Epidemiological Forecasting with Statistical Models

Epidemiological forecasting is a young and current area of research with great potential benefit to society. In particular, seasonal epidemics cause considerable and repeated mortality and economic burden to the afflicted community. An analogy can be made to weather forecasting which has matured over several decades into a well-understood field with a rich toolbox of methods widely used by institutions and public alike. Similarly, accurate forecasts of upcoming seasonal epidemics can help inform the public and aid policy makers in designing and implementing countermeasures. The main difficulty lies in making far-ahead predictions for such a highly variable, complex process.  In this presentation, we describe three flexible statistical modelling techniques that are readily deployable to a wide array of seasonal epidemics -- an empirical Bayes procedure on epidemiological trajectories, a spline-based regression approach, and Gaussian process regression with spectral kernels. Strengths and weaknesses of each method are presented to aid comparison. Of the three methods, the first two have been successfully applied to Dengue fever in Brazil (for the 2014 World Cup), Puerto Rico and a region of Peru, and influenza in the United States.