Esomeprazole Magnesium With or Without Aspirin in Preventing Esophageal Cancer in Patients With Barrett Esophagus: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Esomeprazole Magnesium With or Without Aspirin in Preventing Esophageal Cancer in Patients With Barrett Esophagus (NCT00474903), a randomized phase 2 study. The primary outcome is Change in Mean Tissue Prostaglandin E2 (PGE2) Concentration as Determined From Barrett's Research Mucosal Biopsy Samples.

2. Clinical Question

3. Trial Design

This was a randomized, triple-masked, parallel-group phase II trial. The registry lists enrollment of 122 and a completed status. The recorded start was 2007-04, and primary completion was 2011-03. The sponsor was the National Cancer Institute (NCI). The registry identifies Barrett esophagus and esophageal cancer as conditions of interest.

Three groups received esomeprazole magnesium. Arm I received esomeprazole magnesium (40 mg, twice daily) and two oral placebos once daily. Arm II received esomeprazole magnesium (40 mg, twice daily), an oral placebo, and acetylsalicylic acid at a dose of 81 mg once daily. Arm III received esomeprazole magnesium (40 mg, twice daily), an oral placebo, and acetylsalicylic acid at a dose of 325 mg once daily. The registry’s group labels and descriptions are reproduced in the posted results; they identify a placebo comparison and two aspirin-dose comparisons.

The primary outcome was change in mean tissue PGE2 concentration, measured in Barrett’s research mucosal biopsy samples. The timeframe was baseline to 30 days after completion of study treatment. The registry reports mean changes and standard deviations for each arm. It also posts two pairwise analyses comparing each aspirin arm with Arm I.

Toxicity is listed as a secondary outcome. The registry record does not describe the randomization procedure, how masking was maintained, the planned duration of treatment, or the protocol’s original sample-size justification. Further details of those design features are not reported in the registry record.

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.

Before assignment: Of the 120 participants in the intervention cohort, 114 were evaluable for the primary endpoint (n = 29, 42, and 43 in Arm I, Arm II, and Arm III, respectively).

Overall Study

Reasons for not completing this period:

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.

Population: Two patients from ARM II withdrew due to an inadequate number of biopsies and were not included in baseline measures.

6. Primary Endpoint

The registry lists the primary outcome as Change in Mean Tissue Prostaglandin E2 (PGE2) Concentration as Determined From Barrett's Research Mucosal Biopsy Samples, measured over Baseline to 30 days after completion of study treatment.

The primary endpoint is change in mean tissue prostaglandin E2 (PGE2) concentration in Barrett’s research mucosal biopsy samples. The registry gives the unit as pg/mL and defines the timeframe as baseline to 30 days after completion of study treatment. It reports the mean change for each arm, together with a standard deviation.

PGE2 is a tissue biomarker. In general, researchers may use a biomarker to assess a biological response to an intervention without measuring a clinical event directly. Here, the endpoint describes the amount of change in a laboratory measure in tissue samples; it does not itself record whether a participant developed esophageal cancer. The registry’s summary places the trial in a chemoprevention context, but the primary outcome posted for this analysis is PGE2 change, not cancer incidence.

The arm means are negative: -67.6 for placebo with esomeprazole, -123.9 for low-dose aspirin with esomeprazole, and -174.9 for higher-dose aspirin with esomeprazole. A negative change indicates a decrease from baseline under the endpoint’s change convention. The registry does not report a clinically important threshold for PGE2 change or explain how a particular change should be interpreted clinically. Such interpretation is therefore not reported in the registry record.

7. Results Posted in the Registry

• Arm II (Low-dose Aspirin, Placebo, Esomeprazole Magnesium): mean -123.90 versus -67.60 on control; difference -56.30 (95% CI -176.15 to 63.55), standardized effect d = -0.21; z-test p = 0.357.
• Arm III (Higher-dose Aspirin, Palcebo, Esomeprazole Magnesium): mean -174.90 versus -67.60 on control; difference -107.30 (95% CI -222.18 to 7.58), standardized effect d = -0.43; z-test p = 0.067.

8. Statistical Interpretation

9. Statistical Methods Explained

The endpoint is continuous: each participant’s tissue PGE2 measurement can take a range of values, and the registry summarizes change using a mean and standard deviation. In general, a mean describes the central value of measurements, while a standard deviation describes their spread around that mean. A mean difference compares the average changes between two arms. Its sign depends on which arm is subtracted from which; here, the posted re-calculations compare each aspirin arm with Arm I.

The re-calculated comparisons use the posted arm means and standard deviations to estimate simple, unadjusted mean differences. For low-dose aspirin versus placebo, the estimate is -56.3. For higher-dose aspirin versus placebo, it is -107.3. These estimates describe the difference in mean PGE2 change under that calculation. They do not account for other variables, and the registry does not report an adjusted model for these estimates.

In general, a confidence interval gives a range of values compatible with the data and the method used to construct the interval. It helps show the uncertainty around an estimate; it is not a guarantee that the true value lies within the particular interval. The re-calculated interval for low-dose aspirin versus placebo runs from -176.2 to 63.55. For higher-dose aspirin versus placebo, it runs from -222.2 to 7.575. Both intervals include zero, the value corresponding to no mean difference. The posted sponsor analyses use a different statistical approach, so these mean-difference intervals should not be treated as intervals for the sponsor’s rank-based results.

The sponsor reports Wilcoxon (Mann-Whitney) tests for Arm I versus Arm II and Arm I versus Arm III. In general, this rank-based test compares the distributions of observations between groups using their ordering rather than relying directly on a model of mean differences. It can be useful when observations are not well represented by assumptions underlying a standard mean-based analysis. The registry does not state why this method was selected, whether it was specified before the trial began, or what additional assumptions were checked.

The re-calculated p-values are unadjusted: 0.357 for the low-dose comparison and 0.067 for the higher-dose comparison. In general, a p-value describes how unusual data at least as inconsistent with a specified null model would be if that model were correct. It is not the probability that the null hypothesis is true, and it does not measure the size or clinical importance of an effect. These unadjusted p-values belong to the re-calculated mean comparisons, not to the sponsor’s Wilcoxon tests.

The posted means and standard deviations also show why interpreting group averages requires care. The standard deviations are large relative to the mean changes, indicating variation in participant-level changes. In general, variability can make estimates less precise, and a small study can yield wide intervals even when an estimated difference is not zero. The registry does not provide individual participant data, distribution plots beyond the calculated displays, or a detailed account of missing measurements. Those omissions limit further assessment of distributional shape, outliers, and the influence of incomplete data.

There are two sponsor-reported pairwise analyses, one for each aspirin dose against the placebo group. In general, when multiple comparisons are made, the chance of observing an apparently unusual result across the set can differ from the chance associated with a single pre-specified comparison. Methods such as adjusted thresholds or multiplicity corrections may address that issue. Whether the trial’s analysis plan specified a primary comparison, a correction, or a testing hierarchy is not reported in the registry record. The posted p-values should therefore be read as reported tests, without assuming an adjustment that is not described.

The registry also lists toxicity as a secondary outcome and posts adverse-event information. A secondary outcome can provide additional descriptive or inferential information, but its role and analysis depend on the protocol. The number of secondary outcomes in the posted results is not a substitute for a full analysis plan. Details such as outcome definitions, analysis populations, handling of missing values, covariate adjustment, and interim analyses are not reported in the registry record.

The comparison calculations and planning illustrations elsewhere on this page serve different purposes. The re-calculated mean differences summarize observed arm-level results using simple unadjusted methods. The planning figures illustrate participant counts under stated assumptions and are not observed trial results. They do not establish that the original trial was designed with those assumptions. The original sample-size justification and design assumptions are not reported in the registry record.

10. Confidence Intervals and Precision

The confidence intervals around the re-calculated mean differences are wide. For low-dose aspirin versus placebo, the mean difference is -56.3, with an interval from -176.2 to 63.55. For higher-dose aspirin versus placebo, the mean difference is -107.3, with an interval from -222.2 to 7.575. Each interval includes zero and also includes values on either side of zero. This means the posted re-calculations do not pin down the direction or size of the mean difference precisely.

In general, a confidence interval reflects both the estimated difference and the uncertainty in that estimate under the method used. A wide interval can be compatible with a meaningful difference in either direction as well as little or no difference. The interval does not by itself identify which values are more likely, and it does not describe the probability that a specific treatment effect is true.

The arm-level standard deviations are 229.7 in Arm I, 284.0 in Arm II, and 263.6 in Arm III. The comparison calculations use posted group sizes of 29.0 and 42.0 for Arm I and Arm II, and 29.0 and 43.0 for Arm I and Arm III. These relatively modest analysis-group counts and the observed spread are relevant context for the uncertainty shown. The registry does not explain why the outcome counts differ from the overall enrollment or give a participant flow for this endpoint.

Precision concerns mean PGE2 change, not cancer outcomes. The registry does not report an interval for a cancer-incidence contrast in the results. Nor does it provide a threshold for a meaningful PGE2 difference. The displayed estimates and intervals should therefore be interpreted narrowly as estimates of biomarker change under the stated calculations.

11. 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 sample-size table and curve are planning illustrations based on the observed mean differences, not a report of the trial’s original design calculation. The figures are participants per arm for 80% power and participants per arm for 90% power. They are not total event counts: the event-count convention applies to time-to-event outcomes, whereas this continuous endpoint uses participants per arm.

The planning assumptions 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 to the stated fraction of the observed effect. The table presents planning for 100%, 75%, and 50% of the observed effect. For example, the higher-dose comparison’s row for planning at 100% of the observed effect lists 86 participants per arm for 80% power and 115 participants per arm for 90% power. The low-dose comparison’s corresponding figures are 344 and 460 participants per arm.

In plain terms, the 80% power column shows the estimated number of participants per arm under which the planning method would have an 80% chance of detecting the specified effect, if its assumptions hold. The 90% column expresses the same idea using a higher target chance of detection. The entries are planning targets, not guarantees about a future study’s result.

Smaller effects require more participants under these assumptions. That pattern is visible in the rows for planning at 75% and 50% of the observed effect. The calculations are illustrations from the observed effects and do not establish the trial’s original sample-size justification, which is not reported in the registry record. This is not observed power; it is a prospective planning illustration, not a measure of whether the completed trial had enough power after its results were known.

12. 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 = -56.300; pooled standard deviation σ = 263.312.
2. n = 2 × (1.96 + 0.842)² × 263.312² / -56.300² = 343.4, rounded up to 344 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 <- -56.3                      # observed difference in means
sigma <- 263.311609832274                     # 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(474903)
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 = 344 (80% power), 460 (90% power)
 75% of observed effect: n per arm = 611 (80% power), 818 (90% power)
 50% of observed effect: n per arm = 1,374 (80% power), 1,839 (90% power)
power.t.test: 344.3 
Simulated power at n = 344 per arm: 0.796 (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.

13. Analyses Reported by the Sponsor

The sponsor’s posted analyses compare Arm I with each aspirin arm using the Wilcoxon (Mann-Whitney) method. The registry gives a p-value for each comparison but does not provide a sponsor-reported mean difference or confidence interval for those tests. The sponsor’s own analysis takes precedence over the simple re-calculated comparisons because the posted results contain a discrepancy.

The re-calculations are unadjusted comparisons of means derived from the posted arm summaries. They give mean differences of -56.3 for low-dose aspirin versus placebo and -107.3 for higher-dose aspirin versus placebo. Their p-values are 0.357 and 0.067, respectively, and their confidence intervals include zero. These statistics do not reproduce the sponsor’s rank-based tests: the estimand and method are not identical.

The sponsor’s reported p-value is lower for the higher-dose comparison than for the low-dose comparison. It should not be interpreted as a direct measure of effect size, and a difference between p-values is not itself evidence that the aspirin doses differ from each other. The registry does not report a direct comparison between the aspirin-dose arms or whether the sponsor adjusted the pairwise tests for multiplicity. Read the sponsor’s results as the trial’s reported analyses and the re-calculations as supplementary, unadjusted summaries—not as competing replacements.

14. Multiplicity, Missing Data and Analysis Populations

The registry posts two primary-outcome analyses: Arm I versus Arm II and Arm I versus Arm III. Both use the Wilcoxon (Mann-Whitney) method. These are two comparisons against the same placebo-plus-esomeprazole arm. The record does not state whether one comparison was designated as primary, whether a testing hierarchy was used, or whether the p-values were adjusted for multiple comparisons. Details of any multiplicity strategy are not reported in the registry record.

In general, multiplicity matters when several statistical tests are considered together. If each is judged using the same nominal criterion, the overall chance of at least one result appearing unusual can differ from that for a single test. Adjustment procedures can address this issue, but their choice depends on the trial question and analysis plan. Without information about the planned comparison structure, these posted p-values should not be assumed to have a particular family-wise interpretation.

The registry lists toxicity as a secondary outcome and identifies one posted secondary outcome. It does not provide a comprehensive description of all planned secondary or exploratory analyses, so the full multiplicity context is not reported in the registry record. The record also does not describe interim analyses or whether any interim findings affected conduct or interpretation.

Missing-data handling and analysis populations are also not described in sufficient detail. The primary endpoint summaries include 29.0 participants in Arm I, 42.0 in Arm II, and 43.0 in Arm III, while overall enrollment is 122. The record does not explain the difference between overall enrollment and the numbers contributing to these outcome summaries, identify an intention-to-treat or per-protocol population, or state how missing biopsies or measurements were handled. It is therefore not possible from the registry record to determine whether the analysis included all randomized participants or how missing outcome data may have affected estimates.

The adverse-event groups list 30, 45, and 45 participants at risk, which are separate posted counts for those summaries. The record does not explain the relationship between those counts and the primary-outcome analysis counts. No assumptions about missingness, exclusions, protocol deviations, or analysis-set membership should be inferred from the displayed counts alone.

15. 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 0% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: hypertension (2.2%).

The registry posts serious and other adverse-event counts by arm. For serious adverse events, the listed numbers at risk are 30 in Arm I, 45 in Arm II, and 45 in Arm III. The counts affected are 0, 0, and 1, respectively. For other adverse events, the listed numbers at risk are 30, 45, and 45, and the counts affected are 8, 9, and 15, respectively. These figures are descriptive counts from the posted safety information.

The registry identifies hypertension at 2.2% among the top serious-event terms. It does not give deaths for these groups; death counts are not reported in the registry record. The summary does not provide enough detail to infer how long participants were observed for adverse events or how each event was attributed. Counts should not be treated as comparative risks without considering the safety definitions, follow-up, and analysis procedures, which are not described in the posted information.

These safety figures are not the primary endpoint, which is tissue PGE2 change. They also do not by themselves establish whether a treatment caused or prevented an adverse event. They indicate what was recorded in the registry’s adverse-event summary and should be read alongside the limitations in event definitions and reporting.

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

This trial’s posted primary results concern changes in tissue PGE2 concentration in biopsy samples from participants with Barrett esophagus. Mean changes were negative in all arms, with the largest negative arm mean in the higher-dose aspirin group. The re-calculated mean differences were -56.3 for low-dose aspirin versus placebo and -107.3 for higher-dose aspirin versus placebo. Their confidence intervals include zero, and the arm-level standard deviations indicate considerable variation in measured changes.

The sponsor reported Wilcoxon (Mann-Whitney) comparisons and p-values for the same two pairwise contrasts. Because the registry record contains a discrepancy between those analyses and the simple re-calculations, the sponsor’s own analysis takes precedence. The re-calculations remain useful as a separate description of unadjusted mean differences, but they should not be substituted for the sponsor’s reported analysis. The registry does not explain the discrepancy.

The posted results do not directly answer whether esophageal cancer was prevented. PGE2 is a tissue measure, and the registry’s primary endpoint is not a cancer diagnosis or a cancer-incidence comparison. The record does not report whether a change in this biomarker predicts later cancer outcomes or what size of change would be clinically important. The results therefore support interpretation about the measured biomarker only, not a conclusion about cancer prevention.

Nor do the posted data establish that either aspirin dose is superior or inferior to the other. The two reported comparisons use the same control arm, and no direct comparison between aspirin doses is posted. The registry also does not describe multiplicity handling, missing-data methods, or the analysis population in enough detail to resolve those uncertainties. The appropriate reading is bounded: the record reports arm-level PGE2 changes, sponsor-reported rank-based tests, and adverse-event counts, with important limitations in precision and analysis detail.

17. Why This Trial Matters Statistically

This trial illustrates a common statistical distinction between a biological measure and a clinical outcome. The primary endpoint is mean change in tissue PGE2, while the trial’s stated context concerns chemoprevention of esophageal cancer. These are related in the trial’s rationale, but they are not interchangeable outcomes. Statistical interpretation must stay with what was measured and posted.

It also shows why the choice of summary and test matters. The arm summaries describe means and standard deviations, while the sponsor’s pairwise analyses use a rank-based method. A simple mean difference and a Wilcoxon test do not necessarily answer the same statistical question. Presenting them together can be informative, provided their different roles are made clear and a re-calculation is not mistaken for the sponsor’s primary analysis.

Precision is another central issue. The re-calculated intervals extend across zero, and the observed changes vary substantially. A point estimate alone could make the comparisons seem more definite than the data allow. The intervals make uncertainty visible, while the sponsor’s reported p-values need to be understood in light of two pairwise tests and the absence of reported multiplicity details.

Finally, the sample-size illustrations show how planning depends on the effect size one aims to detect. They are not evidence of the trial’s original design rationale or a retrospective verdict on its adequacy. Separating planning assumptions, observed data, sponsor analyses, and unadjusted re-calculations is essential to a clear statistical reading of this registry record.

18. 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 main limitation is that the posted primary outcome is a biomarker rather than an esophageal cancer outcome. The results cannot establish whether cancer incidence differed between groups. The registry does not report a clinically meaningful threshold for PGE2 change or explain how the magnitude of change relates to later clinical outcomes.

The primary-outcome analysis groups are modest in size, and the reported standard deviations are large in relation to the arm means. The re-calculated confidence intervals are broad and include zero. The registry does not provide individual-level data or enough information to assess the distribution of changes, influential observations, or the effect of missing measurements.

The sponsor’s Wilcoxon analyses and the simple mean-based re-calculations are not identical analyses, and the record contains a discrepancy. The sponsor’s own analysis takes precedence, but the registry does not explain why the two approaches yield different inferential results. The record also does not state whether the two comparisons were adjusted for multiplicity.

Other relevant design and analysis details are not reported in the registry record. These include the original sample-size justification, analysis populations, missing-data procedures, interim analyses, and a detailed protocol-based explanation of outcome definitions. The posted adverse-event counts do not supply follow-up duration or attribution detail. These limits mean that interpretation should remain descriptive and focused on the results and analyses actually posted.

19. Related Tutorials

20. Related Statistical Calculators

21. Sources

• ClinicalTrials.gov: Esomeprazole Magnesium With or Without Aspirin in Preventing Esophageal Cancer in Patients With Barrett Esophagus — NCT00474903.
• Registry record last updated 2014-07-02; status last verified 2014-01.
• Results first posted 2014-07-02.
• 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.

22. Record Summary

This completed randomized, triple-masked, parallel-group phase II trial enrolled 122 participants with Barrett esophagus. All three arms received esomeprazole magnesium; two arms also received once-daily aspirin at 81 mg or 325 mg, while the comparison arm received placebo.

The primary outcome was change in mean tissue PGE2 concentration from baseline to 30 days after completion of treatment. The registry posts arm means, standard deviations, and Wilcoxon (Mann-Whitney) comparisons. Simple unadjusted re-calculations show mean differences with confidence intervals that include zero; because the record contains a discrepancy, the sponsor’s own analysis takes precedence.

The findings describe a biomarker, not cancer incidence. Multiplicity handling, missing-data methods, analysis populations, and the original sample-size justification are not reported in the registry record.