Outcome variables
Sedentary behavior
A validated Japanese questionnaire14The questionnaire, which measures sedentary activity in six specific behaviors across three domains (work, transport, leisure) was used. Participants were asked to report their average daily sedentary time for each behavior over the past week. As the outcome measures, we used three specific sedentary behaviors on workdays: sitting time at work, sitting time in cars, and sitting time using public transport. These behaviors were defined as occurring in or near participants’ work environments, since workers performed them on workdays. This questionnaire has moderate to high testretest reliability (intraclass correlation coefficient). [ICC]=0.83 for the Work Domain with a 1-week recall Period14. The criterion validity (rho=0.57 for workdays, p0.001) as well as the whole week (rho=0.49, p0.001) was moderate when comparing the questionnaire to accelerometer.14.
Exposure variables
Built environment in the workplace neighbourhood
The Abbreviated Neighborhood Energy Walkability Scale Japanese edition (ANEWSJ) was used for measuring environmental perceptions in the workplace area. Workplace neighbourhood was defined to be within a 10–15-minute walk of the workplace. Six subscales were evaluated: land use mix diversity (16 items), access to land use (6 items), street connectivity (three items), quality and availability of walking/cycling infrastructures (4 pieces), aesthetics (4 pieces) and safety (5 items). Cronbachs, which is an indicator of internal consistency for land use mix diversity and land use mix accessibility, street connectivity availability and quality of walking/cycling infrastructures as well as aesthetics and safety, were 0.91, 0.65 and 0.64, 0.72 and 0.73, respectively. We did not include subscales for residential density because they were not applicable to this study. Traffic safety was also not included (=0.26).15. Supplementary Table 2 provides details on the modified ANEWSJ that was used in this study. Except for the subscales to assess land use mix diversity, which were rated on a six-point scale (six-point scale), all items were rated. Scoring the subscales was done using the ANEWSJ published online procedures (http://www.tmu-ph.ac/pdf/ANEWS_Jpn_ver3.pdf). Higher scores indicate greater walkability. ANEWS-J was found to be acceptable in testretest reliability (ICCs=0.760.96 for residential neighborhoods).16. We examined the reliability of ANEWSJ for workplace neighborhood in a subsample (n=200). Within two weeks, participants reported their perceptions about their workplace environment twice. ANEWS-J’s testretest reliability was moderate to high for all subscales. (Supplementary table 3).
Walkability in the workplace neighbourhood measured objectively
Walk Score was used for estimating the level of walkability in work-place neighbourhoods. It is a measure that measures access to local destinations. It uses a distance decay function to reach destinations such as grocery stores and restaurants. It is adjusted using two street connectivity metrics, intersection density and block height.17. You can assign Walk Score to specific locations (e.g. postcodes or addresses) and it is normalized between 0 and 100. A higher Walk Score signifies that there are more destinations within walking range. Walk Score uses open-source data like Google, Education.com and Open Street Map to identify relevant destinations17. Japan’s Walk Score has been confirmed to be a valid measure of neighbourhood walkability18. Around 60% of participants provided their seven digit workplace postcodes (n=1360), while 777 were unable to provide their full workplace postcodes. Each workplace postcode was entered manually into the Walk Score website (www.walkscore.comTo obtain the score in JulyAugust 2020, please click (). Walk Score was available to 1163 participants. Because of the limited data available from Japan, Walk Score was not generated for 197 participants who provided a postcode for their workplace. Walk Score was negatively skewed (median score=82; 25th percentile=63; 75th percentile=94); so we used Walk Score to categorize the results. Participants were divided into three groups based on Walk Score: car-dependent (069), somewhat walkable (7089) and very walkable (9091).
Covariates
Individual-level covariates included gender, age (2029, 3039 and 4049 years), marital status (not divorced or married), educational level(tertiary or below tertiary), individual annual income (4,000,000 yen) and physical activity duration. We used the Global Physical Activity Questionnaire, (GPAQ).19to determine the amount of physical activity in each of three domains (work/transport/lounge). The World Health Organization standardised procedures were used to verify that the GPAQ data provided valid responses.20. As a covariate, the total physical activity in these domains was used. Because of missing data on total activity, four more participants were excluded. Information about the possession of a driving license (yes/no), was also collected to assess transport-related sedentary behavior. The question “How many hours have worked in the past 7 days?” was used to assess work hours. The workplace-level covariate, workplace size, was calculated by the self-reported number workers in the participants workplace (29 to 3099,100 employees or missing).
Statistic analysis
Pearsons Chi-square tests were used to examine differences in characteristics between subsample groups. Independent t-tests were used for continuous variables. Spearmans Correlation was used to examine correlations between Walk Score and perceived workplace built-environment attributes. The latter was biased.
To investigate the association of neighbourhood attributes and sedentary behavior at work, we used linear regression models. The unstandardised regression coefficients ()The associations were given 95% confidence intervals, which correspond to one standard deviation (SD), increment in perceived environmental attributes. We also calculated a 95% confidence interval (CI) for the Walk Score category using the mid category (somewhat walkingable) as the reference. Each attribute of a workplace neighbourhood was individually examined in the models. All regression analyses were done using Stata 15 (Stata Corp College Station, Texas, USA). The level of significance was set to p0.05.