Bayesian Network Modeling Applied to Feline Calicivirus Infection Among Cats in Switzerland

Abstract

Bayesian network (BN) modeling is a rich and flexible analytical framework capable of elucidating complex veterinary epidemiological data. It is a graphical modeling technique that enables the visual presentation of multi-dimensional results while retaining statistical rigor in population-level inference. Using previously published case study data about feline calicivirus (FCV) and other respiratory pathogens in cats in Switzerland, a full BN modeling analysis is presented. The analysis shows that reducing the group size and vaccinating animals are the two actionable factors directly associated with FCV status and are primary targets to control FCV infection. The presence of gingivostomatitis and Mycoplasma felis is also associated with FCV status, but signs of upper respiratory tract disease (URTD) are not. FCV data is particularly well-suited to a network modeling approach, as both multiple pathogens and multiple clinical signs per pathogen are involved, along with multiple potentially interrelated risk factors. BN modeling is a holistic approach-all variables of interest may be mutually interdependent-which may help to address issues, such as confounding and collinear factors, as well as to disentangle directly vs. indirectly related variables. We introduce the BN methodology as an alternative to the classical uni-and multivariable regression approaches commonly used for risk factor analyses. We advise and guide researchers about how to use BNs as an exploratory data tool and demonstrate the limitations and practical issues. We present a step-by-step case study using FCV data along with all code necessary to reproduce our analyses in the open-source R environment. We compare and contrast the findings of the current case study using BN modeling with previous results that used classical regression techniques, and we highlight new potential insights. Finally, we discuss advanced methods, such as Bayesian model averaging, a common way of accounting for model uncertainty in a Bayesian network context.

Biosecurity risk factors for highly pathogenic avian influenza (H5N8) virus infection in duck farms, France

“…The search can however be guided by incorporating causal knowledge to limit the amount of computations (de Campos & Castellano, 2007). Moreover, ABN face challenges similar to classical regression frameworks, such as those related to trade‐off between fit and complexity (Lewis & McCormick, 2012; Lewis & Ward, 2013; Kratzer et al., 2020). The network score (Bayesian log marginal likelihood), that is used to select the best model, was based on very weakly informative priors and did assume additive effect of the parent on the link function scale.…”

“…First, an exact search was run using a parent limit constraint of 1 (a maximum of one arc pointing to each node). This process was then repeated until increasing the parent limit did not result in a model with an improved goodness of fit metric (Bayesian log marginal likelihood, also called network score, that represents an approximation of the Bayesian Information Criteria (Kratzer et al., 2020)). A uniform prior distribution was used for the network structure meaning that all eligible model structures were assumed equally plausible regardless of their complexity.…”

Calicivirus Infection in Cats

“…The prevalence is broadly proportional to the number of cats in a household. It is lowest in healthy household cats kept in small groups of fewer than four cats (2.5%), and higher in groups of healthy cats of four or more cats (32%) [ 2 , 44 ]. The prevalence within individual colonies and shelters is variable, ranging from low to high (50–90%) values [ 11 , 41 , 43 , 45 , 46 ].…”

Additive Bayesian network analysis of the relationship between bovine respiratory disease and management practices in dairy heifer calves at pre-weaning stage

“…Although the directed graph gives us an impression of the causal web, DAG derived from ABN analysis merely describes the statistical relationships between variables in observational data; therefore, the results should be interpreted carefully with expert knowledge and biological understanding to examine causal relationships [10]. Nevertheless, as the ABN analysis helps to disentangle the interrelationships between multiple dependent variables, it is useful for analysing disease risk factors [14][15][16].…”

“…ABN analysis provides a graphical representation of the structure of the associations between all variables of a model, which allows us to distinguish indirect and direct associations between variables [ 12 ]. While a causal diagram can be constructed using prior knowledge [ 13 ], ABN construction is data-driven [ 14 ], making it appropriate for the analysis of interrelating factors for which there is limited information. Although the directed graph gives us an impression of the causal web, DAG derived from ABN analysis merely describes the statistical relationships between variables in observational data; therefore, the results should be interpreted carefully with expert knowledge and biological understanding to examine causal relationships [ 10 ].…”