Application of techniques for the quantitative analysis of complex surveillance systems to disease and wildlife management

Scenario tree modelling and associated methods provide tools to quantify the combined value of multiple complex surveillance activities over time. The outputs of this analysis are an estimate of the sensitivity of a surveillance system, and the cumulative probability of disease freedom that surveillance provides over time. The ability to quantify the performance of complex surveillance systems provides a number of new opportunities in the design and application of search and detection activities. One of these is the ability to objectively compare alternative detection strategies. The sensitivity of a strategy (the probability that the target biota would be detected, given that they are present at a defined level) may be balanced against cost and practicality considerations to determine the most effective strategy for a given situation. The paper provides an example of the comparison of two surveillance strategies (structured surveys and abattoir surveillance) for disease detection in animal health (Classical Swine Fever). These techniques may also play an important role in the certification of success in pest eradication operations. The ability to use multiple sources of evidence to evaluate success means that a higher level of confidence can often be achieved at lower cost. Examples of the application of scenario tree modelling in plant health and invasive pests are provided to illustrate its use within eradication programs.     

Spatially-explicit wildlife surveillance to prove freedom from diseases or pests

Surveillance undertaken to prove freedom from a disease has traditionally focussed on sampling a specified proportion of the population to determine the presence of the infective agent. Based on a sample of individuals testing negative for disease, standard probability theory using the binomial or hypergeometric distribution can be used to make inference about the probability of disease being absent. However, if the disease agent involves a wildlife host, then determining what proportion of the population was sampled becomes problematic as it requires an estimate of population abundance, which may be difficult and expensive to obtain. Surveillance during the eradication of pests such as feral ungulates often make use of the Judas technique, whereby radio-collared individuals are released and followed in the hope that the sociality of the individuals will betray the locations of conspecifics, which can then be dispatched. Data collected during such surveillance is explicitly spatial, containing information about the ‘search area’ of the Judas animal. However, there has been no attempt to use such data to make inference about the probability of the area being free of the pest, given Judas animals fail to detect any conspecifics. We present a framework for making specific use of the spatial nature of such wildlife surveillance data to make inference about the probability of freedom from the disease or pest. Underlying the sampling framework is a model of the detection process by sampled individuals. Estimates of individual detection probability are spatially smoothed using the extent of individual movements to produce a spatially- explicit detection surface. Bayes theorem is then used to combine this 2-dimensional surface with prior information on the probability of pest or disease presence, prior to sampling, to estimate the probability that the area is free from the disease or pest, given surveillance fails to detect evidence of their presence. We illustrate the method with examples of the detection of bovine Tb in wildlife in New Zealand and the detection of pigs (Sus scrofa) using the Judas technique during the feral pig eradication programme on Santa Cruz Island, California.     

Bio-economic optimisation of surveillance to confirm broadscale eradications of invasive pests and diseases

Although pest eradications from islands have been successful and impart biodiversity benefits, eradications at regional/national scales are more challenging. Such broadscale eradications incur high repeated costs (e.g. control and surveillance effort) because the entire area cannot be treated at one time, and a progressive ‘treat-evaluate-move on’ approach must be employed. We describe a two-stage model to analyse surveillance data for assessing progress and declaring success of broadscale eradications, and to identify optimal cost-efficient surveillance strategies. Stage I modelling coincides or follows population control within a subset area or management zone (MZ). Surveillance data are analysed to quantify the probability of freedom for a treated MZ (i.e. local eradication), which is used to inform an operational decision to reallocate resources to other MZs, and progress across the region. Importantly, freedom declared individually in all MZs is not necessarily equivalent to a high probability of eradication over the broadscale area, because each MZ will have a probability of being erroneously declared free. After a MZ has been operationally declared free, Stage II surveillance commences to detect MZ-level failures, and to estimate the broadscale surveillance sensitivity and a corresponding probability of eradication. We developed a computer algorithm to identify cost-optimal Stage I and II surveillance strategies for a hypothetical large area. We assessed the following: (1) the balance between local surveillance intensity and spatial coverage; (2) the number of years to declare success in Stages I and II; (3) the stopping probability of freedom (Stage I); and (4) the optimal strategy given variation in the starting-over cost, should a MZ be erroneously declared free. This two-stage approach provides an objective basis for decision-making in wildlife pest/disease eradication, and guidance for implementing optimal bio-economic surveillance strategies.     

A framework for estimating the sensitivity of eDNA surveys

Imperfect sensitivity, or imperfect detection, is a feature of all survey methods that needs to be accounted for when interpreting survey results. Detection of environmental DNA (eDNA) is increasingly being used to infer species distributions, yet the sensitivity of the technique has not been fully evaluated. Sensitivity, or the probability of detecting target DNA given it is present at a site, will depend on both the survey method and the concentration and dispersion of target DNA molecules at a site. We present a model to estimate target DNA concentration and dispersion at survey sites and to estimate the sensitivity of an eDNA survey method. We fitted this model to data from a species‐specific eDNA survey for Oriental weatherloach, Misgurnus anguillicaudatus, at three sites sampled in both autumn and spring. The concentration of target DNA molecules was similar at all three sites in autumn but much higher at two sites in spring. Our analysis showed the survey method had ≥95% sensitivity at sites where target DNA concentrations were ≥11 molecules per litre. We show how these data can be used to compare sampling schemes that differ in the number of field samples collected per site and number of PCR replicates per sample to achieve ≥95% sensitivity at a given target DNA concentration. These models allow researchers to quantify the sensitivity of eDNA survey methods to optimize the probability of detecting target species, and to compare DNA concentrations spatially and temporarily.     

Modelling studies to estimate the prevalence of foot-and-mouth disease carriers after reactive vaccination

Foot-and-mouth disease (FMD) is a highly contagious and economically significant viral disease of cloven-hoofed animals. Vaccination can be used to help restrict the spread of the infection, but evidence must be provided to show that the infection has been eradicated in order to regain the FMD-free status. While serological tests have been developed, which can identify animals that have been infected regardless of vaccination status, it is vital to know the probable prevalence of herds with FMD carriers and the within-herd prevalence of those carriers in order to design efficient post-epidemic surveillance strategies that establish freedom from disease. Here, we present the results of a study to model the expected prevalence of carriers after application of emergency vaccination and the impact of this on the sensitivity of test systems for their detection. Results showed that the expected prevalence of carrier-containing herds after reactive vaccination is likely to be very low, approximately 0.2%, and there will only be a small number of carriers, most likely one, in the positive herds. Therefore, sensitivity for carrier detection can be optimized by adopting an individual-based testing regime in which all animals in all vaccinated herds are tested and positive animals rather than herds are culled.     

Validation of in situ hybridisation and histology assays for the detection of the oyster parasite Marteilia refringens

An in situ hybridisation technique has been developed for the detection of infection in oysters with Marteilia refringens with particular emphasis on light infections or confirmation of suspected cases by means of histology. Although validation of new diagnostic methods is usually achieved by comparison with standard techniques, in our case the sensitivity and specificity of the standard (histology) had not previously been established. Another point to consider is that surveillance and monitoring frequently target populations displaying different levels of prevalence under different field conditions. The objective of our study was to evaluate the sensitivity and specificity values of in situ hybridisation and histology for the detection of M. refringens, based on 3 populations of flat oysters, free of the disease and with mild and high levels of prevalence. A blind assay of 200 individuals from each population was performed using both techniques. Results were analysed by means of the classical approach and latent models (maximum likehood and Bayesian approach). Assumptions and results were found to vary slightly with the different statistical approaches. The more realistic estimate by the Bayesian approach shows a link between the level of prevalence and the sensitivity of the techniques. Values of sensitivity and specificity for histology were 0.7 and 0.99 respectively, and 0.9 and 0.99 respectively in the case of in situ hybridisation. Some uncertainty remains regarding these values because the study does not take into account the severity of infection or the developmental stages of the parasite actually present in each individual. This work provides valuable information with regard to the choice and potential use of those 2 diagnostic methods currently recommended by international standards.     

Adult mosquito trap sensitivity for detecting exotic mosquito incursions and eradication: a study using EVS traps and the Australian southern saltmarsh mosquito, Aedes camptorhynchus

Adult mosquito traps are commonly used in biosecurity surveillance for the detection of exotic mosquito incursions or for the demonstration of elimination. However, traps are typically deployed without knowledge of how many are required for detecting differing numbers of the target species. The aim of this study was to determine the sensitivity (i.e., detection probability) provided by carbon dioxide-baited EVS traps for adult female Australian southern saltmarsh mosquitoes, Aedes camptorhynchus, a recent biosecurity problem for New Zealand. A mark-release-recapture study of three concurrently released cohorts (sized 56, 296, and 960), recaptured over four days with a matrix of 20 traps, was conducted in Australia. The detection probability for different numbers of traps and cohorts of different sizes was determined by random sampling of recapture data. Detection probability ranged from approximately 0.3 for a single trap detecting a cohort of 56 mosquitoes to 1.0 (certainty of detection) when seven or more traps were used. For detection of adult Ae. camptorhynchus around a known source, a matrix of traps provides a strong probability of detection. Conversely, the use of single traps deployed over very large areas to detect mosquitoes of unknown entry pathway is unlikely to be successful. These findings have implications for the design of mosquito surveillance for biosecurity.     

Application of rapid enzyme assay techniques for monitoring of microbial water quality

Rapid enzyme assay techniques based on direct measurement of beta-d-galactosidase (GALase) or beta-d-glucuronidase (GLUase) activity without selective cultivation are used for rapid estimation of the level of coliform bacteria and Escherichia coli in water samples. Reported detection limits using fluorogenic substrates correspond to culturable target bacteria concentrations that can be appropriate within present guidelines for recreational waters. The rapidity, that is detection within one hour, compromises the specificity of the assay; enzyme activity contributions from other than target bacteria need to be considered, particularly at low levels of target bacteria. Enzyme activities are more persistent than the culturability of target bacteria to environmental and disinfection stress, thus water samples may express enzyme activities of both culturable and viable non-culturable cells.     

Risk-targeted selection of agricultural holdings for post-epidemic surveillance: estimation of efficiency gains

Current post-epidemic sero-surveillance uses random selection of animal holdings. A better strategy may be to estimate the benefits gained by sampling each farm and use this to target selection. In this study we estimate the probability of undiscovered infection for sheep farms in Devon after the 2001 foot-and-mouth disease outbreak using the combination of a previously published model of daily infection risk and a simple model of probability of discovery of infection during the outbreak. This allows comparison of the system sensitivity (ability to detect infection in the area) of arbitrary, random sampling compared to risk-targeted selection across a full range of sampling budgets. We show that it is possible to achieve 95% system sensitivity by sampling, on average, 945 farms with random sampling and 184 farms with risk-targeted sampling. We also examine the effect of ordering samples by risk to expedite return to a disease-free status. Risk ordering the sampling process results in detection of positive farms, if present, 15.6 days sooner than with randomly ordered sampling, assuming 50 farms are tested per day.     

An appraisal of PCR-based technology in the detection of Mycobacterium tuberculosis

Tuberculosis is an under-recognized yet catastrophic health problem, particularly in developing countries. The HIV pandemic has served to increase the number of susceptible individuals, and multidrug-resistance and poor socioeconomic conditions also augment the prevalence and the consequences of the disease. To control the disease and its spread, it is vital that tuberculosis diagnostics are accurate and rapid. Whereas microscopy and culture have several limitations (low sensitivity is a problem for the former, while the latter has a delayed turnaround time), PCR-based techniques targeting regions of the Mycobacterium tuberculosis genome such as IS6110 have proved to be useful. The purpose of this review is to assess the use of PCR-RFLP, nested PCR and real-time PCR protocols and the choice of target regions for the detection of M. tuberculosis. Real-time PCR for the detection of M. tuberculosis target genes in clinical specimens has contributed to improving diagnosis and epidemiologic surveillance in the past decade. However, targeting one genome sequence such as IS6110 may not by itself be sufficiently sensitive to reach 100% diagnosis, especially in the case of pulmonary tuberculosis. Additional testing for target genome sequences such as hsp65 seems encouraging. An interesting approach would be a multiplex real-time PCR targeting both IS6110 and hsp65 to achieve comprehensive and specific molecular diagnosis. This technology needs development and adequate field testing before it becomes the acceptable gold standard for diagnosis.     

Causality in Scale Space as an Approach to Change Detection

Kernel density estimation and kernel regression are useful ways to visualize and assess the structure of data. Using these techniques we define a temporal scale space as the vector space spanned by bandwidth and a temporal variable. In this space significance regions that reflect a significant derivative in the kernel smooth similar to those of SiZer (Significant Zero-crossings of derivatives) are indicated. Significance regions are established by hypothesis tests for significant gradient at every point in scale space. Causality is imposed onto the space by restricting to kernels with left-bounded or finite support and shifting kernels forward. We show that these adjustments to the methodology enable early detection of changes in time series constituting live surveillance systems of either count data or unevenly sampled measurements. Warning delays are comparable to standard techniques though comparison shows that other techniques may be better suited for single-scale problems. Our method reliably detects change points even with little to no knowledge about the relevant scale of the problem. Hence the technique will be applicable for a large variety of sources without tailoring. Furthermore this technique enables us to obtain a retrospective reliable interval estimate of the time of a change point rather than a point estimate. We apply the technique to disease outbreak detection based on laboratory confirmed cases for pertussis and influenza as well as blood glucose concentration obtained from patients with diabetes type 1.     

Monitoring and surveillance for rare health-related events: a review from the veterinary perspective

Disease monitoring and surveillance systems (MOSSs) have become one of the major components of veterinary activity. Such systems are used to assess the existing levels of prevalence, the effectiveness of control programmes and, after disease eradication, to document the continued absence of disease from a given region or zone. With decreasing disease or infection prevalence, traditional approaches become less reliable and increasingly costly. The objective of this work was to summarize and discuss methodological issues related to veterinary (animal health) MOSSs. There are considerable inconsistencies in the use of the terms 'monitoring' and 'surveillance'. Passive as well as active MOSS have their disadvantages when used for rare health-related events such as emerging and re-emerging diseases. There is a need for evaluation and improvement of these approaches. Integrated systems that call for the use of several parallel surveillance activities seem to be the favoured approach, and analytical methods to combine MOSS data from various sources into a population prevalence, or probability of disease freedom, are under development. The health and safety of the animal and human generations depends on our continuous ability to detect, monitor and control newly emerging or re-emerging livestock diseases and zoonoses rapidly. Uniform surveillance definitions, sound scientifically based approaches that use the resources and data available, and a pool of researchers and veterinary public health officials with sufficient training in epidemiology, are critically important to handle this challenging task.     

Developing noninvasive methodologies to assess koala population health through detecting Chlamydia from scats

Wildlife diseases are a recognized driver of global biodiversity loss, have substantial economic impacts, and are increasingly becoming a threat to human health. Disease surveillance is critical but remains difficult in the wild due to the substantial costs and potential biases associated with most disease detection methods. Noninvasive scat surveys have been proposed as a health monitoring methodology to overcome some of these limitations. Here, we use the known threat of Chlamydia disease to the iconic, yet vulnerable, koala Phascolarctos cinereus to compare three methods for Chlamydia detection in scats: multiplex quantitative PCR, next generation sequencing, and a detection dog specifically trained on scats from Chlamydia‐infected koalas. All three methods demonstrated 100% specificity, while sensitivity was variable. Of particular interest is the variable sensitivity of these diagnostic tests to detect sick individuals (i.e., not only infection as confirmed by Chlamydia‐positive swabs, but with observable clinical signs of the disease); for koalas with urogenital tract disease signs, sensitivity was 78% with quantitative PCR, 50% with next generation genotyping and 100% with the detection dog method. This may be due to molecular methods having to rely on high‐quality DNA whereas the dog most likely detects volatile organic compounds. The most appropriate diagnostic test will vary with disease prevalence and the specific aims of disease surveillance. Acknowledging that detection dogs might not be easily accessible to all, the future development of affordable and portable “artificial noses” to detect diseases from scats in the field might enable cost‐effective, rapid and large‐scale disease surveillance.     

The Limits of Test-Based Scrapie Eradication Programs in Goats

Small ruminant post-mortem testing programs were initially designed for monitoring the prevalence of prion disease. They are now considered as a potential alternative to genetic selection for eradicating/controlling classical scrapie at population level. If such policy should be implemented, its success would be crucially dependent on the efficiency of the surveillance system used to identify infected flocks. In this study, we first determined the performance of post-mortem classical scrapie detection in eight naturally affected goat herds (total n = 1961 animals) according to the age at culling. These results provided us with necessary parameters to estimate, through a Monte Carlo simulation model, the performance of scrapie detection in a commercial population. According to this model, whatever the number of tests performed, post mortem surveillance will have limited success in identifying infected herds. These data support the contention that scrapie eradication programs relying solely on post mortem testing in goats will probably fail. Considering the epidemiological and pathological similarities of scrapie in sheep and goats, the efficiency of scrapie surveillance in both species is likely to be similar.     

Streamlining 'search and destroy': cost-effective surveillance for invasive species management

Invasive species surveillance has typically been targeted to where the species is most likely to occur. However, spatially varying environmental characteristics and land uses may affect more than just the probability of occurrence. Biodiversity or economic value, and the ease of detection and control are also likely to vary. We incorporate these factors into a detection and treatment model of a low-density invader to determine the surveillance strategy that minimizes expected management costs. Sites with a high probability of invader occurrence and great benefits associated with detection warrant intensive surveillance; however, the optimum investment is a nonlinear function of these factors. Environments where the invader is relatively easy to detect are prioritized for surveillance, although only a moderate investment is necessary to ensure a high probability of detection. Intensive surveillance effort may be allocated to other sites if the probability of occurrence, budget and/or expected benefits is sufficiently high.     

Testing the Independence of Two Diagnostic Tests

Consider two diagnostic procedures having binary outcomes. If one of the tests results in a positive finding, a more definitive diagnostic procedure will be administered to establish the presence or absence of a disease. The use of both tests will improve the overall screening sensitivity when the two tests are independent, compared with employing two tests that are positively correlated. We estimate the correlation coefficient of the two tests and derive statistical methods for testing the independence of the two diagnostic procedures conditional on disease status. The statistical tests are used to investigate the independence of mammography and clinical breast exams aimed at establishing the benefit of early detection of breast cancer. The data used in the analysis are obtained from periodic screening examinations of three randomized clinical trials of breast cancer screening. Analysis of each of these trials confirms the independence of the clinical breast and mammography examinations. Based on these three large clinical trials, we conclude that a clinical breast exam considerably increases the overall sensitivity relative to screening with mammography alone and should be routinely included in early breast cancer detection programs.     

Modelling tropical fire ant (<Emphasis Type="Italic">Solenopsis geminata</Emphasis>) dynamics and detection to inform an eradication project

Invasive species threaten endangered species worldwide and substantial effort is focused on their control. Eradication projects require critical resource allocation decisions, as they affect both the likelihood of success and the overall cost. However, these complex decisions must often be made within data-poor environments. Here we develop a mathematical framework to assist in resource allocation for invasive species control projects and we apply it to the proposed eradication of the tropical fire ant (Solenopsis geminata) from the islands of Ashmore Reef in the Timor Sea. Our framework contains two models: a population model and a detection model. Our stochastic population model is used to predict ant abundance through time and allows us to estimate the probability of eradication. Using abundance predictions from the population model, we use the detection model to predict the probability of ant detection through time. These models inform key decisions throughout the project, which include deciding how many baiting events should take place, deciding whether to invest in detector dogs and setting surveillance effort to confirm eradication following control. We find that using a combination of insect growth regulator and toxins are required to achieve a high probability of eradication over 2 years, and we find that using two detector dogs may be more cost-effective than the use of lure deployment, provided that they are used across the life of the project. Our analysis lays a foundation for making decisions about control and detection throughout the project and provides specific advice about resource allocation.     

Sampling Considerations for Disease Surveillance in Wildlife Populations

Abstract Disease surveillance in wildlife populations involves detecting the presence of a disease, characterizing its prevalence and spread, and subsequent monitoring. A probability sample of animals selected from the population and corresponding estimators of disease prevalence and detection provide estimates with quantifiable statistical properties, but this approach is rarely used. Although wildlife scientists often assume probability sampling and random disease distributions to calculate sample sizes, convenience samples (i.e., samples of readily available animals) are typically used, and disease distributions are rarely random. We demonstrate how landscape-based simulation can be used to explore properties of estimators from convenience samples in relation to probability samples. We used simulation methods to model what is known about the habitat preferences of the wildlife population, the disease distribution, and the potential biases of the convenience-sample approach. Using chronic wasting disease in free-ranging deer (Odocoileus virginianus) as a simple illustration, we show that using probability sample designs with appropriate estimators provides unbiased surveillance parameter estimates but that the selection bias and coverage errors associated with convenience samples can lead to biased and misleading results. We also suggest practical alternatives to convenience samples that mix probability and convenience sampling. For example, a sample of land areas can be selected using a probability design that oversamples areas with larger animal populations, followed by harvesting of individual animals within sampled areas using a convenience sampling method.     

A Comparison of Methods for Estimating Northern Bobwhite Covey Detection Probabilities

Abstract: We compared the time-of-detection and logistic regression methods of estimating probability of detection for northern bobwhite (Colinus virginianus) coveys. Both methods are unusual in that they allow estimation of the total probability of detection (i.e., the product of the probability that a covey is available for detection [i.e., that a covey vocalizes] and detection given availability). The logistic regression method produced an average detection probability of 0.596 (SE = 0.020) and the time-of-detection method produced a detection probability estimate of 0.540 (SE = 0.086), and the 2 estimates were not significantly different. This is the first evaluation of the time-of-detection method with empirical field data. Although the time-of-detection and logistic regression method each have advantages, both can be used under appropriate conditions to improve estimates of bobwhite abundance by allowing for the estimation of detection probabilities. Improved estimates of bobwhite abundance will allow land managers to make more informed management decisions.     

Accuracy of using visual identification of white sharks to estimate residency patterns

Determining the residency of an aquatic species is important but challenging and it remains unclear what is the best sampling methodology. Photo-identification has been used extensively to estimate patterns of animals' residency and is arguably the most common approach, but it may not be the most effective approach in marine environments. To examine this, in 2005, we deployed acoustic transmitters on 22 white sharks (Carcharodon carcharias) in Mossel Bay, South Africa to quantify the probability of detecting these tagged sharks by photo-identification and different deployment strategies of acoustic telemetry equipment. Using the data collected by the different sampling approaches (detections from an acoustic listening station deployed under a chumming vessel versus those from visual sightings and photo-identification), we quantified the methodologies' probability of detection and determined if the sampling approaches, also including an acoustic telemetry array, produce comparable results for patterns of residency. Photo-identification had the lowest probability of detection and underestimated residency. The underestimation is driven by various factors primarily that acoustic telemetry monitors a large area and this reduces the occurrence of false negatives. Therefore, we propose that researchers need to use acoustic telemetry and also continue to develop new sampling approaches as photo-identification techniques are inadequate to determine residency. Using the methods presented in this paper will allow researchers to further refine sampling approaches that enable them to collect more accurate data that will result in better research and more informed management efforts and policy decisions.