TeenCove: Complete Statistical Analysis of mRNA-1273 in Adolescents for SARS-CoV-2

1. Trial at a Glance

TeenCove was a randomized, parallel, quadruple-masked phase 2/3 study evaluating the safety, reactogenicity, immunogenicity, and effectiveness of mRNA-1273 and mRNA-1273.222 in adolescents 12 to <18 years old for prevention of SARS-CoV-2 infection.

2. Clinical Question

The central clinical question was whether mRNA-1273 could provide an acceptable safety and immune-response profile in adolescents 12 to <18 years old, and whether later study parts evaluating mRNA-1273 and mRNA-1273.222 produced prespecified immunogenicity and effectiveness measures relative to the specified comparison groups.

3. Trial Design

Study-part structure

4. Endpoints

The registry lists 12 primary endpoints. They span solicited local and systemic adverse reactions, unsolicited adverse events, antibody geometric means, seroresponse rates, and longer-term safety outcomes. Their statistical interpretation depends strongly on whether the endpoint is a binary proportion or a continuous antibody measure summarized on a geometric scale.

Safety endpoint definitions

Solicited local reactions included injection-site pain, erythema, swelling or induration, and ipsilateral axillary swelling or tenderness. Solicited systemic reactions included fever, headache, fatigue, myalgia, arthralgia, nausea or vomiting, and chills. Solicited reactions causally related to injection were graded from 1 to 4 using the specified toxicity grading scale, with lower scores indicating lower severity and higher scores indicating greater severity.

An unsolicited AE was an event not specified as a solicited reaction or a solicited reaction beginning outside the protocol-defined reporting period. The registry also defines SAEs, AESIs, and medically attended AEs separately. Importantly, the classification of SARS-CoV-2 infection differed by study part: COVID-19/SARS-CoV-2 infections were treated as AEs in Parts 1 and 2 but as efficacy clinical events rather than AEs in Part 3.

Immunogenicity endpoint definitions

Seroresponse was defined relative to baseline as a change from below the lower limit of quantification to at least 4 times the lower limit, or at least a 4-fold rise when the baseline value was already at or above the lower limit. The lower and upper quantification limits varied by assay and endpoint.

For antibody measurements, values below the lower limit of quantification were replaced by 0.5 times the lower limit. Values above the upper limit were replaced by the upper limit when actual values were unavailable. These transformations are important because they affect the distribution used for the geometric-mean analysis.

5. Statistical Methodology

Non-inferiority framework for geometric mean measures

Several primary immunogenicity comparisons used a geometric mean ratio (GMR). The registry describes non-inferiority criteria based on the lower bound of the 95% confidence interval and a minimum point-estimate threshold.

For the Part 1A geometric mean value comparison, non-inferiority was considered demonstrated when the lower bound of the 95% CI was greater than 0.667 and the GMR point estimate was at least 0.8. The registry describes this as being based on a non-inferiority margin of 1.5.

The Part 3 ancestral-strain GMC analysis used the same lower-bound criterion of greater than 0.667 and point-estimate threshold of at least 0.8. By contrast, the Part 3 Omicron BA.4/BA.5 analysis used a superiority criterion: the lower bound of the 95% CI for the GMR had to be greater than 1.

Risk differences for seroresponse

Seroresponse-rate comparisons were expressed as differences in percentages. For Part 1A and Part 1C-1, the registry specified a 10% non-inferiority margin. The lower bound of the 95% confidence interval had to exceed -10%; the Part 1A endpoint also specified a point-estimate threshold greater than -5%.

ANCOVA for antibody levels

The registry explicitly describes an analysis of covariance (ANCOVA) model for the Part 1A antibody-level endpoint, with study group as a fixed effect. The Part 3 antibody endpoints against Omicron BA.4/BA.5 and the ancestral strain also specify an ANCOVA model with study group as a fixed effect.

ANCOVA is useful when the outcome is continuous and the analysis is intended to compare adjusted group means. Here, the reported effect measure is a ratio of geometric means, consistent with the highly skewed nature of antibody titers and the use of geometric summaries.

Exact methods for infection endpoints

The Part 1A efficacy analyses used vaccine efficacy (VE), defined as 1 minus the ratio of incidence rates for mRNA-1273 versus placebo. The registry states that the 95% CI of the incidence-rate ratio was calculated using an exact method conditional on the total number of cases, with adjustment for person-years.

For the first occurrence of symptomatic COVID-19, the registry analysis is normalized as an Exact / Clopper-Pearson binomial method and also identifies covariate adjustment in the analysis text. The resulting estimate was expressed as vaccine efficacy rather than as a direct risk difference or risk ratio.

Confidence intervals

The posted primary analyses use two-sided 95% confidence intervals. For non-inferiority, the direction of the lower confidence limit is central: the question is whether the entire relevant lower confidence boundary remains above the prespecified non-inferiority threshold. For superiority of the Part 3 Omicron BA.4/BA.5 GMC, the lower confidence limit being above 1 establishes the registry's stated criterion.

Analysis populations

The immunogenicity analyses used prespecified immunogenicity subsets rather than simply all enrolled participants. The Part 1A PPIS consisted of randomized participants selected for the Immunogenicity Subset who received planned doses, complied with the immunogenicity testing schedule, and had no major protocol deviations affecting key or critical data; participants seropositive at baseline were excluded.

The Part 1C-1 PPIS-Neg required baseline SARS-CoV-2 negativity, antibody assessments at booster Day 1 and Day 29, no major protocol deviations, no off-study COVID-19 vaccination before the Day 29 visit, receipt of two mRNA-1273 doses during the blinded phase, receipt of booster, and pre-booster SARS-CoV-2 negativity.

The Part 3 PPIS-Pos consisted of participants who received Dose 1 of mRNA-1273.222, had Day 29 antibody assessments, had no major protocol deviations, did not receive off-study COVID-19 vaccination before Day 29, and were SARS-CoV-2 positive at baseline.

6. Primary Results: Part 1A Immunogenicity

Geometric mean value of serum pseudovirus nAb ID50

The GMR of 1.078 indicates that the estimated geometric mean antibody titer in the Part 1A adolescent group was 1.078 times that of the specified young-adult comparator group.

Part 1A seroresponse rate

The point estimate of -0.2 indicates a very small difference in seroresponse rate, with the adolescent group lower by 0.2 percentage points on the reported scale.

7. Primary Results: Part 1C-1 Booster Immunogenicity

Geometric mean concentration after booster

The reported GMR of 5.071 means that the estimated geometric mean concentration in the adolescent booster group was 5.071 times the geometric mean in the specified young-adult comparator group.

Part 1C-1 seroresponse rate

8. Primary Results: Part 3 mRNA-1273.222

Omicron BA.4/BA.5 neutralizing antibody GMC

The reported GMR of 48.191 indicates that the estimated geometric mean antibody concentration against Omicron BA.4/BA.5 in the specified Part 3 adolescent analysis population was 48.191 times that of the specified young-adult comparator group.

SARS-CoV-2 ancestral-strain neutralizing antibody GMC

9. Primary Safety Results

The registry reports primary safety endpoints for solicited local and systemic adverse reactions, unsolicited adverse events, and longer-term safety events. The posted statistical analyses do not include numerical effect estimates for those safety endpoints among the 6 primary analyses; the numerical safety information available in the record includes serious adverse events by reported study part and arm.

These counts should be interpreted within their respective study parts rather than pooled across the entire trial. The denominators differ substantially, and the study parts evaluated different regimens and objectives. A raw comparison of counts across parts would therefore not be a valid comparison of safety rates.

What constitutes an SAE?

The registry defines a serious adverse event as an AE resulting in death, being life-threatening, requiring inpatient hospitalization or prolongation of an existing hospitalization, causing disability or permanent damage, constituting a congenital anomaly or birth defect, or representing an important medical event.

The longer-term primary safety endpoint also includes AESIs, medically attended AEs, and AEs leading to study discontinuation. Medically attended AEs are defined around unscheduled healthcare encounters, including unscheduled study-site assessments and visits to healthcare practitioners outside the study site.

10. Secondary Efficacy Results

Four secondary efficacy analyses are posted for Part 1A. These analyses compare mRNA-1273 100 µg with placebo beginning at Day 43, defined as 14 days after the second injection, and extending to the specified follow-up periods.

SARS-CoV-2 infection, symptomatic or asymptomatic

Asymptomatic SARS-CoV-2 infection

First occurrence of symptomatic COVID-19

CDC secondary case definition of COVID-19

11. Statistical Methods Explained

Why use a geometric mean ratio for antibody concentrations?

Antibody titers are commonly summarized on a multiplicative scale because their distributions can span a wide range. A ratio of geometric means describes the relative level between two groups and is naturally suited to log-scale modeling. A GMR of 1 indicates equality of the two geometric means.

Why was ANCOVA used for the antibody endpoints?

The registry explicitly specifies ANCOVA for several antibody-level primary endpoints, with study group as a fixed effect. ANCOVA provides a framework for comparing continuous outcomes between groups while incorporating prespecified model structure. In TeenCove, the reported effect measure is a ratio of geometric means rather than a simple arithmetic mean difference.

Why is non-inferiority judged against a margin?

Non-inferiority is not simply a test of whether two groups are statistically different. The question is whether the new or alternative regimen remains within an acceptable loss of the prespecified reference effect. TeenCove's posted immunogenicity analyses therefore use explicit thresholds such as a lower GMR confidence bound greater than 0.667 or a lower seroresponse-rate difference confidence bound greater than -10%.

What does a risk difference of -0.2 mean?

For the Part 1A seroresponse endpoint, the reported risk difference was -0.2 percentage points. On the registry's stated scale, this means the adolescent group's estimated seroresponse proportion was 0.2 percentage points lower than the specified comparator group's estimate. The associated 95% CI of -2.1 to 1.9 percentage points quantifies uncertainty around that difference.

Why use an exact method for the infection endpoint?

The posted Part 1A efficacy analyses describe an exact confidence-interval calculation conditional on the total number of cases and adjusted for person-years. Exact methods can be useful when the number of observed events is limited because large-sample approximations may be less reliable in sparse-event settings.

Why does the analysis population matter?

The antibody analyses use narrowly defined immunogenicity populations. For example, Part 1A excludes participants who were seropositive at baseline, while Part 3 PPIS-Pos specifically includes participants who were SARS-CoV-2 positive at baseline. Consequently, estimates from different study parts should not be interpreted as though they came from the same underlying population.

12. Understanding the Non-Inferiority Results

Non-inferiority is the central statistical concept connecting several of TeenCove's primary immunogenicity analyses. The important distinction is between an estimate being numerically close to or above a comparator and the confidence interval demonstrating that the treatment does not fall below a prespecified acceptable boundary.

These results illustrate why confidence intervals are central to non-inferiority interpretation. A point estimate can appear favorable while still being too imprecise to exclude an unacceptable loss. Conversely, a confidence interval that remains comfortably beyond the prespecified boundary provides the statistical evidence required by the registry's stated criterion.

13. Superiority for Omicron BA.4/BA.5 Antibody Response

The Part 3 Omicron BA.4/BA.5 GMC endpoint used a different hypothesis framework. The registry states that superiority was considered demonstrated when the lower bound of the 95% CI for the GMR was greater than 1.

The statistical distinction is important. For this endpoint, the relevant reference value is 1 because the effect measure is a ratio. A confidence interval entirely above 1 is evidence that the geometric mean in the numerator group exceeds that in the comparator under the stated model and population.

That statistical conclusion remains narrower than a clinical effectiveness conclusion. An antibody concentration comparison is an immunogenicity endpoint; it is not itself a direct estimate of protection against symptomatic disease or infection.

14. Secondary Efficacy and Exact Methods

The four posted secondary efficacy analyses provide a useful contrast between broad infection, asymptomatic infection, symptomatic COVID-19, and the CDC secondary case definition.

The asymptomatic-infection estimate demonstrates why effect size and precision must be considered together. Its point estimate is 43.5%, but the confidence interval extends below zero. In contrast, the confidence intervals for the broader infection endpoint and CDC secondary case definition remain above zero over their reported ranges.

15. Safety and Statistical Interpretation

Safety endpoints are fundamentally different from immunogenicity ratios. A serious adverse event is a binary participant-level outcome, and the registry reports affected participants and participants at risk for each study part and regimen.

16. Limitations and Interpretation Issues

• Different study parts: TeenCove is not a single two-arm comparison throughout its full registry structure. Several immunogenicity and safety analyses belong to specific study parts with different regimens and analysis populations.
• Part 1C-2 enrollment: enrollment was discontinued before the planned number of participants, limiting the amount of information available from that part.
• Part 2 discontinued early: the registry states that Part 2 was discontinued early because an updated variant vaccine became available, and no hypothesis testing was done for its primary endpoint.
• Comparator differences: the Part 1A efficacy analyses compare mRNA-1273 with placebo, whereas several primary immunogenicity analyses compare adolescent vaccine recipients with selected young adult vaccine recipients from Study P301.
• Analysis-population differences: PPIS, PPIS-Neg, PPIS-Pos, and the Part 1A PP Set for Efficacy represent different inclusion criteria and therefore should not be treated as interchangeable populations.
• Non-inferiority interpretation: the conclusion depends on the prespecified margin and confidence-interval criterion rather than on whether a conventional p-value crosses a generic threshold.
• Endpoint-specific interpretation: antibody GMRs, seroresponse-rate differences, and vaccine efficacy are different effect measures and should not be directly ranked against one another.
• Missing p-values: the posted statistical analyses provide estimates and confidence intervals but do not provide p-values for these reported analyses. The confidence intervals and prespecified hypothesis criteria are therefore the principal inferential information available in the registry entries.
• Early discontinuation: when a study part stops early, the resulting sample size and follow-up may be smaller than originally planned, affecting precision and limiting the amount of information available for that part.

17. Why This Trial Matters Statistically

TeenCove is a useful statistical teaching case because it combines several distinct clinical-trial analysis frameworks within one registry record. The same overall study includes randomized safety comparisons, exact binary-outcome methods, vaccine-efficacy estimation, geometric mean ratios, seroresponse-rate differences, ANCOVA, and both non-inferiority and superiority criteria.

18. A Deeper Look at Confidence Intervals

Confidence intervals are particularly important in TeenCove because the registry's principal inferential decisions are built around them. The same type of interval can support different conclusions depending on the effect measure and hypothesis framework.

The Part 1A GMR of 1.078 has a 95% CI of 0.940 to 1.237. The Part 1C-1 GMR of 5.071 has a 95% CI of 4.477 to 5.745. The difference in these intervals reflects both the different estimates and the different study-part comparisons; it should not be interpreted simply as one endpoint being more clinically important than another.

19. Interpreting Vaccine Efficacy

The registry defines vaccine efficacy as 1 minus the incidence-rate ratio for mRNA-1273 versus placebo. This definition makes VE a relative measure of incidence rather than an absolute probability of infection avoided.

The four secondary analyses demonstrate how the same statistical framework can produce different estimates for different clinical definitions. Symptomatic COVID-19, asymptomatic infection, all SARS-CoV-2 infection, and the CDC secondary case definition are not interchangeable outcomes. Their estimates and confidence intervals should therefore be interpreted separately.

20. Statistical Interpretation vs Clinical Interpretation

The distinction is especially important for the Part 3 Omicron BA.4/BA.5 analysis. A GMR of 48.191 is an immunogenicity result on the antibody-concentration scale. It is not itself a vaccine-efficacy estimate, a risk difference, or a measure of symptomatic COVID-19 prevention.

21. Registry Caveats and Study Evolution

22. What the Main Effect Measures Do — and Do Not — Mean

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Calculators

25. Sources

• ClinicalTrials.gov: NCT04649151.
• PubMed: PubMed record 39332418.
• PubMed: PubMed record 34379915.

26. Record Summary

TeenCove provides a particularly useful example of how a single clinical trial can require several distinct statistical frameworks. The registry combines randomized safety assessment, immunogenicity comparisons using geometric mean ratios, seroresponse-rate differences, exact binary-outcome methods, vaccine-efficacy estimation, ANCOVA, and both non-inferiority and superiority criteria.

The most important interpretive principle is to keep each estimate connected to its endpoint, comparator, analysis population, and hypothesis framework. A GMR of 1.078, a seroresponse-rate difference of -0.2 percentage points, a GMR of 48.191, and vaccine efficacy of 60.3% are all valid statistical measures, but they answer different questions and should not be interpreted as though they were interchangeable measures of the same treatment effect.