| Background. The majority of vaccine trials have focused on vaccine-related incidence reduction, which is the Food and Drug Administration (FDA) criterion required for licensure. However, studies focused on the association between vaccine efficacy and severity of disease are lacking. The primary goal of this project is to jointly model disease incidence and severity as a new approach for vaccine evaluation. This study examined the model's performance as compared with other candidate models.; Methods. In this project, the study outcome of interest is a multivariate random vector composed of elements describing both incidence and severity of infection. We refer to this problem as dual modeling of incidence and severity. Derived from the finite mixture model, an incidence and severity dual model was proposed to be a new method to examine the protective effects of vaccine associated with incidence and severity simultaneously. The within-subject correlation and dependence of severity on incidence has been considered during the dual modeling process. Through instructive scenarios, large sample simulation studies, and practical examples, the incidence and severity dual model, is compared with individual model, ordinal logit model, sieve analysis, and a two-part model. The bias and true parameter coverage percentage of 95% confidence intervals were used as the indicators to compare the model estimates. In order to study the model performance in hypothesis testing for significance of vaccine-outcome association, the type I error and power of the incidence and severity dual model were compared with a two-part model in simulation studies.; Results. Estimates from the dual model were found to be more robust to sensitivity of disease definition compared to the other models. Giving the same threshold for disease definition, simulation studies indicated that the dual estimates had the smallest mean bias and the highest true parameter coverage percentage of the 95% confidence intervals among all the models when a vaccine was assumed to have a low to medium level of efficacy in combating infection. The incidence and severity dual model worked the best when a vaccine acts to reduce both incidence and severity or is only efficacious with respect to severity. In the situations when a disease definition has low sensitivity to infection, different disease definitions have been applied to various vaccine trials interested on the same vaccine or infection, or when a subclinical infection is of great importance in public health, the incidence and severity dual model provides the closest estimates of vaccine parameters than the individual model, the ordinal logit model, the sieve analysis, and the two-part model. (Abstract shortened by UMI.)... |
