This page separates reported trial results from statistical interpretation. Numerical results are restricted to the ClinicalTrials.gov record. Where the ClinicalTrials.gov record does not provide an effect estimate or confidence interval, none is inferred or reconstructed.
1. Trial at a Glance
PREVAC was a randomized, parallel-group, quadruple-masked phase 2 prevention trial in Ebola Virus Disease. The trial enrolled 4789 participants and evaluated five intervention arms involving Ad26.ZEBOV, MVA-BN-Filo, rVSVΔG-ZEBOV-GP, placebo, and an rVSV boost strategy.
| Feature | PREVAC |
|---|---|
| Trial name | Partnership for Research on Ebola VACcinations |
| NCT ID | NCT02876328 |
| Condition | Ebola Virus Disease |
| Phase | Phase 2 |
| Status | COMPLETED |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | QUADRUPLE |
| Primary purpose | PREVENTION |
| Enrollment | 4789 |
| Primary endpoint type | Binary |
| Formal statistical analysis | Cochran-Mantel-Haenszel test |
| Hypothesis type | Superiority |
| Lead sponsor | National Institute of Allergy and Infectious Diseases (NIAID) |
| Sponsor type | NIH |
2. Clinical Question
The central statistical question was whether the randomized Ebola vaccination strategies differed in the percentage of participants achieving the prespecified GP-EBOV antibody-response definition through Month 12.
Population
Participants enrolled in a phase 2 prevention trial for Ebola Virus Disease. The registered analysis separately evaluates adults and children.
Interventions
Ad26.ZEBOV, MVA-BN-Filo, rVSVΔG-ZEBOV-GP, placebo, and an rVSV boost strategy were represented in the five-arm trial design.
Comparator
The primary statistical analysis compared protocol Version 4.0 Ad26/MVA, protocol Version 4.0 rVSV, protocol Version 4.0 rVSV Boost, and protocol Version 4.0 pooled placebo.
Primary question
Does the distribution of the binary antibody-response endpoint differ among the randomized vaccine-strategy groups and pooled placebo when the comparison is stratified by site?
3. Trial Design
The design is statistically important because randomization creates the basis for comparing groups without assigning participants according to observed baseline characteristics. Quadruple masking adds another layer of protection against differential expectations or assessment behavior. The parallel design means participants remain associated with their assigned strategy rather than being intentionally exposed to each intervention in sequence.
Intervention structure
Ad26/MVA and rVSV strategies
- Ad26.ZEBOV (biological)
- MVA-BN-Filo (biological)
- rVSVΔG-ZEBOV-GP (biological)
- rVSV boost (biological)
Placebo
- Placebo (biological)
- Pooled placebo was the comparator group in the posted primary statistical analysis
- The analysis was conducted separately in adults and children
4. Trial Timeline
Trial start
The registered trial start date was March 27, 2017.
Primary completion
The registered primary completion date was December 24, 2019.
Registry status
The trial is listed as COMPLETED, with results posted in the ClinicalTrials.gov record.
5. Primary Endpoint
| Endpoint | Registered definition | Time frame | Type |
|---|---|---|---|
| Percentage of Participants With Ebola Virus Glycoprotein (GP-EBOV) Antibody Response | Antibody responder at 12 months is defined as a participant who experiences a 4-fold increase in antibody level from baseline and for whom the antibody level at 12 months is greater than or equal to 200 EU/mL. | Measured through Month 12 | Binary |
This definition combines two conditions. A participant must have at least a four-fold increase from baseline and an antibody level at Month 12 of at least 200 EU/mL. The endpoint is therefore not simply a continuous antibody concentration and is not equivalent to asking whether the average antibody level increased.
The primary analysis therefore operates on a participant-level binary outcome. The reported measure is the percentage of participants meeting the prespecified responder definition.
6. Statistical Methodology
Cochran-Mantel-Haenszel test
The posted primary analysis used the Cochran-Mantel-Haenszel test, a categorical-data method designed to compare groups while accounting for a stratification variable. In PREVAC, the analysis was stratified by site.
This is an important distinction from an unadjusted comparison of pooled percentages. If response rates vary across sites, a simple pooled comparison can mix the treatment comparison with differences in the composition of participants across sites. A stratified analysis instead evaluates the treatment-group relationship within the site strata and then combines that information using the Cochran-Mantel-Haenszel framework.
The vertical bar represents conditioning on the stratification variable. The goal is not to ignore site differences, but to account for them while evaluating the treatment-response association.
Why a categorical method fits this endpoint
The primary endpoint is explicitly binary. Each participant is classified according to whether the Month 12 antibody-response definition was achieved. This makes a categorical-data framework natural: the analysis compares responder versus non-responder classifications across the randomized groups rather than treating the underlying antibody measurement as the primary outcome.
Stratified analysis
The analysis notes state that analyses are stratified by site and apply to both adults and children. Stratification is especially relevant when the study is conducted across multiple sites because site can affect the observed response distribution for reasons unrelated to the randomized intervention.
Superiority hypothesis
The registered hypothesis type is superiority. Conceptually, a superiority analysis asks whether the randomized groups differ in the outcome rather than merely whether one strategy remains within a prespecified non-inferiority margin. The ClinicalTrials.gov record does not provide a non-inferiority margin, so none is applicable to the reported primary analysis.
7. Analysis Population
The posted primary analysis states that the analysis population includes participants from protocol Version 4.0. Adults and children were analyzed separately. The groups compared in the formal analysis were:
| Comparison group | Role in primary analysis |
|---|---|
| Protocol Version 4.0 Ad26/MVA | Active vaccination strategy |
| Protocol Version 4.0 rVSV | Active vaccination strategy |
| Protocol Version 4.0 rVSV Boost | Active vaccination strategy |
| Protocol Version 4.0 pooled placebo | Comparator |
8. Primary Results
The registry reports a formal statistical analysis for the primary antibody-response endpoint. The reported method was the Cochran-Mantel-Haenszel test, with analyses stratified by site and conducted separately for adults and children.
Primary statistical test
Cochran-Mantel-Haenszel test · Superiority hypothesis · Stratified by site
Comparison: protocol Version 4.0 Ad26/MVA vs protocol Version 4.0 rVSV vs protocol Version 4.0 rVSV Boost vs protocol Version 4.0 pooled placebo
What the reported P-value establishes
The reported P < .001 indicates that the formal Cochran-Mantel-Haenszel test produced a result inconsistent with the null hypothesis at the stated P-value threshold. In the context of the registered superiority analysis, the test provides statistical evidence of a difference among the compared groups after accounting for site strata.
What it means: the primary categorical comparison produced a very small P-value under the specified stratified analysis. The result indicates evidence of a difference in the binary antibody-response outcome across the analyzed groups.
What it does not mean: P < .001 does not tell us the size of the treatment effect, the percentage of responders in any arm, or how much better one intervention was than another. Those questions require effect estimates and their uncertainty intervals.
Why the confidence interval matters: no confidence interval is reported in the ClinicalTrials.gov record. Without an effect estimate and interval, the magnitude and precision of the group differences cannot be quantified from the posted statistical-analysis entry alone.
Why the P-value is not an effect size: the P-value measures evidence against the null hypothesis under the specified statistical model and sampling framework. It is affected by sample size and does not measure the magnitude or practical importance of a difference.
Key caution: this was a stratified analysis, so the P-value should be interpreted in the context of the site-adjusted comparison and the stated analysis population rather than as an unadjusted comparison of all randomized participants.
9. Understanding the Cochran-Mantel-Haenszel Result
The most important statistical feature of PREVAC is that the primary endpoint is binary while the trial comparison is not simply a two-group unstratified test. The analysis compares four protocol Version 4.0 groups and incorporates site as a stratification variable.
Binary outcome
Each participant is classified as an antibody responder or not according to the registered Month 12 definition.
Multiple groups
The posted formal comparison contains Ad26/MVA, rVSV, rVSV Boost, and pooled placebo groups.
Site stratification
The analysis is stratified by site rather than treating every participant as belonging to one homogeneous population.
Separate age groups
The analysis applies to both adults and children, which are analyzed separately according to the registry analysis description.
Statistically, this structure is useful because the same overall P-value can arise from very different underlying response patterns. Without the arm-specific percentages or an effect estimate, the ClinicalTrials.gov record cannot establish the magnitude of the difference between any particular vaccine strategy and placebo.
10. Statistical Methods Explained
Why was a Cochran-Mantel-Haenszel test used?
The primary endpoint is categorical: each participant either meets the antibody-response definition or does not. The Cochran-Mantel-Haenszel framework is designed for categorical comparisons across strata, making it suitable for a binary outcome when the analysis also needs to account for site.
Why stratify by site?
Participants enrolled at different sites can have different response patterns for reasons unrelated to treatment assignment. Stratification allows the analysis to account for the site structure rather than assuming that the pooled participant population is statistically homogeneous with respect to the outcome.
What does P < .001 mean here?
It means the reported CMH test generated a P-value below .001 under the specified null hypothesis and analysis framework. It is evidence against the null hypothesis of no association in the stratified comparison. It is not a percentage improvement and it is not a probability that a vaccine is effective.
Why can't the P-value tell us which arm was best?
The ClinicalTrials.gov record is a formal comparison across the listed groups, but it does not provide the arm-specific response percentages or pairwise effect estimates. A global categorical test can establish evidence of a group difference without, by itself, quantifying every pairwise contrast.
Why are adults and children analyzed separately?
The registry analysis explicitly states that adults and children are analyzed separately. Separate analyses can preserve interpretability when the participant populations may have different underlying antibody-response distributions. The ClinicalTrials.gov record does not provide the resulting arm-specific estimates for either population.
What would an effect estimate add?
An effect estimate would quantify the magnitude of a comparison rather than simply indicate whether the statistical test detected evidence of a difference. For a binary endpoint, useful measures can include risk differences, risk ratios, odds ratios, or adjusted versions of those measures. None is reported in the ClinicalTrials.gov record, so none is reconstructed.
Why does randomization matter for interpretation?
Randomization determines treatment assignment before the outcome is observed. When implemented appropriately, it supports causal comparison by balancing measured and unmeasured prognostic factors in expectation. The statistical analysis then uses the randomized comparison to evaluate differences in the prespecified endpoint.
11. Randomization and Blinding
PREVAC is registered as randomized and quadruple masked. These design features are central to interpreting the statistical comparison.
| Design feature | Statistical relevance |
|---|---|
| Randomized allocation | Creates the primary basis for comparing outcome distributions across intervention groups. |
| Parallel design | Participants remain associated with their assigned strategy rather than receiving sequential randomized interventions in a crossover framework. |
| Quadruple masking | Reduces the opportunity for knowledge of assignment to influence participant behavior, treatment administration, assessment, or other trial processes. |
| Prevention purpose | The primary endpoint evaluates an immune-response outcome rather than a time-to-event disease outcome. |
Randomization and masking do not eliminate every possible source of bias. They instead address important mechanisms through which treatment assignment or knowledge of assignment could influence trial conduct and outcome assessment.
12. Endpoint Interpretation
What exactly is an antibody responder?
The registered definition requires both a 4-fold increase in antibody level from baseline and an antibody level at 12 months of greater than or equal to 200 EU/mL. A participant who satisfies only one of these conditions does not meet the stated responder definition.
This construction makes the endpoint a composite rule for binary classification. It is different from reporting a mean antibody concentration or a mean fold-change as a continuous endpoint.
Why the endpoint is measured through Month 12
The registered time frame is "Measured through Month 12." The analysis therefore focuses on whether participants meet the prespecified antibody-response definition by the stated follow-up point. It should not be silently reinterpreted as a lifetime protection endpoint or as direct evidence of clinical disease prevention.
13. Secondary and Other Posted Outcomes
the ClinicalTrials.gov record states that 3 outcome measures were posted. The detailed trial data provided for this page supplies a formal statistical analysis for the primary antibody-response endpoint but does not provide the numerical results or formal statistical analyses for the other posted outcome measures.
14. Safety Results
the ClinicalTrials.gov record includes serious adverse events by arm. The registry extract provided for this page contains the following complete affected/at-risk entries:
| Arm / group | Serious adverse events |
|---|---|
| Ad26.ZEBOV (rHAd26) Vaccine + MVA-BN-Fil | 14/396 |
| Pooled Placebo (0.5 mL or 1 mL) - Adults | 5/412 |
| rVSVΔG-ZEBOV-GP (rVSV) Vaccine + Placebo | 6/395 |
| rVSVΔG-ZEBOV-GP (rVSV) Vaccine + rVSV Bo | 1/197 |
| Ad26.ZEBOV (rHAd26) Vaccine + MVA-BN-Fil | 5/403 |
| Pooled Placebo (0.5 mL or 1 mL) - Children | 8/389 |
The safety counts should also be kept conceptually separate from the primary antibody-response analysis. A serious-adverse-event count describes an adverse-event outcome, whereas the primary endpoint classifies antibody response. Neither endpoint substitutes for the other.
15. Multiplicity and the Primary Hypothesis
The ClinicalTrials.gov record identifies one registered primary endpoint and one posted primary-endpoint statistical analysis. The hypothesis type is superiority.
| Component | Registered / posted information |
|---|---|
| Primary endpoints | 1 |
| Primary endpoint type | Binary |
| Formal primary analysis | Cochran-Mantel-Haenszel test |
| Hypothesis type | Superiority |
| Stratification | Site |
| Analysis populations | Adults and children analyzed separately |
Because the ClinicalTrials.gov record identifies a single registered primary endpoint, there is no basis here for constructing a broader multiplicity framework involving additional primary endpoints. Likewise, the ClinicalTrials.gov record does not identify an interim analysis, alpha-spending procedure, or a multiplicity-adjusted hierarchy beyond the stated primary analysis.
What is documented
A superiority hypothesis, one registered primary endpoint, a CMH test, and site-stratified analysis.
What is not reported
No interim-analysis procedure, alpha-spending rule, or additional multiplicity-adjustment procedure is provided in the ClinicalTrials.gov record.
16. Missing Data and Analysis Interpretation
Missing antibody measurements matter because the primary outcome requires both a baseline value and a Month 12 value. Without the baseline measurement, the four-fold increase cannot be evaluated. Without the Month 12 measurement, neither the fold-change criterion nor the Month 12 threshold can be fully assessed.
The distinction between the 4789 enrolled participants and the denominator used for the primary antibody-response analysis is therefore important. Enrollment is a trial-level design quantity; the analysis denominator is an endpoint-specific quantity. They need not be identical when eligibility for the endpoint requires observed measurements at particular time points.
17. What the Primary P-value Does — and Does Not — Mean
The reported P < .001 means the observed categorical pattern was statistically inconsistent with the null hypothesis under the specified Cochran-Mantel-Haenszel framework. The analysis accounted for site strata.
A small P-value does not establish whether the difference between groups is small, moderate, or large. That requires the actual responder percentages and an effect measure with an uncertainty interval.
P < .001 is not the probability that the null hypothesis is true or false. It is a probability calculated under the statistical framework represented by the null hypothesis and the assumed sampling process.
The registry-reported statistical-analysis entry does not report a confidence interval. Consequently, the precision of the underlying group difference cannot be quantified from the ClinicalTrials.gov record beyond the reported P-value.
18. Limitations
- Effect-size reporting: the registry-reported primary statistical analysis reports P < .001 but does not provide arm-specific antibody-response percentages, a treatment-effect estimate, or a confidence interval.
- Multiple intervention groups: the posted formal comparison contains four analysis groups, so a global test does not by itself quantify every pairwise comparison.
- Age-group analyses: adults and children are analyzed separately, meaning the overall enrollment number should not be treated as the denominator for either age-specific analysis.
- Site stratification: the primary result is a stratified categorical analysis. It should not be interpreted as an unadjusted pooled comparison.
- Endpoint specificity: the antibody-response definition is a prespecified immunologic binary endpoint. It should not automatically be interpreted as a direct measure of clinical protection from Ebola Virus Disease.
- Unreported procedures: the ClinicalTrials.gov record does not document an interim-analysis procedure, alpha-spending rule, Bayesian method, or non-inferiority margin.
19. Why This Trial Matters Statistically
PREVAC is a useful teaching example because it demonstrates how the statistical method should follow the structure of the endpoint and design rather than defaulting to a generic two-group comparison.
| Concept | How it appears in PREVAC |
|---|---|
| Randomization | The trial is registered as RANDOMIZED. |
| Blinding | The trial is registered as QUADRUPLE masked. |
| Parallel design | The registered design model is PARALLEL. |
| Binary endpoint | The primary endpoint classifies participants according to a prespecified antibody-response definition. |
| Cochran-Mantel-Haenszel test | The posted primary statistical analysis uses the CMH test. |
| Stratified analysis | The primary analysis is stratified by site. |
| Superiority testing | The registered hypothesis type is superiority. |
| Separate populations | Adults and children are analyzed separately. |
| P-value interpretation | The primary analysis reports P < .001 without an accompanying effect estimate in the ClinicalTrials.gov record. |
| Missing-data considerations | The endpoint requires both baseline and Month 12 antibody results. |
The central statistical lesson is that a statistically significant result is only one component of an interpretable clinical-trial result. For a binary endpoint, the reader ultimately wants to know the response proportion in each group, the contrast between groups, and the uncertainty around that contrast. When those quantities are absent from the ClinicalTrials.gov record, the correct approach is to preserve the limitation rather than reconstruct them from assumptions.
20. A Deeper Look at Stratified Categorical Analysis
Suppose a trial has several sites and each site contains participants from more than one randomized group. The response rate can differ from site to site because of participant characteristics, study procedures, laboratory factors, or other site-specific features. A stratified analysis asks whether the treatment-response relationship persists after considering those site-specific strata.
The second formulation reflects the structure of the reported primary analysis. The CMH method combines information across strata rather than simply collapsing all observations into one contingency table.
This distinction becomes especially important when the distribution of participants across sites is uneven. A pooled analysis can obscure the within-site treatment relationship. A stratified analysis instead preserves the site's role in the inferential framework.
21. Why the Four-Group Comparison Requires Careful Interpretation
The posted primary analysis compares four groups: Ad26/MVA, rVSV, rVSV Boost, and pooled placebo. A global P-value below .001 tells the reader that the overall categorical distribution differs across the compared groups under the CMH analysis.
It does not, by itself, answer every possible pairwise question. For example, the global result alone does not supply the numerical difference between Ad26/MVA and placebo, the difference between rVSV and placebo, or the difference between the two active strategies. Those questions require corresponding effect estimates or pairwise analyses.
Global question
Do the compared groups differ in the binary response outcome after accounting for site?
Pairwise question
How large is the difference between two particular strategies, and what is the uncertainty around that difference?
Keeping these questions separate prevents a common statistical mistake: treating evidence of any group difference as if it were a complete quantitative description of every treatment contrast.
22. Planned Interpretation of the Endpoint
The registered antibody-response definition has two clinically and statistically distinct components. The first is relative to the participant's own baseline, while the second imposes an absolute antibody-level threshold at Month 12.
| Component | Question being asked |
|---|---|
| 4-fold increase from baseline | Did the participant's antibody level increase substantially relative to their own starting level? |
| Month 12 level ≥ 200 EU/mL | Did the participant reach the registered absolute antibody threshold at Month 12? |
| Combined responder definition | Did the participant satisfy both criteria? |
This construction avoids reducing the endpoint to either a relative change alone or an absolute level alone. It also means that baseline antibody status can matter to whether a participant satisfies the four-fold component, which is one reason the exact endpoint definition must accompany the reported percentage.
23. No Non-Inferiority Margin or Bayesian Analysis
The ClinicalTrials.gov record identifies the hypothesis type as superiority. It does not provide a non-inferiority margin. The primary analysis is a frequentist Cochran-Mantel-Haenszel test, and the ClinicalTrials.gov record does not identify a Bayesian model or posterior probability analysis.
| Design topic | PREVAC ClinicalTrials.gov record |
|---|---|
| Superiority | Yes |
| Non-inferiority margin | Not reported |
| Bayesian method | Not reported |
| Interim analysis | Not reported |
| Alpha-spending method | Not reported |
| Site stratification | Yes |
These distinctions matter because the interpretation of a P-value depends on the hypothesis being tested. A superiority analysis asks whether the groups differ; a non-inferiority analysis asks whether an intervention is not unacceptably worse than a comparator by a prespecified margin. PREVAC is documented here as a superiority analysis.
24. Overall Statistical Interpretation
PREVAC provides a clear example of the relationship between trial design, endpoint construction, and analysis method. The trial randomized participants in a five-arm parallel design and used quadruple masking. Its primary endpoint was binary and required both a four-fold antibody increase from baseline and a Month 12 antibody level of at least 200 EU/mL. The formal primary analysis used a Cochran-Mantel-Haenszel test stratified by site, with adults and children analyzed separately.
The ClinicalTrials.gov record was P < .001. That result provides statistical evidence of a difference in the antibody-response outcome among the analyzed groups under the specified stratified framework. However, the ClinicalTrials.gov record does not contain the responder percentages, an effect estimate, or a confidence interval. Consequently, the magnitude and precision of the treatment differences cannot be established from the registry-reported numerical result alone.
This distinction is fundamental in clinical-trial statistics. A hypothesis test addresses whether the observed data are compatible with a null hypothesis. An effect estimate describes how large the observed difference is. A confidence interval describes uncertainty around that estimate. A complete statistical interpretation benefits from all three.
25. Related Tutorials
Learn more about the methods used in this trial:
26. Related Calculators
27. Sources
- ClinicalTrials.gov: PREVAC, NCT02876328.
- PubMed: PMID 33485369.
- PubMed: PMID 36516078.
- PubMed: PMID 42326817.
- PubMed: PMID 41616038.
- PubMed: PMID 39559990.
Continue through the Clinical Biostats statistical pathway
Explore the underlying clinical-trial and categorical-data methods through focused tutorials and statistical calculators.
28. Record Summary
PREVAC is a randomized, parallel, quadruple-masked phase 2 prevention trial with 4789 participants and five intervention arms. Its registered primary endpoint is a binary GP-EBOV antibody-response measure assessed through Month 12, with responder status requiring both a 4-fold increase from baseline and a Month 12 antibody level of at least 200 EU/mL.
The posted primary statistical analysis used a Cochran-Mantel-Haenszel test, with the comparison stratified by site and adults and children analyzed separately. The analysis compared protocol Version 4.0 Ad26/MVA, rVSV, rVSV Boost, and pooled placebo groups under a superiority hypothesis and reported P < .001.
The most important interpretive limitation is that the ClinicalTrials.gov record does not report the arm-specific antibody-response percentages, an effect estimate, or a confidence interval. The P-value therefore establishes statistical evidence of a group difference under the specified analysis, but it cannot by itself describe the magnitude or precision of that difference. This is precisely why clinical-trial interpretation should distinguish hypothesis testing from effect estimation.