|
|
Validation and assessment of diagnostic
protocols |
|
Although investigation of serology
has indicated that antibody titres against BDD-associated Treponema spp.
are elevated in BDD-affected animals compared to nondiseased animals, case
definition is currently limited to clinical inspection of feet.
Although efforts have been made to standardize the description of
lesions (Döpfer and Willemen, 1998), the variability in presentation is
such that most authors in the published literature have applied their own
system.
During the observational field studies, assessment of clinical BDD
status was performed using a modified borescope, which negated having to lift
the feet. This was a time- and labour-saving device. To investigate its
performance, the agreement between observation with the borescope and the
corresponding lesion inspection of the lifted feet, which was considered the
'Gold Standard' for clinical BDD status, was determined using Cohens
kappa statistic. Agreement was found to be excellent.
On the basis of our serologic results, and in the
absence of other detailed information on the underlying infection processes
which eventually result in clinical manifestation of BDD lesions, we assumed
that animals were exposed to the causative organisms and were infected prior to
clinical disease becoming manifest. Under this assumption, clinical inspection
is not a Gold Standard diagnostic test and such a test is
currently not available.
We
therefore defined infection with BDD-associated treponemes as a latent
variable, meaning that given the current tools at our disposal, an
animals true infection status cannot be directly measured. We
investigated whether serology was applicable as an additional, possibly more
sensitive and specific, diagnostic criterion. This was performed using Bayesian
techniques; we chose not to dichotomize the test outcome to prevent the
inherent loss of information, following the example of Choi et al. (2006).
Rather, we estimated the conditional probability of infection (CPI) of an
animal given its serologic test result and BDD lesion status. This was very
similar regardless of whether disease was present or absent; the model produced
a set of curves which, assuming that the study population was representative,
can be used predictively in other studies.
|
|
|
Fig. W4. Probability of infection plot as a function of serology,
given the data (i.e. Pr (I | S, data)). Percentage positive is the ELISA
optical density result expressed as a percentage of the plate positive
reference sample. The blue dashed line represents the probability of infection
of lesion negative animals; the dashed black line represents that of lesion
positive animals. The red line represents the probability of infection of all
animals, regardless of lesion status. The grey points represent the serologic
test result and corresponding probability of infection for every individual
animal |
|
The relationship between the humoral response and
clinical disease is complex, and probably relies on pathogenic and immunogenic
mechanisms we have not yet uncovered. The large overlap in the serological
distributions of non-diseased and diseased animals implies that substantial
proportions of BDD-negative animals have high titres, or conversely, that
BDD-positive animals have low titres. Our data indicated that this was
predominantly due to the former, i.e. animals with high titres that had
regressing lesions or that were clinically negative. These animals were
classified as infected by the Bayesian model, and were considered to be exposed
in the deterministic mathematical model.
Basing our case definition on
clinical status as well as serology allowed us to improve the case definition
of BDD, and formulate different models to assess infection status and dynamics.
It is clear that while serology has great potential for prospective
epidemiological study of BDD, interpretation is not straightforward; further
study and refinement of the models developed here is required before they can
be practically applied. |
|
|
|
|
©
Willem Daniel Vink 2006 |