Assessing effects of structural zeros on models of canine cancer incidence: a case study of the Swiss Canine Cancer Registry

Abstract

Epidemiological research of canine cancers could inform comparative studies of environmental determinants for a number of human cancers. However, such an approach is currently limited because canine cancer data sources are still few in number and often incomplete. Incompleteness is typically due to under-ascertainment of canine cancers. A main reason for this is because dog owners commonly do not seek veterinary care for this diagnosis. Deeper knowledge on under-ascertainment is critical for modelling canine cancer incidence, as an indication of zero incidence might originate from the sole absence of diagnostic examinations within a given sample unit. In the present case study, we investigated effects of such structural zeros on models of canine cancer incidence. In doing so, we contrasted two scenarios for modelling incidence data retrieved from the Swiss Canine Cancer Registry. The first scenario was based on the complete enumeration of incidence data for all Swiss municipal units. The second scenario was based on a filtered sample that systematically discarded structural zeros in those municipal units where no diagnostic examination had been performed. By means of cross-validation, we assessed and contrasted statistical performance and predictive power of the two modelling scenarios. This analytical step allowed us to demonstrate that structural zeros impact on the generalisability of the model of canine cancer incidence, thus challenging future comparative studies of canine and human cancers. The results of this case study show that increased awareness about the effects of structural zeros is critical to epidemiological research.

The importance of regional models in assessing canine cancer incidences in Switzerland

“…Using the dogs registered in the Swiss canine population database, we also derived variables associated with known biological risk factors for several canine cancers [ 36 – 38 ] ( Table 1 ). These variables were studied in previous spatial analyses using the SCCR data through conventional regression models [ 13 – 15 ]. The variables are Average Age (in months), Females per Male (in percent), and Average Weight (in kilograms) of the dogs registered in the different municipal units each year, during the period 2008–2013.…”

“…Case-control studies of the SCCR have highlighted important relationships between canine cancers and a number of biological risk factors [ 11 , 12 ]. The same biological risk factors were also studied through spatial analyses, using conventional regression models, but the model coefficients revealed very different relationships to canine cancers [ 13 – 15 ]. While these results have evinced that risk factors for individuals may be difficult to detect among populations [ 16 , 17 ], partly as a consequence of the modifiable areal unit problem (MAUP) [ 18 ], as noted by the original authors, a number of issues still needed to be addressed in modeling canine cancer incidences [ 13 – 15 ].…”

Application of Bayesian Regression for the Identification of a Catchment Area for Cancer Cases in Dogs and Cats

“…Misspecification of the underlying model may also have a significant impact on the CA estimation, which is why we tested different likelihood distributions in order to find the best performance on a spatially autoregressive conditional zero-inflated negative binomial model. The parameterization of this model takes into account the spatial correlation between CT and differentiate between zeros representing under-ascertainment and lack of cases ( 55 ), which have been considered relevant issues when assessing cancer incidence ( 10 , 12 ). The better performance of the CAR model suggests the existence of spatial effects in the data and is consistent with the effects later associated by the logistic regression with the inclusion in the CA (distance and income), as they are likely to have a spatial component.…”

“…However, these data are not exempt from biases. Cases are not randomly obtained from the population but rather different factors, such as the distance to the center or the socioeconomic status of owners, among others, may influence first, the decision of looking for veterinary care and seek specialty care and the choice of a center instead of others in a competitive environment ( 10 – 12 ). Misidentification of the underlying population that is providing the data and the degree of underreporting affect the accuracy and reliability of any subsequent analysis ( 7 , 13 ).…”

The Global Initiative for Veterinary Cancer Surveillance (GIVCS): Report of the first meeting and future perspectives

“…5 In Switzerland, the Swiss Canine and Feline Cancer Registries have been established, with the former taking advantage of the mandatory official registration of dogs since 2007, allowing calculation of population-based tumour incidences. [6][7][8][9][10][11][12][13] The University of Queensland is developing a pilot study on cancer occurrence in dogs using retrospective data from the Veterinary Laboratory Services. In Kenya, a collaborative work is ongoing with the human Kenya National Cancer Registry, investigating cancers affecting humans and dogs in Nairobi.…”