A Phase III Study to Test the Benefit of a New Kind of Anti-cancer Treatment in Patients With Melanoma, After Surgical Removal of Their Tumor: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Phase III Study to Test the Benefit of a New Kind of Anti-cancer Treatment in Patients With Melanoma, After Surgical Removal of Their Tumor (NCT00796445), a randomized phase 3 study. The primary outcome is Disease Free Survival (DFS).

2. Clinical Question

3. Trial Design

This was a Phase III, randomized, triple-masked, parallel-group study of an antigen-specific cancer immunotherapeutic as adjuvant therapy in patients with resected melanoma. The registry lists GlaxoSmithKline as sponsor, melanoma as the condition, and GSK 2132231A and placebo as the interventions. Total enrollment is reported as 1351. The status is terminated; the reason for termination is not reported in the registry record.

The two principal groups were the MAGE-A3 group and the placebo group. Their posted primary-outcome analysis populations are 893 and 452, respectively. The analysis description says that the Total Treated population was analyzed as randomized, meaning participants were assigned to the groups allocated by the randomization system at the start of the study. The registry does not report further details about adherence to assigned treatment or the handling of participants who did not complete the study.

The MAGE-A3 group received up to 13 doses of recMAGE-A3 + AS15 ASCI, and placebo participants received up to 13 doses of placebo. Both descriptions specify intramuscular administration in the deltoid or lateral thigh, avoiding anatomical regions where lymph nodes had been excised. The MAGE-A3 arm description is incomplete after its account of the initial dosing phase, so the full schedule is not reported in the registry record.

The posted primary endpoint is disease-free survival at final analysis, with the timeframe stated as Month 30 = Year 2.5. The record also lists secondary outcomes covering overall survival, disease-free specific survival, distant metastasis-free survival, health-related quality of life, immune response measures, and abnormal laboratory parameters. The study start and primary completion dates are recorded as 2008-12-01 and 2016-01-27. The registry does not report the original design assumptions or the reason the study was terminated.

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: Out of 1351 patients enrolled in the study, 6 patients did not receive treatment and were excluded from study start, hence 1345 patients were included in the Total treated population-as treated (895 in the MAGE-A3 Group and 450 in the Placebo Group). Between the final and follow-up analyses, 1 patient was found to have an invalid ICF and was not included in the follow-up analysis, which included 1344 patients: 894 in the MAGE-A3 Group and 450 in the Placebo Group.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: The analysis was performed on the Total Treated population - as randomized, which included patients in the treatment groups as allocated by the randomization system at the start of the study.

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 Disease Free Survival (DFS), measured over At Final analysis (Month 30 = Year 2.5).

The primary outcome is disease-free survival (DFS), defined as the time from randomization to the first investigator-assessed disease recurrence or death from any cause, whichever came first. Recurrences could be loco-regional or distant metastases. A death without previously documented tumor recurrence counted as an event rather than being censored. A new primary cancer at another site, including a second primary melanoma, was not considered a recurrence.

The posted measure is a rate: the number of patients with at least one event divided by the total follow-up time in years until the first event. It is reported as first events per person-year, with the final analysis timeframe stated as Month 30 = Year 2.5. If no event had occurred by the analysis, follow-up was censored at the participant’s last tumor assessment or visit. The registry states that counting death as an event was considered less prone to introduce bias than censoring such deaths.

General statistical context: a time-to-event outcome incorporates both whether an event occurred and when it occurred. Expressing events relative to accumulated person-time accounts for differing observation periods in the rate calculation. This can be useful when participants have varying follow-up, but the rate does not by itself show the timing pattern of events or the shape of survival over time. The record also reports a Cox-model hazard ratio, a different summary of the relative event experience over follow-up.

7. Results Posted in the Registry

• MAGE-A3 Group: 0.51 versus 0.48 (First events per person-year) on control; rate ratio 1.06 (95% CI 0.95–1.17, approximate, Poisson model); p ≈ 0.294.
• GS+ Mage-A3 Sub-Group: 0.50 versus 0.48 (First events per person-year) on control; rate ratio 1.05 (95% CI 0.90–1.22, approximate, Poisson model); p ≈ 0.556.
• GS+ Placebo Sub-Group: 0.46 versus 0.48 (First events per person-year) on control; rate ratio 0.96 (95% CI 0.80–1.16, approximate, Poisson model); p ≈ 0.691.
• GS- Mage-A3 Sub-Group: 0.44 versus 0.48 (First events per person-year) on control; rate ratio 0.91 (95% CI 0.79–1.06, approximate, Poisson model); p ≈ 0.224.

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.

Disease-free Specific Survival (DFSS)

• Endpoint type: Event rate: events per unit of follow-up, compared as a rate ratio.
• Test: Poisson rate ratio, based on the estimated number of events and follow-up time.
• Result:
  • MAGE-A3 Group: 0.50 versus 0.48 (First events per person-year) on control; rate ratio 1.04 (95% CI 0.94–1.16, approximate, Poisson model); p ≈ 0.412.
  • GS+ Mage-A3 Sub-Group: 0.50 versus 0.48 (First events per person-year) on control; rate ratio 1.05 (95% CI 0.90–1.22, approximate, Poisson model); p ≈ 0.556.
• Interpretation:
  • The 95% confidence interval includes no difference (a ratio of 1), so these data do not show a statistically significant difference between MAGE-A3 Group and the control. The interval gives the range of effects that remain compatible with the data.
  • The 95% confidence interval includes no difference (a ratio of 1), so these data do not show a statistically significant difference between GS+ Mage-A3 Sub-Group and the control. The interval gives the range of effects that remain compatible with the data.

Distant Metastasis-free Survival (DMFS)

• Endpoint type: Event rate: events per unit of follow-up, compared as a rate ratio.
• Test: Poisson rate ratio, based on the estimated number of events and follow-up time.
• Result:
  • MAGE-A3 Group: 0.39 versus 0.34 (First events per person-year) on control; rate ratio 1.13 (95% CI 1.00–1.28, approximate, Poisson model); p ≈ 0.043.
  • GS+ Mage-A3 Sub-Group: 0.39 versus 0.34 (First events per person-year) on control; rate ratio 1.13 (95% CI 0.95–1.35, approximate, Poisson model); p ≈ 0.152.
• Interpretation:
  • The 95% confidence interval excludes no difference (a ratio of 1), so the event rate is statistically higher for MAGE-A3 Group than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.
  • The 95% confidence interval includes no difference (a ratio of 1), so these data do not show a statistically significant difference between GS+ Mage-A3 Sub-Group and the control. The interval gives the range of effects that remain compatible with the data.

Number of Subjects With Potential Immune-mediated Diseases (pIMDs)

• Endpoint type: Binary: the number of participants with the event, compared as proportions.
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • MAGE-A3 Group: 3.7% (33/895) versus 5.1% (23/450) on control; risk difference -1.4 percentage points (95% CI -3.8 to 1.0), risk ratio 0.72 (0.43–1.21); unadjusted two-proportion z-test p = 0.217.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between MAGE-A3 Group and the control. The interval gives the range of effects that remain compatible with the data.

The registry lists secondary outcomes spanning survival and recurrence-related measures, health-related quality of life, anti-MAGE-A3 antibody concentrations and responses, and abnormal hematological and biochemical parameters. It also posts results for disease-free specific survival, distant metastasis-free survival, and potential immune-mediated diseases. Taken together, these outcomes address several distinct aspects of follow-up and biological response rather than one single confirmatory question.

The broad pattern is that secondary estimates are not uniformly aligned in direction or precision. Some posted comparisons have intervals that include the null value, while the recalculated distant metastasis-free survival comparison has an interval just above 1 and an unadjusted p-value of 0.043. This should be read in the context of the primary DFS result, whose recalculated interval includes 1, and the many outcomes and subgroup comparisons in the registry. The secondary findings do not by themselves resolve the primary comparison or establish a consistent treatment effect across outcomes.

Unadjusted secondary results carry less weight for confirmatory claims because multiple outcomes and subgroup analyses create more opportunities for apparently notable findings. General statistical context: when many hypotheses are tested, the chance of at least one small p-value can increase, even if no underlying effect exists. Adjustments or a clearly pre-specified hierarchy can help control this issue. The registry does not report a multiplicity adjustment or enough detail about outcome prioritization to determine how these analyses were intended to support formal conclusions.

Some secondary outcomes are not accompanied by sponsor-reported comparative analyses in the record. The posted summary statistics and recalculations can be descriptive, but they should not be treated as independent confirmation of the primary finding. The registry’s full analysis plan and the handling of missing secondary-outcome data are not reported.

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

• Overall Survival (OS)
• Disease-free Specific Survival (DFSS)
• Distant Metastasis-free Survival (DMFS)
• Health-related Quality of Life
• Number of Subjects With Anti-MAGE-A3 Antibody Concentrations Above the Cut-off Value
• Anti-MAGE-A3 Antibody Geometric Mean Concentrations
• Number of Subjects With Anti-MAGE-A3 Antibody Response
• Number of Subjects With Abnormal Haematological and Biochemical Parameters

9. Statistical Interpretation

10. Statistical Methods Explained

The primary outcome is expressed as an event rate, using first events per person-year. In general, an event rate divides the number of participants with an event by the accumulated time at risk. Here, the registry’s definition specifies the number with at least one qualifying first event divided by the sum of follow-up periods in years until recurrence or death. The event definition and the way death is handled matter because they determine what enters the numerator and when follow-up ends.

The main rate comparison is an unadjusted recalculation from the posted group rates and exposure information, using a Poisson rate-ratio approach. It reports a rate ratio of 1.056, a 95% confidence interval of 0.954 to 1.171, and p = 0.294. General statistical context: a rate ratio compares the event rates between groups, with 1 representing equal rates. A confidence interval describes the range of effect sizes reasonably compatible with the data under the method’s assumptions. A p-value measures how surprising data at least as discrepant as those observed would be under a specified null hypothesis; it is not the probability that the null hypothesis is true.

The registry separately reports Cox regression analyses. A Cox model is a time-to-event method that estimates a hazard ratio, comparing the instantaneous event rates between groups over follow-up under the model’s assumptions. The sponsor’s main-group result is a hazard ratio of 1.013, with a 95% confidence interval of 0.879 to 1.169 and p = 0.857. A hazard ratio is not the same quantity as a rate ratio: the former is derived from time-to-event analysis, while the latter compares event counts relative to person-time. Their estimates and p-values need not match.

The sponsor analysis is explicitly labelled non-inferiority. General statistical context: a non-inferiority analysis assesses whether an intervention is not worse than a comparator by more than a pre-specified margin. The margin and the details needed to evaluate the sponsor’s non-inferiority analysis are not reported in the registry record. A p-value alone cannot establish non-inferiority without the margin and the relevant decision rule.

What does censoring mean?

The outcome definition says that participants without an event by the analysis were censored at their last tumor assessment or visit. In general, censoring means that the exact event time is not observed within the period used for analysis; the participant contributes information up to the last known assessment. Valid time-to-event interpretation depends on assumptions about how censoring relates to event risk. The registry does not report diagnostic checks or additional methods for assessing those assumptions.

Why can event rates be useful?

Person-time rates can accommodate differing observation lengths by relating event counts to accumulated follow-up. General statistical context: count-based rate models may require attention to over-dispersion, which occurs when the variability in event counts exceeds the model’s assumptions. The record does not report a dispersion estimate or a model adjustment for over-dispersion. It also does not provide confidence limits for the separate arm rates, so precision is available in the posted comparison estimates rather than for each arm rate individually.

What can subgroup analyses tell us?

The record reports comparisons for GS+ and GS- subgroups, including rate-based recalculations and some sponsor Cox analyses. Subgroups can describe whether estimates differ across categories, but they generally use fewer participants and are less precise than the full-group comparison. General statistical context: a claim that treatment effects differ by subgroup calls for a direct interaction test, not simply separate within-subgroup p-values. Such a test and the subgroup analysis plan are not reported in the registry record.

The recalculations are unadjusted. The registry does not report whether or how the sponsor’s primary analysis adjusted for prognostic factors, nor does it report the full statistical analysis plan. As a result, the recalculated rate ratio is best read as a descriptive, unadjusted comparison of the posted summary statistics, alongside—not in place of—the sponsor’s analysis.

11. Confidence Intervals and Precision

For the primary DFS comparison, the unadjusted rate-ratio estimate is 1.056, with a 95% confidence interval from 0.954 to 1.171. The interval includes 1, the value corresponding to equal rates under this measure. Thus, the posted recalculation is compatible with values on either side of equal rates; it does not identify a precise direction or magnitude of a difference. The unadjusted p-value is 0.294.

The sponsor’s Cox estimate for the main groups is a hazard ratio of 1.013, with a 95% confidence interval from 0.879 to 1.169 and p = 0.857. This interval also includes 1. Because the hazard ratio and rate ratio are different measures, their intervals should be interpreted in relation to their respective models rather than compared as if they were interchangeable.

General statistical context: a confidence interval reflects sampling uncertainty under the analysis method and its assumptions. Wider intervals indicate less precision about the effect size; an interval that includes the null value does not prove that the groups are identical. It indicates that the data do not rule out the null value at the confidence level represented by that interval. Confidence intervals also do not account for every source of uncertainty, such as unreported analysis choices or potential bias.

The subgroup estimates are also uncertain. The rate-ratio intervals for the reported subgroup comparisons include 1, and the sponsor’s subgroup Cox intervals include 1. Smaller subgroup populations can produce less precise estimates, though the registry does not report subgroup-specific precision calculations beyond the estimates and intervals. The posted results do not establish that subgroup effects differ from one another; the relevant interaction analysis is not reported in the registry record.

Precision should therefore be judged from the interval ranges and the methods behind each estimate, not from a p-value alone. The registry does not provide confidence intervals for the arm-specific event rates or the reason for termination. These omissions limit what can be inferred about the full range of uncertainty and study conduct.

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 sample-size figures are planning illustrations based on the observed effect, not the trial’s original design calculation. The stated assumptions are a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation. The effect to detect is set at the stated fraction of the observed effect. The original trial’s design assumptions and sample-size justification are not reported in the registry record.

For time-to-event comparisons, the planning convention gives the total number of events required, using the Schoenfeld approach. For other outcomes, figures are participants per arm. Read each outcome’s planning values according to that convention; an event total is not the same as a participant count, because participants may have different follow-up and not all participants necessarily have an event.

The 80% power column describes the number of events or participants the illustration estimates would be needed to have an 80% chance of detecting the specified effect under its assumptions. The 90% power column gives the corresponding planning target for a 90% chance. These are design probabilities under an assumed effect and model, not probabilities that the observed result is true. This is not observed power.

The rows use the full observed effect or smaller fractions of that effect. Smaller effects require more information to distinguish them from no difference, so the planning figures increase as the target effect becomes smaller. The values are hypothetical planning requirements based on the observed estimates and assumptions, not an explanation of why the trial enrolled the reported number of participants or why it was terminated. They also should not be treated as a recommendation for a future study without independent design work and a justified target effect.

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, λ₀ and λ₁ are event rates per patient-year, and T is follow-up per participant in years. This assumes Poisson counts and equal follow-up.

Worked calculation for this trial

1. Control rate λ₀ = 0.4780; treatment rate λ₁ = 0.5050 events per patient-year; follow-up T = 2.50 years.
2. n = (1.96 + 0.842)² × (1/0.4780 + 1/0.5050) / (2.50 × [ln(1.056)]²) = 4,234.5, rounded up to 4,235 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
lambda0 <- 0.478                      # control rate, events per patient-year
lambda1 <- 0.505                      # treatment rate
tt      <- 2.5                       # follow-up per participant, years
n_per_arm <- function(power, frac = 1) {
  l1 <- lambda0 * (lambda1 / lambda0)^frac   # smaller effect = fraction of the observed log rate ratio
  ceiling(round((z_a + qnorm(power))^2 * (1 / lambda0 + 1 / l1) / (tt * log(l1 / lambda0)^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 = ",")))

# Simulation 1: Poisson counts. Simulation 2: over-dispersed counts (variance = 1.5 x mean), analysed with the matching variance.
set.seed(796445)
n    <- n_per_arm(0.80)
reps <- 20000
m0 <- lambda0 * n * tt; m1 <- lambda1 * n * tt
x0 <- rpois(reps, m0); x1 <- rpois(reps, m1)
z  <- log(x1 / x0) / sqrt(1 / x0 + 1 / x1)
cat(sprintf("Simulated power, Poisson counts, n = %s per arm: %.3f\n", format(n, big.mark = ","), mean(abs(z) > z_a, na.rm = TRUE)))
phi <- 1.5
y0 <- rnbinom(reps, mu = m0, size = m0 / (phi - 1)); y1 <- rnbinom(reps, mu = m1, size = m1 / (phi - 1))
z2 <- log(y1 / y0) / sqrt(phi * (1 / y0 + 1 / y1))
cat(sprintf("Simulated power, over-dispersed counts (phi = 1.5), same n: %.3f\n", mean(abs(z2) > z_a, na.rm = TRUE)))

Output from running the code

100% of observed effect: n per arm = 4,235 (80% power), 5,669 (90% power)
 75% of observed effect: n per arm = 7,579 (80% power), 10,146 (90% power)
 50% of observed effect: n per arm = 17,168 (80% power), 22,983 (90% power)
Simulated power, Poisson counts, n = 4,235 per arm: 0.806
Simulated power, over-dispersed counts (phi = 1.5), same n: 0.634

The first simulation draws Poisson counts and should land close to the target power. The second allows extra variability (variance 1.5 times the mean, a purely illustrative value), which lowers the power at the same sample size, so rate trials usually need more participants than the Poisson formula suggests.

14. Analyses Reported by the Sponsor

The sponsor’s reported primary comparison uses Cox regression and is labelled non-inferiority. It gives a hazard ratio of 1.013, a 95% confidence interval of 0.879 to 1.169, and p = 0.857. The unadjusted recalculation from the posted DFS rates uses a rate ratio instead and gives 1.056, with a 95% confidence interval of 0.954 to 1.171 and p = 0.294. The methods and effect measures differ, so the two results are not expected to be numerically identical.

The discrepancy flag is false, so the posted sponsor analysis does not conflict with the recalculation in the sense identified by the registry analysis. The sponsor’s own analysis takes precedence as the trial’s reported analysis, while the unadjusted recalculation is a supplementary way to examine the summary data. The registry does not report the non-inferiority margin or enough details to independently evaluate the sponsor’s decision rule. The recalculation therefore should not be used to replace or reinterpret the sponsor’s formal analysis.

For the subgroup analyses, the sponsor reports Cox-model hazard ratios for the overall GS+ and GS- comparisons. Some of the recalculated subgroup comparisons do not correspond exactly to the sponsor’s reported comparison pairings. The registry’s arm labels and subgroup presentation should be read as posted rather than used to infer an unreported subgroup analysis plan.

15. Multiplicity, Missing Data and Analysis Populations

The registry lists 13 outcome measures and six analyses, and the posted results cover multiple outcomes and subgroup comparisons. It does not report whether the trial used a formal multiplicity strategy, such as a testing hierarchy or adjustment of significance levels. Without that information, unadjusted p-values for secondary or subgroup comparisons should be regarded as exploratory rather than as a set of controlled confirmatory tests. The record also does not report whether the sponsor designated any secondary endpoints as confirmatory.

The primary outcome description states that participants with no event by the time of analysis were censored at their last tumor assessment or visit. This specifies one aspect of follow-up, but broader missing-data procedures are not reported in the registry record. In particular, the record does not describe missing tumor assessments, reasons for loss to follow-up, sensitivity analyses for censoring, or approaches to missing data for other outcomes. The assumptions used for analyses of quality of life and laboratory measures are also not reported.

The primary analysis population is identified as Total Treated, analyzed as randomized. The description says this includes patients in the treatment groups allocated by the randomization system at study start. The registry does not provide further details on exclusions, protocol deviations, treatment exposure, or the populations used for every secondary analysis. Those details matter when comparing the posted population definition with the population required for a formal non-inferiority assessment.

The sponsor’s main analysis is labelled non-inferiority, but the margin and the full analysis plan are not reported. The registry also does not report the rationale for termination or whether planned analyses changed after termination. These are limits on interpretation, not evidence that a particular method was or was not used. No assumptions about the unreported procedures can be made from the posted results alone.

16. Safety Results

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

Deaths

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: basal cell carcinoma (2.9%), cellulitis (1.1%), erysipelas (1.1%), malignant melanoma (1.0%), squamous cell carcinoma of skin (0.7%).

The serious adverse event summary reports 129 affected participants among 895 at risk in the MAGE-A3 group and 64 among 450 at risk in the placebo group. Deaths are reported as 5 and 1, respectively. The record also lists serious event terms, including basal cell carcinoma at 2.9%, cellulitis at 1.1%, erysipelas at 1.1%, malignant melanoma at 1.0%, squamous cell carcinoma of skin at 0.7%, and cholelithiasis at 0.4%.

These are descriptive counts and listed percentages from the registry. They do not establish that an event was caused by treatment, and the record does not provide enough information to assess timing, duration, severity gradations, or participant-level overlap among event terms. Serious adverse event counts should not be confused with the number of all adverse events or with the number of participants who discontinued treatment.

The at-risk counts in the safety summary differ from the primary DFS analysis populations, and the record does not explain the population accounting in detail. The figures therefore should be read within their stated safety summary rather than combined with the primary endpoint counts. No safety conclusion about comparative benefit or harm follows from these counts alone.

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

The posted primary results describe DFS event rates of 0.505 first events per person-year in the MAGE-A3 group and 0.478 in the placebo group. The unadjusted rate-ratio estimate is 1.056, with a 95% confidence interval of 0.954 to 1.171 and p = 0.294. The sponsor’s Cox analysis reports a hazard ratio of 1.013, with a 95% confidence interval of 0.879 to 1.169 and p = 0.857. Both intervals include 1, and the estimates are not precise enough to establish a difference in the posted comparisons.

These results do not show that the treatments are identical. An interval including the null value means that the analysis does not rule out that value at the stated confidence level; it does not demonstrate equivalence. Likewise, the sponsor’s non-inferiority label cannot be fully evaluated without the margin and the complete analysis details, which are not reported in the registry record.

The data also do not establish that the immunotherapeutic prevents relapse, or that it fails to do so. The trial record describes a comparison of groups on a specified endpoint and timeframe. Interpretation must account for the endpoint definition, follow-up and censoring, differences between the sponsor’s hazard-ratio method and the unadjusted rate-ratio recalculation, and the multiplicity of secondary analyses.

Secondary results offer additional descriptions but do not erase the uncertainty in the primary comparison. Nor does the terminated status explain the results: the reason for termination is not reported. The registry summary alone does not establish why the study ended or how termination affected the planned analyses. The appropriate statistical reading is limited to the estimates, intervals, and methods actually recorded, with unreported design and analysis details left unresolved.

18. Why This Trial Matters Statistically

This trial illustrates why the reported endpoint and effect measure matter. The registry’s primary outcome is expressed as first events per person-year, while the sponsor reports a Cox-model hazard ratio. Both relate to time-to-event experience, but they summarize it differently. A reader should not treat a rate ratio and a hazard ratio as interchangeable simply because they concern the same clinical endpoint.

It also demonstrates the role of precision in interpretation. A point estimate alone gives an incomplete picture; the confidence interval shows the range of effect sizes compatible with the analysis. The primary recalculated interval includes 1, as does the sponsor’s reported interval. The estimates therefore require a cautious reading rather than a conclusion based only on whether one number appears higher than another.

The registry combines a primary endpoint with several secondary outcomes and subgroup comparisons. That breadth can help describe different aspects of the study, but it raises the importance of knowing which hypotheses were confirmatory and how multiplicity was handled. Those details are not reported. A nominally notable secondary p-value should consequently be read in context rather than treated as a standalone conclusion.

Finally, the study record distinguishes the sponsor’s formal analysis from a supplementary, unadjusted recalculation. Keeping those roles separate is important: the recalculation can help readers understand summary data, but it does not replace the trial’s own analysis or supply missing protocol details. The case underscores how transparent reporting of analysis populations, censoring, non-inferiority margins, and multiplicity is needed to assess clinical trial results fully.

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.
• Rate comparisons use a Poisson approximation unless the sponsor posted confidence limits; real event counts are often over-dispersed.

Several limitations specific to this record constrain interpretation. The study is marked terminated, but the reason for termination is not reported in the registry record. The record does not say whether termination was related to results, recruitment, operational issues, or another cause, so no explanation should be inferred.

The original sample-size rationale, target effect, and design assumptions are not reported. Planning illustrations based on observed effects do not reconstruct the original design. Likewise, the sponsor’s primary analysis is labelled non-inferiority, but its margin and decision criteria are not included. This prevents an independent assessment of the non-inferiority claim from the registry summary alone.

The trial description identifies the primary analysis population as Total Treated, analyzed as randomized, but does not give detailed accounting of exclusions, protocol deviations, or participant flow. The MAGE-A3 arm description is truncated, leaving part of the dosing schedule unclear. The record also does not report adherence, treatment discontinuation, or duration of actual exposure.

For the time-to-event endpoint, the censoring rule for participants without an event is stated, but the reasons for censoring, completeness of assessments, and sensitivity analyses are not reported. The record does not describe checks of model assumptions or the effect of potential over-dispersion in the count-rate recalculation. Multiplicity handling and the prespecification of subgroup analyses are also not reported. These gaps do not establish that the corresponding procedures were absent; they mean that the registry record does not document them.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: A Phase III Study to Test the Benefit of a New Kind of Anti-cancer Treatment in Patients With Melanoma, After Surgical Removal of Their Tumor — NCT00796445.
• Registry record last updated 2021-03-05; status last verified 2021-02.
• Results first posted 2019-03-19.
• 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

NCT00796445 is a randomized, triple-masked, parallel Phase III study of GSK 2132231A versus placebo in patients with melanoma after surgical removal of their tumor. The registry reports enrollment of 1351 and a terminated status. The primary outcome is disease-free survival, defined by recurrence or death, expressed as first events per person-year at final analysis, Month 30 = Year 2.5.

The main-group recalculation gives a rate ratio of 1.056, with a 95% confidence interval of 0.954 to 1.171 and p = 0.294. The sponsor’s Cox analysis reports a hazard ratio of 1.013, with a 95% confidence interval of 0.879 to 1.169 and p = 0.857, labelled as non-inferiority. The record does not report the non-inferiority margin, the reason for termination, or several key analysis details, including multiplicity procedures and the original sample-size justification.