Dogslife: a cohort study of Labrador retrievers in the UK

13 Studies of animals that visit primary and secondary veterinary centres dominate companion 14 animal epidemiology. Dogslife is a research initiative that collects data directly from owners 15 about the health and lifestyle of Kennel Club (KC) registered Labrador Retrievers (LR) in the 16 UK. The ultimate aim is to seek associations between canine lifestyle and health. A selection 17 of data from Dogslife regarding the height, weight and lifestyle of 4,307 LR up to four years 18 of age is reported here. taller than the UK breed standard. Dog weights continued to increase between one and four 1 years of age. Those with chocolate coloured coats were heavier than their yellow and black 2 counterparts. Greater dog weight was also associated with dogs whose owners reported 3 restricting their dog’s exercise due to where they lived. 5 These findings highlight the utility of wide public engagement in the collation of phenotypic 6 measures, providing a unique insight into the physical development and lifestyle of a cohort 7 of LRs. In combination with concurrently collected data on the health of the cohort, 8 phenotypic data from the Dogslife Project will contribute to understanding the relationship 9 between dog lifestyle and health.

meal frequency decreased as the dogs aged. Working dogs spent more time exercising than 23 pets, and dogs in Wales and Scotland were exercised more than their counterparts in England.  year of their dogs' lives and quarterly thereafter. Individual questions are detailed in 20 Appendix 1. All questionnaire answers or 'data entries' were automatically date-stamped.

21
With the exception of dog weight, all questions required an answer before the owner could 22 continue through the questionnaire. However, if the owner chose 'other' from a drop-down 23 list, a free-text box would be generated and this could be left blank. 24 25 Measurements taken by owners included the height of their dog to the shoulder until the dog 26 was 18 months of age (demonstrated via an online video). They were also asked to weigh 27 their dog when possible, irrespective of age. Owners were asked to weigh their dogs' meals then report the average daily food intake in addition to meal frequency and type of diet (for 1 example 'dried' or 'home-prepared'). Use of SI units in the UK is inconsistent so owners 2 were given the option to enter a measurement and choose their preferred units from a drop 3 down box (centimeters (cm) or inches for height, kilograms (kg) or pounds for dog weight 4 and grams (g) or ounces (oz) for food weight). Entries made in inches were automatically 5 multiplied by 2.54 and stored in cm. Entries made in pounds were divided by 2.20 and stored 6 in kg. Entries made in ounces were multiplied by 28.3 and stored in g.

26
Conservative Bonferroni corrections were applied to account for multiple testing. Return intervals were examined and time to assumed loss from the project was investigated 8 with a Cox proportional hazards model (Cox, 1972), using the survival package in R 9 (Therneau, 2014). For dogs under one year of age, this was considered to be two months after 10 their last questionnaire answer and for dogs over one year, four months. After model fitting,

11
the proportional hazards assumption was tested. The percentages of dogs aged over one, two 12 and three years that were retained within the project were reported. 13 14 Exercise 15 A weighted average of weekday (5/7) and weekend day (2/7) exercise levels was created.

16
Total daily times spent exercising (TDE) were generated by taking the midpoints of the 17 relevant exercise time categories (the 'over 2 hours' category was assumed to be '2-4 hours') 18 and summing. These times were square-root transformed (tTDE) before further analysis.

19
Univariable plots were created comparing tTDE in different groups. A multivariable, linear  (Figure 1). There were two distinct growth curves and it was hypothesised that the 2 lower curve, which was approximately 2.5 times shorter than the main curve, was generated 3 by owners who had taken measurements in inches but reported them as cm. It was also 4 thought possible that some of the very high heights were measured in cm and reported in 5 inches. 6 7 A probabilistic model was used to estimate whether entries might have been made in the 8 correct or incorrect units. Equations 1-3 describe the heights which were assumed to be the pups were born. Parameter pi is the prior probability of a measurement belonging to each 1 different error class: i.e. estimated 10% chance of being subject to each type of inches-cm 2 error and 80% chance of having the correct units. Once identified, the mis-reported heights 3 were corrected using a multiplier of 2.54 or 1 /2.54. 4 (4) 5 (5) 6 (6) 7 (7) 8 (8) 9 (9)

26
Households that did not report keeping any other pets (41%; 1,719/4,148) were in the 27 minority. A simplified description of other pets kept in participating households is shown in 28 Table 1. Families were disproportionately less likely to have another dog ( 2 = 13.7 (1df) , P 1 < 0.001) and disproportionately more likely to have a cat ( 2 = 48.4 (1df), P < 0.001) 2 compared to other households. By contrast, households comprising more than one adult were 3 disproportionately like to have no other pets (χ 2 = 22.4 (1df), P < 0.001).

5
The results of an investigation into factors associated with assumed loss to the project are 6 shown in Table 2. It should be noted that return intervals were irregular and many owners 7 assumed to be lost were instead late. The maximum return interval was nearly three years,   (Table 3). Working dogs 1 were found disproportionately in households comprising more than one adult when compared 2 to other household types (χ 2 = 14.6 (1df), P < 0.001).   The total time spent exercising was not associated with dog weight but working dogs, a group 10 that typically spent more time exercising than pets, were more than 2kg lighter than pets. The 11 mean weight of a two-year-old Dogslife LR was 26.8 kg for females and 31.6 kg for males.   to the project and dog age was correlated with many of the lifestyle factors. People who 2 described their households as 'families' or whose household included a tobacco smoker were 3 more likely to be lost to follow-up (Table 2). By contrast, retired households and those 4 including another dog were more likely to be retained. Indeed, these two factors were 5 themselves positively correlated within the cohort. In their examination of biases in a Spanish 6 cohort study, Alonso et al. (2006) found a similarly increased risk of loss with regard to 7 tobacco smokers and also that older people were more likely to be retained. With regard to the 8 excess loss of families, it is possible that time constraints were a contributing factor because 9 families were also a group who spent less total time exercising their dogs.

11
Of the data reported in this publication, the proportions neutered were likely to be the only 12 measures that might be adversely affected by retention bias. For dogs whose owners ever 13 answered the neutering question, just 28.1% of dogs were apparently neutered, but the 14 denominator includes many dogs whose owners were effectively lost to the project before 15 their dogs were old enough to be neutered. One would expect the prevalence of neutered dogs 16 in the cohort to increase with age, as shown in Figure 4, and the prevalence of neutering in 17 Dogslife registered dogs over three years of age reached 0.67 for females and 0.55 for males.   Nearly 30 years ago, LR were identified as the most likely breed to be overweight in the UK 25 vet visiting dog population (Edney and Smith, 1986) and it is of concern that the average 26 weight of the cohort continued to increase, approximately linearly, at 0.89 kg per year 27 between one and four years of age. Whilst it is not possible to extrapolate beyond the age range of the data, if weight continues to increase markedly with age, an expanding proportion 1 of the cohort will become subject to the health consequences of obesity. For example, it has 2 been demonstrated in Elkhounds that there is an association between dogs that were