Using Bayesian inference to research wildlife reservoirs of bTB
Bovine tuberculosis (bTB) infections—caused by zoonotic bacteriaMycobacterium bovis— are globally relevant, difficult to
control, and scrutinised by disease ecologists across many host species.
Research on mammals maintaining bTB reservoirs over wildlife-livestock
boundaries dominate the literature, is high-profile and economically
important. Yet evidence indicating that no capable bTB host
should be disregarded in the search for bTB controls suggests that the
current body of research on bTB infection reflects a tradeoff between
economic and ecological importance. It is possible that Bayesian
approaches could help bridge the data gaps between less and well-studied
bTB hosts by enabling information on host ecology from non-disease
studies to inform future epidemiological models.
In badger bTB research, a better understanding of transmissions within
and among badger reservoirs, as well as between badgers and cattle, or
other non-reservoir host species, is required. Like all disease systems,
the understanding of the badger-bTB system is constantly shifting with
new pieces of information, which can act to better inform priors with
expert knowledge and improve our beliefs. For example, the rapid
serological Dual-Path Platform VetTB test has recently been validated
for bTB testing in badgers and a new badger behaviour called
super-ranging has been detected, which is potentially responsible for
long-distance bTB transmissions . Consequently, these specific pieces of
information could help provide updated estimates of disease
transmission and disease progression within a badger bTB system .
Although the number of “how-to” papers describing the power of
Bayesian inference in the context of wildlife epidemiology is increasing
, research incorporating Bayesian modelling strategies specifically
interested in the ante-mortem badger bTB system is limited.
Consequently, we are only able to only touch on six specific examples of
this following a comprehensive Web of Science search using combinations
of the terms: “Bayesian”; “bovine”; “tuberculosis”; “model”;
“badger”; “bTB”; and “testing”. Of these six publications , all
used Bayesian methods to explore badger bTB transmissions on a
“landscape-scale”, here defined as the geographical extent of
Woodchester Park, Gloucestershire, UK, where the capture-mark-recapture
data common to all six studies was collected. We consider that in the
“data-poor” environment of wildlife disease modelling, our
understanding of host-pathogen systems over a diversity of geographical
scales is particularly limited.
In South Island, New Zealand—where brushtail possum (Trichosurus
vulpecula ) were speculated to be the keystone reservoir species of bTB
for ca . three decades (Trichosurus vulpecula ) —recent
Bayesian research has provided confirmation that its possum population
is responsible for South Island’s bTB maintenance; rather than its
cattle population. In the UK, although it has been confirmed via
Bayesian Integrated Population Models why Woodchester Park
badgers are an efficient bTB reservoir , the directionality of bTB
transmissions between badgers and cattle remains debated, and it is
suspected that badgers are responsible for roughly half of bTB
infections in cattle within high cattle-bTB incidence areas . Another
analysis concluded that badger to cattle transmissions were
~10.4 times more frequent than vice versa . The
geographic scale of study is just one aspect of ecological hierarchy
which requires consideration within a whole-system model. Yet until we
have whole-system modelling capabilities, we must be careful with
general comparisons between studies over varying geographic, spatial,
temporal and or statistical scales because they inevitably suffer from
the change-of-support problem : they all require complex inference of
variables at values that have not been observed. One solution to the
change-of-support problem is Bayesian hierarchical modelling.
A whole-system model of the bTB systems, capable of linking information
throughout an ecological hierarchy, is required, and an example of what
this model may look like, is presented in Figure 1. A particular
limitation in the development of such a model is the ability to
incorporate individual badger heterogeneities; individual traits are
often ignored in disease models since detailed longitudinal datasets of
individuals—such as the Woodchester Park dataset—within diseased
populations are rare.
Within the badger bTB system, complex interactions among individual
badger heterogeneities —such as between sex, inbreeding, disease and
ageing —and among-individual variation in traits, act as proxies for
infectiousness or “risk” , and are thought to drive fine-scale bTB
dynamics. Fundamentally, an understanding of fine-scale disease
processes in combination with Bayesian methodologies arms ecologists
with the ability to parameterise previously unobservable processes, such
as actuarial senescence . In addition to Hudson’s study, Bayesian
analyses of the Woodchester badger population have contributed knowledge
on the increased susceptibility of male badgers to bTB and on the
diagnostic accuracies of badger bTB tests : information which improves
our capability to model badger heterogeneities in the future. The
importance of latent variables and individual heterogeneities to a
“whole-system” model of wildlife disease is further discussed in the
final section of this review.
Overall, the idea of achieving a better understanding of the bTB system
in wildlife hosts using a whole-system method is not a new one.
Analogous to the “whole-system” model suggested within this review,
proposed the need for a novel modelling framework, and recommended a
comprehensive epidemiological model. In addition, suggested that
combining contact networks with Bayesian inference is the future
direction for understanding wildlife epidemics. The inclusion of a
hierarchy of scale within whole-system ecological models in general have
been recommended by several authors .