Design and study sample
The data used in this cross-sectional study were collected between November 2008 and November 2009 in Stockholm, Sweden as part of the Swedish Neighborhood and Physical Activity (SNAP) study. The SNAP study was originally designed to investigate the association between neighborhood walkability and physical activity [24]. A total of 32 neighborhoods were sampled based on walkability (based on data provided by Statistics Sweden, the City Planning Administration in Stockholm and the company Teleadress) and neighborhood income (based on data provided by Statistics Sweden) in order to ensure variation in neighborhood-level walkability and socio-economic status. Data were collected throughout the study period, except between 9 December 2008 and 12 January 2009 and between 16 June and 17 August 2009 (these two time periods correspond to the winter and summer holidays in Sweden, respectively).
The sampling procedure has been described in detail elsewhere [24]. Briefly, neighborhood walkability and income were calculated for all 408 basic areas (neighborhoods) in the city of Stockholm. Geographic Information Systems (GIS) were used to calculate walkability as an index comprising z-scores for residential density, street connectivity and land use mix. Neighborhoods in the first to fourth walkability index deciles were considered less walkable, and those in the seventh to tenth deciles where considered highly walkable. Neighborhood income in each area was calculated as the median disposable family income, taking the number and age of family members into account. Neighborhoods in the second to fourth neighborhood income deciles were considered to be of low income, and those in the seventh to ninth deciles of high income. Four neighborhood categories were created: high walkability/high income, high walkability/low income, low walkability/high income and low walkability/low income. A total of 32 neighborhoods (eight from each category) were sampled for the study.
The SNAP study aimed to recruit 75 participants from each of the 32 neighborhoods, i.e. 2,400 in total. Simple random sampling of 8,000 individuals aged 20 to 65 (200 from each neighborhood) was performed by the Stockholm Office of Research and Statistics. Immigrants who had arrived in Sweden after 2003 were excluded since knowledge of Swedish was an inclusion criterion (see below). A total of 6,089 individuals had a listed landline or mobile phone number and were included in the recruitment procedure. Of the 4,747 individuals who were reached, 4,369 met the three inclusion criteria: (1) being able to read and write Swedish, (2) having lived in the neighborhood for at least three months, and (3) having no serious impaired ability to walk. The final study population for analyses, after exclusion due to missing data, consisted of 2,037 individuals, which gave a response rate of 47% (2,037/4,369). Recruitment of participants was performed concurrently in all included neighborhoods by the telemarketing company Markör AB (Örebro, Sweden). Markör AB has previously been involved in the recruitment of participants for large-scale research studies. Lists of enrolled participants were delivered to us on a weekly basis and a package containing an accelerometer, an accelerometer logbook, a questionnaire and a prepaid return envelope was sent to the residential address of each participant.
Availability of exercise facilities
Availability of exercise facilities was objectively measured using GIS. To assess area of exposure, neighborhoods were defined by creating a buffer zone originating from the residential address of each participant using the Network Analyst extension in ArcGIS/ArcInfo 9.2 (ESRI Inc., Redlands, California, USA). Data on the road network, including cycle paths and footpaths, was obtained from the City Planning Administration in Stockholm. Line-based network buffer zones were created by following the road network in all possible directions from each residence for 950 meters, and then creating a 50-meter buffer zone in all directions from the center of the road (Figure 1). 1,000-meter buffer zones are likely to represent areas that can be reached in daily life by a large majority of the adult population and have been used to define neighborhoods in previous research [25, 26]. Data from 2008 on the locations and business names of exercise facilities were provided by Teleadress, a company created when the government-owned telecoms agency was privatized and one of the leading providers of geo-coded data on businesses and private individuals in Sweden. The data from Teleadress included privately and publicly owned exercise facilities that have a registered telephone number and/or those that had provided information about their existence to Teleadress. The database is updated continuously and inclusion is free of charge. The data included nine categories of exercise facilities: “gym/fitness center”, “sport facility”, “tennis court”, “dance class center”, “public ice rink”, “squash court”, “sports hall”, “public baths” and “badminton court”. Most facilities were indoor facilities; only a few in the category “tennis court” were outdoor facilities. A vast majority of the exercise facilities were charged. Exercise facilities located within buffer zones were manually screened to identify those that did not offer exercise to the adult population. These facilities, as well as those not offering any exercise opportunities on site, were excluded. We identified 341 exercise facilities; 58 of these were excluded because they did not offer exercise to the adult population on site. Individual exercise facilities offering more than one activity received a count for each activity. For example, an exercise facility listed in both the “gym/fitness center” and “squash court” categories was counted as two facilities. The category “sport facility” was often present as a general description together with a more specific category. For example, gyms often appeared in both the “sport facility” and “gym/fitness center” categories. “Sport facility” was thus only counted when the only category present, and not when accompanied by another exercise facility category.
Time spent in moderate to vigorous physical activity
Actigraph GT1M accelerometers (ActiGraph, Pensacola, Florida, USA) were used to objective measure participants’ physical activity. Participants were asked to wear the accelerometer on the hip or lower back for 7 consecutive days and to remove it only when sleeping or engaging in water-based activities. A study comparing placement of accelerometers on the hip or lower back under free-living conditions found that the position of the accelerometer had no effect on the estimation of time spent in moderate to vigorous physical activity [27]. Four standardized text messages were sent to each participant’s cell phone during the 7-day measurement period to improve compliance. The Actigraph GT1M measures acceleration in the vertical axis at a frequency of 30 times per second (30 Hertz). These accelerations are summed within 60-second periods (epoch) and the output is referred to as “counts”. Non-wear time was defined as 30 or more consecutive minutes with zero counts, and 10 h of wear time was required to constitute a valid day. Accelerometer wear time was calculated by subtracting non-wear time from 24 h. Variance analysis of our own accelerometer data showed that 6 or 7 valid days were required for inclusion in the analysis [28]. Time spent in moderate to vigorous physical activity was determined using Freedson’s cut-off point for accelerometer counts [29], which is ≥1,952 counts/min. This cut-off was applied to each minute of wear time for the valid days. The mean time per day spent in moderate to vigorous physical activity during all valid days was used as the outcome variable.
Meeting physical activity recommendations
According to the global physical activity recommendations of the World Health Organization, adults should engage in ≥150 min of moderate physical activity or ≥75 minutes of vigorous physical activity per week, or an equivalent combination of the two. Activities should be performed in bouts of ≥10 min [30]. In the present study, participants were considered to have met these recommendations if they accumulated ≥150 min of moderate to vigorous physical activity in bouts of ≥10 min within a week. Bouts of moderate to vigorous physical activity were identified as 10 or more consecutive minutes with ≥1,952 counts per minute. During each bout of physical activity, the number of counts per minute was permitted to dip below this cut-off for 1-2 min. This approach, which allows for brief pauses in physical activity (for example when stopping at a red light or tying a shoelace), is recommended [31] and has been used previously [5]. Bouts of physical activity were identified during wear time on valid days as defined above. Weekly time spent in bouts of moderate to vigorous physical activity for participants with 6 valid days were extrapolated to 7 days using the mean of the six valid days (mean value for the 6 valid days multiplied by 7).
Time of year
The year was divided into four periods: January-March, April-June, July-September and October-December. The Swedish climate offers substantial weather variation. According to the Swedish Meteorological and Hydrological Institute (http://www.smhi.se/en/services), daily mean air temperature varied between -7°C and +19°C in the city of Stockholm during the data collection period. January-March was the coldest period with a daily mean temperature of -1°C.
Socio-demographic information
Participants’ socio-demographic information was based on self-report. Age was categorized as 20-30 years, 31-40 years, 41-50 years and 51-66 years. Marital status was dichotomized as married/cohabiting or single. Income was calculated by dividing the gross family income by number of people living in the household, with children under the age of 18 being given a consumption weight of 0.5. Income was then categorized as low (<150,000 SEK/year), middle (150,000-349,999 SEK/year) and high (≥350,000 SEK/year).
Statistical analysis
The association between availability of exercise facilities and time spent in moderate to vigorous physical activity was analyzed by linear regression. Non-parametric cluster bootstrap estimates with 1,000 replications were applied due to the skewed distribution of the physical activity data. It is a method that constructs a number of resamples of the original dataset, each obtained by random replacements of the original dataset and assuming an identically distributed population. Bootstrapping techniques have been used in previous studies of the association between environmental attributes and physical activity [24, 32]. Two models were created: a crude model including only availability of exercise facilities and physical activity, and a full model also including sex, age, income, marital status and time of year. The full model was also adjusted for accelerometer wear time since it was found to be a potential confounder (inclusion of this variable in the model resulted in a 10% change of the regression coefficients). Standard errors were corrected for clustering effects as the data were collected within 32 neighborhoods. The regression coefficients represent differences in minutes per day compared to the reference group. Interactions and multicollinearity between the explanatory variables in the full model were examined.
The association between availability of exercise facilities and whether or not participants met the physical activity recommendations (yes/no) was analyzed by logistic regression. Two models were created: a crude model including only availability of exercise facilities, and a full model also including sex, age, income, marital status and time of year. Accelerometer wear time was not a confounder and was not included in this model. Standard errors were corrected for clustering effects in the data. Interactions between explanatory variables in the full model were examined. Goodness of fit was estimated by the Hosmer-Lemeshow test [33].
All statistical analyses were performed using STATA 10.1 (StataCorp, College Station, Texas, USA) and statistical significance was determined at α < 0.05.
Non-response analysis
Results from a telephone-based non-response analysis of 205 randomly selected non-responders showed that the proportion of females was slightly higher among participants compared to non-participants. Participants were slightly older than non-participants. There was no significant difference in income between participants and non-participants.
Ethics
Ethical approval for this study was granted by the Ethics Committee of Karolinska Institutet, Stockholm. Written informed consent was obtained from all participants.