Felix Weidemann: A Bayesian approach to dynamic transmission modelling: Rotavirus in Germany

Rotavirus is the most common cause of acute gastroenteritis in children under 5 years of age with annually about 110 million cases worldwide. In Germany many aspects concerning transmission of rotavirus and reporting of new cases are still considered unknown and difficult to investigate in clinical studies.

To capture the German rotavirus activity and draw inference on epidemiological quantities we developed a system of ordinary differential equations (ODE) modelling the transmission and age-stratified expected number of weekly rotavirus cases. Our stochastic observational model assumed the reported number of new cases to follow a negative binomial distribution with mean equal to a region-specific proportion of the total number of expected cases, differentiating between East and West Germany. Within a Bayesian inference framework we estimated those parameters which were associated with high uncertainty, e.g. contact pattern, relative infectiousness of symptomatically infected and reporting rates. Prior distributions were based on literature evidence where possible. For model fitting we used age-stratified rotavirus surveillance data for East and West Germany from the Robert Koch Institute (RKI), Berlin. The overall data impact was reweighted by adjusting the likelihood subject to the residual autocorrelation of the data. Posterior distributions were based on asymptotic normality around the posterior mode. To compute prediction intervals for the age stratified incidence we sampled from the joint posterior and the negative binomial distribution of the observations. The final prediction bands captured the rotavirus incidence very well among all age groups, including the strong seasonal variations. By also integrating vaccination mechanisms into the model, we estimated vaccine effectiveness, which enabled the prediction of the potential impact of routine vaccination against rotavirus.

Altogether, we could quantify the difference in the reporting behaviour with a fivefold increased reporting rate in East compared to West Germany. Our model also suggests relatively low infectiousness of asymptomatically infected. Vaccination impact was most distinct in children under 5 years of age, even exceeding the expected direct effect . Our findings help to better understand the transmission of the virus while also providing a fitted model for the rotavirus activity and vaccination in Germany. The presented work is joint work with Manuel Dehnert, Judith Koch and Ole Wichmann from the RKI and Michael Höhle from Stockholm University