Measuring progress and declaring ‘success’ in broad-scale eradications of wildlife pests and disease
Image - Tom Fraser.
Broad spatial-scale eradications are necessarily broad temporal-scale operations because large areas occupied by wildlife pests or diseased animals cannot all be treated at one time. This makes the process logistically complicated and expensive, if not prohibitive. A quantitative method for assessing progress and success is required to justify ongoing investment. However, these methods have yet to be developed. A case in point is New Zealand’s attempt to eradicate bovine tuberculosis (TB) from wildlife and livestock. This is an ambitious goal requiring decades of investment, but an analytical framework for assessing success at regional and country-wide scales that incorporates rigorous probability theory, epidemiology, possum biology and bio-economics has not yet been developed. This is a disease example but the same analytical principles, basic data requirements and spatio-temporal complexities apply to a growing list of vertebrate-pest-eradications: foxes on Tasmania and vertebrate pests on Stewart Island to name just two.
In broad-scale eradication operations such as these the targeted area is subdivided into zones, across which pest or disease control is organised in a spatio-temporal fashion, using strategies such as a ‘rolling front’ (Fig.). To assess (1) the progress and (2) success of such a strategy, Dean Anderson and colleagues are developing a two-stage analytical process that incorporates probability theory, biology, and economic and political constraints.
Stage one operates at a relatively fine-spatial scale and follows control in a given zone (Fig.). The aim is to quantify the probability the zone is free of pests or disease, which infl uences the decision to advance the control to new zones. This probability can be obtained in two ways. The first and inexpensive way is to use precedence, or the proportion of times that previous similar control efforts have succeeded. While this may appear risky, if the economic and political cost of being wrong is low, then this may be the optimal decision (i.e. low cost to re-control). If prior information on the probability of freedom in the zone does not exist, then surveillance data should be collected to calculate the sensitivity.
Surveillance sensitivity, or the probability of detecting the agent in the targeted area given that it is present, is the essential element for using surveillance data to quantify a probability of freedom from the pest or disease. Importantly, a probability of freedom for an area cannot be calculated using surveillance data without a sensitivity of detection. Intuitively and quantitatively, surveillance sensitivity increases with increasing search effort.
The objective in stage two is to provide a probability of ongoing freedom from pests or disease (given no detection) in the area no longer under active control. As in stage one, this requires a measure of surveillance sensitivity over the corresponding area. Ongoing surveillance over all previously controlled areas is critical in broad-scale eradications because of the unavoidable risk of reintroduction through ‘backfill’ (dispersal into a treated zone from an untreated adjacent area). This process continues until all control has been completed in all zones and the target probability of freedom for the entire area has been reached.
Stage-two surveillance to detect pests or disease in previously controlled areas, conducted over multiple zones, must be very inexpensive. A novel and important area of research is the development of surveillance models that are capable of incorporating multiple sources of low-cost detection data, as well as biology and epidemiology (in the case of diseases), to quantify zone-level sensitivities. These are then aggregated statistically for calculating the probability of freedom for the entire area. While low-cost and broad-scale-surveillance data may result in low zone-level sensitivities immediately following control, the modelling approach results in increasing sensitivities over the course of eradication operations.
To illustrate a hypothetical eradication of TB from possums in the South Island, a sensitivity measure would be calculated for all zones following the end of control in 2028 (Fig.). Intuitively, the most risky zones are those last controlled in 2010 because of elapsed time and the potential for reintroduction. However, the modelling approach incorporates the expectation that the possum population has recovered, and if TB was present a decade previously, it would be easily detected through surveillance of TB in livestock, captured sentinel species (e.g. pigs or ferrets) or direct surveillance of possums.
It is possible that it could take several years of low-cost and broad-scale surveillance following control to reach a target probability of freedom of a pest or disease for the entire area. The continuation of low-cost surveillance beyond control will be a small fraction of the total financial and political investment in the eradication programme. Further, the modelling approach can be used to predict, prior to the eradication effort, the amount of low-cost surveillance that will be required throughout and after control ends.
This work was funded by the Ministry of Business, Innovation and Employment through core funding to Landcare Research.
