Here we present the findings obtained from a canine population-based cancer registry that we conducted in northwestern Italy for a 90-month period from 2001 to 2008. The Piedmont canine cancer registry is one of few of its kind whose operations have been consistently supported by long-term public funding. The incidence rate of malignant and benign tumours was about 800 and 900 cases per 100,000 dog-years at risk, respectively, with the largest impact on females and purebred dogs. The highest incidence rates by sex and tumour site were observed for mammary and cutaneous tissues in females and cutaneous tissues and testicles in males. These results were obtained by paying particular attention to the validity of the data collection process, thus minimizing the potential for bias.
Given the constraints derived from working with a canine population, considerable effort was expended to estimate the relevant denominators in terms of population size and structure. Moreover, we were able to obtain detailed reports of cases of incidental tumours by selecting a small-medium size catchment area and through public awareness campaigns coupled with enhanced participation of veterinary practices offered scientific support and histological examinations free of charge.
Though our study may have limitations, they do not necessarily compromise its internal or external validity. By mobilizing the available human and economic resources and restricting the cancer registration to a relatively small area (canine population of about 10,000 units), we were able to maintain the activities for a 90-month period, with a total of about 70,000 dog-years at risk which were then used as the overall denominator. The resulting overall incidence rates were found to be stable; however, less precise estimates could also have been obtained by looking at subgroups (e.g., municipality, breed, age class). For this reason, we carried out few subgroups analyses within our dataset.
As mentioned, there remains the potential for some residual bias in the quantification of the denominators and the numerators. Selection bias is less likely to be a problem with population-based registries than, for example, hospital-based registries [18, 19] or canine registries using databases from pet insurance companies where cross-breed dogs or aged dogs are apt to be underrepresented [6, 20].
With regard to the capture-recapture strategy that was applied to estimate the unknown canine population size, some of the assumptions this approach  requires may not have been fully met, e.g., working on a closed population and using random samples (assumptions that are unlikely to be met in a field situation); however, there are no evident reasons that preclude that the recruited dogs were equally likely to be captured in each of the two samples. The midterm census of 44 of the 46 municipalities in the catchment area indicated no large deviation from the direct enumeration of the existent dogs.
Additionally the population structure by breed- or age-strata may not be as precise as the population size. The structure was obtained from our second capture (the anonymous questionnaire survey) only. Unfortunately, the data on the first capture (i.e., the canine identification and registration system) and the midterm census were available only in hard copy format, precluding any practical way to further analyse the data. Therefore, the incidence rates are likely to be more robust when the denominator was the entire population (e.g., crude malignant rate or site-specific rates) rather than when the denominators referred to individual breeds or age classes.
Finally, with regard to the cases of incidental tumours, a certain degree of underdetection may have occurred. In particular, a proportion of tumours may have gone undiagnosed (e.g., a deep organ tumour), or not reported as not requiring histological confirmation (e.g., osteosarcoma, lymphoma) or may have been diagnosed by laboratories other than ours and not reported by the collaborating veterinary practitioners. In an attempt to minimize these problems, the collaborating veterinary practitioners were given standardized case report forms and local laboratories were contacted.
Assuming an unbiased estimation of cancer occurrence, the incidence rates in our registry are higher than those reported by similar population-based registries in Italy [12, 13] where the crude incidence rates for all malignant cancer were less than 200 cases per 100,000 dog-years at risk. The differences, albeit evident, are smaller when the comparison is carried out with data from international population-based registries ([7, 8, 21]). Rates higher than ours were reported in a registry that used as a denominator for cancer incidence all dogs insured with a pet U.K. insurance company . The higher cancer rates we found cannot be convincingly linked to exposure of the dogs to as yet undetected environmental risk factors in the catchment area. Instead, the excess incidence may more likely stem from the effect of a different distribution of confounding factors such as age and breed, which are known determinants of cancer in dogs [22–24]. Unfortunately, in most of the comparable studies, data on the population structure by age and breed were not available, preventing any appropriate comparison of rates. Moreover unlike other studies, in ours the denominators were reduced to take into account the real proportion of dogs under veterinary care, an adjustment not reported elsewhere.
Another explanation for the high cancer rates we found may be due to the way we managed the data on multiple primary tumours, which are a common finding in dogs [25–27]. Based on the histopathological diagnosis, each type of neoplasm was classified as a separate cancer and added to the numerator. Other canine registries may not have done this. Finally, the potential risk of overestimation due to the inclusion of cases sourced from outside the registry catchment area was minimized by applying the preliminary case inclusion/exclusion criteria established in the registration protocol.
The distribution of cancer types in our registry is fairly consistent with the literature. As reported elsewhere [7–13, 21], the most prevalent malignant neoplasms were mammary carcinoma and cutaneous mastocytoma. The high prevalence of seminoma found in the Norwegian dog population  is analogous to the high incidence estimated in our registry. With regard to breed, the higher incidence rates in purebred rather than in cross-breed dogs were evident for Yorkshire terrier and Boxer, breeds known [5, 7, 27] to be particularly at risk of developing neoplasms, for which a genetic predisposition has been suggested. In the female dogs, the observed cancer trend by age matches those seen in other studies, with an exponential increase in elderly females for all tumours  or specifically for mammary tumours .
Comparison with the recent findings from two other Italian cancer registries [12, 13] shows that the distribution of tumours is substantially similar. In our registry, the ratio of malignant to benign tumours was 0.90:1, as compared to 0.96:1 in Genoa and 1:1 in the Veneto Region. The incidence of malignant cancer was higher in females than in males (2.5 times, 2.7 times, and 1.7 times in the study area, Genoa, and Veneto, respectively) and in purebred dogs than in cross-breed dogs (1.5 times in Piedmont and 2 times in Veneto where data were available). Finally, the distribution by tumour site (mammary and cutaneous/soft tissues tumours in females and cutaneous/soft tissues and genital tract tumours in males) was quite similar for all three registries. The exceptionally higher incidence of lymphoma in dogs of both sexes in the Genoa registry may have been due to the inclusion of cancers diagnosed by cytology or because of exposure to unknown risk factors present in a large city.
The Piedmont canine cancer registry is based on a large target population that combine rural and urban environments and on a population-based study design; for this reason our findings have the necessary external validity to be generalized, after adjustment for breed and age, to the canine population of northwestern Italy.
Assuming that the incidence rates from a population-based registry have external validity, national and international standards should be developed and shared to be able to make any meaningful comparison across studies. The adoption of an international classification and coding system, like the International Classification of Diseases for Oncology used in human cancer registries , would facilitate the exchange of comparable data and the possibility to carry out multicenter studies .
Finally, basic epidemiological methods (e.g., standardization techniques) to account for the confounding effect of age and breed should always be applied: data on population structure should be collected and made available in each study and, as proposed by Thrusfield , an international normal dog population should be established and shared as an external standard.