Estimating the burden of multiple endemic diseases and health conditions using Bayes’ Theorem: A conditional probability model applied to UK dairy cattle

Phillip Rasmussen, University of Zurich, has recently had his comorbidity paper published online. You can find this here Phil is one of our Postdoctoral Fellows in the section of Epidemiology.”

The Global Burden of Animal Diseases (GBADs) is an international collaboration aiming, in part, to measure and
improve societal outcomes from livestock. One GBADs objective is to estimate the economic impact of endemic
diseases in livestock. However, if individual disease impact estimates are linearly aggregated without consideration for associations among diseases, there is the potential to double count impacts, overestimating the total
burden. Accordingly, the authors propose a method to adjust an array of individual disease impact estimates so
that they may be aggregated without overlap. Using Bayes’ Theorem, conditional probabilities were derived from
inter-disease odds ratios in the literature. These conditional probabilities were used to calculate the excess
probability of disease among animals with associated conditions, or the probability of disease overlap given the
odds of coinfection, which were then used to adjust disease impact estimates so that they may be aggregated. The
aggregate impacts, or the yield, fertility, and mortality gaps due to disease, were then attributed and valued,
generating disease-specific losses. The approach was illustrated using an example dairy cattle system with input
values and supporting parameters from the UK, with 13 diseases and health conditions endemic to UK dairy
cattle: cystic ovary, disease caused by gastrointestinal nematodes, displaced abomasum, dystocia, fasciolosis,
lameness, mastitis, metritis, milk fever, neosporosis, paratuberculosis, retained placenta, and subclinical ketosis.
The diseases and conditions modelled resulted in total adjusted losses of £ 404/cow/year, equivalent to herdlevel losses of £ 60,000/year. Unadjusted aggregation methods suggested losses 14–61% greater. Although
lameness was identified as the costliest condition (28% of total losses), variations in the prevalence of fasciolosis,
neosporosis, and paratuberculosis (only a combined 22% of total losses) were nearly as impactful individually as
variations in the prevalence of lameness. The results suggest that from a disease control policy perspective, the
costliness of a disease may not always be the best indicator of the investment its control warrants; the costliness
rankings varied across approaches and total losses were found to be surprisingly sensitive to variations in the
prevalence of relatively uncostly diseases. This approach allows for disease impact estimates to be aggregated
without double counting. It can be applied to any livestock system in any region with any set of endemic diseases,
and can be updated as new prevalence, impact, and disease association data become available. This approach
also provides researchers and policymakers an alternative tool to rank prevention priorities.

Leave a Reply