A study in Nature scientific reports explores the use of machine learning in predicting vaccine ingestion rate among wild boars in support of efforts to control classical swine fever. Using Random Forest modelling based on vaccine dissemination data, the authors created prediction surfaces for the probability of vaccine ingestion by wild boar through “spatial interpolation techniques”. They found that the distance from the vaccination point to a water source was the “most important variable”, with areas of high probability identified in northern, eastern, and western areas in Gifu.
Classical swine fever (CSF)
The paper states that classical swine fever (CSF), caused by classical swine fever virus of the genus Pestivirus in the Flaviviridae family, is an infectious viral disease of domestic and wild pigs. The disease is “one of the most transboundary swine diseases” because of its potential to “severely impact the swine industry”.
Transmission most commonly occurs through direct contact between a “healthy susceptible host and infected animals”, but it can also be spread through the “discharge of infected animals and contaminated pork products”. Therefore, contaminated food residues, vehicles, and clothing are “important sources of indirect transmission routes”. The disease has both acute and chronic forms, and ranges from severe with high mortality to mild or no symptoms. Clinical presentations are “very similar to those of African swine fever (ASF)”.
CSF in Japan
Although Japan successfully eradicated CSF through vaccination in the twentieth century, it re-emerged in 2018. Since then, the infected area has expanded due to large outbreaks in wild boar populations and “sporadic” outbreaks at pig farms.
“The current CSF virus epidemic strain in Japan is considered moderately virulent, with a mix of individuals dying from infection and those surviving, contributing to the large-scale expansion of infection.”
Vaccination is “encouraged” across most regions, yet the control of the epidemic among wild boars “remains a major challenge”.
Wild boar vaccination
Wild boars are omnivorous and have a “wide range of food choices”; they are “widely distributed” in Japan. They dig up the ground to feed on plant roots and rhizomes but also eat acorns and insects and reptiles. A study by the Ministry of the Environment identified that the distribution range of wild boars has expanded about 1.9 times over the past 40 years, from which the authors infer that CSF could “further expand and have a more serious impact of the Japanese swine industry”.
The introduction of oral vaccination has been recognised as a “key strategy for controlling diseases” in populations where interventionist management is “more challenging compared to livestock”. In Gifu Prefecture, ongoing outbreaks have demanded efforts to control the disease “diligently”. The authors draw on insights from local hunters and wild boar experts to suggest that the oral vaccine was dispersed at around 14,000 sites over the three seasons each year from 2019 to 2020.
At each site around 20 vaccines were spread and buried in holes about 10-15cm deep with lure food. To understand the effectiveness of the vaccination application, the leftover vaccine or vaccine packet was collected 5 days after application. Vaccine ingestion was confirmed in approximately 30% of the dispersal sites.
“The expansion of infected areas and the prolonged infection period will burden the economy as will the cost of control measures. Therefore, the development of effective vaccination strategies is an urgent priority.”
Research in context
A previous study presented a generalised linear mixed model (GLMM) analysis on data from sensor cameras across around 10% of dispersal points to identify areas where the animals were most likely to appear. This revealed a “positive correlation” between the emergence of other wildlife and the emergence of wild boar. Furthermore, road density and vegetation were believed to influence wild boar emergence. However, areas of high wild boar emergence don’t necessarily coincide with areas of high vaccine feeding.
Machine learning algorithms can analyse large, complex data sets and identify patterns and trends that are harder to detect for humans. Thus, more studies are applying them to the prediction of infectious disease outbreaks. For this approach, Random Forest is a “commonly used” and “powerful” machine learning technique.
For this study the authors predicted the wild boar vaccine ingestion rate at each site based on a Random Forest model. The results of this were combined with spatial interpolation techniques to “output a prediction surface showing the vaccine ingestion probability of wild boars” in Gifu Prefecture.
“Effective vaccination strategies for wild animals can be summarised in two aspects: high efficacy of the vaccine itself and efficient vaccine delivery to each individual. This study aimed to contribute to the control of CSF in wild boars by focusing on the latter.”
The results of the model identified “important variables” such as distance to water source, elevation, season, road density, and slope, with distance to the water source “chosen as the most important variable”. The probability maps estimated “high probability areas” in the northeastern and southeastern parts of Gifu Prefecture, as well as the western region.
How does the study contribute to the progress of incorporating technological advances into vaccine research and development? For more on vaccine technologies don’t forget to get your ticket to the Congress in Washington this April or subscribe to our newsletters here.
