The J-ECOH Study is an ongoing collaborative study to elucidate the risk factors for non-communicable diseases among workers in Japan from more than 10 large-scale companies. We analysed longitudinal data of its subcohort (one major company) where data on primary commuting mode has been collected at the annual health checkup. This company is located in Hitachi city, Ibaraki prefecture, which is a suburban area of East Japan. The type of workplace is a mixture of office and factories. The details of the J-ECOH Study  and the present subcohort  have been described elsewhere. In Japan, the law requires workers to undergo at least one health examination annually. In this prospective analysis, we included participants who underwent health examinations from April 2006 through March 2011, and the earliest data obtained were considered as baseline (mostly in 2006). Health checkup data obtained 5 year after the baseline examination were collected from April 2011 to March 2016. Prior to the data collection, the conduct of the J-ECOH Study was announced in each participating companies using posters to explain the purpose and procedures of the study. Participants did not provide their verbal or written informed consent to join the study but were allowed to refuse their participation. This consent procedure conforms to the Japanese Ethical Guidelines for Epidemiological Research. The study protocol was approved by the Ethics Committee of the National Center for Global Health and Medicine, Japan.
A total of 38,329 participants (32,486 men and 5843 women) aged 30 to 64 years at baseline underwent a baseline health examination and a follow-up examination 5 years after baseline measurement. Of these, 478 participants who had a history of cancer or cardiovascular disease at baseline were excluded because these diseases may influence both commuting mode and BMI. Of the remaining 37,851 participants, we further excluded 7102 (18.6%) who did not have data on commuting mode (n = 4326), lifestyle habits (smoking, n = 4266; alcohol, n = 2783; sleep, n = 4272), and occupational factors (overtime work, n = 5511; occupational physical activity, n = 4444; shift work, n = 5225; job position, n = 5275) at baseline, leaving 30,749 workers. Some participants met two or more exclusion criteria. Three participants did not have data on body height at baseline, and it was imputed from the following health checkup data. Then, we additionally excluded 991 workers without data on commuting mode after 5-year follow-up, remaining 29,758 participants (25,808 men and 3950 women) aged 30 to 64 years (mean [SD]: 43.2 [8.2] years) for the main analysis. Of these, approximately 0.1 to 3.5% of them did not have 5-year follow-up data of covariates (smoking, n = 43; alcohol, n = 113; sleep, n = 195; leisure-time exercise, n = 279; occupational physical activity, n = 142; overtime work, n = 323; shift work, n = 162; job position, n = 1040); these missing data of covariates were imputed by using last observation carried forward method. Most of the missing data were replaced by 4-year follow-up data.
Exposure (primary commuting)
The primary commuting mode to work was self-reported at the annual health examination according to 4 response options of walking, bicycling, train or bus, and car or motorbike. We categorized the 4 types of commuting mode into two groups as (1) inactive commuting (i.e., car or motorbike) and (2) active commuting (i.e., walking or bicycling) or public transportation (train or bus) [20, 30, 31]. We combined active commuting and public transport use as one category because public transport use typically involves in walking [9, 32]. Then, we further reclassified the participants into 4 groups according to the baseline and follow-up commuting mode: (1) maintained inactive commuting, (2) switching from inactive commuting at baseline to active or public transportation mode at 5 years after the baseline examination, (3) switching from active or public transportation mode at baseline to inactive commuting at 5 years after the baseline examination, and (4) maintained active or public transportation mode.
For sensitivity analysis, commuting mode during 1 to 4 years after the baseline period was categorized into the following three categories: persistent inactive commuting, intermittent, and persistent active commuting or public transport use.
Outcome (obesity indicators)
The body height and weight were measured during annual health examinations, and the BMI of each participant was calculated using the formula kg/m2. For the main outcome, we obtained the 5-year changes in BMI by subtracting the baseline BMI from the BMI obtained 5 years after baseline.
Visceral and subcutaneous fat areas and waist circumference were measured in a subgroup of the present participants with additional measurements of these variables (n = 4322). Single slice imaging was performed at the umbilical level under fasting condition using a Redix Turbo CT scanner (Hitachi Medico, Tokyo, Japan) and visceral fat area, subcutaneous fat area, and waist circumference were estimated by the PC software application fatPointer (Hitachi Medico, Tokyo, Japan) as described elsewhere . Changes from the baseline examination to 5-year follow-up were also calculated for these three obesity indicators as secondary outcomes.
During examination, the lifestyle habits (smoking, alcohol use, sleep duration, and leisure-time exercise), work-related factors (physical activity at work, overtime work hours, and shift work) and socioeconomic status (job position) of the participants were assessed using a questionnaire. Details of this questionnaire have been explained previously . The metabolic markers for blood pressure, glucose, and lipids were measured following the procedures described elsewhere .
We calculated the means (SD) and number (%) to illustrate baseline characteristics of the participants according to the 5-year changes in commuting mode.
Multivariable linear regression was used to quantify the association of 5-year changes in commuting mode with simultaneous BMI changes. We treated maintained inactive commuting mode as reference. First, we adjusted for baseline variables including age (year, continuous), sex, and BMI (kg/m2, continuous) as model 1. Then, we adjusted for all factors in model 2, namely, smoking transition (continued no smoking, started smoking, quit smoking, and continued smoking), baseline alcohol use (go [unit for Japanese sake; 1 go of Japanese sake contains about 23 g of ethanol, which approximates 2 units of alcohol] per day, continuous), 5-year changes in alcohol use (go, continuous), baseline sleep duration (< 5 h, 5 to < 6 h, 6 to < 7 h, and ≥ 7 h per day), changes in the sleep duration (decrease, no change, or increase), baseline weekly exercise duration during leisure (min, continuous), 5-year changes in the exercise duration (min, continuous), occupational physical activity transition (continued sedentary work, started sedentary work, started physically active work [standing, walking, or physically active], continued physically active work), job position transition (continued low-rank position, promoted, demoted, and continued high-rank position), shift work transition (continued no shift work, started shift work, quit shift work, and continued shift work), and overtime work transition (continued low overtime work hours, increased overtime work hours, decreased overtime work hours, and continued high overtime work hours). The estimated changes in BMI according to commuting mode were calculated from these models. Subgroup analysis was performed for leisure-time exercise transition (persistently no exercise, quit exercise, started exercise, and persistently active) and occupational physical activity (persistently sedentary work, switched to active work, switched to sedentary work, and persistently active work) in model 2.
We performed the following analyses to confirm the robustness of the results of BMI. First, as about 20% of participants were excluded due to missing data, we performed a sensitivity analysis on the association between commuting and BMI using multiple imputation with chained equations (MICE) for missing data of exposures and covariates at baseline and follow-up. A total of 100 imputed data sets were generated. All analyses were performed on each imputed data set; the 100 estimates were combined into an overall estimate using the rules from Rubin. Second, we conducted an analysis considering the transitions in commuting mode during 1 to 4 years after the baseline. Lastly, we performed an analysis while treating public transport use as an independent category, separated from active commuting.
We repeated the main analysis for other obesity outcomes (visceral fat, subcutaneous fat, and waist circumference) where the baseline obesity outcome was adjusted for in each analysis (e.g., adjustment for baseline visceral fat areas if the outcome is visceral fat change). Two-sided p-values of less than 0.05 was considered as statistically significant. All analyses were performed using Stata 14.2 (Stata Corp, College Station, Texas, USA).