Risk Factors Associated with Lifespan in Pet Dogs Evaluated in Primary Care Veterinary Hospitals
ABSTRACT
The objective of this population-based retrospective cohort study was to identify factors associated with lifespan in pet dogs evaluated at primary care veterinary hospitals. Dogs ≥3 mo of age that visited any of 787 US hospitals at least twice from January 1, 2010, through December 31, 2012, were included. Survival curves were constructed for dogs by reproductive status, breed, body size, and purebreed (versus mixed-breed) status. Multivariate Cox proportional hazard regression was performed to identify factors associated with lifespan. There were 2,370,078 dogs included in the study, of whom 179,466 (7.6%) died during the study period. Mixed-breed dogs lived significantly longer than purebred dogs, and this difference was more pronounced as body size increased. Controlling for other factors, dogs of either sex had a greater hazard of death over the study follow-up period if sexually intact rather than gonadectomized. For dogs who lived to 2 yr of age, the hazard of death decreased with increasing frequency of dental scaling. Our findings support previous reports of the impact of body size and gonadectomy on lifespan and provide new evidence in support of ultrasonic dental scaling and mixed breeding.
Introduction
Over the past 150 yr, selective breeding of dogs derived from traditional foundation stock has led to the modern breeds, which are genetically isolated and fairly inbred.1 An adverse effect of this practice has been the emergence of a wide range of breed-specific genetic diseases and disease predispositions that are as diverse among breeds as are other aspects of their phenotype.2 Some diseases are less breed-specific but appear to be associated with certain body sizes and shapes; examples of these include hip dysplasia or gastric dilation and volvulus, which are more common in large-breed dogs than in smaller breeds.3,4 It is not surprising, consequently, that breed can also influence a dog’s aging process and, therefore, life expectancy. The general belief is that large-breed dogs have shorter lives than small-breed dogs, with body weight being a better predictor of life expectancy than height at the withers.5,6
Prior studies of lifespan in dogs have typically relied on data from dead animals and the calculation of life expectancy on the basis of age at death.7–9 This approach results in an underestimation of lifespan as a result of right censoring; that is, all dogs in a birth cohort who are still alive at the end of the study period will not be included as they should be. Such bias is obvious in longitudinal studies of individuals after birth but becomes obscured in cross-sectional studies of dead individuals, which constitute the majority of studies on lifespan performed to date.10 Given that the degree of bias due to right-censored data can vary considerably among different study populations, a comparison of findings among studies becomes difficult, if not impossible.11
Statistical methods exist that omit bias attributable to right-censored data, including Kaplan-Meier estimation (product limit estimator) and Cox proportional hazards regression.10 Other methodologic approaches have been used and various canine populations studied to estimate lifespan and elucidate causes of death, each with limitations unique to the populations used such as insured populations or primary care versus tertiary care patients.12–15
The purpose of the present study was to use a primary care practice population database of pet health records to investigate our hypothesis that the lifespan in pet dogs in the United States would be associated with breed status (pure versus mixed), breed, body size, mean adult body weight (because body weight has been proposed to be a better predictor of lifespan in dogs than height, breed, or breed group), and reproductive status.6 Another objective was to use data collected on live dogs as recorded in their health records to estimate survival curves for the most common dog breeds and other dog groupings. In addition, because the medical records also included information on veterinary care received, and because little evidence exists regarding any influence of that care on lifespan, we also sought to investigate our hypothesis that lifespan would be associated with the frequency of certain veterinary services for which full, consistently recorded data was available, namely veterinary visits and ultrasonic dental scaling (under general anesthesia).
Materials and Methods
Banfield Pet Hospital represents a national network of primary care veterinary hospitals in the United States, joined by a common proprietary electronic medical records system. These hospitals are visited annually by ∼4% of the American Veterinary Medical Association–reported pet dog population in the United States.16 All patient data entered into this system is uploaded on a daily basis to a central database, which contains the medical histories of several million patients including information on age, breed, reproductive status, and death date. Wellness care is a large focus of the veterinary service provided in these hospitals (prepaid medical care plans that include preventive care and unlimited hospital visits), intended to promote regular visits and health monitoring.
Data Collection
Dogs considered for inclusion in this retrospective cohort study were those evaluated between January 1, 2010, and December 31, 2012, at any Banfield Pet Hospital. To be included in the study, dogs were required to have had at least two hospital visits during the study period. Dogs who died or were euthanized during this period were required to have survived initially until at least 3 mo of age to avoid the potential influence of death as a result of congenital problems. Dogs were excluded when their birth date, breed, or gonadectomy status was unknown. Dogs with an age exceeding 30 yr at euthanasia, death, or last visit were considered to have had their birth date inaccurately reported and were also excluded.
Data was extracted or calculated from the electronic medical records for each dog from the point of first visit on record (which, in many situations, was prior to January 1, 2010) through December 31, 2013 (which provided at least 1 yr of follow-up for dogs who might have had their first visit near the end of the inclusion period on December 31, 2012). This data included signalment variables, gonadectomy status, breed, body size category, mean adult body weight, and age at last visit, euthanasia, or death as well as the veterinary service variables frequency of veterinary visits, frequency of ultrasonic dental scalings performed (under general anesthesia), and frequency of anal gland expression. Mean, rather than median, body weight was chosen because of the desire to capture the full range of body weights on record, which the median would not have reflected. Frequencies of veterinary services represented rates over time and were calculated as the number of each service recorded divided by the time elapsed from the first recorded visit to study exit. Frequency of anal gland expression, which we hypothesized a priori would not be associated with lifespan because of a lack of biological plausibility, was included with the intention that if it was found through multivariate analysis to have an association with lifespan, then that finding would suggest the presence of a source of bias in the data that had not been accounted for.
Mean adult body weight was calculated on an individual basis from weights on record from 2 yr of age (by which point dogs of any breed were presumed to have stopped growing on the basis of unpublished growth curve data for the hospital population) to the end of the study period; thus, to have a value for this variable, dogs were required to have lived for ≥2 yr. Because there is no peer-reviewed standard for canine body size classification, we used the classification system used in a lay publication involving the same patient population.17 That system had been derived from cutoffs identified through internet searches of breed association and pet food manufacturer websites as follows: small, <9.1 kg (20 lb); medium, 9.1 to <22.7 kg (20 to <50 lb); large, 22.7 to <40.9 kg (50 to <90 lb); and giant ≥40.9 kg (90 lb). Then, an internal standard was developed using data from all canine patients in our entire database (no time restriction; excluding duplicate visits, dogs <2 yr of age, and dogs with recorded weights on file <1 lb or >300 lb), by which each purebred breed in the entire database (485 total breeds recorded) was assigned to one of the aforementioned body size categories on the basis of the median population body weight for the breed.
For analyses comparing purebred and mixed-breed dogs within body size categories, dogs whom veterinary personnel had flagged in the record as being a mixed version of a particular breed (e.g., dogs for whom personnel had entered a specific breed but had also checked a mixed-breed flag option) were grouped into body size categories corresponding with the indicated breed. For example, a dog for whom the recorded breed was Labrador retriever (a breed classified as large) but with a “mixed-breed” flag in its record was treated as a mixed-breed large dog rather than a purebred large dog. All other mixed-breed dogs (i.e., those lacking any mention of a pure breed designation) were excluded from these analyses.
Statistical Analysis
All statistics were computed with the aid of statistical softwarea. Values for signalment variables such as breed, breed status (pure versus mixed), and reproductive status are summarized as counts and percentages. Values for veterinary service variables, such as frequency of veterinary visits, dental scaling, or anal gland expression, are summarized as median (interquartile range), and correlations among these variables were evaluated by calculation of the Pearson correlation coefficient (r).
To provide lifespan estimates for dogs in various signalment groupings, without controlling for other factors, Kaplan-Meier survival curves were constructed.18 The outcome of interest was age at euthanasia or reported death. The endpoint of the follow-up period was December 31, 2013. Dogs who survived to the final visit within the study period were included in the analysis and censored at the point of the final visit. Dogs who had visits beyond the study endpoint were censored at December 31, 2013. Lifespan estimates are reported as median and 95% confidence interval (CI) of the median.
To simultaneously evaluate associations of signalment and veterinary service variables with the hazard (or risk) of death over the study follow-up period, multivariate Cox proportional hazards regression was used.18 Because of the large size of the data set and the recognition that high power can result in significant but spurious findings, as well as balancing complexity of modeling and relevance of results, a single interaction term was included in the model between purebred (versus mixed-breed) and body size.
Model building was performed manually (rather than through a stepwise selection process) to identify candidate models.19 The proportional hazard assumption was tested for each candidate model by examination of the log-cumulative hazard plot for parallel lines for the groups for each model, and the goodness-of-fit test was performed.20 The final model was chosen by comparing values of the Akaike information criterion among candidate models. Because of the large sample sizes and the high likelihood that a significant effect would be detected by chance alone, only values of P < .001 were considered significant.21
Results
Animals
A total of 2,504,518 dogs visited a Banfield Pet Hospital at least twice between January 1, 2010, and December 31, 2012. Of these, 2,370,078 (94.6%) dogs met the inclusion criteria, including 85,630 (3.6%) dogs who were reported as having died and 93,836 (4.0%) who were euthanized on or before December 31, 2013, the end of the follow-up period. A total of 787 hospitals in 43 states were represented. States without hospitals included Hawaii, Alaska, Wyoming, North Dakota, West Virginia, Vermont, and Maine.
Distributions of dogs by reproductive status were as follows: sexually intact female, 230,974 (9.8%); sexually intact male, 322,958 (13.6%); spayed female, 906,252 (38.2%); and neutered male, 909,894 (38.4%). Small-sized breeds made up the largest proportion of the study sample for which a specific breed (other than mixed) was recorded (1,208,085; 52.8%), followed by medium-sized breeds (545,219; 23.8%), large-sized breeds (493,539; 21.6%), and giant-sized breeds (42,882; 1.9%). Approximately one-third (706,000; 29.8%) of dogs were recorded as mixed-breed. For these dogs, either the breed entered in the record by veterinary personnel was mixed (n = 80,353) or the personnel had entered a breed other than mixed but also selected a mixed-breed flag (n = 625,647). The median number of dental scalings and anal gland expressions per year for all dogs was 0 (interquartile range, 0.2) and 0 (interquartile range, 0.4), respectively. The median number of visits per year was 2.7 (interquartile range, 3.1). The mean adult body weight was calculated for 2,289,725 (96.6%) dogs.
A total of 1,848,824 (78.0%) included dogs had been enrolled in a wellness plan at some point in the past, and 521,254 (22.0%) dogs had never been enrolled in such a plan. Because wellness plan enrollment (duration of enrollment or whether a dog had ever been enrolled) was strongly correlated (r > 0.80) with the frequencies of dental scaling, anal gland expression, and veterinary visits, that variable was not evaluated further. These frequency variables had a minimal degree of correlation (r = 0.01–0.21) with each other.
Survival Estimates
Kaplan-Meier survival analysis revealed that, without controlling for other factors, survival rates differed significantly (P < .001) by body size category for all dogs for whom a body size could be determined (i.e., excluding dogs for whom the recorded breed was mixed), with giant-sized dogs having the lowest survival rate (Figure 1). The median (95% CI) estimated lifespan of dogs in the giant, large, medium, and small categories was 11.11 yr (11.01–11.25 yr), 13.38 yr (13.35–13.40 yr), 13.86 yr (13.82–13.89 yr), and 14.95 yr (14.92–14.98 yr), respectively, and all estimates differed significantly (P < .001). Mixed-breed dogs had a significantly (P < .001) higher survival rate than purebred dogs. The median (95% CI) estimated lifespan of mixed-breed dogs was 14.45 yr (14.42–14.49 yr) and differed significantly (P < .001) from that of reportedly purebred dogs, which was 14.14 yr (14.12–14.15 yr).
Citation: Journal of the American Animal Hospital Association 55, 3; 10.5326/JAAHA-MS-6763
Gonadectomized dogs also had a significantly (P < .001) greater survival rate than sexually intact dogs up to 15 yr of age (Figure 2). However, beyond that point, the difference between the groups was no longer significant (P ≥ .07). The median (95% CI) estimated lifespan of sexually intact females, sexually intact males, neutered males, and spayed females was 13.77 yr (13.68–13.86 yr), 14.09 yr (14.03–14.16 yr), 14.15 yr (14.13–14.18 yr), and 14.35 yr (14.33–14.37 yr), respectively. For both sexes, survival rates and estimated lifespans were significantly (P < .001) higher for gonadectomized versus sexually intact dogs. Kaplan-Meier survival curves and lifespan estimates were also generated for the five most common breeds within each body size category (Figure 3, Table 1).
Citation: Journal of the American Animal Hospital Association 55, 3; 10.5326/JAAHA-MS-6763
Citation: Journal of the American Animal Hospital Association 55, 3; 10.5326/JAAHA-MS-6763
Multivariate Hazard Analysis
Pearson correlation analysis revealed that none of the service frequency variables evaluated were strongly correlated with each other (all values of r ≤ 0.21), so the relationship of all frequency variables on lifespan was evaluated via multivariate proportional hazard regression, controlling simultaneously for the other variables already identified as influential (purebred status, body size, mean adult body weight, and reproductive status) as well as for the interaction between purebred status and body size. The Spearman correlation coefficient between mean body weight and body size was 0.85 and significant (P < .001). Because of this high degree of collinearity, model building required a choice between the two variables. Body size, rather than mean weight, was included in the fitted model as model fit criterion values were comparable, and body size would lend itself to more clear interpretation and communication of results, especially for the interaction between purebred status and body size.
All variables in the final model were significant (P < .001; Table 2). Independent of the other factors, frequency of ultrasonic dental scaling had a large, positive (decreased hazard) association with lifespan that increased with increasing number of cleanings per year. Anal gland expression had a smaller but still significant positive association as well. Visit frequency, on the other hand, appeared to increase the hazard of death over the study period.
Discussion
To our knowledge, the present study included the largest sample of dogs ever used to investigate factors associated with lifespan. Use of data from a national, primary care veterinary practice population with standardized recordkeeping and daily record submission to a central database yielded results that we believe can be more easily extrapolated to the general pet dog population than results from referral hospital or insured pet populations. A survival analysis approach yielded more valid results than calculations of mean or median ages at death, which fail to account for dogs still alive at the time of those calculations, potentially resulting in an underestimation of lifespan.10 Our methodology accounted for right-censored data and included dogs from primary care hospitals rather than tertiary care facilities or insured populations, which would explain the longer lifespans identified in our study than in previous studies.7–9,13,14,22–24
Large dogs are generally believed to have a shorter life expectancy than smaller dogs, as our findings confirmed. Scientific consensus has leaned toward a somewhat shorter life expectancy in purebred dogs.1,12 The present study offered a novel approach to answering this question by comparing lifespan between mixed-breed and purebred dogs within different body size groups, showing that the mixed-breed advantage in lifespan was proportional to body size, with the difference between mixed-breed and purebred dogs being most pronounced for giant-sized dogs and least pronounced for small-sized dogs.
It should also be noted that body size was a more important predictor of lifespan than purebred status; mixed-breed large and giant dogs did not live as long as purebred small-breed dogs. Interestingly, when the five most common breeds in all five size categories were evaluated, considerable breed differences were identified in Kaplan-Meier survival curves and lifespan estimates. These differences among purebred dogs of similar size belonging to different breeds may have been attributable to breed-specific diseases, varying degrees of inbreeding depression among breeds, or other factors.
Overall, spayed females had the longest life expectancy of all dogs (∼30 wk longer than sexually intact female dogs), suggesting a clear and clinically important advantage to gonadectomy for female dogs. A lesser (∼3 wk difference) but still significant positive relationship with gonadectomy was also identified for male dogs, although the clinical importance of this finding is likely low. A beneficial association of gonadectomy with lifespan has previously been described for female dogs.7,25,26 The usual explanation for this effect is that the increase in lifespan reflects a lack of pregnancy- and birth-associated hazards as well as a significant reduction in the risk of mammary tumors and uterine infections in spayed females. In contrast, the situation is less clear in males, for which the literature is scant. Hoffman et al. evaluated the association of gonadectomy with canine lifespan and found a protective association for both sexes in a teaching hospital population.13 This was also found to be true for gonadectomy in our primary care practice population, with significant associations with lifespan identified in both sexes and females having the greater advantage.
Studies of the impact of veterinary services on patient longevity are theoretically difficult to perform in a prospective or experimental manner, particularly in pets versus research animals, because of the challenge in ensuring uniformity of care within the treatment group and the ethical concerns that would arise when attempting to establish a control group that lacks such services. To our knowledge, the influence on lifespan of veterinary visit, vaccination, prophylactic deworming, or dental cleaning frequency has not been investigated in pet dogs, although such practices are strongly advocated by professional veterinary associations such as the AAHA. 27
Three interesting findings arose from the multivariate modeling that controlled for factors associated with lifespan in dogs. The first was that increasing visit frequency increased the hazard of death over the study period. A supposition made when these variables were chosen for the study was that visit frequency could be used as an indicator of general level of care provided by owners that was independent of dental scaling frequency. Given the increasing focus on preventive medicine within the veterinary profession, one would expect that the more a dog visited the veterinarian, the sooner disease could be detected and treated, thereby extending its lifespan. However, given the results reported here, our interpretation would be that visit frequency increases with increasing health problems that could portend a decrease in lifespan. We believe this explanation likely to be valid given that, in our experience analyzing health data of the source population, the risk of illness diagnosis also increases with increasing visit frequency, which we attribute to the increased probability of disease detection.
Frequency of ultrasonic dental scaling was selected as a variable in our analysis because the data could be more easily and reliably extracted from the electronic medical records history than other aspects of preventive care history such as vaccination or deworming. In addition, the techniques and equipment used for dental scaling and anal gland expression remained fairly constant over the study period, whereas other services changed in both manner of recording and manufacturers. The magnitude of the beneficial association of dental scaling frequency with lifespan (hazard ratio = 0.817) identified in the study was surprising. When comparing two dogs, all other factors being equal, a difference of one dental cleaning per year is associated with a nearly 20% decreased risk for death. However, this finding cannot be interpreted as indicating that dental scaling directly increases lifespan in dogs because of the observational nature of the study design. Indeed, dental scaling may be in whole or in part a surrogate indicator for some other factor such as general degree of veterinary (or owner) care dogs receive, owner attachment, or dog health status. One might also argue that dogs who live longer have more opportunity to have their teeth cleaned than dogs who die earlier and therefore that scaling has no association; however, we believe this possibility was controlled for through the use of dental scaling frequency rate (number of scalings per year over the dog’s lifespan) rather than a simple count of scalings in the modeling process.
Frequency of anal gland expression was chosen as a control variable because of our a priori hypothesis that it would not be associated with lifespan. It was therefore surprising that this variable also had a beneficial association with lifespan, although the magnitude of the effect was far smaller (hazard ratio = 0.916) than for dental scaling frequency. Again, as for dental scaling, frequency of anal gland expression may have been a reflection of general care provided or other unidentified factors. Because the correlation between frequencies of anal gland expression and dental scaling was weak (r = 0.22), we believe the overlap between the actual factors underlying the observed beneficial association for both variables was likely small.
In the present study, potential for misclassification of purebred versus mixed-breed status existed because either the dogs’ owners or the veterinary personnel assessed and recorded a specific breed (including an option of mixed-breed) and could have assigned a mixed-breed flag in the record if indicated. Given the high proportion of study dogs who were deemed purebred by this method (71.2%) and therefore treated as such in the study, it is likely that some of the reportedly purebred dogs were actually of mixed breeding. We believe this limitation should not impact generalization of our findings to the US pet dog population; however, the findings may be less generalizable to confirmed purebred dogs.
A unique feature of the data used in this study is that a large number of dogs remained alive at the end of the study period, leading to right censoring of a high percentage (92.4%) of individuals; nonetheless, the Kaplan-Meier and Cox proportional hazards techniques that we used take the bias introduced by right censoring into account.28 An uncommon but potential problem was the large number of dogs, which could have led to statistically significant, but clinically or biologically negligible, associations attributable to the high power to detect a difference in lifespan between groups. For this reason, we focused solely on results pertinent to tests of hypotheses established a priori rather than report incidental findings. We believe the mentioned limitations had minimal impact on the major study findings that a significant lifespan advantage exists of small-sized over large-sized, neutered female over sexually intact female, neutered male over sexually intact male, and mixed-breed over purebred dogs as well as dogs who undergo dental scaling or anal gland expression over those who do not. Additionally, mixed-breed dogs have a lifespan advantage over purebred dogs that is proportional to body size.
Conclusion
Although additional research is necessary to elucidate the specific mechanisms underlying the differences in lifespan by purebred status, breed size, and body weight, the knowledge of these differences as well as the association with gonadectomy and other veterinary services can assist practitioners in counseling dog owners about gonadectomy, pure versus mixed breeding, and life expectancy.
Survival curves and 95% CIs (shaded regions) for dogs evaluated at US primary care veterinary hospitals by body size category. Note that small standard errors result in small (and poorly visible) CIs. CI, confidence interval.
Survival curves and 95% CIs (shaded regions) for dogs evaluated at US primary care veterinary hospitals by reproductive status. CI, confidence interval.
Survival curves and 95% CIs (shaded regions) for the most common (A) giant dog breeds, (B) large dog breeds, (C) medium dog breeds, and (D) small dog breeds evaluated at US primary care veterinary hospitals. CI, confidence interval.
Contributor Notes
