Review 2: "Efficacy, Public Health Impact and Optimal Use of the Takeda Dengue Vaccine"

RR\ID Evidence Scale rating by reviewer:

• Reliable. The main study claims are generally justified by its methods and data. The results and conclusions are likely to be similar to the hypothetical ideal study. There are some minor caveats or limitations, but they would/do not change the major claims of the study. The study provides sufficient strength of evidence on its own that its main claims should be considered actionable, with some room for future revision.

***************************************

Review: The study by Daniels et al. (2024) examines the impact of Takeda Pharmaceuticals’ new Qdenga vaccine on dengue infection rates and severity. Dengue vaccines carry the risk of worsening the disease due to cross-reactive immunity among different dengue serotypes. Therefore, understanding the benefits and risks of new dengue vaccines is crucial for public health officials, medical practitioners, researchers, and the general public. To assess the Qdenga vaccine, the authors use phase III clinical trial data. They estimate vaccine efficacy based on age, dengue serotype, and individual serostatus at vaccination. The authors then explore the potential population-level impacts of a Qdenga vaccination program through simulations based on a mechanistic model (i.e., a dynamic model with scientifically interpretable parameters) describing dengue transmission.

The study assesses vaccine efficacy by fitting a Bayesian cohort survival model to Qdenga phase III clinical trial data. This approach allows for estimation across different age groups, dengue exposures, and serotypes, thus adding to the understanding of Qdenga vaccine efficacy. There is considerable uncertainty in some parameter estimates; for instance, efficacy estimates beyond 12 months for non-DENV2 serotypes range from negative to near-perfect efficacy (Supplement Figure 8). As noted in the article, these large uncertainties are due to the small number of cases observed during the trial, which “highlights the need for continued monitoring of Qdenga’s efficacy profile and collection of additional data.”

To estimate the population-level effects of a large vaccination program, the authors integrate their vaccine efficacy model into an existing mechanistic dengue transmission model. This model accounts for transmission dynamics between humans and mosquitoes, stratified by serotype, age, and vaccination status. Sensitivity analysis was performed, exploring different levels of serotype prevalence, modes of vaccination impact, and durations of efficacy. Mathematically, the mechanistic model is expressed as a system of ordinary differential equations; however, it is also stated that transitions between human compartments are drawn from binomial distributions. Therefore, the model is actually a Markov counting system with rates defined by the right-hand side of the provided equations.

It is unclear how thoroughly the transmission model has been validated as a quantitatively accurate dengue transmission model. Conclusions drawn from a model are more reliable when the model is statistically validated against data. The primary evidence supporting the transmission model is its qualitative success in predicting antibody-dependent enhancement, as demonstrated in a previous study. Our recent retrospective analysis of the cholera outbreak in Haiti suggests comparing the likelihood-based criteria of the mechanistic model to that of a simple statistical model called a benchmark (Wheeler et al., 2024). Essentially, a mechanistic model used for quantitative predictions should have statistical predictive value not much worse than simple statistical models. In the absence of such evidence, the output from the transmission model should be considered qualitative rather than quantitative; while the assumptions appear reasonable, the model’s predictive skill has not been tested.

In this study, the wide confidence intervals invite the reader not to draw strong quantitative conclusions, reducing the danger of misinterpretation. However, the wide confidence intervals speak only about the small sample size for the vaccine efficacy study, and not about uncertainty about the transmission model.

In addition, we note that the model does not include a mechanism for environmental stochasticity, which has been found to be necessary for population-level transmission models to provide a statistically adequate fit to data (Stocks et al., 2020). The use of benchmarks would help to clarify whether this omission is significant in the current context.

As a related point, the biologically plausible model for vaccine efficacy proposed in this paper has only been compared to nested versions of the same model. While these tests of nested hypotheses are useful, none of them are readily able to test against the hypothesis that the entire class of models under consideration has substantial misspecification. One would expect some misspecification, for example, due to unmodeled country-level effects within the clinical trial data. The total amount of misspecification may be large or small. Benchmarking against a simple non-mechanistic statistical model can address this issue more effectively than any of the checks carried out in the present version of this manuscript.

This study is carefully constructed, clearly described, and demonstrates a high standard of reproducibility through the provided code. It adds to the current understanding of the potential individual and population level impacts of the Qdenga vaccine. Nevertheless, there is room for improvement in current standards for inference on disease transmission models in the context of informing vaccination policy (Wheeler et al., 2024).

References

1. Daniels, B. C., Ferguson, N., and Dorigatti, I. (2024). Efficacy, public health impact and optimal use of the Takeda dengue vaccine. medRxiv, 2024.08.10.24311393.

2. Stocks, T., Britton, T., and HÅNohle, M. (2020). Model selection and parameter estimation for dynamic epidemic models via iterated filtering: Application to rotavirus in Germany. Biostatistics, 21(3):400–416.

3. Wheeler, J., Rosengart, A., Jiang, Z., Tan, K., Treutle, N., and Ionides, E. L. (2024). Informing policy via dynamic models: Cholera in Haiti. PLOS Computational Biology, 20(4):e1012032.