Phase II Study of MEDI4736, Tremelimumab, and MEDI4736 in Combination w/ Tremelimumab Squamous Cell Carcinoma of the Head and Neck: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Phase II Study of MEDI4736, Tremelimumab, and MEDI4736 in Combination w/ Tremelimumab Squamous Cell Carcinoma of the Head and Neck (NCT02319044), a randomized phase 2 study. The primary outcome is Objective Response Rate at 6 Months.

2. Clinical Question

3. Trial Design

This was a randomized, open-label, parallel-group Phase II study with no masking. The registry lists a total enrollment of 267 and three treatment arms: MEDI4736 monotherapy, tremelimumab monotherapy, and MEDI4736 plus tremelimumab. The sponsor was AstraZeneca. The study is recorded as completed.

The registry describes the intended population as patients with recurrent or metastatic squamous cell carcinoma of the head and neck who had progressed during or after treatment with a platinum-containing regimen for recurrent or metastatic disease. The primary analysis used an evaluable analysis set, defined as patients who received at least one dose of study treatment, had a baseline tumor assessment, and had measurable disease. The posted arm sizes for that endpoint were 130 for the combination, 65 for MEDI4736, and 63 for tremelimumab.

The primary outcome was objective response rate at 6 months. The registry specifies a primary analysis based on blinded independent central review according to RECIST v1.1 and excludes unconfirmed responses from the response count. Posted secondary outcomes include best objective response, duration of response, time to response, disease control rate, and progression-free survival. The registry lists 13 outcome measures and 11 posted secondary outcomes.

Details of the original sample-size justification, allocation procedure, stratification, and any planned interim analyses are not reported in the registry record. The original statistical analysis plan is also not reported in the registry record. The posted planning illustrations are based on the observed comparisons and do not describe the trial's original design assumptions.

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

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Evaluable analysis set - all patients who received at least one dose of study treatment, who had a baseline tumor assessment, and had measurable disease

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: Full analysis set - all randomized patients

6. Primary Endpoint

The registry lists the primary outcome as Objective Response Rate at 6 Months, measured over After 6 months.

The primary endpoint is the objective response rate at 6 months, expressed as the percentage of participants with a response. It counts responses assessed by blinded independent central review using RECIST v1.1, and the registry says unconfirmed responses are excluded. The analysis set includes patients who received at least one dose of treatment, had a baseline tumor assessment, and had measurable disease.

A response-rate endpoint provides a proportion of patients meeting a prespecified tumor-response criterion within the stated assessment framework. It can be described by arm and compared across groups, while confidence intervals convey the statistical uncertainty around the estimated proportion or difference. As a binary outcome, each patient contributes to the response or non-response category for this endpoint; the measure does not describe the duration of response or the time until progression.

The 6-month response outcome therefore addresses whether a qualifying response was recorded at the specified time point, not every aspect of disease course. Other outcomes listed in the registry, including duration of response and progression-free survival, address different questions. The registry record does not report the complete assessment schedule or further endpoint-handling details beyond the stated review method, criteria, population, and exclusion of unconfirmed responses.

7. Results Posted in the Registry

• MEDI4736: 9.2% (6/65) versus 7.7% (10/130) on control; risk difference 1.5 percentage points (95% CI -6.9 to 9.9), risk ratio 1.20 (0.46–3.16); unadjusted two-proportion z-test p = 0.712.
• Tremelimumab: 1.6% (1/63) versus 7.7% (10/130) on control; risk difference -6.1 percentage points (95% CI -11.6 to -0.6), risk ratio 0.21 (0.03–1.58); unadjusted two-proportion z-test p = 0.086.

8. Statistical Interpretation

9. Statistical Methods Explained

What is a response rate?

For a binary endpoint, a response rate is the proportion of participants classified as having the event of interest. Here the event is an objective response recorded at 6 months under the stated RECIST v1.1 assessment and central-review approach. The reported percentages are arm-specific estimates in the evaluable analysis set. They do not capture how long a response lasted; duration of response is listed separately among the outcomes.

What does the risk difference describe?

A risk difference compares the response proportions by subtracting one group’s response probability from the other’s, using the comparison direction specified in the analysis. It is expressed on the absolute risk scale, so zero represents equal estimated proportions. The posted risk-difference estimate for MEDI4736 versus the combination is 0.015, with a 95% confidence interval from -0.069 to 0.099. For tremelimumab versus the combination, the estimate is -0.061, with a 95% confidence interval from -0.116 to -0.006. These are the registry’s reported values; no further recalculation is needed to interpret their scale or uncertainty.

How do risk ratios and odds ratios differ?

A risk ratio compares the probability of the event between groups as a ratio. An odds ratio compares the odds, where odds are the event probability relative to the probability of no event. Both are relative measures and can look different from an absolute risk difference, particularly when events are uncommon. The registry reports a risk ratio of 1.2 and an odds ratio of 1.22 for MEDI4736 versus the combination. Their confidence intervals are 0.456 to 3.157 and 0.423 to 3.519, respectively. For tremelimumab versus the combination, the reported risk ratio is 0.206 (95% CI 0.027 to 1.577) and the odds ratio is 0.194 (95% CI 0.024 to 1.547).

What does a confidence interval convey?

In general, a confidence interval describes uncertainty around an estimate under the statistical method used. A wider interval means the data are compatible with a broader range of values. An interval that includes zero for a risk difference is compatible with no difference under that interval calculation; an interval that excludes zero does not, on its own, resolve questions about multiplicity, bias, or clinical relevance. The registry record does not report the full methods needed to assess all those issues.

Why do sparse events matter?

General statistical explanation: when an outcome occurs infrequently, estimates can be sensitive to a small number of outcome classifications. The number of responses was 10 in the combination group, 6 in the MEDI4736 group, and 1 in the tremelimumab group. This limited event count helps explain why relative-effect confidence intervals are broad, particularly for comparisons involving the tremelimumab arm. Sparse data can make asymptotic approximations less reliable, depending on the method; the registry record does not report whether alternative small-sample methods were used.

What does the p-value add?

A p-value summarizes how unusual data at least as discrepant as those observed would be under a specified null model and analysis. It is not the probability that the null hypothesis is true, nor is it a measure of the size or importance of a difference. The posted unadjusted p-values are 0.712 for MEDI4736 versus the combination and 0.086 for tremelimumab versus the combination. The registry record does not report the detailed test specification or any multiplicity correction for these comparisons.

What is over-dispersion?

General statistical explanation: over-dispersion occurs when observed variability is greater than a model assumes. It is particularly relevant for some count or clustered-data models, where ignoring extra variability can make uncertainty appear too small. The primary endpoint is binary, and the registry record does not report an over-dispersion assessment or a model-based analysis of the response endpoint. It would not be appropriate to infer that such an analysis was conducted.

What does the analysis population mean?

The primary endpoint population is defined by treatment received, baseline tumor assessment, and measurable disease. This is the population to which the posted response percentages apply. It differs conceptually from an analysis of every randomized participant. The registry record does not report additional details on missing assessments, exclusions by arm, or sensitivity analyses for the primary outcome.

10. Confidence Intervals and Precision

The arm-specific 95% confidence intervals indicate substantial uncertainty around the estimated response proportions. For the combination group, the interval is 3.75% to 13.69%; for MEDI4736, it is 3.46% to 19.02%; and for tremelimumab, it is 0.04% to 8.53%. The intervals overlap, and the estimates should not be interpreted as exact underlying response probabilities. Their width reflects limited information in the evaluable groups, particularly when the number of observed responses is small.

For the risk difference comparing MEDI4736 with the combination, the 95% confidence interval runs from -0.069 to 0.099. It spans zero, so the data are compatible with values on either side of no difference under the reported interval. For the tremelimumab comparison, the interval runs from -0.116 to -0.006. The posted interval and unadjusted p-value should both be considered, while recognizing that the registry record does not give enough detail to reconstruct the test or any adjustment strategy.

Relative-effect intervals are also broad: the risk-ratio interval is 0.456 to 3.157 for MEDI4736 versus the combination, and 0.027 to 1.577 for tremelimumab versus the combination. The odds-ratio intervals are 0.423 to 3.519 and 0.024 to 1.547, respectively. This breadth means the data do not pin down the relative effects precisely. Confidence intervals quantify statistical uncertainty under an analysis; they do not account automatically for selection, missing outcome data, or multiplicity.

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 effect, not the original trial design. The registry record does not report the trial’s own sample-size justification or original planning assumptions. The illustrations use a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. Each row considers detecting the stated fraction of the observed effect. The figures are participants per arm for these binary outcome comparisons; the note about total events applies to time-to-event planning calculations, not these comparisons.

The 80% power column gives the number of participants per arm that the planning calculation estimates would be needed to have an 80% chance of detecting the chosen effect under the listed assumptions, if that effect were the true one. The 90% power column gives the corresponding per-arm number for a 90% chance. These are design probabilities under assumptions, not guarantees for a future study. Smaller effects require more participants because they are harder to distinguish from random variation.

For MEDI4736 versus the combination, planning for 100% of the observed effect gives 5,136 participants per arm for 80% power and 6,876 for 90% power. Planning for 75% gives 8,943 and 11,972, respectively; planning for 50% gives 19,696 and 26,367. For tremelimumab versus the combination, the corresponding values are 186 and 248 at 100%, 382 and 511 at 75%, and 974 and 1,303 at 50%. These illustrations depend on the observed effect and the stated assumptions. This is not observed power.

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, p₀ and p₁ are the control and treatment proportions, and p̄ is their average.

Worked calculation for this trial

1. Control proportion p₀ = 0.077; treatment proportion p₁ = 0.092; average p̄ = 0.085.
2. First term: z_1−α/2 × √[2p̄(1−p̄)] = 1.96 × 0.3936 = 0.7714.
3. Second term: z_1−β × √[p₀(1−p₀) + p₁(1−p₁)] = 0.842 × 0.3934 = 0.3311.
4. n = (0.7714 + 0.3311)² / (0.0154)² = 5,135.9, rounded up to 5,136 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
p0 <- 0.0769230769230769                          # control proportion
p1 <- 0.0923076923076923                          # treatment proportion
n_per_arm <- function(power, frac = 1) {
  q1 <- p0 + (p1 - p0) * frac               # smaller effect = fraction of the observed difference
  pbar <- (p0 + q1) / 2
  z_b <- qnorm(power)
  ceiling(round((z_a * sqrt(2 * pbar * (1 - pbar)) + z_b * sqrt(p0 * (1 - p0) + q1 * (1 - q1)))^2 / (q1 - p0)^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 (unrounded n per arm)
cat("power.prop.test:", round(power.prop.test(p1 = p0, p2 = p1, power = 0.80)$n, 1), "\n")

# Simulation: how often does the z-test reject at the 80%-power sample size?
set.seed(2319044)
n    <- n_per_arm(0.80)
reps <- 20000
x0 <- rbinom(reps, n, p0); x1 <- rbinom(reps, n, p1)
ph <- (x0 + x1) / (2 * n)
z  <- ((x1 - x0) / n) / sqrt(ph * (1 - ph) * 2 / n)
cat(sprintf("Simulated power at n = %s per arm: %.3f (%d replicates)\n", format(n, big.mark = ","), mean(abs(z) > z_a, na.rm = TRUE), reps))

Output from running the code

100% of observed effect: n per arm = 5,136 (80% power), 6,876 (90% power)
 75% of observed effect: n per arm = 8,943 (80% power), 11,972 (90% power)
 50% of observed effect: n per arm = 19,696 (80% power), 26,367 (90% power)
power.prop.test: 5135.9 
Simulated power at n = 5,136 per arm: 0.798 (20000 replicates)

The simulation draws many pairs of event counts at the planned sample size and counts how often the two-proportion z-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. Multiplicity, Missing Data and Analysis Populations

The study compares three treatment groups and posts two pairwise comparisons against the combination group for the primary response outcome. The registry record does not report whether the original statistical plan controlled the family-wise error rate or used another multiplicity strategy. The reported p-values are explicitly unadjusted. When several comparisons are considered, the chance of at least one apparently unusual result can increase if no multiplicity procedure is used; this is a general statistical concern, not evidence that a particular adjustment was planned or omitted from the study’s original analysis.

Missing-data handling for the primary endpoint is not reported in the registry record. The endpoint population is defined as patients who received at least one dose, had a baseline tumor assessment, and had measurable disease, but the record does not describe how patients without an evaluable follow-up assessment were classified. It also does not report sensitivity analyses or the extent and pattern of missing outcome data. These details matter because response proportions can depend on how non-evaluable participants are treated.

The posted analyses use the evaluable analysis set, while the study itself was randomized and had enrollment of 267. The registry record does not report a detailed intention-to-treat analysis for this endpoint or explain how the evaluable population relates to the randomized population by arm. The endpoint arm sizes are 130, 65, and 63, whereas the safety denominators differ for some groups. This underscores that outcome-specific populations and denominators need to be kept distinct.

Other analysis details are also not reported in the registry record, including stratification variables, covariate adjustment, any interim analyses, and the full statistical analysis plan. No claim about their use or absence can be made from the record alone.

14. 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: diarrhoea (7.7%), pneumonia (6.8%), dysphagia (3.1%), pneumonia bacterial (3.1%), dizziness (3.1%).

The serious adverse event figures are counts of participants affected and the corresponding at-risk group sizes, not measures of treatment effect. For the combination group, 59 of 133 participants were recorded as affected by a serious adverse event; for MEDI4736, 18 of 65; and for tremelimumab, 25 of 65. The registry also reports other adverse events affecting 110 of 133, 52 of 65, and 59 of 65 participants, respectively. Death counts are 86, 44, and 51, respectively.

The most frequently listed serious adverse event terms include diarrhoea at 7.7%, pneumonia at 6.8%, dysphagia at 3.1%, bacterial pneumonia at 3.1%, dizziness at 3.1%, and dyspnoea at 3.1%. Those terms are recorded as percentages in the registry; the listed data do not establish that a treatment caused an event. The record does not give further details in the listed summary about severity grading, timing, attribution, or event overlap. Participants may have experienced more than one event, so affected-participant counts and term-specific counts should not be treated as interchangeable.

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

The posted primary results describe objective responses recorded by blinded independent central review at 6 months in the evaluable analysis set. The response estimates were 7.7% for the combination, 9.2% for MEDI4736, and 1.6% for tremelimumab. The confidence intervals are broad, and the pairwise estimates carry substantial uncertainty. They are useful descriptions of the observed data, but they do not establish a definitive ordering of the regimens.

Randomization is a design feature intended to support fair comparisons between assigned groups. However, the endpoint analysis set is defined by receipt of treatment and having baseline and measurable disease assessments. The registry record does not report enough detail about missing outcomes, exclusions, or sensitivity analyses to assess fully how the evaluable-set restriction may affect interpretation. Nor does the record report a multiplicity strategy for the two posted comparisons.

These response results do not, by themselves, answer questions about durability, progression-free survival, overall survival, symptoms, or quality of life. Some related outcomes are listed in the registry, but their numerical results are not included in the registry record described for this analysis. They should not be inferred from the response endpoint. Likewise, the safety counts are descriptive and do not independently establish causation or comparative safety.

The record identifies the study as a Phase II randomized, open-label, multicenter, global study and lists enrollment of 267. That context is important: a Phase II comparison can provide estimates for further evaluation, but the posted summaries remain limited by the available sample sizes and response counts. The findings should be interpreted as the specific registry results for this population, endpoint, and analysis, not as medical advice or a conclusion about treatment choice.

16. Why This Trial Matters Statistically

This trial is statistically informative because it places three regimens in a randomized, parallel-group comparison and reports an objective response endpoint assessed by blinded independent central review. The use of a common endpoint definition and central review provides a defined basis for describing response across the arms, even though the study was open-label.

The results also illustrate why estimates should be read together with their precision. The arm-specific response proportions differ, but the observed response counts are limited and the confidence intervals are wide. Relative measures amplify that uncertainty: the posted risk-ratio and odds-ratio intervals span broad ranges. Reading only the point estimates would understate how much uncertainty remains about the comparisons.

Another important statistical feature is the distinction between a randomized study population and an endpoint-specific evaluable population. The primary analysis population required treatment exposure, a baseline assessment, and measurable disease. That definition helps identify who contributed to the response analysis, while also raising questions about participants who did not meet those criteria or lacked follow-up assessments. The registry record does not report the details needed to resolve those questions.

Finally, the posted pairwise p-values are unadjusted, and the record does not report the original multiplicity strategy. Together, the limited event counts, broad intervals, and incomplete reporting of some analytic details counsel a cautious reading. The central statistical contribution is not a simple ranking, but a set of response estimates and uncertainty measures that show what these data can and cannot determine.

17. 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.
• Participant counts were reconstructed from posted percentages and arm sizes, so small rounding differences are possible.

The primary endpoint analysis uses an evaluable population rather than a stated all-randomized analysis. The registry definition requires at least one dose, a baseline tumor assessment, and measurable disease. The record does not report the number or characteristics of randomized participants excluded from this endpoint population, nor the handling of missing or unevaluable follow-up assessments. This limits assessment of possible selection effects.

The number of observed responses is small, especially in the tremelimumab group, where the posted response count is 1. Estimates based on few events can be unstable, and the wide relative-effect confidence intervals reflect considerable uncertainty. The study’s Phase II status and the posted arm sizes also limit precision for these comparisons.

The trial was open-label, although the primary endpoint description specifies blinded independent central review. The registry record does not report details of the central-review process, adjudication procedures, assessment timing beyond the stated 6-month endpoint, or whether any sensitivity analyses tested alternative outcome handling. It also does not report the original sample-size rationale, detailed statistical analysis plan, interim analyses, or multiplicity strategy.

The available primary results focus on response rate. Numerical results for the listed secondary outcomes are not included in the record details used for this interpretation. Safety information is also limited in the listed summary: event counts and selected serious-event terms do not provide full timing, duration, severity, attribution, or participant-level overlap. These limitations restrict the scope of conclusions that can be drawn from the posted summaries.

18. Related Tutorials

19. Related Statistical Calculators

20. Sources

• ClinicalTrials.gov: Phase II Study of MEDI4736, Tremelimumab, and MEDI4736 in Combination w/ Tremelimumab Squamous Cell Carcinoma of the Head and Neck — NCT02319044.
• Registry record last updated 2020-09-29; status last verified 2020-09.
• Results first posted 2018-06-19.
• 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.

21. Record Summary

The registry describes a completed randomized, open-label, parallel-group Phase II study of MEDI4736, tremelimumab, and their combination in recurrent or metastatic squamous cell carcinoma of the head and neck. Enrollment was 267. The primary endpoint was objective response rate at 6 months, assessed by blinded independent central review according to RECIST v1.1, with unconfirmed responses excluded.

Posted response rates were 7.7% for the combination, 9.2% for MEDI4736, and 1.6% for tremelimumab. The comparisons include confidence intervals and unadjusted p-values, with considerable uncertainty around several estimates. The primary analysis population was evaluable participants meeting the stated treatment and tumor-assessment criteria. Details of the original sample-size justification, missing-data strategy, and multiplicity approach are not reported in the registry record.