Julie Dudasova 1, Garrett T. Nieddu 2, Jeffrey R. Sachs 2
1 MSD (Prague, Czech Republic), 2 Merck & Co., Inc. (Rahway, USA)
Objectives: The durability of vaccine-induced protection and its variability across different demographic groups are key questions in vaccine development. Traditional phase 2b and phase 3 clinical trials, which assess overall vaccine efficacy (VE) at the study’s end, can lack the power to provide insights into the persistence of vaccine-induced protection over time and/or within specific demographically or clinically defined subgroups. This limitation arises because clinical endpoint accrual is often sparse within individual years of follow‑up or within subgroups of interest, making year‑specific or subgroup‑specific VE estimates imprecise or entirely non‑estimable. Prior work has shown that precision of VE estimated in demographic subgroups can be substantially increased by quantifying the relationship between immunogenicity (measured shortly post vaccination) and the probability of disease, and then using that relationship to predict immunogenicity-based VE [1]. Building on this foundation, the objective of this work was to extend this model-based framework to incorporate longitudinal immunogenicity data in order to predict the durability of VE in predefined demographic subgroups, rather than relying on a single fixed immunogenicity measurement. To illustrate the utility of the framework, we applied it to a dengue vaccine dataset with rich longitudinal antibody data and clinically relevant heterogeneity in VE by baseline serostatus. These characteristics provide a realistic setting for assessing model performance under known patterns of waning VE and subgroup differences, and for comparing predictions with established external evidence.
Methods: The extended logistic regression-based approach was illustrated using participant-level data from the longitudinal immunogenicity sub-study of a phase 3 clinical trial of the dengue vaccine CYD-TDV (n = 611; >6 years of follow-up) [2]. Time‑dependent neutralizing antibody titers were measured annually and during virologically confirmed dengue disease. To assess the relationship between time-dependent immunogenicity and the probability of disease, we conducted correlate of risk (CoR) and correlate of protection (CoP) analyses using a weighted cross-sectionally pooled logistic regression model. The weighting scheme accounted for the length of the risk interval following each immunogenicity measurement, allowing the model to approximate a discrete‑time hazard relationship. Baseline serostatus was treated as a prespecified covariate given its known clinical relevance. Yearly immunogenicity measurements and the best-fitting logistic model were then used to predict yearly VE in baseline seronegative and baseline seropositive subgroups. Model outputs were compared with results from larger, independent studies to assess external consistency.
Results: When applying the framework to this dataset, the analysis showed that the time-dependent PRNT₅₀ antibody titer is both a CoR and a CoP for serotype 2 dengue (DENV-2), indicating that time-dependent immunogenicity drives the vaccine’s protective effect against DENV-2. The immunogenicity-based predictions of VE decreased over time in the baseline seropositive group, with VE predicted to remain significantly different from zero (α = 0.05) for up to two years post-vaccination. In contrast, all yearly predictions for the baseline seronegative group were not significantly different from zero, highlighting lack of VE in this subgroup. Importantly, immunogenicity‑based VE estimates were more precise than case‑counting VE estimates and were directionally consistent with observations from independent, larger studies, supporting the reliability of the model.
Conclusions: Incorporating annual immunogenicity measurements into logistic regression models improves the precision of VE estimates within demographic subgroups and provides earlier insight into the durability of vaccine‑induced protection. Implementing such methodologies during the clinical trial stage can accelerate vaccine development and support design of vaccines that provide robust and long-lasting protection.
References:
[1] Dudasova, J., Valenta, Z. & Sachs, J. R. Elucidating vaccine efficacy using a correlate of protection, demographics, and logistic regression. BMC Medical Research Methodology 24, 101 (2024).
[2] Salje, H. et al. Evaluation of the extended efficacy of the Dengvaxia vaccine against symptomatic and subclinical dengue infection. Nature Medicine 27, 1395-1400 (2021).
Reference: PAGE 34 (2026) Abstr 12234 [www.page-meeting.org/?abstract=12234]
Poster: Methodology - Covariate/Variability Models