3.3 Discussion concerning causal inference
Establishing causal inference requires a comprehensive evaluation of the
available literature and appropriate interpretation of that literature
that enables either to establish evidence of an association or to
identify what research is missing so that gaps can be filled. Astrovirus
is an emerging virus in swine production and there is a hypothesis that
it may be causally related to neurological diseases. Establishing
causation is difficult and discussed extensively by Awadh etal., 2017 and Frank et al., 2016 for Zika virus and an
excellent example of how research evidence can be used to “build” a
case for causation is provided by Waddell et al., 2016. To assess
the potential of PoAstVs to cause disease in swine, the Bradford Hill
guidelines were used, collecting and organizing the epidemiological
evidence in a structured fashion:
1. Strength of association:
The strength of association is the statistical or clinical significance
of the association. The important component to determine strength of
association including case definition, sample size and statistical
power. The stronger the association, the more likely the relationship
between risk factor and outcome is to be causal. For strength of
association, we included studies that calculated measures of association
for PoAstVs (Figure 2).
The magnitude of odds ratio is a measure of “strength of association”.
Forty-three percent (6 of 14) of the case-control studies had OR
>1 (1.24-36) (Boros et al., 2017; Shan et al., 2011; Cai et
al., 2016; Monini et al., 2015; Kumthip et al., 2018; Xiao et al., 2017)
indicating the odds of detection of PoAstVs among cases were greater
than odds of detection of PoAstVs among controls. The remaining 57% (8
of 14) (Blomström et al., 2010; Xiao et al., 2013; Goecke et al., 2017;
Zhou et al., 2016; Ito et al., 2017; Salamunova et al., 2018; Lee et
al., 2013; Wallgren et al., 2014) had OR < 1, indicating that
odds of detection of PoAstVs among cases were lower than odds of
detection of PoAstVs among controls. The inference from these studies
would be that detection of PoAstVs might be the protective factor
against the disease. Hence, the P-value of heterogeneity was low
(P < 0.01) and the random effect model resulted in an
overall OR of 0.95 (95% CI: 0.65, 1.39) indicating a poor association
between PoAstV detection and disease condition. Hence, this criterion
was not met.
It is noteworthy to discuss the findings from Boros et al.(2017), who reported OR of 36 strongly associating detection of PoAstV3
with neurologic disease in nursery pigs. The cases were defined as pigs
showing clinical signs including posterior leg weakness or paraplegia
and pitching (stage 1); later paralysis of both legs and skin pain
(stage 2); or loss of consciousness, paresis, and serious flaccid
paralysis of muscles (stage 3) without any gastroenteric symptoms. The
controls were asymptomatic pigs. There was difference in the time of
collection samples from cases and controls. Out of n = 5 cases, (n = 2
in March 2016, n = 2 in July 2016 and n = 1 in November 2015) and out of
n = 5 controls, (n = 1 in July 2016 and n = 4 in June 2017). Four of the
cases were 25 days old and one was 35 days old. On other hand three of
the cases were 25 days old and two of them were 35 days old. Although
the cases were selected objectively based on the severity of the
disease, the sample size used was small which impairs representativeness
of cases and controls. Moreover, the wide 95% CI (1.80 to 718.68)
indicated a low level of precision of the reported OR in that study.
2. Consistency of the association:
If the association of PoAstVs and disease outcomes is causal, we would
expect that PoAstVs detection would consistently be associated with
disease incidence. In situations where we are not able to measure
incidence, we would expect consistent reporting of OR greater than one,
which was not the case here as shown in (Figure 2) the forest plot.
Hence, this criterion was partially met.
3. Specificity:
Originally, when Sir Bradford Hill proposed the criterion for
specificity, he referred to a single cause (or exposure) leading to a
single effect, and vice-versa. However, this criterion is rarely
considered these days because it is known that it does not always hold
true even for most pathogens. For example, it is well established that
PRRSV may cause more than one disease syndrome (respiratory,
reproductive), and there are other pathogens causing similar clinical
conditions (i.e., Influenza virus A, Mycoplasma hyopneumoniae ).
Thus, for infectious diseases specificity would be very rarely
considered a factor.
A causal relationship between PoAstVs detection and gastroenterologic or
neurologic disease cannot be specific, because other causes for
gastroenterologic or neurologic disease may not be excluded including
Rotavirus A, B and C, TGEV, PEDV, PCV2, PTV, PSV, APPV, and others.
Therefore, as later recognized by Sir Bradford Hill, this criterion
should be assessed in combination with the strength of association.
Hence, this criterion was not met.
4. Temporality:
For an exposure-disease relationship to be causal, exposure must precede
the onset of disease. This is a fundamental criterion to postulating a
cause-and-effect relationship and fits an intuitive understanding of
causality. Cross-sectional studies or surveys measure both the exposure
and outcome in a sample of the population at a point in time. It allows
reporting the prevalence of disease in the population being studied but
it is not possible to know whether the exposure preceded the effect.
Hence, cross-sectional studies help to produce causal hypotheses but
cannot prove causality.
Case-control studies may reveal associations, but they do not
irrefutably demonstrate causation. The temporal relationship between the
supposed cause and effect cannot be determined by a case-control study.
Hence, all the relevant papers were either cross-sectional or
case-control studies and thus this criterion was not met.
5. Biological gradients:
Also known as dose-response relationship. It refers to the incremental
changes in disease rates with changes in exposure. This would be either
document in a prospective cohort study or challenge study which was not
the case here. Hence, this criterion was not applicable for this scoping
review.
6. Plausibility:
There must be a biological mechanism explaining how the exposure causes
the outcome. There is limited knowledge regarding pathogenesis of
PoAstVs in general. Although neuroinvasive ability has been seen in
PoAstV3 (Arruda et al., 2017; Boros et al., 2017) with the detection of
the virus in lesions of the brain and the spinal cord by using histology
and in situ hybridization. Hence, this criterion was met.
7. Coherence:
The cause-and-effect interpretation of data should not seriously
conflict with generally known facts of natural history and biology of
disease. Six of the case-control studies having OR > 1
(1.24-3.25) (Shan et al., 2011; Cai et al., 2016; Monini et al., 2015;
Kumthip et al., 2018; Xiao et al., 2017) were associated with
gastroenterologic diseases but were not able to back up the association
with further diagnostic evidence including histologic evidence of
disease and ruling out presence or absence of bacterial cause of
disease. Hence, this criterion was met.
8. Experimental evidence:
This required data from animal models or natural experiment on the
population level. Possible reasons regarding the lack of experimental
evidence for PoAstVs include the difficulty in isolation of virus,
limited knowledge regarding pathogenesis with the exception of PoAstV3
(MAstV22) which was recently associated with outbreaks of
polioencephalomyelitis in swine in the United States (Arruda et al.,
2017; Matias et al., 2018; Rawal et al., 2019d) and Hungary (Boros et
al., 2017) and there is always financial constraint there with
experimental studies as it is relatively expensive in comparison with
other study designs. Hence, none of the relevant papers reported
experimental or challenge study, and thus this criterion was not
applicable for this scoping review.
9. Analogy:
More recently AstVs have been linked to neurological signs in different
species including human, mink, bovine, ovine, and porcine. Astroviruses
in these species are clustered together phylogenetically and defined as
the HMO clade. PoAstV3 is closely related to the HMO clade. Hence this
criterion was met.