Impact of TAK-003 on epidemiologic, economic and quality-of-life burden of dengue: comparison of the results from a dynamic transmission model and a static model with a dynamic component

Introduction: Dynamic transmission models are commonly used to evaluate the impact of vaccination as they capture both direct and indirect effects; however, they tend to be complex and computationally intensive. Static models do not capture indirect effects and may underestimate vaccine benefit, but are easier to use. A static Markov model developed for the dengue vaccine TAK-003 included a dynamic component to approximate indirect effects. This study aims to validate this innovative “dyna-static” model by comparing its results with that of the dynamic model of TAK-003. Methods: The dynamic model explicitly modelled hosts and vectors, and included all four serotype-specific infections. The dyna-static model included 16 health states capturing four consecutive infections. Both models differentiated infection severity into asymptomatic, symptomatic non-hospitalized and symptomatic hospitalized, and incorporated cross-protection after each infection and higher probability of severe disease with secondary infections. Wherever possible, the assumptions and inputs were aligned between the two models. The dyna-static model simulated indirect effects by making the probability of infection in the future cycles dependent on the number of infections predicted at previous cycles. For comparing the dyna-static model to the dynamic model, 12 scenarios were examined using Thailand epidemiology as an example. Different magnitude of indirect effects was approximated by varying vaccination strategy, relative transmissibility of symptomatic dengue, and efficacy against asymptomatic dengue. Eleven combinations of cycle length and lag (the number of cycles before the impact on the probability of infection) were analyzed for the aforementioned 12 scenarios, resulting in 132 scenarios. The goodness of fit was assessed by visual assessment and two formal goodness-of-fit metrics (mean absolute error and root mean squared error). Results: The combination of 2-month-cycle and 2-month-lag was found to provide the closest fit with the dynamic model for all timeframes between 1–30 years, on outcomes of predicted cases averted, costs savings and disability-adjusted-life-years (DALYs) averted. Over 30 years, the number of averted symptomatic infections predicted by the dyna-static model under this combination of cycle and lag was, on average, 1% lower than the number predicted by the dynamic model; for hospitalized infections it was 2% lower. The predictions of the dyna-static model for total cost savings from societal perspective and DALYs averted were 4% and 2% lower, respectively. The dyna-static model tends to underestimate when indirect effects are excluded and overestimate when included. Nevertheless, the results aligned well with the dynamic model. Discussion: The dyna-static model yielded results that were comparable to those from the dynamic model, thereby affirming its applicability for assessing the cost-effectiveness of the impact of TAK-003 as a simpler tool. Conclusion: The findings are specific to the epidemiological context of Thailand and may best apply to settings with similar characteristics.