Abstract
Bayesian network (BN) modeling is a rich and flexible analytical framework capable ofelucidating complex veterinary epidemiological data. It is a graphical modeling techniquethat enables the visual presentation of multi-dimensionalresults while retaining statisticalrigor in population-level inference. Using previously published case study data aboutfeline calicivirus (FCV) and other respiratory pathogens in cats in Switzerland, a full BNmodeling analysis is presented. The analysis shows that reducing the group size andvaccinating animals are the two actionable factors directly associated with FCV statusand are primary targets to control FCV infection. The presence of gingivostomatitis andMycoplasma felisis also associated with FCV status, but signs of upper respiratorytract disease (URTD) are not. FCV data is particularly well-suited to a network modelingapproach, as both multiple pathogens and multiple clinicalsigns per pathogen areinvolved, along with multiple potentially interrelated risk factors. BN modeling is aholistic approach—all variables of interest may be mutuallyinterdependent—whichmay help to address issues, such as confounding and collinear factors, as well as todisentangle directly vs. indirectly related variables. Weintroduce the BN methodology asan alternative to the classical uni- and multivariable regression approaches commonlyused for risk factor analyses. We advise and guide researchers about how to use BNsas an exploratory data tool and demonstrate the limitationsand practical issues. Wepresent a step-by-step case study using FCV data along with all code necessary toreproduce our analyses in the open-source R environment. Wecompare and contrastthe findings of the current case study using BN modeling with previous results thatused classical regression techniques, and we highlight newpotential insights. Finally,we discuss advanced methods, such as Bayesian model averaging, a common way ofaccounting for model uncertainty in a Bayesian network context.
Kratzer, Gilles