|
Abstract:
|
Particle filtering has found applications in a wide range of settings. We evaluated this method in a complex disease system where immunological interactions between multiple pathogens give rise to challenges in drawing inference for parameters. We used an established epidemiological model with 80 states and 13 parameters that represents the transmission process of four serotypes of dengue fever, and used an adaptive particle filtering approach across increasingly smaller two-dimensional grids of parameter space in order to estimate two parameters of interest. This approach enabled us to characterize the likelihood surface of these parameters and we tested it on a set of simulated data as well as real dengue case data. In datasets simulated from realistic parameter values, we were able to estimate accurately the impact of enhancement (avg. bias = 0.16; 95% CI coverage probability = 0.9). However, our ability to estimate the cross-protection parameter was limited (avg. bias = 0.4 year; 95% CI coverage probability = 0.4). Additionally, the method showed little sensitivity to misspecification of the parameters assumed to be known. Our analyses quantify the challenges of implementing a full-scale analysis using these methods for complex disease transmission models.
|
