Abstract
In epidemiological surveillance it is important that any unusual increase of reported cases be detected as rapidly as possible. Reliable forecasting based on a suitable time series model for an epidemiological indicator is necessary for estimating the expected non-epidemic indicator and to elaborate an alert threshold. Time series analyses of acute diseases often use Gaussian autoregressive integrated moving average models. However, these approaches can be adversely affected by departures from the true underlying distribution. The objective of this paper is to introduce a bootstrap procedure for obtaining prediction intervals in linear models in order to avoid the normality assumption. We present a Monte Carlo study comparing the finite sample properties of bootstrap prediction intervals with those of alternative methods. Finally, we illustrate the performance of the proposed method with a meningococcal disease incidence series.
Acknowledgements
The authors are grateful to Professor J. Lambert and R. Roy for providing the data. The authors would like to thank Mike Wiper for carefully reading that greatly improved the paper. This work was partially supported by the Dirección General de Investigación Científica y Técnica of Spain and Cátedra de Calidad BBVA.