A method for statistical design of complex biosecurity surveillance systems

22 - 24 September 2009

Biosecurity surveillance problems are typically complex, with multiple threats, heterogeneous risk surfaces and arrayed detection methods. Available design techniques do not deal with complexity and instead attempt to reduce it, by designing for single threats by a single detection method, or ignoring statistics and designing to budget or other pragmatic considerations. We were challenged to design a statistically based surveillance system (SS) for a complex application and developed a design method that has potential for broader use.

The context is a high-value island nature reserve where a major industrial development has been approved, contingent inter alia on a detection program having power (probability of detection given presence) of 80% for non-indigenous species (NIS) of invertebrates, vertebrates and plants. The method addresses typical design complexity as follows:

• To account for a great range of threats, design SSs for several exemplar species, each comprising several detection methods (‘SS components’ (SSCs)), including traps, surveys and worker observations.
• Use expert elicitation to address data gaps.
• Calculate detection power of each SSC, for exemplar population sizes of acceptable risk.
• Array the SSCs together, taking account of their individual power and broad cost, into an optimised exemplar SS meeting the 80% power specification.
• Use risk-mapping to narrow the sampling frame and deploy each SS efficiently.
• Overlay exemplar SSs to develop an efficient, integrated SS for all potential NIS.
• Give responsibility to surveillance staff for temporal and spatial deployment at local level, to achieve further risk-targeting.

We designed integrated SSs for each of invertebrate, vertebrate and plan NIS. Implementation has commenced and further desktop and field studies will examine robustness and validity. Designs will be adapted as data are acquired and analysed. We describe the design method and its application in this case study.