Three of the four breeds identified in this study as having an increased risk of surgery for ED are the same as in several other studies. The German shepherd dog, Labrador retriever, and the Rottweiler are well-known breeds at risk [8,9,10,11]. An interesting finding is that the Staffordshire bull terrier had a high OR for ED. To the authors’ knowledge, there is only one other study available reporting this breed among breeds predisposed for ED . In Scandinavia, the Staffordshire bull terrier has gained great popularity over recent years and from being a rare breed has now become one of the most common breeds in both Norway and Sweden.Footnote 4,Footnote 5 If this is true also for other countries, it may help explain why this is the only other study to date concerning this breeds’ predisposition to ED. The Staffordshire bull terrier shares a common ancestry with Mastiff breeds, which are reported to have the disease .
It should be mentioned that the collective diagnosis ED used in this study comprises three common developmental disorders in the dog, UAP, MCP, and OC. Joint incongruity and articular cartilage damage are also included in the group of conditions known as elbow dysplasiaFootnote 6 but have not been evaluated in our study. However, since all conditions sorted under the collective term are believed to be highly interrelated  and articular cartilage damage and joint incongruity are unlikely to be seen as a separate entity, we believe this to be a minor limitation to the study. Moreover, conclusions about prevalence of the particular diagnoses in each breed has been addressed in previous studies [8,9,10].
Labrador retrievers, Rottweilers and Staffordshire bull terriers are reported to be at increased risk for CCLD, while Chihuahuas, GSDs, and Shetland sheepdogs have been claimed to be at lower risk [1, 2, 13,14,15]. Our study detected an increased risk of disease in the Rottweiler, and decreased in GSDs and Chihuahuas, which are consistent with the earlier reports. For some breeds the literature provides inconsistent results. Cocker spaniels were found to have a decreased risk of CCLD in one study , but not in another . The risk among Golden retrievers have been described both as increased , same as in the reference population  and decreased [2, 15]. Despite the Labrador retriever being one of the most common breeds presenting with CCLD in our material, the combined OR was identical to mixed-breed dogs. Though mixed-breeds have been reported to have a slightly higher OR for CCLD than purebred dogs , this finding highlights the importance of having a comparable control population when reporting breed susceptibility. The country-specific OR for CCLD in the Labrador was lower than for mixed-breed dogs in Sweden, but higher in Norway. As for several other breeds originally bred for hunting, and the retriever breeds in particular, there are two quite different types of Labradors; a slim, lighter working type and a heavier built show type. It is not known whether the likelihood of orthopaedic diseases is the same for both types. Moreover, the relative frequencies of show and field bred Labradors in Norway and Sweden are unknown. This could be a contributing factor to the deviating results observed in the two countries and illustrates that breed susceptibility reported from single-centre studies and/or studies with a limited caseload should not be overemphasised. In general, minimally/borderline significant results in relation to breed susceptibility should be viewed with caution.
Medial patellar luxation is far more common than lateral luxation . Among the breeds reported to have a higher prevalence are the CKCS, JRT and the Chihuahua [10, 16,17,18]. The results are conflicting for Staffordshire bull terriers [17, 18]. Even though the CKCS had a slightly higher OR than mixed-breed dogs in our study, the Chihuahua was the only breed where an increased risk of surgically treated MPL was identified. This is in concordance with a recent study reporting the prevalence of patellar luxation among Swedish Chihuahuas to be 23% . The Labrador retriever is reported with an increased prevalence of MPL in some studies [17, 20, 21], but Labrador retriever is also the most common purebred dog registered in the UK Kennel Club . Two of the aforementioned studies were conducted in the UK, but since neither included a comparable control population, no conclusions about breed predispositions in the source population should be drawn. Even though the Labrador retriever is one of the most popular breeds in Norway and Sweden as well, no Labrador retrievers presented with MPL in our material. It may therefore seem that Scandinavian Labrador retrievers have a decreased rather than increased risk of MPL.
Considering the low bodyweight of the dogs with fractures of the radius and ulna in our material (Table 3), it is not surprising that the Chihuahua, CKCS and the Shetland sheepdog were the only breeds under study with the diagnosis. The absence of fractures of the radius and ulna in larger breeds was expected since these are more common in small and miniature dogs [23, 24].
The discrepancy between earlier studies and our results could be attributed to several factors such as genetic variation between different geographical areas and genetic drift as a consequence of breeding strategies over time , but it could also be due to the lack of an appropriate control group in previously published studies. In addition, a change of breed popularity over time, as discussed for the Staffordshire bull terrier, needs to be taken into account. Breed predispositions reported in studies conducted decades ago should be viewed with caution since they are likely to lack validity today. Comparing breed susceptibility with a control population adjusted to match the geographical distribution of the case population could be extended to larger caseloads from different geographical regions to increase the external validity of the results and to be able to calculate odds ratios for breeds where the diagnosis of interest is rare. A larger case population would improve the accuracy of the estimations and make it a better tool to study breeds with decreased risks, without the need for more advanced statistical methods. The method described in our study provides a framework with a potential for exploring breed-specific disease predispositions further. It is not limited to orthopaedic disorders but could be extended to all diseases where breed predisposition is suspected.
Most studies that report breed predispositions acknowledge the lack of a representative control population as a limitation. The control group is often either completely missing with only raw prevalence being described or limited to randomly selected hospital controls. Hospital populations, in particular referral populations, are mostly composed of sick dogs. Since sick dogs can acquire a different condition of interest, the dogs being sick is not in itself a justified reason for excluding them as controls. However, a variety of different diseases in dogs are breed-related. This introduces selection bias since some breeds are likely to be overrepresented in a study population comprised of sick dogs, and hospital populations are therefore not the most representative population for control selection in regard to breed composition. A source population is defined as the population from which the study subjects are drawn . In some cases, the source population is well-defined, but more often, as in the case of hospital populations, where some animals might come from afar, while others live nearby, the actual source population from which the cases originate is unknown . Some studies have utilised larger clinical databases, such as the VetCompass system in the UK  and the Veterinary Medical Databases in the USA . Although these databases include large numbers of animals, they only contain information about dogs admitted to veterinary care, and not the actual source population (the population of dogs that were likely to be included as cases if they had got the disease in question). Even when large clinical databases are used, the reported risk of disease can appear too high if the breed under investigation has a lower than average disposition for other diseases, and therefore is less frequently represented in the clinical database than in the source population. In recent years, a Swedish database of insured dogs has been used to compare breed predisposition to different diseases [25,26,27]. A limitation of using insured dogs as the reference is that the uninsured dogs are not included and there is a possibility that breeds with more health problems are more likely to be insured. Common for all the large databases is that the information recorded for each case and the details about the diagnostic workup can be sparse.
The reasoning behind calculating a geographically adjusted control group came from observations of different breed profiles at the two VTHs. SLU is situated in a middle-sized Swedish town, Uppsala, while NMBU is located in the city centre of Oslo, the capital of Norway. Registration numbers from different national kennel clubs reveal that breed distribution varies between countries. Even though the overall breed distribution is quite similar in NorwayFootnote 7 and Sweden,Footnote 8 there are large regional variations (Additional file 1). Since both VTHs have a substantial number of referred patients, using the unadjusted registration numbers from the counties of Uppsala and Oslo, or the total numbers for each country, would create bias and not be representative of the actual source population. Adjusting the registration numbers from each of the counties by their relative contribution to the database of eligible cases, ensures this bias is kept at a minimum. The results from the logistic regression analysis (Table 5) show that the risk of becoming a case at NMBU is generally lower than at SLU. Since there are several other large small animal hospitals located near NMBU, while SLU is the largest hospital in Uppsala county, it is not surprising that the relative percentage of the control population seen at NMBU is smaller than at SLU.
Several limitations for this study must be acknowledged. Most importantly, only information from dogs examined at one of the participating VTHs were included. Therefore, information regarding dogs that were referred to other veterinary hospitals in the areas and for dogs whose owners did not pursue surgical treatment at the participating VTHs is lost. It is not unlikely that the treatment and referral strategies of dogs with the same orthopaedic disease might differ between breeds due to factors such as the complexity of the surgical procedure, size and temperament. It is therefore feasible that referral caseloads show a selection bias towards more complicated cases. For example, it is possible that small breed dogs with CCLD are underrepresented in our material because a substantial percentage of these dogs were treated conservatively or not referred in the first place. The information in the database cannot be retrospectively confirmed or rejected; therefore, all results rely on correct reporting of data. While ID-marking is mandatory for all Swedish dogs and for pure-breed Norwegian dogs to be registered in the national kennel club, it is voluntary for mixed-breed dogs in Norway. This discrepancy is a potential selection bias in the control group. However, the general Swedish and Norwegian dog populations are quite similar, and this is most likely true for mixed-breed dogs as well as pure-breeds. Moreover, a variety of cross-breeds (poodle mixes) have gained popularity over the last decades and are bred by breeders in a similar manner as pure-bred dogs. In addition, stray and shelter dogs are uncommon in Scandinavia; most dogs belong to an owner. Since the percentage of mixed-breed dogs in the Norwegian control population was higher than in the Swedish (Table 4), and with the aforementioned factors in mind, we believe the difference in ID-marking policy between Norway and Sweden to be of minor importance to our results. In addition, the control groups have been calculated separately for each country, and the logistic regression model adjusted for hospital.
Even though studies comparing the use of different control populations are available in the human literature , veterinary studies are lacking. The implications of using different control groups (i.e. hospital controls, insurance data, adjusted and unadjusted ID-registry data) in relation to breed susceptibility for disease should be addressed in future studies.