Available online at www.sciencedirect.

com

Preventive Veterinary Medicine 86 (2008) 244249

www.elsevier.com/locate/prevetmed

The use of epidemiology to enhance

production animal research
Catherine Dewey *
Department of Population Medicine, Ontario Veterinary College, University of Guelph,
Guelph, Ont. N1G 2W1, Canada

Abstract
The approach to understanding the impact of management and disease in production animal systems has
evolved with the advent of both routine on-farm data collection and new analytic epidemiology techniques.
Epidemiology provides a tool to describe the hostagentenvironment triad and the impact of multiple
variables on productivity and health recognized by production animal veterinarians in their day-to-day
work. Field trials enable veterinarians to systematically test whether or not a new treatment improves the
health of the animal populations in their geographic region and under their production systems. Hypothesis-
specific coding techniques, such as hierarchical variables, are used in a systematic manner to understand
well-defined biological phenomenon. Clustering at multiple levels has provided the challenges of measuring
management changes in each level. Using random effects models allow us to determine the relative
importance of each level on the dependent variable. As epidemiologists, we have taken advantage of
analytic techniques used in other fields of science. Geo-spatial statistics has been used to understand the
clustering and spread of diseases and more recently, to interpret the laboratory findings related to the
introduction of an exotic strain of the influenza virus. Dr. Martin, through his work as a veterinary
epidemiologist and that of people he has influenced, has been an international leader in promoting the
optimal health and productivity of animal populations and of ensuring the safety of foods of animal origin
and preventing animal-related disease in humans.
# 2008 Elsevier B.V. All rights reserved.

Keywords: Production animal; Epidemiology; Swine; Hierarchical; Random variables

Epidemiologists with a production animal focus use analytic tools to enhance their
understanding of biological phenomenon. Dr. Wayne Martin is an avid analytical epidemiologist.
He reads a broad range of scientific work and is willing to adopt new analytic tools to answer
veterinary questions. Epidemiologists with a production animal focus use analytic methods to

* Tel.: +1 519 824 4120x54746.

E-mail address: [email protected].

0167-5877/$ see front matter # 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.prevetmed.2008.02.010
 C. Dewey / Preventive Veterinary Medicine 86 (2008) 244249 245

understand the associations between productivity and/or disease and putative causative factors.
Often these production measures and causal factors are measured at multiple levels. This paper
will illustrate how the approach to understanding food production animal systems has changed as
analytic tools have evolved. The citations in this manuscript will be from work done by
epidemiologists at the University of Guelph although there are many others that apply.
Dr. Martin was the founding chair of the Department of Population Medicine at the University
of Guelph. The department was created to bring together faculty members who dealt with
populations of animals or people. The vision was to be international leaders in promoting the
optimal health and productivity of animal populations, ensuring the safety of foods of animal
origin and preventing animal-related disease in humans. Dr. Martin has had a huge influence on
the connection between production animal medicine and epidemiology both within and beyond
the Department of Population Medicine. Beginning with fundamental epidemiologic concepts to
more involved analytic epidemiology, I will illustrate some of this impact.
The Department of Population Medicine was populated with faculty members and graduate
students who focused on cows, sheep, chickens, pigs, fish and people. Every graduate student, no
matter their area of interest, took the general epidemiology course and most took the analytic
epidemiology course. This creates a common bond, and a common way of approaching research
questions and the practice of food animal production medicine. Dr. Martin was responsible for
teaching both epidemiology courses for many years, even after he retired! We all cut our
epidemiology teeth on Daisy, a data set based on dairy herds. Daisy taught us about metritis,
mastitis, calving to conception interval, confounding, interaction, outliers, regression and the
application of epidemiology to real life production animal problems. Now we had a real tool to
start to understand the problems in production animal medicine that we dealt with in our herd
health visits.
Epidemiology has been used to benchmark production. Drs. McDermott and Alves worked
with Dr. Martin to establish data collection and analysis of cowcalf producers in Ontario
(McDermott et al., 1991). These researchers encouraged producers to keep records and with
those, they set industry standards. This enabled producers to set and work towards attaining
production goals. In the early 1990s, many swine producers began keeping production records.
They used the information to assist with weekly management decisions, determining which sows
to move to the farrowing room and which sows to cull. With summary statistics, the producers
were then able to set production goals. These data allowed epidemiologists the opportunity to
identify individual farm problems that were reducing herd productivity. We then amalgamated
data from multiple farms to identify management factors limiting reproductive productivity on a
regional basis (Dewey et al., 1994, 1995a,b). We learned that records kept in one software
program could not be merged with those kept in another program because there were
fundamental differences in how the programs calculated production parameters. In one program
cost was determined by the number of breeding females in the herd, so producers did not record
gilts until they farrowed. This biased the farrowing rate in gilts, causing it to reach close to 100%.
We also recognized the extensive data cleaning required when we as epidemiologist wished to
use data collected for production purposes as the basis for research (Dewey et al., 1995a).
Veterinarians and researchers in production animal medicine visit the barn and observe
biological phenomena. We develop a fundamental understanding of how things work and the
impact of management and disease on production. As epidemiologists, we want to know whether
or not our assumptions are real. Is there an association between a putative management factor and
disease or production and further, what is the strength of that association? Should we encourage
producers to change what they do, and if so, what result can we anticipate? If we can show that
246 C. Dewey / Preventive Veterinary Medicine 86 (2008) 244249

there is an association using analytic epidemiology then we can use the information to promote
optimal health and productivity of animal populations, ensure the safety of foods of animal origin
and prevent animal-related disease in humans.
The hostagentenvironment triad is a fundamental epidemiologic concept that we use as
researchers in production animal medicine. As an example, we were interested to determine
whether or not Porcine circovirus type II (PCV II) and post-weaning multisystemic wasting
syndrome (PMWS) was associated with nursery pig mortality in Manitoba, Canada (Dewey
et al., 2006). We did a herd-level, casecontrol study comparing herds with high mortality to
those with low mortality. Factors that were associated with high losses included having a previous
diagnosis of PMWS, farms in-which PCV II was identified from at least one of three pigs
submitted for post-mortem examination, producers having a low minimum body weight cut-off
for weaned pigs and farms that had at least one other pig farm within 3 km of the location. This
project clearly identified an agent, clinical disease and management all associated with poor
production measures. Veterinarians can use this information to show producers that if we do not
address all three factors, we will be less successful in solving problems.
Another basic tenant of general epidemiology is the well-designed field trial. We learn how to
randomly assign animals to treatment, the importance of controls and how to blind the observer
and the person doing the analyses. Testing products on commercial farms allows us to examine
the efficacy of the drug in the face of management and disease challenges. Field trials are rarely
conducted by veterinarians in private practice, where their mandate is to find the best treatment
and use it on all animals. Those of us doing field trials will tell you that it seems easier in the
textbook description than in real life. However, our fieldwork provides the validity to use a
treatment in the field. As an example, the work by Chernysheva et al. (2003) showed the industry
that eggyolk antibody was unlikely to control post-weaning diarrhea due to Escherichia coli
even though it was used widely at the time (Chernysheva et al., 2003). Another study found that
levamisole, used at low doses enhanced the immune function of nursing piglets (Kumar et al.,
1999).
We broadened our scope from univariate to multivariate analyses. If the dependent variable is
litter size in pigs and we know that parity is associated with litter size, we can include parity as a
co-variate as we analyse the other putative factors. In one study, we determined the association
between the use of a modified live Porcine Reproductive and Respiratory Syndrome virus vaccine
and litter size, after controlling for parity. When the vaccine was first available, producers used a
variety of different vaccination protocols. This allowed us to compare sows that farrowed before
the vaccine was used in the herd to those vaccinated at different times during gestation and also to
those gestating but not vaccinated (Dewey et al., 2004). Compared to sows that farrowed before
the vaccine came into the herd, litter size in sows vaccinated early in gestation was reduced by
0.20 pigs and that of sows vaccinated later in gestation was reduced by 0.12 pigs (Dewey et al.,
2004). By analyzing data from multiple herds, some of which had experienced clinical problems
due to the PRRS virus and others that had not, we were able to advise the producers about the
most appropriate way to use the vaccine in sows. We develop a hypothesis, then we use analytic
epidemiology to test the hypothesis and finally, we go back to the field to share the results with the
producers and veterinarians.
Not only have production records been used for research but also for day-to-day work of
production animal veterinarians. From the benchmarking project in cattle to the analyses of
swine production records, it became clear that the work of veterinarians practicing food animal
production medicine would be enhanced by skills in data analysis. Dr. Martin and others
presented pre-conference workshops to members of the American Associations of Bovine
 C. Dewey / Preventive Veterinary Medicine 86 (2008) 244249 247

Practitioners (AABP) and Swine Veterinarians (AASV). We started with basic descriptive
statistics, distributions and progressed to simple analyses. Veterinarians enhanced their power of
persuasion through the use of histograms and scatter graphs. Before long, swine veterinarians
were insisting that all of their clients keep records. Without records, the veterinarians believed
they could not effectively advise their clients. Commodity groups also understood the value of the
analysis of production records. In Ontario, Dairy Herd Improvement and the Ontario Pork
Producers Marketing board both provided research funds for graduate students interested in using
their databases to answer important research questions.
Dr. Martin is well read and he thinks outside the box. He reads about new applications of
statistics in other fields and in epidemiology as it pertains to human research and brings those
concepts to the work in production animal research. Veterinarians observed that sows
reproductive performance changed as revealed by the weaning to breeding interval. Sows that
lost a lot of weight during lactation by entering a catabolic state had a longer weaning to breeding
interval than sows that lost only a small amount of weight. Sows that were bred on the second
observed estrus after weaning had better reproductive performance than sows bred on the first
estrus after weaning. We assumed that there would be threshold levels of the weaning to breeding
interval variable that would be associated with either lower or higher subsequent reproductive
productivity. Dr. Martin suggested the use of hierarchical dummy variables in a method described
by Walter et al. (1987) to examine this association. We used day one as the referent group when
examining the association between weaning to breeding intervals and litter size. For day 2, the X2
variable is coded 1 but is coded 0 for all other days. For day 3, the X3 variable is coded 1 for either
day 2 or day 3 and then coded 0 for all other days. Finally, the X14 variable, the last interval
included in the analysis is coded 1 for all days from day 2 to day 14. This coding system identifies
changes in litter size as the weaning to breeding interval changes by 1-day increments. Litter size
was reduced on a daily basis for weaning to breeding intervals of 5, 6 and 7 days (Dewey et al.,
1994). Then, from 7 to 10 days, litter size remained low. Finally, litter size increased with
weaning to breeding intervals of 11 or more days. By using these hierarchical dummy variables,
as proposed by Dr. Martin, we were able to detect changes in litter size as the weaning to breeding
interval changed by just 1-day increments.
Many concepts in production animal medicine are complicated by the fact that there are two or
more levels of clustering impacting the parameter of interest. Recently, we researched the factors
associated with in-transit deaths in finisher pigs in Ontario. In-transit mortality includes all pigs
that die from the time they leave the farm until they are moved out of lairage at the packing plant.
By identifying the level of clustering that explains most of the variation in this mortality, we are
then able to focus future research work on reducing the losses. We applied variance analysis to
random variables depicting farm, truck and packing plant to identify levels of clustering that were
responsible for the largest component part of the variation in the outcome of interest. The records
kept by Ontario Pork indicate that half of the deaths (56%) occur on the truck, 4% were identified
as problem pigs at the farm, 17% occurred at the assembly yard and 23% occurred at the packing
plant (Haley, 2005). The industry typically suggests that the truck drivers are to blame. However,
our variance analysis showed that the farm explained 25% of the variation, the transporter 8% of
the variation and the packing plant 16% of the variation. We concluded that changing trucker
behaviour may not reduce the mortality whereas changing management on the farm will likely
reduce losses.
Dr. Martins ability to use analytic techniques for novel applications has been repeated by
several graduate students from Population Medicine including, but not limited to, Dr. Ian Dohoo,
Dr. John McDermott and more recently Dr. Zvonimir Poljak. We were interested in determining
248 C. Dewey / Preventive Veterinary Medicine 86 (2008) 244249

which serotypes of swine influenza were present in Ontario swine populations (Poljak et al.,
2007). The H1N1 serotype has traditionally been present but the H3N2 strain was causing
morbidity in swine populations in the United States in the early 2000s. Serum samples from 20
sows on each of 65 herds distributed across Ontario were tested serologically for the H1N1 and
H3N2 strains. Dr. Poljak did a cluster analysis of the serological results for H1N1 and found three
clusters of herds based on prevalence; 35%, 4075% and 80%. These clusters likely
represent: (1) herds where the virus circulated in the past but incoming gilts were negative and
remained nave, (2) herds where the virus circulated through the sow herd intermittently and
finally, (3) herds that experienced a recent outbreak. This application of cluster analysis to herd
prevalence data enabled us to better understand the variation in herd infection status rather than to
simply designate herds as positive or negative.
There were only 13 sows from a total of 3 herds where sows were serologically positive for
H3N2. The initial concern was that these tests were false positive results. However, geo-spatial
clustering analytic techniques were applied to the data to further understand the results.
Specifically, the CuzickEdwards analysis was used, which looks at the nearest neighbours and
the spatial Bernoulli statistics tested whether there were clusters after adjusting for the density of
pig farms in the area. The results indicated that there was a cluster of H1N1 positive sow herds in
southwestern Ontario, which is the pig dense area of the province. There was also a cluster of
farms that were positive for H3N2 located in another area of Ontario. As all of the positive farms
were clustered in one small geographic area of Ontario close to a lake, it was concluded that these
illustrated true positive cases. As the tools of analytic epidemiology improve, they will further
our understanding of problems facing food animal production systems.
Dr. Martin has enhanced food animal production through his own work and that of his
graduate students and their graduate students. We have come a long way from Daisy to
benchmarking and routine production animal records. Record keeping led us to multivariable
analyses as part of the day-to-day practice of production animal veterinarians. Epidemiologists
continue to use novel analytic approaches to understand the biology that is observed on the farm.
Dr. Martin, through his work and that of people he has influenced, has been an international
leader in promoting the optimal health and productivity of animal populations and of ensuring the
safety of foods of animal origin and preventing animal-related disease in humans.

Conflict of interest statement

Dr. Dewey does not have a financial or personal relationship with other people or
organisations that could inappropriately influence or bias the paper entitled The use of
epidemiology to enhance production animal research.

References

Chernysheva, L.V., Friendship, R.M., Gyles, C.L., Dewey, C.E., 2003. Field trial assessment of the efficacy of specific
egg-yolk antibody product for control of postweaning E. coli diarrhea. Vet. Ther. 4, 279284.
Dewey, C.E., Martin, S.W., Friendship, R.M., Wilson, S., 1994. The effects on litter size of previous lactation length and
previous weaning-to-conception interval in Ontario swine. Prev. Vet. Med. 18, 213223.
Dewey, C.E., Martin, S.W., Friendship, R.M., Kennedy, B., Wilson, S., 1995a. Associations between litter size and
specific herd level management factors in Ontario. Prev. Vet. Med. 22, 89102.
Dewey, C.E., Martin, S.W., Friendship, R.M., Kennedy, B., Wilson, S., 1995b. The associations between litter size and
specific sow level management factors in Ontario swine. Prev. Vet. Med. 23 (12), 101110.
 C. Dewey / Preventive Veterinary Medicine 86 (2008) 244249 249

Dewey, C.E., Wilson, S., Buck, P., Leyenaar, J.K., 2004. Effects of porcine reproductive and respiratory syndrome
vaccination in breeding-age animals. Prev. Vet. Med. 62, 299307.
Dewey, C.E., Johnston, W.T., Gould, L., Whiting, T.L., 2006. Postweaning mortality in Manitoba swine. Can. J. Vet. Res.
70, 161167.
Haley, C., 2005. Factors associated with in-transit losses in the Ontario swine industry. Ph.D. thesis. University of Guelph.
Kumar, S., Dewey, C.E., Friendship, R.M., Bowland, S.L., Shewen, P.E., 1999. Improved weight gain in pigs using
levamisole as an immunomodulator. Swine Health Prod. 7, 103107.
McDermott, J.J., Alves, D.M., Anderson, N.G., Martin, S.W., 1991. Benchmarka large observational study of Ontario
beef breeding herds: study design and collection of data. Can. Vet. J. 32 (7), 407412.
Poljak, Z., Dewey, C.E., Martin, S.W., Christensen, J., Carman, S., Friendship, R.M., 2007. Spatial clustering of swine
influenza in Ontario on the basis of herd-level disease status with different misclassification errors. Prev. Vet. Med. 81,
236249.
Walter, S.D., Feinstein, A.R., Wells, C.K., 1987. Coding ordinal independent variables in multiple regression analyses.
Am. J. Epidemiol. 125, 319323.