While many methods have been proposed for detecting disease
outbreaks from pre-diagnostic data, their performance is usually not
well understood. We argue that most existing temporal detection methods
for biosurveillance can be characterized as a forecasting component
coupled with a monitoring/detection component. In this paper, we
describe the effect of forecast accuracy on detection performance. Quantifying this effect allows one to measure the benefits of improved
forecasting and determine when it is worth improving a forecast
method’s precision at the cost of robustness or simplicity. We quantify
the effect of forecast accuracy on detection metrics under different
scenarios and investigate the effect when standard assumptions are
violated. We illustrate our results by examining performance on
authentic biosurveillance data.