Optimizing surveillance for early disease detection: Expert guidance for Ostreid herpesvirus surveillance design and system sensitivity calculation
Introduction
Variants of the Ostreid herpesvirus 1 (OsHV-1), including the reference strain and microvariants, are best known as causal agents of mortality events of consequence for Crassostrea gigas (Pacific oyster) populations in many locations globally since the early 1990s (EFSA AHAW Panel, 2015; Castinel et al., 2015; Arzul et al., 2017; Burge et al., 2018; OIE, 2019). Microvariant genotypes of OsHV-1 were first identified in 2008 associated with large-scale C. gigas mortality in France (Segarra et al., 2010), and are distinguished from the reference genotype (Davison et al., 2005) by sequence variation in select regions of the genome (OIE, 2019). OsHV-1 variants have also been associated with events of lesser impact (Burge et al., 2006; Cáceres-Martínez and Vásquez-Yeomans, 2013; Bai et al., 2015). Though typically linked to large-scale losses in C. gigas, microvariants have been found in apparently healthy C. gigas, both during post-outbreak recovery (e.g., in Australia, Evans et al., 2017) and alongside mortality in other species (e.g., in Asia, Bai et al., 2015). Failure to apply efficient detection systems, as well as movement controls, contributed to dissemination throughout many affected regions (Carnegie, 2012; Carnegie et al., 2016; Fuhrmann et al., 2019). As such, early detection and timely response are critical to the goal of curtailing damage otherwise associated with rapid spread. However, the pathogen’s short incubation period and rapid transition from extremely low to high prevalence, alongside a tendency for spatiotemporal clustering of infection (Paul-Pont et al., 2013), make OsHV-1 difficult (and costly) to detect prior to a large outbreak using conventional surveillance modalities (Whittington et al., 2019).
Pathogen surveillance in aquaculture is largely built around intermittent submission of samples for laboratory testing (termed “active surveillance” when coordinated externally). While active surveillance can offer very strong evidence of disease absence at test-time, it is less well-suited for early detection of incursions that can occur at any time or location (Vennerström et al., 2017). Often designed to support translocation of stock for production or trade, active surveillance can be patchy in both space and time. Though bolstering active systems (via increased frequency, for example) is one way to improve early detection capacity, it may not be the most efficient. Rather, surveillance built to address both disease freedom and early detection objectives may need to couple periodic testing with observational systems for the most cost-effective designs (Cameron et al., 2020). Risk-based surveillance, targeting animals with the highest risk of exposure, infection, or consequence, may further improve detection capability, for both active and passive systems (Oidtmann et al., 2013).
Frequent observation of animal appearance, behavior, or mortality (here termed “observational surveillance”, also known as “passive surveillance” if led by the producer) can signal abnormalities in population health. Mortality investigations, for example, were key to the initial diagnosis of OsHV-1 μvar in France (Lupo et al., 2014). When observers are versed in disease recognition and networked with health professionals and response agencies, unexplained morbidity or mortality can trigger disease investigations, targeted laboratory diagnostics, and subsequent reporting. Similarly, the absence of reports from these same networks can provide baseline assurance of disease absence in the monitored systems if trust and mutual confidence are present between stakeholders (World Bank, 2010). However, the degree of assurance generated will vary by pathogen, population, and system, and its quantification requires an estimate of the sensitivity of observational surveillance.
Surveillance system sensitivity (SSe), the ability of a system to identify disease or infection in a population, is similar in construct to diagnostic test sensitivity (Se) but focused on populations rather than individual animals. Used in disease freedom calculations, SSe can influence decisions regarding disease control, animal movement, and trade. However, SSe for observational surveillance is notoriously difficult to estimate well, and SSe geared to early (rather than any) disease detection is rarely discussed (Cameron et al., 2020). Several methods are available, e.g., using capture-recapture (Lupo et al., 2012) or mixed qualitative/quantitative (Limon et al., 2014) analyses, to identify traits predicting observational surveillance success. However, data, time or resources required for empirical analyses of observational surveillance systems are often lacking. Therefore, SSe is more commonly derived from estimation of a series of subjective probabilities that are not necessarily data-driven: probabilities that the pathogen will cause clinical signs, the clinical signs will be noticed by a producer, the producer will report the disease, and so on. Expert elicitations, in contrast, offer a replicable mechanism to estimate epidemiologic parameters like SSe when empirical field data are limited (Gustafson et al., 2018). If structured to identify system traits that predict detection capacity, results can be used as guidance for surveillance design. Further, by replicating case-control studies and generating likelihood ratios (LRs, epidemiologic measures of association), the results are portable, informing models that compare the value of different systems or designs.
Expert elicitation accuracy is driven by several processes including whether experts have first-hand knowledge of the topic, whether experts have an opportunity to clarify the wording/meaning of identified factors, and whether elicited parameters are count rather than probability-based (Kynn, 2008; Burgman et al., 2011; McBride et al., 2012). Certain types of group processes, e.g., the integrative group process (IGP) used here, also minimize “group think” (or unjustified peer influence) by eliciting responses from individuals independently prior to group discussion (Gustafson et al., 2003). Finally, group discussion provides a forum for self-calibration as experts with specific knowledge have a chance to share experiences that may, or may not, influence independent revisions by the rest of the group.
Here we describe an expert elicitation to identify and weight factors predicting the capacity of observational surveillance for the early detection of OsHV-1. The expert-identified traits are considered key to early detection and thereby offer guidance to producers and governments aiming to improve surveillance. Results also parameterize a model to compare (1) the sensitivity of different surveillance system designs toward early detection, and (2) the confidence these systems can generate toward disease freedom. For simplicity, we focused this elicitation on the pathogen OsHV-1, regardless of genotype. However, many of the tenets, as well as the estimation process, likely generalize to other pathogens of concern for oyster production systems. Similarly, factors focus on oyster farming sites and their surveillance systems, but the resulting model should also generalize to geographic regions.
Section snippets
Terms and assumptions
We use the term “factor” throughout to describe possible predictors (risk factors or protective factors) of a surveillance system’s early detection capability, and the term “trait” to classify response or sub-categories of the identified factors. Factor describes a general concept (e.g., color) and trait describes its sub-category (e.g., blue, green, orange, etc.). We use the term “detection” to refer to the full sequence of observing, notifying, diagnosing, and reporting a case to the proper
Panelist participation
The twelve invited panelists included 3 industry representatives from the United States, and 9 shellfish health professionals from the U.S. (5), Mexico (1), Australia (1) and France (2). Two of the panelists (one from the U.S. and one from Mexico) invited an additional expert each to provide shared input, bringing the total number of experts involved in discussion and review to fourteen. The twelve core panelists completed the interview and ranking steps, all twelve completed the weighting
Discussion
As aquaculture evolves to meet global food security challenges, the resulting intensification and diversification of production systems may heighten the emergence and international spread of aquatic pathogens (Burge et al., 2014; Feist et al., 2019; King et al., 2019). Aquaculture must balance the pathogen risks inherent to natural settings with the costs and complexities of altered rearing environments that support their bioexclusion (Lafferty et al., 2015). Similarly, surveillance should
Declaration of Competing Interest
There are no conflicts of interest to declare for this article (“Optimizing surveillance for early disease detection: Expert guidance for Ostreid herpesvirus surveillance design and system sensitivity calculation”) focused on optimizing surveillance for the early detection of an oyster pathogen.
Acknowledgements
We would like to acknowledge the expertise, time, and energy of panelists (K. Humphrey, T. Sawyer) that contributed greatly to the content of the study but elected not to participate as co-authors in the presentation of results. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. This is contribution 4031 of the Virginia Institute of Marine Science, William & Mary.
References (51)
- et al.
Viruses infecting marine molluscs
J. Invertebr
(2017) - et al.
Emerging and endemic types of Ostreid herpesvirus 1 were detected in bivalves in China
J. Invertebr. Patho.
(2015) - et al.
People matter in animal disease surveillance: challenges and opportunities for the aquaculture sector
Aquaculture
(2017) - et al.
Detection of Ostreid herpesvirus-1 microvariants in healthy Crassostrea gigas following disease events and their possible role as reservoirs of infection
J. Invert. Pathol.
(2017) - et al.
Hydrographics and the timing of infectious salmon anemia outbreaks among Atlantic salmon (Salmo salar L.) farms in the Quoddy region of Maine, USA and New Brunswick, Canada
Prev. Vet. Med.
(2007) - et al.
Combining surveillance and expert evidence of viral hemorrhagic septicemia freedom: a decision theoretic approach
Prev. Vet. Med.
(2010) - et al.
Stability of Ostreid herpesvirus-1 (OsHV-1) and assessment of disinfection of seawater and oyster tissues using a bioassay
Aquaculture
(2016) - et al.
Oyster disease in a changing environment: decrypting the link between pathogen, microbiome and environment
Coast. Manage.
(2019) - et al.
Using mixed methods to investigate factors influencing reporting of livestock diseases: a case study among smallholders in Bolivia
Prev Vet Med .
(2014) - et al.
Improving early detection of exotic or emergent oyster diseases in France: identifying fators associated with shellfish farmer reporting behavior of oyster mortality
Prev. Vet. Med.
(2014)