Comments on 'Some methodological issues in biosurveillance'
Statistics in Medicine
The comments that follow on Fricker's paper examining methodological issues in biosurveillance  are primarily from the perspective of public health practice. As Fricker noted, a recent directive  identified the two main objectives of biosurveillance as early event detection (EED) and situational awareness (SA). The focus of Fricker's paper was EED and I begin by commenting on the evidence for this objective. I also consider the evidence for SA as practice patterns suggest that this
... n of biosurveillance, even if it remains poorly defined, is particularly valuable to public health organizations [3, 4] . Early event detection I agree wholeheartedly with Fricker that poor EED performance in current systems does not imply biosurveillance should not be used for this purpose. Concerted research in this area is relatively recent and there is every reason to believe that innovations will occur in data processing and biostatistical methods. From a practical perspective, there will also remain a need in public health to detect disease outbreaks through surveillance, even if many outbreaks will be detected through other routes. However, research in EED should be informed by a sober assessment of the current evidence and by pragmatic consideration of the additional evidence required to advance early detection in public health practice. As noted by Fricker, results are difficult to synthesize from much of the research on this topic due to wide variation in the data sources used, inconsistency in the performance measures assessed, and infrequent comparison of new methods to existing ones. Consequently, published reviews have tended to be qualitative  , although reviews do suggest that EED as currently performed can detect some types of disease outbreaks rapidly with high accuracy. Standardizing algorithm performance studies as Fricker suggests will certainly facilitate identification of when and how EED should be used. Agreement on elements of the study design, including the role of simulation, the selection of performance metrics, and comparison to existing algorithms, is critical to develop evidence on this topic and to translate the evidence into practice. Although this field of research is still young, it may be time to consider the development of guidelines for reporting such studies, as has been done other areas of epidemiology and biostatistics  . Another important aspect of standardization is the naming and classification of detection algorithms. Although Fricker focussed on methods drawn from statistical process control (SPC), public health organizations use a much wider variety of methods in practice. Work to develop standard definitions of detection algorithms based upon their function as opposed to their origins will also facilitate the distillation of evidence from results [7, 8] . Consistent use of terminology about algorithms, data, and outbreak signals, will help to advance this field of research and build evidence about the performance of different algorithms when operating under different conditions.