The insured population
The large database used in this study has several advantages, but of course extrapolation of the information must proceed with some cautions. In previous studies, it has been shown that the accuracy of the demographic data in the insurance database was excellent  and that the population insured at Agria is quite similar to the general population of dogs in Sweden . The major differences between the insured population and the total Swedish population are that a lower proportion of mongrels are insured, and there are, of course, more older dogs in the general population.
Some of the earlier studies were on limited years of data, and the earliest one used only breed-level data [3, 12, 13, 15]. In the present study six years of data have been included, dogs could be followed individually and it was possible to calculate the actual time of observation for each dog. This is a dynamic population (dogs may enter or leave the database) and therefore mortality is best expressed with MRs that express the occurrence of death based on the actual time at risk. Compared to most of the published studies on death in breeds of dogs, the database used in this study provides a large number of cases (for example deaths by breed or by diagnostic category) over several years, making the results less prone to random error. It is now possible to present the mortality with respect to the effect of for example age, breed and diagnosis simultaneously. Comparisons are facilitated across breeds as to the relative burden of disease. For example, golden retrievers are one of the most common breeds of dog in Sweden, with over 68,000 dog years at risk in this database (6% of the insured population, data not shown). Irish wolfhounds provide less than 2,000 DYAR. However, based on 10,000 DYAR there were over 1,500 deaths in the Irish wolfhound and only 243 in golden retrievers in insured dogs under 10 years of age. One might postulate differences due to size, i.e. a giant breed vs. a large one, however, breeds of a similar size to golden retrievers, e.g. Bernese mountain dogs, dobermanns and boxers also have a much higher mortality.
As space does not allow presentation of information on all breeds in the database, the decision was made to include the 10 most common breeds, a group of breeds with both high rates of death and a sufficient number of dogs to ensure good validity for calculated rates and then to group all other breeds in one group. The 20 specific breeds accounted for approximately 45% of the insured population and one-half of the deaths. The information provided on the group of all other breeds and for the whole population offers a useful comparison.
It is important to note that puppy mortality is not included in the present study and dogs are only included up to 10 years of age. However, if one considers that the first 10 years represents the major part of an expected lifespan for many breeds, and that it is preferable to target prevention strategies at diseases occurring relatively early in life, the information presented is highly relevant to veterinarians in clinical practice, researchers and for dog organisations, breeders and owners. Although the mortality due to most causes increased with age, the rate of tumour deaths started to rise earlier in the high-risk breeds. This may be related to a faster rate of aging, i.e. certain breeds may be biologically older at a given chronological age compared to other breeds [1, 15]. On the other hand, the rate of trauma is relatively higher in younger ages in common and other breeds, perhaps related to behaviour. It is important to take into account the age pattern of disease and possible differences in age distribution across groups when examining competing causes of death, as in this and the companion paper . In order to die of causes of death that are a result of aging (for example some cancers) one must survive to an increased age. This must also be considered when interpreting proportional estimates for mortality, especially from certain sources of data (e.g. ). Cox regression, assuming proportional hazards over the analysed period, is used to analyse time to an event, in this case death, rather than simply proportions of a breed that have died.
One drawback with the insurance database is that neuter status is not included. Although, there may be an increasing trend for elective spaying and neutering recently in Sweden, data from 1998 suggest that only about seven per cent of Swedish bitches and less than four per cent of males (all ages) are neutered . This estimate includes dogs spayed both for medical conditions and non-medical reasons.
Because owners tend to terminate their policy when an animal has died, even if they have not submitted a claim, the total mortality rate presumably includes almost all deaths. Any unrecorded dead dogs would at least be censored (in the analysis) when their policy was terminated. Only those for whom a claim is processed, however, have a diagnosis entered for the cause of death (although this may include, dead-no diagnosis, if provided by a veterinarian). A diagnosis was specified for approximately 70% of all deaths. For some breeds over 85% of deaths had a recorded cause, however, for mongrels 45% had a diagnosis. The possible reasons that the death of a life-insured animal would not have a cause recorded include the following. The cause of death might not be covered by the policy, for e.g. behavioural problems and some inherited problems in certain breeds. For example, hip dysplasia (HD) is covered for purebred dogs where the sire and dam are registered with the Swedish Kennel Club (SKK) and have been screened HD-free, but is excluded for mongrels. However, as there are relatively few exclusions for most breeds it may be that most cases of death that do not have a diagnosis (i.e. were not claimed) arise because the owner neglects or chooses not to submit a claim, due to simple omission or, perhaps in some cases, a reluctance to either deal with the issue at a painful time or to benefit financially from the animal's death. For mongrels, life insurance benefits may be quite small. It is likely that for some cases of elective euthanasia (e.g. behaviour problems, owner convenience) a claim would not be submitted.
In a previous study, the agreement between diagnostic data in the database and in the practice records was over 80% in all groups evaluated . The overall consistency and accuracy of diagnoses made by practicing veterinarians remains a concern whenever secondary data are used. Consistency is improved by the use of a standard diagnostic registry both within most veterinary practices in Sweden and when data is recorded in the insurance database. Only one diagnosis can be recorded for any case and in general veterinarians are responsible for choosing that designation. Clinicians often use rather non-specific diagnoses, sometimes reflecting the degree to which the case has been resolved (claims may be submitted before final test results are received or owners may elect euthanasia rather than pursue expensive tests to achieve a definitive diagnosis), but often because of time constraints or practical limitations of using the diagnostic registry. For example, one of the most common diagnoses was 'dead – no diagnosis' or similar variants. For the heart system, five diagnoses captured 77% of deaths. However, some of these designations were very non-specific, e.g. 'signs of disease, no cause defined'. In consideration of all these limitations of veterinary diagnoses, both in general and specifically in these insurance data, rather crude diagnostic categories were used, which have been amalgamated over different specific diagnoses, and for which reasonable accuracy can be presumed. Future studies on specific diseases may require further validation of the diagnostic criteria being used by veterinarians.
Overall, among the general diagnostic categories, tumour and trauma had the highest mortality rates (each accounting for almost one fifth of all deaths). Studies of lifetime risk have suggested a wide range in estimates of deaths due to cancer, for example, from 3% of deaths in military working German shepherds  to 27% of dogs from a German study  and 16% of all deaths in the UK .
Among all dogs the specific diagnoses with the highest mortality rates were car accident (approximately 21 per 10,000 DYAR), dead-no diagnosis (19 per 10,000 DYAR), epileptic and epileptic-like seizures, hip dysplasia, signs of heart failure and disc herniation. In spite of extensive monitoring of HD among purebred dogs, and given that dogs claimed for HD would be from HD-free parents, this disease still remains a common problem in many breeds.
Road traffic accidents were the cause of death in approximately 0.2% (data not shown) of insured dogs (all breeds) under 10 years of age. This constituted five per cent of all deaths. Sweden is a country with remarkably few stray dogs and these insured dogs are a 'cared-for' population (i.e. receiving veterinary care) yet car accidents are the most common single cause of death. In a survey of owner reported causes of death in the UK, including lifetime deaths, 3% were due to car accidents . Given that younger dogs are more prone to accidental death, age composition may partially explain the higher proportion in this population. However, caution must be used in comparing proportional estimates, as it may be that the actual rate of traumatic death might not be different.
True population rates of road traffic accident deaths are not readily available. In general, these findings are in agreement with our previous studies, in spite of some differences in defining diagnostic categories, and using either total deaths  or those with a diagnostic code . As this study incorporates data from six years, these are likely better estimates than the ones previously published. These statistics for traumatic death are important, especially if you consider road traffic accident deaths to be mainly preventable by humans.
Breeds and causes of death
Mortality rates calculated using the exact time at risk are useful for comparison of the occurrence of death between breeds within this population and for comparison with other estimates from the literature. Even breeds with a similar lifespan may have very different age patterns of mortality. Although they must be interpreted carefully, proportional mortality statistics are a useful adjunct to MRs in completing the picture on the pattern of death within various categories. From a population perspective it is interesting to identify those breeds that account for significant proportions of death within that population. For example, German shepherds represent approximately 7% of the total population of insured dogs (in table 1 80,049DYAR/1,098,358DYAR = 7%), but account for almost 12% of the total mortality. The 10 high-risk breeds account for approximately 5% of the population and 11% of the mortality. If these breeds and German shepherd dogs are combined, the 11 breeds represent 12% of the insured population and account for almost one quarter of all deaths. Several breeds account for a lower proportion of the total mortality, for example golden retrievers constitute 6% of the total DYAR and less than 4% of total deaths.
Heart disease in the Cavalier King Charles spaniel accounts for over 50% of deaths in that breed (in dogs under 10 years of age) and for over one-quarter of the heart deaths in the insured population. Although heart disease in Cavalier King Charles spaniels is well recognized , these statistics give further insight into the impact of this cause of death in this breed. By comparison, in Irish wolfhounds, although the actual rate of mortality due to heart disease is higher than in the Cavalier King Charles spaniel, that diagnostic category accounts for only 25% of deaths in Irish wolfhounds as they also have other frequent causes of death. Irish wolfhounds represent only 2.7% of all heart deaths in the population, as they as are a less common breed (data not shown). How should these statistics be most effectively used? Mortality rates can be monitored over time to see if there is an increase or decrease in the actual incidence due to, for example, breeding practices. The perception of the commonness of disease occurrence by veterinarians will be a reflection of the population level proportional mortalities, which are also of interest to the insurance company. Proportional mortalities within a breed should inform health strategies among dog breeders, helping them to focus on those diseases causing the most deaths at "too early an age" in their breed.
Notwithstanding the potential usefulness of proportional mortalities, they must be viewed with caution, especially when presented in the absence of actual mortality rates.  presents proportional statistics from a necropsy database and concludes that golden retrievers have an increased risk of tumours similar to that for boxers. In this study, the actual rate of tumours in boxers is almost four times higher than that in golden retrievers.
Almost 50% of all deaths in drevers and miniature dachshunds and 37% in dachshunds were due to trauma, although the actual rate in drevers was more than twice that for both types of dachshunds. With these similar proportional mortalities, traumatic death is of similar importance within each breed. However, drevers account for 11% of all deaths due to trauma in the insured population (data not shown) although they are only 3% of all insured dogs. Most drevers and many dachshunds are used in hunting in Sweden and the specific behavioural characteristics for which they are selected also make them prone to roaming. Notwithstanding, the degree to which they die due to traumatic causes is somewhat alarming. The influence of usage on causes of death has been previously reported, for example by Anderson & Rosenblatt (1965) who found that field beagles were killed by motor accidents to a very large extent, compared to laboratory beagles that died because of tumours to an overwhelming extent. These findings remind us to be very conscious of the source of data when comparing mortality statistics.
Both the types of trauma and reasons behind an increased risk may vary by breed. Mongrels, miniature and toy poodles and greyhounds also have a high proportional mortality for trauma. For greyhounds, fractures were a common cause of death due to trauma (data not shown). Perhaps smaller dogs may be no more likely to be hit by a car, but may be more likely to die if they are. People may be less likely to pay for expensive veterinary care for mongrels following an accident. Given the human element in traumatic deaths in dogs, these statistics might be considered in discussions relative to responsible pet ownership.
Both Irish wolfhounds and Bernese mountain dogs have approximately 300 deaths due to tumours per 10,000 DYAR in dogs less than 10 years of age. The MR for boxers is 200 per 10,000 DYAR. Within the breed, a dying boxer (under 10 years of age) has a 37% chance that it is due to tumour and a Bernese mountain dog 41%, whereas tumour deaths are 22% of all deaths in Irish wolfhounds. Obviously tumours are an important consideration for these breeds and within the population.  showed a high level of relatedness, inbreeding and litter per sire ratio for Dutch Bernese mountain dogs. Although it is possible to estimate the relative risk for a cause of death by dividing the MR for each breed, it is important to consider other possible factors, such as gender influences and differences in age distribution. Further discussion is presented below, and adjusted estimates of relative risks from multivariable models are presented in the companion paper .
An example of the potential for misinterpretation of breed risk based on anecdote can be seen for the neurologic system. Golden retrievers and German shepherd dogs account for 6% and 8% of neurologic deaths so the impression of veterinarians might be that this is a common disease in these breeds. However, that is at least partially due to their frequency in the population. In fact, they do not have a high MR nor an increased risk for death due to neurologic disorders compared to other breeds. The highest incidence is in the St. Bernard, great Dane and boxer. According to the multivariable results (see companion paper,  these three breeds have a significantly increased risk, whereas Labrador retrievers, and both regular and miniature dachshunds have significantly decreased mortality rates compared to all other breeds.
In general, for total and diagnostic mortality and for trauma, locomotor, heart and neurological categories, females had a significantly reduced risk of death. In certain breeds and for some causes strong effects were seen. For example male St. Bernards and Bernese mountain dogs were at least 2.5 times more likely to die of neurological causes compared to females (MRR for death = 0.4) and male Newfoundlands and greyhounds were over twice as likely to die of locomotor disorders (MRR for female = 0.5). In many breeds males were significantly more likely to die from heart causes, for example, as much as five times more than females for great Danes.
Although the effect was not significant for all breeds, in general females were at up to two times greater risk of dying from tumours, perhaps not surprising considering the incidence of mammary cancer. However, the notable exception was that among Bernese mountain dogs females were significantly less likely to die with a diagnosis of neoplasia. A further exploration of these differences is presented in the multivariable analysis in the companion paper .
Female German shepherd dogs were less likely to die from other diagnoses. Although there were many causes of death in this category, the two most common for that breed were exocrine pancreatic insufficiency and circumanal fistulae (data not shown). An increased risk of males for these conditions has been reported in the literature [7, 25].
Improving the health of the population of purebred dogs should be considered not only from a medical perspective but also in terms of animal welfare . In our role as stewards of animal populations, it could be argued that we have a duty to identify and try to reduce preventable disease. These detailed statistics on mortality in various breeds provide a valuable picture of the important causes of death in a large and well-defined population of dogs. To further elucidate the age pattern of death and the relative risk of death across breeds, adjusting for gender and age distribution, multivariable analysis and survival analysis are presented in the companion paper .