This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
SUPERNOVA was a randomized, quadruple-masked, parallel-group phase 2/3 prevention trial in COVID-19 and SARS-CoV-2. The registry reports 3,882 enrolled participants across 14 arms and evaluates safety, predicted SARS-CoV-2 neutralizing-antibody titers, and confirmed symptomatic COVID-19.
| Feature | SUPERNOVA |
|---|---|
| Trial name | SUPERNOVA |
| Phase | Phase 2/3 |
| Conditions | COVID-19; SARS-CoV-2 |
| Design | Randomized, parallel-group |
| Masking | Quadruple |
| Primary purpose | Prevention |
| Enrollment | 3,882 |
| Arms | 14 |
| Status | Completed |
| Lead sponsor | AstraZeneca |
| Start | December 16, 2022 |
| Primary completion | March 29, 2024 |
| ClinicalTrials.gov | NCT05648110 |
2. Clinical Question
The trial addresses two closely related statistical questions. First, among participants receiving AZD3152 or AZD7442 in the sub-study, how do predicted neutralizing-antibody titers compare at Day 29? Second, in the main cohort, does AZD3152 reduce the occurrence of confirmed symptomatic COVID-19 compared with a combined AZD7442/placebo comparator?
Population
Participants enrolled in a randomized phase 2/3 prevention study of COVID-19 and SARS-CoV-2, including main, sentinel, and sub-study cohorts.
Interventions
The registry lists AZD3152, AZD7442/EVUSHELD, AZD5156 in the sentinel cohort, and placebo across the parent study and sub-study.
Comparator
For the main-cohort symptomatic COVID-19 analyses, AZD7442 and placebo were combined as the comparator group.
Primary questions
Are Day 29 predicted nAb titers non-inferior and superior between the sub-study antibodies, and does AZD3152 reduce symptomatic COVID-19 relative to the combined comparator?
3. Trial Design
AZD3152, AZD7442, and placebo
- AZD3152 - Main Cohort
- AZD7442 - Main Cohort
- Placebo - Main Cohort
- Confirmed symptomatic COVID-19 assessed at an average follow-up of 170 days for the primary analysis
AZD3152 and AZD7442
- AZD3152 - Sub Study
- AZD7442 - Sub Study
- Predicted nAb titers assessed at Day 29
- AEs through 29 days and SAEs, MAAEs, and AESIs through Day 451
4. Allocation, Masking, and Analysis Structure
The registry describes SUPERNOVA as randomized, parallel, quadruple-masked, and designed for prevention. Its 14-arm structure spans the parent study sentinel cohort, parent study main cohort, and sub-study. The statistical analyses therefore do not represent a single two-arm comparison across all 3,882 enrolled participants.
Randomization is the foundation for the comparative analyses reported for the main cohort and sub-study.
The registry identifies the overall trial as quadruple-masked.
The primary comparisons are between treatment groups rather than repeated treatment periods within a conventional crossover design.
The main clinical outcome is occurrence of confirmed symptomatic COVID-19.
5. Analysis Populations
The registry gives explicit definitions for the populations used in the formal statistical analyses. These definitions matter because the treatment effect is estimated only within the corresponding analysis set.
| Analysis population | Registry definition / role |
|---|---|
| Pharmacokinetic analysis set | All randomized participants who received a full dose of IMP per protocol, with sufficient information to estimate at least 1 postdose quantifiable serum concentration for any mAb component. Participants with protocol deviations that may interfere with serum concentrations may be excluded or have data collected after protocol deviations set to missing. Participants without valid serum concentration at Day 29 were excluded from the nAb-titer table. |
| SARS-CoV-2-Negative Set | All randomized participants who received part or all of the study intervention and did not have evidence of a current SARS-CoV-2 infection at baseline. Participants with positive RT-PCR test results at baseline were excluded from this analysis set. |
The distinction is statistically important. The pharmacokinetic analysis is tied to valid exposure and measurable serum concentrations, whereas the symptomatic COVID-19 analysis is anchored to randomized participants who were SARS-CoV-2 negative at baseline and received part or all of the intervention.
6. Primary Endpoints
| Endpoint | Time frame | Statistical approach reported |
|---|---|---|
| (Main Cohort) Occurrence of AEs Collected Through Approximately 90 Days After Each IMP Administration; SAEs, MAAEs, and AESIs Collected Through Day 451 | AEs through 90 days post each IMP administration; SAEs, MAAEs, and AESIs through Day 451 | Registry reports occurrence results; no formal inferential analysis is listed for this endpoint. |
| (Sub-study) Occurrence of AEs Collected Through 29 Days After IMP Administration. SAEs, MAAEs, and AESIs Collected Through Day 451. | AEs through 29 days post IMP administration; SAEs, MAAEs, and AESIs through Day 451 | Registry reports occurrence results; no formal inferential analysis is listed for this endpoint. |
| (Sub-study) Predicted SARS-CoV-2 nAb Titers Derived From Serum PK and in Vitro IC50 for SARS-CoV-2 Variants BA.2.86 Following AZD3152 Administration and Alpha Variant Following EVUSHELD Administration. | Day 29 | ANOVA; GMTR; non-inferiority and superiority |
| (Main Cohort) Confirmed Symptomatic COVID-19 Case | Primary Analysis; average follow-up time of 170 days | Poisson regression; relative risk reduction; superiority |
| (Main Cohort) Confirmed Symptomatic COVID-19 Case Attributable to Matched Variants | Primary Analysis; average follow-up time of 170 days | Poisson regression; relative risk reduction; superiority |
| (Sentinel Cohort) Occurrence of AEs, SAEs, MAAEs, and AESIs Collected Through Day 461 | Through Day 461 | Registry reports occurrence results; no formal inferential analysis is listed for this endpoint. |
The registry therefore contains four formal primary-analytical comparisons: two Day 29 predicted-nAb analyses and two main-cohort symptomatic-COVID-19 analyses. The two nAb analyses use the same estimated GMTR but different confidence levels and hypothesis frameworks.
7. Results
Day 29 Predicted Neutralizing-Antibody Titers
The registry defines the predicted nAb titer for AZD3152 against BA.2.86 as the AZD3152 serum pharmacokinetic concentration divided by the BA.2.86 IC50 of 3.8 ng/mL. For AZD7442 against Alpha, the predicted nAb titer is the AZD7442 serum pharmacokinetic concentration divided by the Alpha IC50 of 2.1 ng/mL.
Geometric mean titer ratio
90% two-sided CI: 3.71–4.46
Pre-specified non-inferiority analysis comparing AZD3152 - Sub Study with AZD7442 - Sub Study.
| Analysis | Estimate | Confidence interval | Hypothesis | Method |
|---|---|---|---|---|
| Day 29 predicted nAb titer | GMTR 4.07 | 90% two-sided CI 3.71–4.46 | Non-inferiority | ANOVA on log10-transformed predicted nAb titers |
| Day 29 predicted nAb titer | GMTR 4.07 | 95% two-sided CI 3.65–4.53 | Superiority | ANOVA on log10-transformed predicted nAb titers |
The reported GMTR of 4.07 describes a geometric-mean comparison of the predicted Day 29 antibody-titer measure between AZD3152 and AZD7442 in the pharmacokinetic analysis set. A ratio above 1 indicates that the numerator group has the higher geometric mean on the analyzed scale.
The 90% confidence interval of 3.71–4.46 describes statistical uncertainty around the estimated ratio under the registry's ANOVA framework. It does not describe the range of individual participants' antibody titers and does not imply that every participant experienced the same relative difference.
The non-inferiority conclusion is based on a pre-specified non-inferiority framework, not simply on whether a confidence interval excludes 1. The registry does not state the numerical non-inferiority margin in the information reported here, so the margin itself is not reproduced.
The separate 95% confidence interval of 3.65–4.53 belongs to the superiority analysis. A confidence interval describes precision; it is not itself a measure of biological importance. Likewise, a p-value, when reported, would address evidence against a null hypothesis rather than quantify the magnitude of the effect.
Confirmed Symptomatic COVID-19
For the main cohort, the registry reports confirmed symptomatic COVID-19 at the primary analysis, with an average follow-up time of 170 days. AZD7442 and placebo were combined into the comparator group for the reported relative-risk analysis.
Relative risk reduction
97.5% two-sided CI: 15.0–50.1 · P < 0.001
AZD3152 - Main Cohort versus combined AZD7442/Placebo comparator.
| Endpoint | Estimate | Confidence interval | P-value | Method |
|---|---|---|---|---|
| Confirmed symptomatic COVID-19 | Relative risk reduction 34.9% | 97.5% two-sided CI 15.0–50.1 | < 0.001 | Poisson regression with robust variance |
| Confirmed symptomatic COVID-19 attributable to matched variants | Relative risk reduction 42.9% | 95% two-sided CI 19.9–59.3 | = 0.001 | Poisson regression with robust variance |
The reported 34.9% relative risk reduction means that the estimated relative risk for confirmed symptomatic COVID-19 in the AZD3152 group was lower than in the combined AZD7442/placebo comparator according to the registry's Poisson regression analysis.
The corresponding 97.5% CI of 15.0–50.1 describes uncertainty around that estimated relative reduction. The interval does not mean that individual participants experienced reductions confined to this range, and it does not provide an absolute reduction in the probability of COVID-19.
The model used robust-variance Poisson regression, included treatment and randomization stratification factors as covariates, and adjusted for follow-up time. This is important because the analysis is estimating a relative risk while accounting for differing amounts of observed follow-up.
The P < 0.001 addresses the statistical evidence under the specified superiority hypothesis. It does not measure the size of the treatment effect, the clinical importance of the effect, or the probability that the treatment is effective.
The analysis population also matters: participants with positive RT-PCR test results at baseline were excluded. Consequently, this estimate applies to the defined SARS-CoV-2-Negative Set rather than automatically to every enrolled participant.
Confirmed Symptomatic COVID-19 Attributable to Matched Variants
Relative risk reduction
95% two-sided CI: 19.9–59.3 · P = 0.001
AZD3152 - Main Cohort versus combined AZD7442/Placebo comparator.
The estimated 42.9% relative risk reduction is specific to confirmed symptomatic COVID-19 cases attributable to matched variants at the primary analysis. It should not be interpreted as a 42.9% absolute reduction in the probability of disease.
The 95% CI of 19.9–59.3 indicates uncertainty around the estimated relative treatment effect. Because the interval remains above zero when expressed as relative risk reduction, it is consistent with a positive relative reduction under the corresponding two-sided framework.
The reported P = 0.001 is evidence against the null hypothesis used for the superiority analysis. It does not establish the magnitude of benefit by itself. The effect estimate and confidence interval provide the information about magnitude and precision.
As with the broader symptomatic-COVID-19 endpoint, the model adjusts for randomization stratification factors and follow-up time. The matched-variant definition also makes this a narrower endpoint than confirmed symptomatic COVID-19 overall.
Safety Primary Endpoints
The registry reports occurrence of adverse events for the main cohort, sub-study, and sentinel cohort, but does not provide a corresponding formal inferential analysis for these safety primary endpoints. Serious adverse events are reported by affected participants and participants at risk as follows.
| Group | Serious adverse events | Affected / at risk |
|---|---|---|
| AZD3152 - Main Cohort | 315 / 1,671 | 315 affected among 1,671 at risk |
| AZD7442 - Main Cohort | 222 / 1,102 | 222 affected among 1,102 at risk |
| Placebo - Main Cohort | 90 / 561 | 90 affected among 561 at risk |
| AZD5156 - Sentinel Cohort | 2 / 41 | 2 affected among 41 at risk |
| Placebo - Sentinel Cohort | 0 / 16 | 0 affected among 16 at risk |
| AZD3152 - Sub Study | 15 / 310 | 15 affected among 310 at risk |
| AZD7442 - Sub Study | 11 / 156 | 11 affected among 156 at risk |
| Crossover - Sub Study | 1 / 2 | 1 affected among 2 at risk |
8. Secondary Endpoint Results
Matched SARS-CoV-2 Variants Through Day 361
The secondary endpoint "(Main Cohort) Symptomatic COVID-19 Case (Negative RT-PCR at Baseline to Positive at Any Time up 12 Months With All Observed Events at Final Readout) Caused by Any SARS-CoV-2 Matched Variant" was assessed through Day 361.
Relative risk reduction
95% two-sided CI: 16.2–56.7 · P = 0.003
AZD3152 - Main Cohort versus combined AZD7442/Placebo comparator.
The 39.8% relative risk reduction summarizes the modeled comparison through Day 361 for symptomatic COVID-19 caused by matched variants. It is a relative measure and does not provide the absolute number of cases prevented.
The 95% CI of 16.2–56.7 provides a range of statistical uncertainty around the estimated relative reduction. Its width reflects the precision of the estimate; it should not be interpreted as a distribution of individual treatment effects.
The P = 0.003 is evidence against the specified null hypothesis for the superiority comparison. It should be interpreted alongside the effect estimate and confidence interval rather than used as a substitute for either.
Symptomatic COVID-19 Through Day 361, Any SARS-CoV-2 Variant
Relative risk reduction
95% two-sided CI: 10.7–40.2 · P = 0.002
AZD3152 - Main Cohort versus combined AZD7442/Placebo comparator.
The estimated 26.9% relative risk reduction applies to symptomatic COVID-19 caused by any SARS-CoV-2 variant through Day 361. Because the endpoint includes any variant, it is broader than the matched-variant endpoint.
The 95% CI of 10.7–40.2 indicates the statistical precision of the estimated relative reduction. The interval does not answer how large the absolute reduction was, because no absolute event rates are reported in the statistical analysis record reproduced here.
The P = 0.002 describes evidence against the relevant null hypothesis under the reported model. It is not a measure of effect size and should not be read as the probability that the observed result occurred by chance.
Severe COVID-19 Through Day 361
Relative risk reduction
95% two-sided CI: −95.1–87.9 · P = 0.308
AZD3152 - Main Cohort versus combined AZD7442/Placebo comparator.
The point estimate is a 51.5% relative risk reduction, but the confidence interval is extremely wide, extending from −95.1% to 87.9%. A relative risk reduction below zero corresponds to a relative risk above 1, so the interval encompasses both directions of association as well as substantially different magnitudes.
The P = 0.308 does not provide evidence against the specified superiority null hypothesis at conventional significance levels. More importantly, the wide confidence interval indicates that the estimate is imprecise.
The registry states that the hierarchical testing procedure stopped at this result. Therefore, subsequent interpretation of outcomes within that testing sequence must respect the prespecified hierarchy rather than treating each later nominal result as an independent confirmatory test.
The Poisson model included study intervention and randomization stratification factors and adjusted for follow-up time. The registry also states that only participants with non-missing covariates were included and that a stratification factor was dropped if there was 0 event at any level of that factor.
9. Statistical Methodology
ANOVA for predicted neutralizing-antibody titers
The Day 29 predicted nAb analysis used an analysis of variance model with the log10-transformed predicted nAb titer as the response variable and treatment arm as the primary explanatory variable.
The reported geometric mean titer ratio is derived from the comparison on the log-transformed scale and then expressed as a ratio of geometric means.
Log transformation is natural for antibody-titer measures because it converts multiplicative differences into additive differences on the analysis scale. The resulting ratio is therefore more interpretable than a raw arithmetic difference when the biological quantity is expected to vary multiplicatively.
Poisson regression for symptomatic COVID-19
The symptomatic-COVID-19 endpoints were analyzed using Poisson regression with robust variance. The model included study intervention and randomization stratification factors as covariates and adjusted for follow-up time.
The registry evaluates the relative risk through a Poisson regression with robust variance and then expresses the treatment effect as a relative risk reduction.
This approach is useful when the analysis needs a relative-risk interpretation while accounting for follow-up time and stratification variables. The robust variance addresses the variance estimation rather than changing the underlying definition of the relative risk.
Covariate adjustment
For the main-cohort symptomatic COVID-19 analyses, randomization stratification factors were included as covariates. Adjustment can improve statistical efficiency and preserve consistency with the factors used in the randomized design.
Stratified analysis
The registry identifies randomization stratification factors as part of the Poisson models. The analysis therefore does not treat the entire main cohort as if it had been randomized without any stratification structure.
Per-protocol features of the pharmacokinetic analysis
The pharmacokinetic analysis set is explicitly defined around participants who received a full dose of IMP per protocol and had sufficient postdose serum concentration information. This differs from a conventional all-randomized intention-to-treat efficacy analysis and makes the analysis population an important part of interpreting the Day 29 GMTR.
10. Statistical Methods Explained
Why was ANOVA used for the Day 29 nAb endpoint?
The registry uses ANOVA on log10-transformed predicted nAb titers. On the logarithmic scale, multiplicative differences between treatment groups become additive. Back-transforming the model comparison naturally produces a geometric mean titer ratio, which is the reported GMTR of 4.07.
What does a GMTR of 4.07 mean?
A GMTR of 4.07 indicates that the geometric mean predicted nAb-titer measure for the AZD3152 group was estimated to be 4.07 times that of the AZD7442 group under the reported comparison. It does not mean that every participant had a titer exactly 4.07 times higher, nor does it directly translate into a corresponding reduction in COVID-19 risk.
Why is non-inferiority different from superiority?
A superiority test asks whether the evidence is inconsistent with equality or the specified null hypothesis in the direction of benefit. A non-inferiority test instead asks whether the treatment is sufficiently close to the comparator according to a pre-specified margin. The registry identifies the nAb analysis as pre-specified non-inferiority, but the numerical margin is not reported in the ClinicalTrials.gov record.
Why use Poisson regression for symptomatic COVID-19?
The registry uses Poisson regression with robust variance to estimate relative risk while adjusting for follow-up time and randomization stratification factors. The robust variance is important because the model is being used for relative-risk inference rather than simply treating the observed event counts as exact Poisson realizations.
What does a 34.9% relative risk reduction mean?
It means the modeled relative risk for confirmed symptomatic COVID-19 in the AZD3152 group was estimated to be lower than that of the combined comparator by 34.9%. It is not a 34.9 percentage-point reduction in absolute risk. Absolute risk reduction would require the corresponding event risks in the compared groups.
Why does follow-up time enter the COVID-19 analysis?
Participants can contribute different amounts of observation time. The registry states that the Poisson regression adjusts follow-up time, which allows the analysis to account for the amount of time over which participants were observed when estimating the relative risk.
Why does the severe-COVID-19 result need particular caution?
The severe-COVID-19 point estimate is 51.5%, but its 95% confidence interval ranges from −95.1% to 87.9%, and P = 0.308. The wide interval indicates substantial uncertainty. The registry also states that the hierarchical testing procedure stopped, which is relevant when interpreting the inferential sequence of secondary endpoints.
11. Non-Inferiority and Superiority: Reading the Day 29 Analysis
SUPERNOVA illustrates how the same continuous endpoint can support different statistical questions. The registry reports a 90% two-sided confidence interval for the pre-specified non-inferiority analysis and a 95% two-sided confidence interval for the superiority analysis.
| Question | Reported estimate | Confidence interval | Interpretive role |
|---|---|---|---|
| Non-inferiority | GMTR 4.07 | 90% two-sided CI 3.71–4.46 | Evaluates the pre-specified non-inferiority hypothesis. |
| Superiority | GMTR 4.07 | 95% two-sided CI 3.65–4.53 | Evaluates superiority of the Day 29 predicted nAb-titer comparison. |
12. Relative Risk Reduction and Confidence Intervals
The COVID-19 analyses report relative risk reduction rather than a hazard ratio. This distinction matters because the estimand has a different interpretation.
Relative risk reduction
Expresses the proportional reduction in relative risk between AZD3152 and the combined comparator under the reported Poisson model.
Not an absolute risk reduction
The percentage does not state how many fewer COVID-19 cases occurred per 100 participants. Absolute risks would be needed for that interpretation.
Confidence interval
Describes statistical uncertainty around the estimated treatment effect under the model and analysis population.
P-value
Addresses evidence against a specified null hypothesis; it is not a measure of effect magnitude or clinical importance.
13. Multiplicity and Hierarchical Testing
SUPERNOVA contains multiple primary and secondary endpoints spanning safety, predicted antibody titers, and symptomatic COVID-19. The registry explicitly states that the severe-COVID-19 test was not significant and that the hierarchical testing procedure stopped at that point.
| Endpoint role | Reported statistical framework | Key interpretation issue |
|---|---|---|
| Predicted nAb titer | Non-inferiority and superiority | Different confidence levels correspond to different inferential questions. |
| Confirmed symptomatic COVID-19 | Superiority | 97.5% CI reported for the relative risk reduction. |
| Matched-variant symptomatic COVID-19 | Superiority | 95% CI and P = 0.001 reported. |
| Severe COVID-19 | Superiority | Testing hierarchy stopped after the non-significant result. |
Multiplicity matters because a long sequence of hypothesis tests can otherwise make isolated nominally significant findings difficult to interpret as confirmatory evidence. A hierarchical procedure establishes an order in which formal testing proceeds, so a later endpoint does not automatically retain the same confirmatory status once the sequence has stopped.
14. Missing Data, Protocol Deviations, and Analysis Set Definition
The registry gives several explicit rules for the pharmacokinetic analysis. Participants without a valid serum concentration at Day 29 were excluded from the corresponding table. Participants with protocol deviations that could interfere with serum concentrations could be excluded, or data collected after the deviations could be set to missing.
The COVID-19 analysis has a different structure. It begins with the SARS-CoV-2-Negative Set and includes only participants with non-missing covariates for the Poisson model. The registry states that if there is 0 event at any level of a stratification factor, that factor is dropped from the model.
15. Safety Results
Serious adverse events were reported separately across the main cohort, sentinel cohort, sub-study, and crossover sub-study. The denominator is important because the registry reports affected participants together with participants at risk.
| Group | SAEs affected | At risk |
|---|---|---|
| AZD3152 - Main Cohort | 315 | 1,671 |
| AZD7442 - Main Cohort | 222 | 1,102 |
| Placebo - Main Cohort | 90 | 561 |
| AZD5156 - Sentinel Cohort | 2 | 41 |
| Placebo - Sentinel Cohort | 0 | 16 |
| AZD3152 - Sub Study | 15 | 310 |
| AZD7442 - Sub Study | 11 | 156 |
| Crossover - Sub Study | 1 | 2 |
The registry's safety structure is more complex than a simple two-arm comparison. In particular, the crossover group contains 2 immunocompromised participants with exposure to both AZD7442 and an optional AZD3152 dose. That exposure history is relevant when interpreting adverse-event counts by arm.
16. Primary Endpoint Time Frames
The registry uses several distinct follow-up windows. Keeping those windows separate is essential because a 29-day safety endpoint, a Day 29 antibody endpoint, a 170-day average follow-up efficacy analysis, and a Day 361 secondary endpoint answer different questions.
Predicted nAb titers
Predicted SARS-CoV-2 nAb titers for AZD3152 against BA.2.86 and AZD7442 against Alpha were analyzed at Day 29.
Sub-study adverse events
AEs were collected through 29 days post IMP administration in the sub-study.
Primary symptomatic COVID-19 analysis
Confirmed symptomatic COVID-19 and matched-variant symptomatic COVID-19 were analyzed at the primary analysis with average follow-up of 170 days.
Secondary COVID-19 outcomes
Secondary analyses assessed symptomatic COVID-19 through Day 361, including matched variants, any SARS-CoV-2 variant, and severe COVID-19.
Longer safety follow-up
SAEs, MAAEs, and AESIs in the main cohort and sub-study were collected through Day 451, while sentinel-cohort AEs, SAEs, MAAEs, and AESIs were collected through Day 461.
17. Interpreting the Primary COVID-19 Model
The main-cohort Poisson regression has three important statistical components: treatment, stratification factors, and follow-up time. The treatment variable identifies the randomized intervention comparison; stratification factors preserve information from the randomized design; and follow-up adjustment accounts for the amount of observation contributed by participants.
The reported effect is therefore a relative measure derived from the model rather than a direct subtraction of observed percentages.
This distinction becomes particularly important when communicating the result. A relative risk reduction of 34.9% does not tell a reader how common symptomatic COVID-19 was in either group. Without the corresponding absolute risks, the number needed to treat and the absolute risk reduction cannot be obtained from the reported estimate alone.
18. Statistical Interpretation of the Severe COVID-19 Result
The severe-COVID-19 secondary endpoint provides a useful example of why point estimates should never be interpreted without their confidence intervals.
| Quantity | Reported value | Statistical meaning |
|---|---|---|
| Relative risk reduction | 51.5% | Point estimate from the Poisson regression model. |
| 95% CI | −95.1–87.9 | Very wide uncertainty around the point estimate, including both negative and positive relative-risk reductions. |
| P-value | = 0.308 | Not statistically significant under the reported superiority test. |
| Testing sequence | Hierarchical procedure stopped | Later endpoints in the hierarchy should not be treated as continuing confirmatory testing after this stopping point. |
A large point estimate is not automatically persuasive evidence when the estimate is imprecise. The severe-COVID-19 result illustrates the difference between magnitude of the point estimate and precision of the estimate. The confidence interval is central to understanding that distinction.
19. Why P-values Do Not Measure Effect Size
SUPERNOVA contains several results with small P-values and one secondary result with P = 0.308. The P-value is determined by the observed data, the statistical model, and the null hypothesis. It does not directly quantify the magnitude of a treatment effect.
Effect size
The relative risk reduction describes the estimated magnitude of the relative treatment difference.
Precision
The confidence interval communicates uncertainty around the effect estimate.
Evidence against a null
The P-value describes the compatibility of the observed result with the specified null hypothesis under the model.
Clinical importance
Clinical importance requires consideration of the absolute effect, endpoint severity, duration, and context rather than the P-value alone.
20. Important Limitations and Interpretation Issues
- Complex multi-arm structure: the trial includes 14 arms spanning sentinel, main-cohort, sub-study, and crossover groups. The formal efficacy analyses therefore apply to specific comparisons and populations rather than the entire enrollment as a single two-arm study.
- Different analysis populations: the pharmacokinetic and symptomatic-COVID-19 analyses use different analysis-set definitions. Results should not be generalized across populations without preserving those definitions.
- Non-inferiority margin: the registry identifies a pre-specified non-inferiority analysis but does not report the numerical margin in the statistical-analysis information summarized here.
- Relative rather than absolute effects: the COVID-19 results are reported as relative risk reductions. Absolute event risks are needed to understand the absolute reduction in disease occurrence.
- Wide uncertainty for severe COVID-19: the 51.5% point estimate has a 95% CI from −95.1 to 87.9, indicating substantial imprecision.
- Hierarchical testing: the registry states that the testing hierarchy stopped after the severe-COVID-19 result was not significant. Later nominal results should not be interpreted as if the hierarchy continued unchanged.
- Safety interpretation: adverse-event summaries span different cohorts and exposure patterns, including a small crossover group. Direct comparisons across all groups are therefore not equivalent to a simple randomized two-arm safety comparison.
- Pharmacokinetic exclusions: participants without valid Day 29 serum concentrations were excluded from the relevant nAb table, and protocol deviations affecting concentrations could lead to exclusions or missing post-deviation data.
- Variant-specific endpoints: matched-variant COVID-19 outcomes represent a narrower endpoint than symptomatic COVID-19 caused by any SARS-CoV-2 variant.
21. Why This Trial Matters Statistically
SUPERNOVA is a useful statistical case study because it combines several distinct clinical-trial methods within one registry record. The antibody endpoint is analyzed with ANOVA on a log-transformed continuous measure, while clinical COVID-19 outcomes use Poisson regression with robust variance, covariate adjustment, stratification factors, and follow-up-time adjustment.
| Concept | How it appears in SUPERNOVA |
|---|---|
| Randomization | Randomized, parallel-group phase 2/3 design. |
| Blinding | Quadruple-masked trial. |
| ANOVA | Used for log10-transformed predicted nAb titers at Day 29. |
| Geometric mean ratio | Reported as the effect measure for predicted nAb titers. |
| Non-inferiority | Pre-specified hypothesis for the Day 29 nAb comparison. |
| Superiority | Used for the Day 29 nAb comparison and symptomatic COVID-19 analyses. |
| Poisson regression | Used for confirmed symptomatic COVID-19 outcomes. |
| Risk ratio / relative risk | Converted to relative risk reduction for the reported COVID-19 effects. |
| Covariate adjustment | Randomization stratification factors included in the main-cohort Poisson model. |
| Stratified analysis | Randomization stratification factors incorporated into the efficacy model. |
| Per-protocol features | Pharmacokinetic analysis set requires full-dose administration per protocol and sufficient serum-concentration information. |
| Multiplicity | Hierarchical testing stopped after the severe-COVID-19 test was not significant. |
| Confidence intervals | 90%, 95%, and 97.5% two-sided intervals are reported for different analyses. |
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: NCT05648110.
- PubMed: PMID 42704596.
- PubMed: PMID 40015292.
- PubMed: PMID 38924429.
Continue through Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical-trial analyses covering study design, regression, inference, and analysis populations.
25. Record Summary
SUPERNOVA provides a detailed example of how different statistical estimands can coexist within a single randomized prevention trial. The Day 29 antibody endpoint uses an ANOVA model on log10-transformed predicted nAb titers and reports a geometric mean titer ratio, with both non-inferiority and superiority analyses. The main-cohort symptomatic COVID-19 endpoints use Poisson regression with robust variance, adjustment for randomization stratification factors, and follow-up-time adjustment, with effects reported as relative risk reductions.
The reported primary symptomatic-COVID-19 analysis estimated a 34.9% relative risk reduction with a 97.5% two-sided CI of 15.0–50.1 and P < 0.001. The matched-variant primary analysis estimated a 42.9% relative risk reduction with a 95% two-sided CI of 19.9–59.3 and P = 0.001. The Day 29 predicted nAb analysis reported a GMTR of 4.07, with a 90% two-sided CI of 3.71–4.46 for the pre-specified non-inferiority analysis and a 95% two-sided CI of 3.65–4.53 for the superiority analysis.
The secondary analyses illustrate why effect estimates, confidence intervals, P-values, analysis populations, and multiplicity must be interpreted together. Through Day 361, the reported relative risk reductions were 39.8% for symptomatic COVID-19 caused by matched variants and 26.9% for symptomatic COVID-19 caused by any SARS-CoV-2 variant. For severe COVID-19, the point estimate was 51.5%, but the 95% confidence interval ranged from −95.1 to 87.9 and P = 0.308; the registry states that the hierarchical testing procedure stopped.