Refining uncertainty about the TAK-003 dengue vaccine with a multi-level model of clinical efficacy trial data

Abstract Background A safe and effective vaccine that can be universally administered would represent a major advance in efforts to control dengue. Takeda’s TAK-003 is a potential candidate for such a vaccine. However, phase-III trial results suggest that TAK-003 confers differential protection by outcome, serotype, and serostatus. Published analyses estimated these stratified vaccine efficacies independently, ignoring the fact that some aspects of the disease process are shared across particular stratifications. Methods We addressed these shortcomings by building a multi-level model that pooled all publicly available trial data to estimate parameters that characterize this vaccine’s profile. Results We found that protection varied by both serotype and serostatus, with initial protection against symptomatic disease ranging from a median of 99.6% (95% credible interval [CrI]: 96.5, 100.0) among seronegatives infected with DENV-2 to 26.7% (95% CrI: −8.2, 54.9) among seronegatives infected with DENV-3. We found that initial protection against hospitalized disease among those with symptomatic disease ranged from 98.7% (95% CrI: 90.1, 100.0) among seropositives infected with DENV-2 to −72.7% (95% CrI: −149.4, −4.0) among seronegatives infected with DENV-3. Importantly, our estimate of the latter is less uncertain than a previously published efficacy estimate (−87.9%, 95% CI: −573.4, 47.6). Conclusions Our results demonstrate that inferences based on complex data from a clinical trial can be sensitive to model structure. Additionally, they reinforce the recommendation from the World Health Organization that use of this vaccine is best suited to high-transmission settings.