Page 106 - Withrow and MacEwen's Small Animal Clinical Oncology, 6th Edition

CHAPTER 4 Epidemiology and the Evidence-Based Medicine Approach 85

25
risk factors for occurrence of cancer. In addition, the Swedish within the Swedish data is quite informative given that the limi-
data only include dogs up to 10 years of age; it is unknown how tations occur equally across breeds. Crudely comparing overall
mortality rates for cancer, BMDs were approximately six times
the statistics would look if dogs of all ages were included. As the
VetBooks.ir authors of this study discuss, for cancer (or any cause of death) more likely to die of cancer compared with all dogs combined
(306 versus 50 per 10,000 DYAR, respectively). Where it was
that occurs at older ages, a dog must live long enough to experi-
24
ence it (i.e., not die at a younger age because of any other cause), possible to do more sophisticated analyses, BMDs were shown to
and deaths before 10 years of age are relevant to focus on for can- be 17 times more likely to die of cancer, compared with baseline,
33
24
cer prevention. and adjusting for age, gender, and breed. Even if specifics of the
population may not be the same as other populations, data such as
Sources of Information on Cancer Occurrence these are important for identifying high-risk breeds. Comparison
across populations and over time is needed, with due consider-
26
One of the largest clinic-based databases is the VMDB. This ation of data issues.
database was started in 1964 by the National Cancer Institute, Studies on Swedish insurance data have also presented statistics
includes patient data from 26 university teaching hospitals in on morbidity and mortality in cats. 38,39 As with dogs, the diagno-
the United States and Canada, and contains more than 7 million ses are made by veterinarians, but further details are unavailable.
records from all species covering the full range of diagnoses, includ- The overall age-standardized mortality rate for death resulting
ing cancer. The VMDB is a widely used source of cancer surveil- from cancer in insured Swedish cats (generally <12 years of age)
lance information for companion animals; however, as discussed was 37 per 10,000 cat-years at risk. The most common types of
previously (and presented in Fig. 4.1), there is no information on cancer in the Swedish data were mammary, stomach/intestinal,
the base population in these studies and only proportional mea- and lymphoma. Siamese breeds were at increased risk of death
sures can be calculated. Given that the data sources are teaching caused by neoplasia; mammary cancer was the most common
hospitals, the patients and disease diagnoses represented are likely type, in agreement with an earlier study. Differences between
40
41
to be influenced by referral bias, resulting in estimates of disease populations and data are no doubt affected by differences in vari-
frequency that may not be typical of those seen in the general dog ous factors (e.g., spay/neuter rates and age structure of the popula-
and cat population. In an analysis using VMDB medical records it tions). Further study of neoplasia in cats is needed.
was concluded that substantial referral bias may indeed exist in the Notwithstanding the previous discussion on the relative pau-
data set, and the authors suggested the accuracy of prevalence esti- city of population-based, histologically confirmed data on the
mates measured from the VMDB could be improved by statistical incidence of cancer, there is no doubt that certain breeds are at
adjustment on the basis of geographic proximity of the patient’s high risk for cancer (e.g., the BMD, flat-coated retrievers, Boxers,
residence to the nearest university teaching hospital. 27 and Scottish terriers). In the section that follows, we will pres-
The use of primary care data for investigations regarding the ent an overview of known and suspected risk factors for specific
occurrence of cancer in companion animals is less affected by cancers, including breed-specific risks, which is not strictly lim-
28
referral bias, but may be influenced by misclassification bias. ited by the level of evidence or quality of studies or data but that
To date, primary care data have been an underused resource for reflects the current state of knowledge. An important concern for
determining cancer incidence, but advances in large-scale data col- the future, however, is that without true incidence data we are
lection and management suggest primary care data may become limited in our ability to track changes in occurrence over time,
increasingly accessible for this purpose. For instance, the Vet- as proportional measures are influenced by changes both in the
erinary Companion Animal Surveillance System (VetCompass), numerator and denominator. In other words, an increase in the
which began collecting clinical data from primary practices in the popularity of a breed may lead to its apparent overrepresentation
United Kingdom in 2009, now holds data on nearly 6 million in proportional cancer measures; a change in the distribution of
animals collected from more than 500 veterinary practices across breeds may affect cancer prevalence, without any actual change in
29
the United Kingdom (as of August 2017). VetCompass began breed risk. Without incidence measures, it would be impossible to
30
collecting clinical data in Australia in 2016 and pilot projects are accurately evaluate the effectiveness of programs aimed at prevent-
underway in Spain, Germany, and New Zealand. Projects related ing or controlling disease.
31
to cancer outcomes in companion animals are already underway,
29
and these data will likely become an increasingly important source Factors Associated with Cancer Risk
of information for research concerning cancer occurrence.
Two well-established insurance databases are from the United Observational studies are the tools of epidemiology used to iden-
Kingdom and from Sweden. 24,32,33 A notable limitation of these tify and characterize the determinants of cancer risk. Informa-
32
databases is that not all cases are histologically confirmed. The tion from descriptive studies such as case series may help generate
benefits and limitations of these data have been discussed exten- hypotheses but is not adequate as a basis for evidence-based cancer
sively in the literature. From the UK database, using data from prevention strategies. Results from case series are also no longer
34
1997 through 1998, cancer incidence among 130,684 dogs at risk accepted for publication in at least one major veterinary medi-
was 747.9 per 100,000 dog-years. From the Swedish data, the cal journal. Analytic observational studies such as case-control
35
6
overall mortality rate for cancer was 50 per 10,000 dog-years-at- and cohort designs, on the other hand, are used to test research
risk (which equates to 500 per 100,000). Osteosarcoma (OSA) hypotheses, and when well designed, can provide valuable infor-
24
incidence rates were 6.1 and 5.0 dogs per 10,000 dog-years for mation for cancer prevention strategies.
36
males and females, respectively, and among females, breast can- The case-control study design is the most commonly used
cer incidence was 111 dogs per 10,000 dog-years. The limita- observational study design in veterinary epidemiology research
37
tions of the Swedish data are that deaths are mainly in dogs 10 and in cancer epidemiology research in general. This is the most
years of age or younger and it is unknown whether the diagnosis efficient study design, in terms of cost and time, when evaluat-
20
has been validated by histology, but comparison across breeds ing associations with relatively rare outcomes, such as specific