Connectivity-based risk ranking of infectious salmon anaemia virus (ISAv) outbreaks for targeted surveillance planning in Canada and the USA

Gautam, R. and Price, D. and Revie, C.W. and Gardner, I.A. and Vanderstichel, R. and Gustafson, L. and Klotins, K. and Beattie, M. (2018) Connectivity-based risk ranking of infectious salmon anaemia virus (ISAv) outbreaks for targeted surveillance planning in Canada and the USA. Preventive Veterinary Medicine, 159. pp. 92-98. ISSN 0167-5877

[img]
Preview
Text (Gautam-etal-PVM-2018-Connectivity-based-risk-ranking-of-infectious-salmon-anaemia-virus)
Gautam_etal_PVM_2018_Connectivity_based_risk_ranking_of_infectious_salmon_anaemia_virus.pdf
Accepted Author Manuscript
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo

Download (915kB)| Preview

    Abstract

    Infectious salmon anaemia (ISA) can be a serious viral disease of farmed Atlantic salmon (Salmo salar). A tool to rank susceptible farms based on the risk of ISA virus (ISAv) infection spread from infectious farms after initial incursion or re-occurrence in an endemic area, can help guide monitoring and surveillance activities. Such a tool could also support the response strategy to contain virus spread, given available resources. We developed a tool to rank ISAv infection risks using seaway distance and hydrodynamic information separately and combined. The models were validated using 2002-2004 ISAv outbreak data for 30 farms (24 in New Brunswick, Canada and 6 in Maine, United States). Time sequence of infection spread was determined from the outbreak data that included monthly infection status of the cages on these farms. The first infected farm was considered as the index site for potential spread of ISAv to all other farms. To assess the risk of ISAv spreading to susceptible farms, the second and subsequent infected farms were identified using the farm status in the given time period and all infected farms from the previous time periods. Using the three models (hydrodynamic only, seaway-distance, and combined hydrodynamic-seaway-distance based models), we ranked susceptible farms within each time interval by adding the transmission risks from surrounding infected farms and sorting them from highest to lowest. To explore the potential efficiency of targeted sampling, we converted rankings to percentiles and assessed the model's predictive performance by comparing farms identified as high risk based on the rank with those that were infected during the next time interval as observed in the outbreak data. The overall predictive ability of the models was compared using area under the ROC curve (AUC). Farms that become infected in the next period were always within the top 65% of the rank predicted by our models. The overall predictive ability of the combined (hydrodynamic-seaway-distance based model) model (AUC = 0.833) was similar to the model that only used seaway distance (AUC = 0.827). Such models can aid in effective surveillance planning by balancing coverage (number of farms included in surveillance) against the desired level of confidence of including all farms that become infected in the next time period. Our results suggest that 100% of the farms that become infected in the next time period could be targeted in a surveillance program, although at a significant cost of including many false positives.