This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical estimates on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
Ad26.COV2.S was a randomized, double-blind, parallel phase 3 prevention trial in healthy adults. The study evaluated Ad26.COV2.S in combination with seasonal influenza vaccination arrangements, with primary immunogenicity comparisons between Groups 1 and 2.
| Feature | Ad26.COV2.S trial |
|---|---|
| Brief title | A Study of Ad26.COV2.S and Influenza Vaccines in Healthy Adults |
| Phase | Phase 3 |
| Status | Completed |
| Condition | COVID-19 Prevention |
| Design | Randomized, double-blind, parallel |
| Primary purpose | Prevention |
| Enrollment | 861 |
| Arms | 4 |
| Lead sponsor | Janssen Vaccines & Prevention B.V. |
| Study dates | Start: 2021-11-02; Primary completion: 2022-06-17 |
| Registered primary endpoints | 2 |
| Outcome measures posted | 14 |
| Statistical analyses posted | 5 |
2. Clinical Question
The statistical question centers on whether administration of Ad26.COV2.S alongside the seasonal quadrivalent influenza vaccination arrangement in Group 1 produces influenza antibody responses that meet the prespecified non-inferiority criterion relative to the corresponding Group 2 arrangement, while also evaluating the SARS-CoV-2 antibody response after Ad26.COV2.S administration.
Population
Healthy adults enrolled in a phase 3 prevention study.
Intervention
Ad26.COV2.S, used within the study's four randomized group configurations.
Comparator
Placebo-containing group configurations, with seasonal quadrivalent influenza vaccination administered according to the registered study design.
Primary question
Do the prespecified influenza and SARS-CoV-2 immunogenicity comparisons satisfy the registry's non-inferiority framework?
3. Trial Design
The trial used randomized allocation, a parallel design, and double masking. Four groups were represented in the registry. The primary analyses in the ClinicalTrials.gov record focus on Groups 1 and 2.
Ad26.COV2.S + quadrivalent standard-dose influenza vaccine
- Ad26.COV2.S
- Quadrivalent (Q) standard-dose (SD) influenza vaccine
- Placebo
Placebo + quadrivalent standard-dose influenza vaccine
- Placebo
- Quadrivalent (Q) standard-dose (SD) influenza vaccine
- Ad26.COV2.S
Ad26.COV2.S + quadrivalent high-dose influenza vaccine
- Ad26.COV2.S
- Quadrivalent (Q) high-dose (HD) influenza vaccine
Placebo + quadrivalent high-dose influenza vaccine
- Placebo
- Quadrivalent (Q) high-dose (HD) influenza vaccine
4. Endpoints
| Registered primary endpoint | Time frame | Analysis reported |
|---|---|---|
| Groups 1 and 2: Geometric Mean Titers (GMTs) of Hemagglutination Inhibition (HI) Antibodies Against Each of the Four Influenza Vaccine Strains 28 Days After the Administration of a Seasonal Quadrivalent Standard-dose Influenza Vaccine | 28 days after vaccination with seasonal quadrivalent standard-dose influenza vaccine (Day 29) | ANOVA; geometric mean ratio; non-inferiority |
| Groups 1 and 2: Geometric Mean Concentrations (GMCs) of Antibodies Measured by Spiked-Enzyme-linked Immunosorbent Assay (S-ELISA) 28 Days After Administration of Ad26.COV2.S Vaccine | 28 days after vaccination with Ad26.COV2.S vaccine (Group 1: Day 29, Group 2: Day 57) | ANOVA; geometric mean ratio; non-inferiority |
The first registered endpoint covers four influenza strains: A/Victoria (H1N1), A/Cambodia (H3N2), B/Victoria (B/Victoria), and B/Phuket (B/Yamagata). The registry data describe the primary outcome as an immunogenicity endpoint and the posted statistical analyses as continuous outcomes analyzed using ANOVA.
5. Statistical Methodology
ANOVA
The registry reports analysis of variance (ANOVA) as the method for all five posted primary analyses. In these analyses, confidence intervals were constructed around the model-based difference and then back-transformed to produce geometric mean ratios.
For these immunogenicity analyses, the reported comparison was expressed on a geometric scale. The registry-reported analysis notes state that confidence intervals around the relevant ANOVA difference were back-transformed by exponentiation to obtain the geometric mean ratio.
Geometric mean ratio
The effect measure reported for all five analyses is a Geometric Mean Ratio (GMR). The registry's analysis notes define the direction as GMTControl / GMTCoAd.
A GMR above 1 therefore indicates that the geometric mean in the control group was higher than that in the co-administration group under the direction specified by the registry. A GMR near 1 indicates similar geometric means on this ratio scale.
Back-transformation
The registry-reported analysis notes state that the ANOVA confidence interval around the difference between Group 2 and Group 1 was back-transformed by exponentiation to produce a GMR. This matters because the final ratio is not obtained by simply subtracting or averaging the original antibody measurements.
Per-protocol analysis
The influenza analyses used the PPII set, described in the registry data as including randomized participants who received Ad26.COV2.S in combination with seasonal influenza vaccine for the co-administration group and corresponding participants in the comparator configuration. The SARS-CoV-2 immunogenicity analysis used the per protocol SARS-CoV-2 immunogenicity set (PPSI).
Per-protocol analyses can be useful for immunogenicity and non-inferiority questions because they focus on participants who adhered sufficiently to the protocol-defined treatment and assessment requirements. They also have a limitation: removing participants after randomization can weaken the protection that randomization provides against selection differences. Consequently, the analysis population itself is part of the interpretation.
Non-inferiority framework
Each of the five posted analyses is labeled Non-inferiority. The registry specifies a common criterion: the upper bound of the two-sided 95% confidence interval for the GMR had to be below 1.5.
Prespecified non-inferiority criterion
The registry identifies this as the criterion for non-inferiority for the reported GMR analyses.
6. Results
The registry contains five posted primary statistical analyses. Four correspond to the four influenza vaccine strains included in the HI-antibody endpoint, and one corresponds to the S-ELISA antibody-concentration endpoint after Ad26.COV2.S vaccination.
Influenza HI Antibody Response
All four strain-specific analyses used ANOVA in the PPII population and reported a two-sided 95% confidence interval for the GMR. The registry defines the GMR direction as GMTControl/GMTCoAd.
| Influenza strain | GMR | 95% CI | NI criterion |
|---|---|---|---|
| A/Victoria (H1N1) | 1.28 | 1.09–1.53 | Upper CI must be below 1.5 |
| A/Cambodia (H3N2) | 1.23 | 1.05–1.45 | Upper CI must be below 1.5 |
| B/Victoria (B/Victoria) | 0.99 | 0.84–1.19 | Upper CI must be below 1.5 |
| B/Phuket (B/Yamagata) | 1.03 | 0.88–1.21 | Upper CI must be below 1.5 |
A/Victoria (H1N1) geometric mean ratio
95% CI: 1.09–1.53 · Two-sided CI
Registry direction: GMTControl/GMTCoAd
The point estimate of 1.28 means that the geometric mean HI antibody titer in the control configuration was estimated to be 1.28 times the geometric mean in the co-administration configuration, using the ratio direction specified by the registry.
The estimate does not mean that individual participants had antibody titers exactly 1.28 times one another, nor does it describe the probability that a particular participant would respond. It is a population-level ratio of geometric means.
The two-sided 95% CI of 1.09–1.53 describes uncertainty around the estimated ratio under the stated analysis. It is not a range containing 95% of individual antibody titers.
The ClinicalTrials.gov record contains no p-value for this analysis, so none is reported here. More importantly, the non-inferiority criterion is defined directly in terms of the upper confidence-limit boundary: 1.5. The reported upper confidence limit is 1.53, which is above that boundary. That is the relevant statistical observation from the ClinicalTrials.gov record; it should not be replaced by an inferred p-value or by a different hypothesis-testing rule.
A/Cambodia (H3N2) geometric mean ratio
95% CI: 1.05–1.45 · Two-sided CI
Registry direction: GMTControl/GMTCoAd
The point estimate of 1.23 means that the geometric mean HI antibody titer in the control configuration was estimated to be 1.23 times that in the co-administration configuration under the registry's stated ratio direction.
The GMR is a summary measure, not an individual-level effect. It does not indicate that every participant experienced a 23% difference, and it does not provide a probability of an individual response.
The 95% CI of 1.05–1.45 gives the uncertainty around the estimated geometric mean ratio. The interval remains below the registry's non-inferiority boundary of 1.5.
No p-value is in the ClinicalTrials.gov record. A p-value would answer a question about incompatibility with a specified null hypothesis; it would not replace the GMR as the effect-size measure or the confidence interval as the stated non-inferiority criterion.
B/Victoria geometric mean ratio
95% CI: 0.84–1.19 · Two-sided CI
Registry direction: GMTControl/GMTCoAd
The estimate of 0.99 is very close to 1. On the registry's GMR scale, this indicates that the estimated geometric mean HI antibody titers were close between the control and co-administration configurations, with the control numerator slightly below the co-administration denominator.
This does not mean that the two groups had identical antibody titers at the individual level. It is a ratio of geometric means and therefore summarizes the groups rather than individual observations.
The 95% CI of 0.84–1.19 describes the precision of the estimated ratio. Its upper limit is below the non-inferiority boundary of 1.5.
No p-value is provided. The non-inferiority interpretation is based on the prespecified confidence-interval rule rather than on inventing or reconstructing a significance test that is not reported in the ClinicalTrials.gov record.
B/Phuket (B/Yamagata) geometric mean ratio
95% CI: 0.88–1.21 · Two-sided CI
Registry direction: GMTControl/GMTCoAd
The point estimate of 1.03 is close to 1, indicating a small estimated difference in geometric mean HI antibody titers on the registry's control-to-co-administration ratio scale.
The estimate is not a statement about the magnitude of effect for every participant. Individual antibody responses can vary substantially even when the ratio of group-level geometric means is close to 1.
The two-sided 95% CI of 0.88–1.21 quantifies uncertainty around the GMR. Its upper bound is below the specified non-inferiority boundary of 1.5.
No p-value is reported in the registry analysis. A p-value is not a measure of effect size or clinical importance; here, the prespecified non-inferiority rule and the estimated GMR provide the principal statistical framework.
SARS-CoV-2 S-ELISA Antibody Concentration
Geometric mean concentration ratio
95% CI: 0.97–1.26 · Two-sided CI
Outcome unit: ELISA Unit per milliliter (EU/mL)
The estimated GMR of 1.11 means that the geometric mean antibody concentration in the control configuration was estimated to be 1.11 times the geometric mean in the co-administration configuration, using the registry's stated control-to-co-administration direction.
The estimate does not mean that individual antibody concentrations were 11% higher in the control group, because a ratio of geometric means is a group-level summary rather than an individual-level treatment effect.
The 95% CI of 0.97–1.26 describes uncertainty around the estimated ratio. The upper confidence limit is below the prespecified non-inferiority boundary of 1.5.
The ClinicalTrials.gov record does not report a p-value for this analysis. The statistical evidence should therefore be described using the reported GMR, its confidence interval, the analysis population, and the stated non-inferiority criterion rather than adding an unreported hypothesis-test result.
7. Results Summary
| Primary analysis | Estimate | 95% CI | Method | NI boundary |
|---|---|---|---|---|
| A/Victoria (H1N1) HI GMT | 1.28 | 1.09–1.53 | ANOVA | Upper CI < 1.5 |
| A/Cambodia (H3N2) HI GMT | 1.23 | 1.05–1.45 | ANOVA | Upper CI < 1.5 |
| B/Victoria HI GMT | 0.99 | 0.84–1.19 | ANOVA | Upper CI < 1.5 |
| B/Phuket (B/Yamagata) HI GMT | 1.03 | 0.88–1.21 | ANOVA | Upper CI < 1.5 |
| S-ELISA antibody GMC | 1.11 | 0.97–1.26 | ANOVA | Upper CI < 1.5 |
Across the five posted analyses, the point estimates range from 0.99 to 1.28. Four of the five reported upper confidence limits are below the stated non-inferiority boundary of 1.5; the A/Victoria (H1N1) analysis has an upper confidence limit of 1.53.
8. Statistical Methods Explained
Why was ANOVA used?
The registry reports ANOVA for all five primary analyses. ANOVA provides a model-based framework for comparing group-level responses. Here, the resulting comparison was expressed as a geometric mean ratio after back-transformation of the model-based difference.
What does a geometric mean ratio of 1.28 mean?
Because the registry defines the ratio as GMTControl/GMTCoAd, a value of 1.28 means the estimated geometric mean in the control configuration was 1.28 times the estimated geometric mean in the co-administration configuration. It does not mean that every participant's antibody titer differed by 28%.
Why use geometric means for antibody measurements?
Antibody measurements can be summarized on a multiplicative rather than purely additive scale. A geometric mean ratio naturally expresses the relative relationship between two geometric means and is therefore the effect measure selected in these registry analyses.
Why was a non-inferiority margin of 1.5 used?
The registry explicitly states that non-inferiority was assessed by requiring the upper bound of the two-sided 95% CI for the GMR to be below 1.5. The margin is therefore part of the trial's prespecified decision framework. It is not a value that should be replaced by a conventional null value of 1.
Why can a confidence interval cross 1 and still be compatible with the non-inferiority criterion?
Non-inferiority and superiority ask different questions. For a superiority comparison of a ratio, the value 1 is often the central reference point. In this trial's stated non-inferiority framework, the relevant boundary is 1.5, so an interval that includes 1 can still have an upper limit below 1.5.
Why does the analysis population matter?
The influenza analyses use the PPII set and the SARS-CoV-2 immunogenicity analysis uses the PPSI. A per-protocol analysis restricts attention to participants meeting protocol-defined requirements. That can be informative for an immunogenicity non-inferiority question, but it also means the result should not automatically be interpreted as if it were an intention-to-treat estimate.
Why is the p-value not shown?
The ClinicalTrials.gov record provides estimates and two-sided 95% confidence intervals but do not provide p-values. A p-value should not be reconstructed from rounded estimates and confidence limits when the registry does not report one. The confidence interval and the explicitly stated non-inferiority criterion are sufficient to describe the reported analysis without inventing an additional result.
9. Confidence Intervals and Non-Inferiority
The most important statistical feature of this trial is that the decision boundary is expressed through the confidence interval rather than through a generic “P < 0.05” rule.
The reported confidence interval therefore needs to be read against 1.5. The point estimate alone is insufficient for the non-inferiority assessment because it does not quantify uncertainty.
| Analysis | Upper 95% CI | Distance from NI boundary |
|---|---|---|
| A/Victoria (H1N1) | 1.53 | Above 1.5 |
| A/Cambodia (H3N2) | 1.45 | Below 1.5 |
| B/Victoria | 1.19 | Below 1.5 |
| B/Phuket (B/Yamagata) | 1.21 | Below 1.5 |
| S-ELISA GMC | 1.26 | Below 1.5 |
This table illustrates why the confidence interval is central to a non-inferiority analysis. The first analysis has a point estimate that is not dramatically different from the other influenza results, but its upper confidence limit extends beyond the specified boundary. The distinction arises from the combination of estimated effect and statistical uncertainty.
10. Analysis Populations
| Population | Role in registry-reported analyses |
|---|---|
| PPII | Used for the four influenza HI antibody analyses. The registry describes it as the per-protocol influenza immunogenicity set. |
| PPSI | Used for the S-ELISA SARS-CoV-2 immunogenicity analysis. The registry describes it as the per protocol SARS-CoV-2 immunogenicity set. |
The use of per-protocol populations is particularly important when interpreting non-inferiority studies. Excluding protocol deviations can make treatment groups more closely reflect the intended treatment contrast, but the exclusions can also introduce selection that would not occur in a pure randomized comparison.
A statistically careful reading should therefore keep three pieces together: the randomized allocation, the analysis population actually used, and the prespecified non-inferiority rule. The five posted estimates should not be interpreted as though all participants enrolled in the trial necessarily contributed to every analysis.
11. Safety Results
The ClinicalTrials.gov record reports serious adverse events by study group. The values below are expressed as affected participants divided by participants at risk, exactly as provided in the trial data.
| Group | Study configuration | Serious adverse events | Rate among listed at-risk participants |
|---|---|---|---|
| Group 1 | Ad26.COV2.S + Quadrivalent (Q) standard-dose influenza vaccine | 9/382 | 9 of 382 |
| Group 2 | Placebo + Q standard-dose influenza vaccine | 7/384 | 7 of 384 |
| Group 3 | Ad26.COV2.S + Q high-dose influenza vaccine | 1/47 | 1 of 47 |
| Group 4 | Placebo + Q high-dose influenza vaccine | 2/46 | 2 of 46 |
The serious-adverse-event figures are descriptive counts by randomized group. The ClinicalTrials.gov record does not provide a formal statistical comparison, confidence intervals, exposure-adjusted rates, or p-values for serious adverse events, so none is inferred here.
Standard-dose groups
Group 1 had 9 serious adverse events among 382 participants at risk; Group 2 had 7 among 384.
High-dose groups
Group 3 had 1 serious adverse event among 47 participants at risk; Group 4 had 2 among 46.
12. What the GMR Does — and Does Not — Mean
A GMR is a relative group-level summary. For example, the A/Cambodia estimate of 1.23 describes the estimated ratio of the control group's geometric mean to the co-administration group's geometric mean. It is not an individual-level percentage change.
A 95% CI describes statistical uncertainty around the estimated GMR under the specified analysis. It does not mean that 95% of individual participants have antibody values inside the interval, nor does it describe biological variability among individuals.
No p-values are posted on ClinicalTrials.gov for the five posted statistical analyses. A p-value would not measure the size of the geometric mean ratio and would not substitute for the trial's stated non-inferiority boundary.
The relevant question is whether the entire two-sided 95% CI satisfies the registry's specified upper-bound requirement of less than 1.5. This is different from asking whether the confidence interval excludes 1.
13. Why the Direction of the Ratio Matters
The registry-reported analysis notes explicitly state that the GMR was constructed as GMTControl/GMTCoAd. This direction is essential when interpreting values above or below 1.
| GMR value | Interpretation under the registry's direction |
|---|---|
| Below 1 | Estimated control geometric mean is lower than the co-administration geometric mean. |
| Equal to 1 | Estimated geometric means are equal on the ratio scale. |
| Above 1 | Estimated control geometric mean is higher than the co-administration geometric mean. |
This is a useful safeguard against a common statistical-reading error: interpreting a ratio without checking which group appears in the numerator. The same numerical ratio can imply the opposite directional relationship if the numerator and denominator are reversed.
14. Back-Transformation and Geometric Means
The registry analysis notes state that the confidence interval around the ANOVA difference was back-transformed by exponentiation to obtain the GMR. Conceptually, the analysis therefore moves between an additive model scale and a multiplicative interpretation scale.
The reported GMR is consequently a multiplicative effect measure. This is why the natural reference value for equality is 1 rather than 0.
The back-transformation also explains why confidence limits are interpreted multiplicatively. A confidence interval such as 1.05–1.45 should be read as a range of plausible values for the ratio, not as an additive interval around the point estimate.
15. Multiplicity and Multiple Primary Comparisons
The registry data identify two registered primary endpoints and five posted statistical analyses. The first primary endpoint is evaluated separately for four influenza strains, while the second concerns S-ELISA antibody concentrations.
| Component | Role |
|---|---|
| Influenza HI endpoint | Primary endpoint covering four vaccine strains |
| A/Victoria (H1N1) | Separate posted primary analysis |
| A/Cambodia (H3N2) | Separate posted primary analysis |
| B/Victoria | Separate posted primary analysis |
| B/Phuket (B/Yamagata) | Separate posted primary analysis |
| S-ELISA antibody GMC endpoint | Second registered primary endpoint and separate posted primary analysis |
The ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure across the five posted analyses. Therefore, the individual confidence intervals should be reported as reported in the registry rather than assigning an unreported familywise error interpretation.
16. Blinding and Randomization
Randomization and double masking are core design features of the trial. Randomized allocation creates the framework for comparing study groups without assigning treatment based on participants' observed characteristics, while double masking is intended to reduce the influence of treatment knowledge on trial conduct and outcome assessment.
17. Trial Timeline
Study start
The registered study start date was 2021-11-02.
Enrollment
The trial enrolled 861 participants across four study groups.
Primary completion
The registered primary completion date was 2022-06-17.
Results posted
The registry contains 14 posted outcome measures and five posted statistical analyses.
18. Limitations
- Per-protocol populations: the primary immunogenicity analyses use PPII and PPSI populations rather than an all-randomized population. Interpretation therefore depends on the protocol-defined analysis sets.
- Non-inferiority boundary: the conclusion is tied to the registry's specified upper-confidence-limit criterion of 1.5. The clinical meaning of a non-inferiority margin depends on its scientific justification, which is not reported in the ClinicalTrials.gov record here.
- One analysis exceeds the boundary: the A/Victoria (H1N1) analysis has an upper 95% CI of 1.53, compared with the stated boundary of 1.5.
- Multiple analyses: five statistical analyses are posted across the two primary endpoint structures, but the ClinicalTrials.gov record does not specify a multiplicity-adjustment strategy.
- No p-values reported: p-values cannot be responsibly added from the registry-reported rounded estimates and confidence intervals.
- No subgroup results reported: the trial data do not provide subgroup estimates, so no subgroup interpretation is attempted.
- No baseline table reported: demographic or baseline clinical characteristics beyond the healthy-adult population are not included in the ClinicalTrials.gov record.
- Safety comparisons are descriptive: serious adverse-event counts are provided by arm, but no formal inferential safety comparison is reported.
- Endpoint-type metadata are inconsistent with posted analysis metadata: the profile labels the primary endpoint types as “Binary; Other / unclear,” while the posted statistical analyses identify the analyzed outcomes as continuous.
19. Why This Trial Matters Statistically
This trial is a useful teaching example because it combines randomized clinical-trial design with immunogenicity endpoints and a non-inferiority framework. Unlike a conventional superiority analysis in which an estimate is often interpreted relative to a null value of 1, the registry's non-inferiority rule uses a prespecified upper confidence boundary of 1.5.
| Concept | How it appears in this trial |
|---|---|
| Randomization | Randomized allocation across four parallel study groups. |
| Blinding | Double-masked trial. |
| ANOVA | Reported method for all five posted statistical analyses. |
| Geometric mean ratio | Primary effect measure for influenza HI titers and S-ELISA antibody concentrations. |
| Confidence intervals | Two-sided 95% CIs accompany all five posted primary estimates. |
| Non-inferiority | Upper bound of the two-sided 95% CI must be below 1.5. |
| Per-protocol analysis | PPII and PPSI populations are used for the registry-reported immunogenicity analyses. |
| Multiple endpoints | Four strain-specific influenza analyses plus one S-ELISA analysis are posted. |
| Safety by arm | Serious adverse events are reported as affected participants divided by those at risk. |
20. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The five posted analyses provide GMR estimates with two-sided 95% confidence intervals. Four upper confidence limits are below 1.5, while the A/Victoria (H1N1) upper confidence limit is 1.53. The registry defines these confidence-limit comparisons as its non-inferiority criterion.
Clinical interpretation
The statistical results describe antibody immunogenicity comparisons. They do not, by themselves, provide a direct estimate of clinical protection, individual benefit, or the probability of preventing COVID-19.
This distinction is particularly important for vaccine immunogenicity studies. A laboratory antibody endpoint is a measured biological outcome. It should not automatically be restated as a clinical outcome unless the trial data explicitly establish that relationship.
21. What the Results Do Not Establish
| Reported result | What can be said | What should not be inferred |
|---|---|---|
| GMR 1.28 for A/Victoria | The estimated control-to-co-administration geometric mean ratio was 1.28. | That individual participants had 28% higher titers, or that a clinical outcome changed by 28%. |
| 95% CI 1.05–1.45 for A/Cambodia | The confidence interval for the estimated GMR was 1.05–1.45. | That 95% of individual antibody responses fall in that interval. |
| GMR 0.99 for B/Victoria | The estimated ratio was close to 1. | That the groups were biologically identical in every participant. |
| Upper CI 1.53 for A/Victoria | The upper confidence limit was above the stated 1.5 non-inferiority boundary. | A p-value or a different formal hypothesis-test result not reported by the registry. |
| Serious AEs 9/382 vs 7/384 | These are the registry-reported descriptive serious-adverse-event counts. | A formal safety difference without a reported inferential analysis. |
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: NCT05091307 — Ad26.COV2.S.
- Linked publication: PubMed PMID 39866854.
Continue through the Clinical Biostats statistical pathway
Explore the statistical methods that connect randomized trial design, ANOVA, confidence intervals, per-protocol analysis, and non-inferiority methodology.
25. Record Summary
The Ad26.COV2.S trial provides a compact example of how immunogenicity endpoints can be evaluated using ANOVA, geometric mean ratios, confidence intervals, and a non-inferiority decision rule. The ClinicalTrials.gov record includes four influenza strain-specific HI antibody analyses and one S-ELISA antibody-concentration analysis. All five use two-sided 95% confidence intervals and report GMR estimates. The stated non-inferiority criterion requires the upper confidence limit to remain below 1.5. Four analyses have upper confidence limits below that boundary, while the A/Victoria (H1N1) analysis has an upper confidence limit of 1.53.
The statistical story is therefore not captured by the point estimates alone. Correct interpretation requires attention to the direction of the GMR, the analysis population, the confidence interval, the non-inferiority boundary, and the fact that the endpoints measure antibody immunogenicity rather than a directly reported clinical outcome. The serious-adverse-event data add a descriptive safety perspective, but the ClinicalTrials.gov record does not provide a formal inferential safety comparison.