Statistical Modelling, Monitoring and Predictive Analytics against Infectious Disease Outbreaks

Project leader

Funding source

Swedish Research Council - Vetenskapsrådet (VR)

Project Details

Start date: 01/01/2016
End date: 31/12/2019
Funding: 2800000 SEK


Statistical modeling has become a key component for understanding the transmission of infectious diseases and for evaluating the impact of control measures. As public health institutions collect more and more data, the need for adequate data analysis beyond pure descriptions is pronounced. This project aims at exploiting advances in statistical modelling and machine learning for making better use of such public health surveillance data in order to quickly detect and track outbreaks while they occur. This could, e.g., be outbreaks of emerging respiratory viruses (such as severe acute respiratory syndrome), pandemic influenza or even foodborne gastrointestinal outbreaks. This project considers variants of the susceptible-infectious-recovered (SIR) compartmental model as well as more black-box oriented count data time series models for describing the dynamics of an outbreak. The aim is to use these models as basis for surveillance algorithms, which prospectively detect aberrations in the data, as well as for real-time tracking and forecasting algorithms analysing, e.g., the epidemic curve during an outbreak. From a statistical point of view the focus is on the sequential nature of the data acquisition process. As a consequence, both the prospective monitoring and the real-time tracking of outbreaks need to be of sequential nature. Furthermore, artifacts by the reporting system, such as under-reporting and reporting delays, also need to be taken into account. In the project a multivariate approach towards the topic is utilized, where multiple data streams originating from neighbouring regions as well as age, symptoms or taxonomy related groups provide an added value compared to the single stream analysis. A Bayesian inference framework will be used for calculating the necessary predictive distributions, because it provides a unified approach for the handling of uncertainty of predictions in such observed dynamic systems. One computational challenge of the project will be to develop fast sequential Bayesian inference methods for adaptive parameter inference. Building on existing open-source software for the early detection and tracking of outbreaks by the project applicant (and in use at several public health institutions), the idea is to disseminate the methodological achievements into an open-source software toolbox. Hence, the outcome of the project could, with time, become an integrated part of the preparedness against outbreaks.

Last updated on 2017-28-07 at 10:49