Page 107 - Withrow and MacEwen's Small Animal Clinical Oncology, 6th Edition

86 PART I The Biology and Pathogenesis of Cancer

cancers. Unfortunately, as data collection is often retrospective, TABLE 4.3 Guidelines for Interpreting Clinical
many potential sources of bias must be considered. The features Relevance from Odds Ratios or Relative
of an ideally conducted study (e.g., with the least opportunity for
Risk Measures
VetBooks.ir systematic bias) include the complete ascertainment of all newly Inverse Association ≈ Positive Association ≈
diagnosed cases with histopathologic confirmation of primary
tumors and a random (or matched) selection of controls from the Decreased Risk Clinical Relevance Increased Risk
same base population as the cases. In a population-based case-
control study design, 42,43 we can assume that if a control subject 1.0 Not evident 1.0
had been diagnosed with the tumor of interest, that control would 0.7 to <1.0 Weak >1.0–1.5
have been a case in the study (i.e., the controls are from the same 0.5 to <0.7 Moderate >1.5–2.0
base population as the cases). The goal of the control group is to
represent the exposure experience of the base population. For this 0.3 to <0.5 Strong >2.0–3.5
reason, we are not interested in selecting the “healthiest” subjects <0.3 Very strong >3.5
as our comparison group.
In a hospital-based case-control study, both cases and con-
trols are selected from the same hospital(s). The limitation with
this design is that we cannot generalize the study results to a
clearly defined base population. However, this design is valid The RR and OR are similarly interpreted. A value greater than
and can still provide meaningful results. When using hospital- 1.0 indicates that the exposure is positively associated with dis-
based case-control study design, it is preferable to randomly ease (increases risk), whereas a value less than 1.0 indicates that
or systematically sample from the noncase population and the exposure is inversely associated with disease (decreases risk).
to not include animals that have been diagnosed with other A value of 1.0 indicates there is no association between expo-
cancers. 44,45 sure and disease. The 95% confidence interval (CI) indicates the
In a prospective cohort study, a group of animals is defined precision of the RR or OR, and if the 95% CI includes 1.0, we
on the basis of exposure and followed over time to compare interpret the RR or OR to be statistically nonsignificant. It must
the incidence of disease (or other specified outcome) among be remembered, however, that statistical significance does not
the exposed and unexposed groups. The results obtained from a necessarily equate with clinical importance. For the latter, the
prospective cohort study are advantageous compared to results magnitude of the effect is also important to consider. Table 4.3
obtained from a case-control study for many reasons. One pri- shows suggested guidelines for interpretation of risk estimates.
mary advantage is that we can assume temporality, or that the When considering whether to implement preventive measures
exposure came before the disease, when associations are observed or health interventions at the population level, the following,
from prospectively collected data. Systematic errors resulting in addition to the risk estimate, are also relevant: prevalence of
from selection bias (e.g., referral bias) and differential recall bias the factor (i.e., likelihood of exposure) and the prevalence of the
(e.g., misclassification of exposure by disease status) are also not disease. These values are used to estimate the attributable risk or
major concerns when interpreting results from a well-performed risk-reduction measures.
prospective cohort study. There is a need for more longitudinal Findings from all observational studies are influenced by sys-
(as opposed to retrospective), preferably population-based stud- tematic error to some degree because there is inherent bias in the
ies to strengthen the quality of evidence in the field of veterinary methods used to select the study population, measure exposures,
oncology. and identify the outcome. The opportunity for any one study to
Regardless of the observational study design used, nondiffer- report an association that is due in part to chance is a real concern,
ential exposure misclassification will be a major concern, and the even with the use of valid study design methods and statistical
possibility and extent of misclassification should be considered analyses. Confidence in the evidence for a particular association
when interpreting observational study results. Methods by which is strengthened when it is observed repeatedly in multiple popula-
exposure misclassification can be reduced include using a precise tions and with the use of more and more rigorous study design
and accurate questionnaire that has been properly validated or methods. Meta-analysis is a technique whereby results from mul-
incorporating the use of biomarkers of exposure that can be quan- tiple, similar studies can be combined to increase the power of
tified into the study design. findings. Several examples are available from the human litera-
To estimate the magnitude of an association between an expo- ture relating to nutritional risk factors associated with pancreatic,
sure and a cancer type, the relative risk, or risk ratio (RR), and breast, and colon cancer. 46–48 Unfortunately, in small animal
odds ratio (OR) are calculated from data collected from cohort oncology, studies have neither been performed nor reported con-
studies or cross-sectional and case-control studies, respectively. sistently enough nor are there an adequate number of studies
The RR is calculated as follows: conducted to support meta-analyses being conducted regularly at
this time. Notwithstanding these limitations, Table 4.4 presents
RR = Incidence among exposed subjects ÷ risk factors, including breed risks, for some of the more common
Incidence among unexposed subjects cancers in dogs and cats for which there are at least reasonable
where incidence is the number of events divided by total ani- estimations of association.
mal-time of follow-up. The OR can be used to estimate the RR The identification of modifiable risk factors for canine can-
when incidence data are not available. The OR is calculated as cers is the first step in eventually reducing incidence. Table 4.5
follows: presents analytic studies used to test hypotheses that selected
factors were either associated with an increased or decreased
OR = (Number of exposed cases ÷ number of unexposed cases) ÷ risk of canine and feline cancers. Characteristics of the study
(Number of exposed controls ÷ number of unexposed controls) design and analytic methods are highlighted as strengths