Introduction
New Zealand has had a simulation model of foot-and-mouth disease (FMD) built within the InterSpread Plus (ISP) simulation system since 1993 (Sanson, 1993; Owen et al. 2011; Stevenson et al. 2013). This model has been periodically improved and refined, both in terms of the capability of the simulation software and in the parameters that represent the epidemiology of the disease and the behaviour of the livestock sectors and has been used to evaluate a range of response options should New Zealand ever experience an outbreak (see for example Sanson et al. 2017).
There is international interest in being able to predict the size and /or duration of outbreaks of FMD using variables or measures available or calculable from data captured early in the response (Tomassenet al. 2002; Hutber et al. 2006; Halasa et al.2013; Sarandopoulos, 2015; Garner et al. 2016).
One of these measures to predict outbreak trend is the estimated dissemination rate (EDR). The dissemination rate of a disease is the average number of herds or premises to which a disease agent is delivered by each infected herd. In practice it is hard to measure, however, Miller (1979) proposed an EDR for FMD as the ratio of cumulative incidence in one week to cumulative incidence in the previous week, based on the premise that the herds detected in a given week were likely infected by herds detected the previous week. Thus EDR provides a simple way to assess the observed rate of disease spread and gain insight as to whether an epidemic is likely to continue (EDR > 1) or whether it is being brought under control (EDR < 1). The time period is based around the inter-farm generation time period for the disease in question, which in the case of FMD is approximately 4-7 days (Hugh-Jones & Tinline, 1976).
These measures have collectively been referred to as early decision indicators (EDIs) (Garner et al. 2016). The benefit of knowing whether an outbreak is likely to be large or small early in the outbreak is that this knowledge can help guide the control and eradication strategy decision making via an adaptive management approach (Geet al. 2010; Halasa et al. 2013; Shae et al. 2014). This may include if vaccination should be adopted as an additional response measure. Various studies have indicated that emergency vaccination, if implemented early for large FMD outbreaks, benefit earlier control (Sanson et al. 2014; Sanson et al. 2017; Rawdon et al. 2018). Vaccination, however, showed little positive effect to control small outbreaks (Dürr et al. 2014), and it could lead to unnecessary competition for resources which are already strained and, a new complex task to monitor and manage vaccinated animals post outbreak for regaining FMD-free status and resuming international trade.
In the study by Garner and colleagues (2016), FMD modelling teams in Australia and New Zealand each generated 10,000 outbreaks of FMD in their respective countries using AusSpread (Garner & Beckett, 2005) and ISP respectively. Linear regression, classification and regression tree, and boosted regression tree analyses were used to quantify the predictive value of a set of parameters on three outcome variables of interest: the total number of IPs, outbreak duration, and the final area under control (AUC). The number of IPs, number of pending culls, AUC, EDR, and cattle density around the index farm at days 7, 14, and 21 following first detection were statistically associated with each of the outcome variables.
ISP supports the definition of trigger points that can be used to invoke specific actions that affect disease transmission or control. These include measures such as the numbers of detected IPs or EDR exceeding specified values by certain time points relative to the start of the simulation or first detection. Given the predictive ability of IP numbers and EDR values early in the response as reported by Garner and co-workers (2016), this study re-analysed the New Zealand dataset at additional time points to define a time-varying complex EDI trigger that could be specified within the ISP environment and triggered in ‘real-time’ during further simulated outbreaks. In a sense, this trigger would behave as a diagnostic test applied to each outbreak during the early stages of the outbreak, and therefore its diagnostic performance could be assessed prospectively by comparing the test results early in the response to the final sizes and durations. This paper reports on the performance of this trigger within a larger study exploring the benefits of emergency vaccination to augment SO within an adaptive management framework.