Meningitis Forecasting
During the 2012 and 2013 meningitis years, the CHICAS (Lancaster) team conducted an empirical evaluation of statistical model-based forecasting algorithms using weekly district-level incidence data from Benin, Burkina Faso, Chad, Nigeria, Senegal and Togo. The team had previously developed the statistical models by analysing an exceptionally long and complete twenty-two year (1986 to 2007) time-series of weekly district-level incidence in Niger.
We used the weekly numbers of new cases in each district and an estimate of the population size of each district to help predict the probability that each district would experience a 'meningitis alert' (5 cases per 100,000), or an 'epidemic alert' (10 cases per 100,000) either within the next few weeks or in the current meningitis season. We used the technique of dynamic linear modelling to deliver the required inferences; our research was shared via teleconferences with the World Health Organisation on a weekly basis.