalized automatically to all outcomes. The record also makes clear that endpoint-specific data availability varies, an important consideration when evaluating an ITT label and the completeness of posted results.

The planning illustrations show how sample-size requirements depend on the target effect and power assumptions. They are not evidence of the completed trial’s observed power, and the trial’s own design assumptions are not reported in the registry record. That distinction helps prevent retrospective planning calculations from being mistaken for the original design rationale.

Finally, the record illustrates the limits of registry-level interpretation. It reports the main estimate, uncertainty, analysis method, and adverse-event counts, yet does not report key details about covariate adjustment, missing-data handling, multiplicity, or the full analysis plan. A sound statistical reading uses the posted result while keeping those limitations visible, rather than treating a small p-value as a complete account of evidence.

19. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The registry record does not include detailed eligibility criteria, demographic characteristics, disease severity, recruitment setting, or participant-flow information. Those omissions limit assessment of which people with COPD the results represent and why outcome data were unavailable for some participants. Although the primary analysis is labeled ITT, the number analyzed is defined by data availability; missing-data methods and sensitivity analyses are not reported.

The sponsor’s ANCOVA is identified, but its covariates, model specification, and diagnostic checks are not reported. The primary recalculation is based on summary statistics and approximate variability, so it cannot verify the adjusted result at participant level. The registry does not report the original sample-size justification, protocol, full statistical analysis plan, multiplicity strategy, or whether interim analyses occurred.

Secondary results are incomplete in the registry record: the SOBQ Total Score is listed, but its results are not reported. The posted secondary outcomes also differ in measurement and statistical model, and the record does not establish how those endpoints were prioritized. This makes broad conclusions across pulmonary function, exacerbations, and symptoms less secure than a focused description of the primary outcome.

For safety, the record supplies adverse-event counts and selected serious-event terms but does not report enough detail about exposure time, event ascertainment, follow-up completeness, or causal assessment to support a comprehensive comparative safety interpretation. More generally, a registry record cannot resolve whether observed results extend to other populations, treatment settings, or durations. These limitations call for a cautious reading of the estimates rather than a claim that the record answers every clinical question.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: Effect of Roflumilast on Lung Function in Chronic Obstructive Pulmonary Disease (COPD) Patients Treated With Salmeterol: The EOS Study (BY217/M2-127) — NCT00313209.
• Registry record last updated 2016-12-14; status last verified 2016-11.
• Results first posted 2011-05-19.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

23. Record Summary

The EOS Study is recorded as a completed phase 3, randomized, double-blind, parallel-group trial of roflumilast versus placebo in people with COPD receiving salmeterol. The primary outcome was change from baseline in pre-bronchodilator FEV1 over 24 weeks. The registry’s ANCOVA analysis reports a mean difference of 49 mL, with a 95% confidence interval of 27 to 71 mL and p < 0.001.

The primary finding is statistically clear under the reported analysis, but its clinical importance is not established by the registry record. Secondary findings are mixed, and the record does not report results for the listed SOBQ outcome or detail key issues such as multiplicity and missing-data handling. The posted safety counts are descriptive and do not establish causation or comparative risk.