To assess the association of dog ownership on PA and SB of older adults, a case-controlled design was used where study participants (DOs and NDOs) were matched on a range of demographic variables. Using activPAL monitors, data was continuously sampled for three one-week data collection periods over the course of a year. This design was employed to reduce the risk of bias from drop-outs (e.g. if the third data collection period was always winter) and thus aimed to create a data set that was representative of a broad range of weather conditions.
Full ethical approval was granted from the School of Life Sciences delegated authority of the University of Lincoln ethical approval committee, with further review and approval given by the WALTHAM animal welfare and ethical review board. All participants provided written, informed consent, and could withdraw from the study at any time without providing a reason.
Sample size calculation
A sample size analysis indicated that 27 older adults per group would be sufficient to have 80% power to detect a difference in time spent walking of 30 min per day (as measured by activPAL), at a 5% significance level (Dall, utilising unpublished data from ). Allowing for a drop-out rate of 25%, the final target sample size was 40 per group.
Recruitment of participants took place between April 2013 and November 2014 until the target sample size was reached. A multi-point recruitment strategy was implemented using advertisement of the study on local radio and press, veterinary surgeries and other locations such as day centres, community groups, and local libraries. Participants were also given the opportunity to recommend others for the study although these participants were excluded from becoming the recommender’s matched pair to prevent social influences on outcome measures. Three distinct geographical regions in the U.K. (Lincolnshire, Derbyshire and Cambridgeshire, selected for convenience) were targeted concurrently.
Participants (both DO and NDO) needed to be aged 65 years or over, reside in a private residence in one of the three chosen geographical areas, have no scheduled health intervention(s) that could alter their PA during the time of data collection (e.g. scheduled surgery) and be able to walk unaided for a minimum of 10 min continuously. For DO, the latter criterion also applied to their dog(s). Participants were not excluded based on the presence or absence of specific mental or physical health conditions. Participants were assigned into matched pairs of DO and NDO based on age [+/− 5 years], gender, ethnicity, and socio-economic status [matching quintile of Townsend index  , derived from home postcode]. An additional matching factor of cat ownership was included since previous research provides conflicting evidence on the influence of this on physical activity [11, 20, 21]. At no stage during the study were participants made aware of any details of their matched pair.
PA and SB were objectively measured using a waterproofed activity monitor (activPAL™). The activPAL monitor has been validated for both postural classification and additional outcome measures in adults and older adults [15, 17].
Data collection took place between April 2013 and November 2014. For each participant, data were gathered during three data collection periods across a period of a year. Each data collection period lasted one week which occurred within one of three designated sampling intervals (March–June, July–October, November–February) to ensure data was collected across a range of seasons for each participant. Within a matched pair, data collection periods for the participants occurred within a four-week period. Initial data collection occurred throughout the year ensuring the first data collection period was not always in the same sampling interval.
Information used for matching was gathered at recruitment. At the first data collection period, participants provided further self-report information about themselves (see Additional file 1), including, height and weight (used to calculate BMI), chronic health conditions (self-reported presence/absence of at least one health condition), and distance that could be walked continuously (0.8, 1.6, 3.2, 4.8, 6.4, 8.0+ km; question asked in units of miles). At each data collection period, participants in the dog-owning group also provided demographic information about their dog(s), including age, type (pedigree, mixed breed, crossbreed), size (giant, large, medium, small, toy; examples were provided in the questionnaire), gender, and length of ownership. They also provided details of their role in caring for the dog(s), for example, what percentage of total responsibility they had for the dog, and whether the dog was usually walked on or off lead (see Additional file 1). In addition, when wearing the activity monitor, all participants completed a diary reporting the times they went to bed/got up, and the estimated times they fell asleep/woke up. This information allowed activity and SB related to waking times to be extracted from the activPALs.
Outputs from the activPAL monitors and information from the walking diaries were processed by a researcher (PD) blind to the groups. Data were downloaded and categorised using proprietary software (PALtechnologies version 7.1.18). Self-reported waking times from the diary were used in a hierarchical manner [(a) estimated sleep/wake times; (b) reported bed/get up times; (c) visual inspection] to isolate waking activity data. When diary data was not available, a second independent (blinded) researcher (BS) visually reviewed the activPAL output and estimated the waking period each day from the first and last activity in the day.
Outcome measures were calculated for the waking day via the event output from the activPAL monitor using a custom Excel macro. An event is defined as a continuous period of a single posture or activity . Waking and sleep times were used exactly as recorded. Any event crossing the wake/sleep time was cut at that point, and only the part within the waking day was included in analysis. PA outcomes were the time spent walking, the time spent walking with a cadence of over 100 steps/min (equivalent to MVPA ), the number of steps taken and the time spent standing. SB measures (see ) were time spent sedentary, number of sedentary events, and the number and time spent sitting in prolonged sedentary events (> 30 mins). Although the duration of the waking day may have varied within and between participants, the proportion of the waking day engaged in an activity were not used for analysis, because choices, such as time of getting out of bed, may have formed an integral part of the lifestyle of participants. Finally, a binary outcome measure based on adherence to current PA guidelines for older adults (150 min per week of moderate PA ) was calculated using the total time spent walking at a moderate cadence across the data collection period. A pro rata threshold for duration of moderate activity was created based on the number of days of data assessed [i.e. 150 min per week* (number of days of assessment/7)], and participants were judged to have met the guidelines if they exceeded this threshold. This outcome measure was calculated separately for each data collection period for each participant.
The same blinded researcher who undertook the data processing performed the statistical analyses. Data from a data collection period for a given participant was included in the analysis if there were at least three waking days at that data collection period. Pairs of participants were included in analysis if there were data from at least one data collection period for each participant in the pair. Baseline demographic variables and matching characteristics and hours awake during the day were compared between groups using paired t-tests or related samples Wilcoxon signed ranks tests as appropriate. Linear mixed effects models, with dog ownership as a fixed effect and a random effects structure of data collection period nested in participant nested in pair, were conducted to assess the effect of dog ownership on all physical activity and sedentary behaviour outcomes. For the proportion of individuals meeting pro-rata PA guidelines, a generalised linear mixed effect model was performed, with binomial distribution and logit link function, using the same random and fixed effects structure. Continuous data were log10 transformed as required based on inspection of the residuals, checking for assumptions of normality and constant variance. Where significant by a likelihood ratio test, the residual variance was weighted by dog ownership group. Mixed effects models using restricted maximum likelihood allow for estimation in the presence of missing data, therefore no imputation of data was deemed necessary. The effect of data collection period (time through study) on the measures was explored as both additive and multiplicative interactions with group. The model was not significantly improved by their inclusion (as tested by likelihood ratio tests) for any outcome, and so data collection period was removed from subsequent analysis. The aim of this study was to assess differences between groups (dog owners and non-dog owners) using matched pairs of participant to account for variability between groups. Matching was extremely successful, and additional testing for confounders was not performed in the models. Data were analysed using R (version 3.3.1) with libraries lme4, nlme and mutcomp, and a p-value of 0.05 was used to indicate significance.