Metformin Hydrochloride in Preventing Esophageal Cancer in Patients With Barrett Esophagus: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Metformin Hydrochloride in Preventing Esophageal Cancer in Patients With Barrett Esophagus (NCT01447927), a randomized phase 2 study. The primary outcome is Percent Change in Median pS6K1 Immunostaining Among Participants With Barrett Esophagus.

2. Clinical Question

3. Trial Design

NCT01447927 is recorded as a completed, randomized phase II trial with a double-blind, parallel-group design. It compared extended-release metformin hydrochloride with extended-release placebo in participants with Barrett esophagus. The registry lists enrollment as 93, a start date of 2012-06, and primary completion in 2013-05. The sponsor is the National Cancer Institute (NCI). These are registry entries; they do not describe how participants were recruited or how allocation concealment was implemented, which is not reported in the registry record.

The metformin group was assigned extended-release metformin by mouth once daily during week 1 and twice daily during weeks 2-12, with morning and evening dosing on week 3. The placebo group followed the corresponding schedule. Both descriptions allow treatment to stop in the absence of unacceptable toxicity or disease progression. The dose amount is not reported in the registry record. The record identifies the masking as double but does not specify which roles were masked or describe any procedures for maintaining the blind.

The primary endpoint was percentage change in median pS6K1 immunostaining from baseline to 3 months. The registry says participants were evaluable for this endpoint when pre- and post-intervention pS6K1 data were available and characterizes this as based on the intent-to-treat principle. The posted endpoint summaries include 36 participants assigned to metformin and 33 assigned to placebo. How participants without both measurements were handled, and whether the analysis included all randomized participants, are not reported in the registry record.

Overall adverse event rates are listed as a secondary outcome. The record lists 2 outcome measures and 1 analysis. Details such as the randomization procedure, allocation ratio, planned sample-size justification, interim analyses, and full statistical analysis plan are not reported in the registry record. The design therefore supports describing the randomized comparison and the outcome results that are posted, but not reconstructing unreported protocol decisions.

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: One subject withdrew post-randomization and did not receive any treatment and 18 subjects were excluded from the trial before assignment to groups: 8 out of range lab values, 4 high grade dysplasia/esophagitis/esophageal stricture, 2 intestinal metaplasia on \<25% of biopsies, and 4 other reasons.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Participants were considered evaluable for primary endpoint if pS6K1 data were available from both the pre- and post-intervention evaluations based on the intent-to-treat principle.

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 Percent Change in Median pS6K1 Immunostaining Among Participants With Barrett Esophagus, measured over Baseline to 3 months.

The primary endpoint is the percentage change in pS6K1 immunostaining between baseline and 3 months among participants with Barrett esophagus. The registry defines the calculation as the month 3 pS6K1 value minus the baseline pS6K1 value, divided by the baseline value, and multiplied by 100. It reports a median and full range for each group. The median describes the middle of the observed participant-level percentage changes, while the full range gives the smallest and largest observed values.

As general statistical context, a tissue marker can be used to assess a biological measurement over time, but a marker result is not itself a direct count of cancer cases. The endpoint is therefore distinct from an outcome such as esophageal cancer incidence. The registry record does not report cancer incidence as the primary endpoint result, nor does it explain the clinical interpretation or validation of pS6K1 immunostaining for predicting cancer outcomes.

There are practical limits to interpreting percentage change. Its calculation depends on the baseline value in the denominator, so the scale of the resulting changes can be sensitive to baseline measurements. This is a general feature of percentage-change outcomes; the registry does not provide individual baseline values or enough detail to assess how that feature affected these results. The reported endpoint is a comparison of distributions of participant-level change, summarized by group medians and ranges. It should not be read as a direct estimate of a difference in cancer risk or as proof of a clinical benefit.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The registry reports a Wilcoxon Rank-Rum Test (1-sided) for the primary two-group comparison. This is a rank-based method: rather than relying directly on the numerical distance between every pair of observations, it orders the observations and assesses how their ranks are distributed across groups. In general, rank-based methods can be useful when data are skewed or contain extreme values, because the analysis does not require the same assumptions about normally distributed measurements as a conventional two-sample t-test. The registry does not report why this test was selected or provide its full implementation details.

What does a rank-based test compare?

In general, a rank-based comparison assesses whether observations in one group tend to be higher or lower in the combined ordering than observations in the other group. It is not automatically a test of medians alone. Interpreting it strictly as a test of a median difference requires additional assumptions about the shapes and spreads of the group distributions. The posted summaries are medians and full ranges, but those summaries do not establish that the underlying distributions have the same shape. The registry record does not report a distributional analysis or a model-based effect estimate to clarify that point.

The registry labels the test one-sided. A one-sided test evaluates evidence in a prespecified direction rather than treating departures in either direction symmetrically. In general, the direction should be specified before examining the results and should align with the scientific question. The registry record does not report the direction of the one-sided alternative or the rationale for using it. The reported p-value, 0.798, should therefore be interpreted as the p-value from the test as registered, without inferring an unreported direction.

How should the medians and ranges be read?

The median is a measure of location that is less influenced by extreme observations than the arithmetic mean. The full range is a measure of the observed spread, spanning the smallest to the largest value. In this trial, the reported median percentage changes were 1.4 for metformin and -14.7 for placebo; the corresponding full ranges were -88.1 to 694.0 and -97.7 to 1646.0. The ranges are broad relative to the central summaries. In general, a range is sensitive to the most extreme observations and does not show where most values lie. It is not a confidence interval and cannot be used as one.

Percentage change is calculated relative to baseline, according to the registry definition. As a general statistical matter, such a transformation can produce highly dispersed values when baseline measurements vary or are small. The record does not give individual baseline values, so the influence of baseline values cannot be examined. It also does not report a transformed-data model or an alternative analysis that would quantify sensitivity to extreme percentage changes.

What does the p-value establish?

In general, a p-value measures how unusual data at least as incompatible with the null hypothesis would be if that hypothesis and the test assumptions were true. It is not the probability that the null hypothesis is true, the probability that results arose by chance, or a measure of clinical importance. The reported p-value of 0.798 does not state the probability that metformin and placebo have identical effects. Nor does it establish equivalence or rule out a potentially meaningful difference. Assessing the range of effects consistent with the data ordinarily requires an effect estimate and confidence interval, neither of which is reported for this comparison.

Confidence intervals, as a general concept, describe a range of estimates compatible with the data under a specified procedure and assumptions. Their width conveys precision: wider intervals represent less precise estimation. There is no primary comparison confidence interval in the registry record, so the precision of the group contrast cannot be characterized that way. The participant-level full ranges are not substitutes for confidence intervals because they describe observed variation, not uncertainty around the estimated contrast.

What can be said about the analysis population?

The registry states that participants were considered evaluable for the primary endpoint if pS6K1 data were available from both pre- and post-intervention evaluations, and it describes this as based on the intent-to-treat principle. The endpoint summaries list 36 in the metformin group and 33 in the placebo group. In general, an intent-to-treat analysis retains participants according to randomized assignment, while an evaluable or complete-measurement analysis may require observed outcome data. The wording alone does not show exactly how the analysis handled participants without both measurements. The number and characteristics of such participants, and any imputation or sensitivity analysis, are not reported in the registry record.

The registry does not report an adjusted model, covariates, test statistic, or the detailed calculation of the p-value. It also does not report an analysis plan explaining whether the test and endpoint were selected before results were examined. These omissions do not invalidate the posted result, but they limit independent assessment of the analysis and its assumptions. The most direct statistical reading is consequently the one the record supports: group medians and full ranges for the primary endpoint, together with a sponsor-reported one-sided rank-based test.

10. Analyses Reported by the Sponsor

The registry's own primary analysis is a one-sided Wilcoxon Rank-Rum Test, with p = 0.798. That is the sponsor-reported inferential result and should be distinguished from any simple recalculation based on summary statistics. A recalculated comparison cannot reproduce the underlying rank test from group medians and ranges alone, because the individual observations and ranks are not reported. The registry record does not report a separate recalculated effect estimate or comparison confidence interval.

The discrepancy flag is false, so there is no reported conflict between a sponsor analysis and a recalculated result that would require resolving in favor of the sponsor's analysis. The primary interpretation should nevertheless remain anchored to the analysis the registry reports. Its p-value does not quantify the treatment contrast or its precision, and the posted summaries cannot supply those missing quantities. The registered test result and the descriptive group summaries answer related but different questions: the summaries describe the observed endpoint values, while the test evaluates a rank-based group comparison under its assumptions.

11. Multiplicity, Missing Data and Analysis Populations

The registry lists 2 outcome measures, with Overall Adverse Event Rates as a secondary outcome, and records 1 analysis. The primary endpoint is percentage change in median pS6K1 immunostaining. The record does not report a multiplicity strategy, such as a hierarchy of outcomes, an adjustment to significance thresholds, or a plan for controlling false-positive findings across multiple tests. Without such information, the one-sided p-value should be understood as the result of the reported test, not as evidence that multiplicity was addressed.

The endpoint population is described as participants with pS6K1 measurements available before and after intervention, based on the intent-to-treat principle. The record does not report how many assigned participants lacked either measurement, why measurements were unavailable, or how missing data were handled. It does not describe imputation, sensitivity analyses, or comparison of participants with and without endpoint measurements. These details matter because an analysis restricted to participants with both measurements can be affected if the availability of measurements is related to outcome or treatment experience. That is a general concern, not a claim that such bias occurred in this trial.

The endpoint summaries list 36 participants in the metformin group and 33 in the placebo group. The record does not explain how these analysis counts relate to the enrollment figure, or give a complete account of randomized and analyzed participants. It also does not report whether the primary analysis followed all randomized participants in their assigned groups. The phrase “intent-to-treat principle” signals the stated approach, but the evaluability requirement means the exact implementation cannot be established from the registry description alone.

The registry record does not report a prespecified statistical analysis plan, sample-size justification, interim analysis plan, or the handling of protocol deviations. These omissions prevent assessment of whether the reported endpoint, test direction, missing-data approach, and multiplicity decisions followed a prespecified plan. They also limit interpretation of the single posted p-value within the full set of analyses that may have been considered.

12. 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: amnesia (2.6%).

The registry lists serious adverse events in the placebo group as 0 affected among 36 at risk, and in the metformin group as 1 affected among 38 at risk. It lists other adverse events as 23 affected among 36 at risk for placebo and 26 affected among 38 at risk for metformin. These are counts of participants affected and at risk as recorded in the safety information; they are not the primary endpoint and do not establish a causal difference between groups.

Amnesia is listed among the top serious adverse event terms at 2.6%. The record does not provide enough detail to determine from this term alone how many events occurred, whether it refers to one participant or multiple events, or how the percentage was calculated. No deaths are listed for either group in the adverse event information.

Safety summaries require attention to their definitions and observation periods. The registry does not report the detailed adverse-event coding rules, severity grading, relatedness assessment, or a full description of event timing in the registry record for this analysis. The groups at risk in the safety listing differ from the endpoint analysis counts, and the record does not explain that difference. These figures describe the posted safety counts; they should not be treated as a complete assessment of comparative safety or as evidence that an event was caused by treatment.

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

The posted primary results describe a short-term biomarker comparison: percentage change in pS6K1 immunostaining from baseline to 3 months. The median was 1.4 in the metformin group and -14.7 in the placebo group, with broad full ranges in both groups. The reported one-sided Wilcoxon Rank-Rum Test gave p = 0.798. These are the findings that can be stated directly from the registry record.

They do not establish that metformin prevents esophageal cancer, because cancer occurrence is not the primary outcome result. They also do not establish that the groups are equivalent, that there is no effect, or that either group had a clinically important change. A large p-value is not proof of no difference; it reflects the result of a particular test and must be interpreted alongside the design, endpoint, data completeness, and uncertainty. The record does not report a confidence interval or a single estimated group contrast with which to judge precision.

The endpoint is an immunostaining measure, and the record does not explain how the observed changes relate to later cancer outcomes. General statistical reasoning cautions against treating a biomarker as interchangeable with a clinical event unless that relationship has been established. No such validation information is reported in the registry record. The trial title and stated prevention aim therefore should not be mistaken for a direct demonstration of a prevention outcome.

Interpretation is further limited by incomplete reporting of missing-data handling, the exact analysis population, the one-sided test direction, and multiplicity procedures. Those gaps do not change the reported medians, ranges, or p-value, but they constrain how confidently the result can be generalized or assessed for robustness. The most defensible conclusion is descriptive: the registry reports these group summaries and this rank-based test for the primary biomarker endpoint, while the magnitude and precision of a treatment contrast and its relationship to cancer prevention remain unclear.

14. Why This Trial Matters Statistically

This trial illustrates why the distinction between a study's stated clinical aim and its measured primary endpoint matters. The registry describes a phase II trial concerned with preventing esophageal cancer in people with Barrett esophagus, while the primary result concerns change in pS6K1 immunostaining from baseline to 3 months. Statistical interpretation must follow the endpoint that was actually measured and reported. A biomarker comparison is not interchangeable with a comparison of cancer occurrence.

It also shows how a median and a p-value answer different questions. The medians describe the centers of the observed percentage-change distributions; the full ranges show that participant-level values varied widely. The one-sided rank-based test provides a test result but not a readily interpretable estimate of the size of the group contrast. A confidence interval or an effect estimate would help characterize precision, but neither is reported for the primary comparison.

The record's evaluability wording highlights another important issue. It invokes the intent-to-treat principle while defining endpoint eligibility through availability of both pre- and post-intervention measurements. Readers need the analysis population and missing-data procedures to understand how randomized comparisons were maintained. Those details are not reported in the registry record, so the result cannot be fully assessed for sensitivity to missing measurements.

Finally, this is a useful example of why statistical evidence should be read in context rather than reduced to whether a p-value crosses a threshold. The reported test, endpoint summaries, and safety counts offer different kinds of information. None alone supplies a complete account of clinical importance, precision, or longer-term outcomes. Clear reporting of estimands, analysis populations, missingness, and uncertainty would support a fuller interpretation.

15. 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 primary endpoint is a surrogate-style tissue measurement rather than a direct cancer outcome. The registry record does not report esophageal cancer incidence as a primary result or explain the relationship between pS6K1 immunostaining and later cancer outcomes. This limits conclusions about the trial's stated prevention aim.

The primary comparison is reported with group medians, full ranges, and a one-sided rank-based p-value, but no confidence interval or single between-group effect estimate. The large ranges indicate substantial observed spread, while the absence of an interval leaves the precision of the group contrast unclear. The individual data and test statistic are not reported, limiting independent reconstruction of the rank test.

The record describes an endpoint population requiring both pre- and post-intervention measurements, while also referring to the intent-to-treat principle. It does not report the number or reasons for missing measurements, the handling of missing outcomes, or sensitivity analyses. The endpoint counts and safety counts also differ, without an explanation in the registry record. These reporting gaps prevent a detailed assessment of the analysis population and possible selection effects.

The test is identified as one-sided, but the direction and rationale are not reported. The record does not report a multiplicity strategy, full analysis plan, sample-size justification, or interim analysis plan. The dose amount for metformin is not reported, and the masking procedures are not described beyond the double-blind designation. The safety information gives limited event summaries without full coding, grading, timing, or relatedness details. Together, these omissions constrain interpretation of the result and prevent a comprehensive evaluation of the planned and conducted analyses.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Metformin Hydrochloride in Preventing Esophageal Cancer in Patients With Barrett Esophagus — NCT01447927.
• Registry record last updated 2014-07-21; status last verified 2014-06.
• Results first posted 2014-07-21.
• 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.

19. Record Summary

NCT01447927 is a completed randomized, double-blind, parallel phase II trial sponsored by the National Cancer Institute (NCI). It compared extended-release metformin hydrochloride with extended-release placebo in participants with Barrett esophagus. The registry lists enrollment as 93, a start date of 2012-06, and primary completion in 2013-05.

The primary outcome was percentage change in median pS6K1 immunostaining from baseline to 3 months. The reported median percentage changes were 1.4 for metformin and -14.7 for placebo, with full ranges of -88.1 to 694.0 and -97.7 to 1646.0, respectively. The registry reports a one-sided Wilcoxon Rank-Rum Test with p = 0.798. Overall adverse event rates are listed as a secondary outcome. The record does not report a primary cancer-incidence result, a comparison confidence interval, or complete details of missing-data and multiplicity handling.