Identifying associations between sedentary time and cardio-metabolic risk factors in working adults using objective and subjective measures: a cross-sectional analysis

This cross-sectional study was conducted among office workers from two Japanese enterprise groups consisting of 12 companies in Japan. The main businesses of the enterprise groups were laboratory testing and food and environmental inspection and information technology. The data were collected between January and March 2010, when an annual health examination was conducted at each enterprise in accordance with the Industrial Safety and Health Act. This study was in accordance with the Declaration of Helsinki, and approved by the ethics committee of the Institute of Health Science, Kyushu University, Fukuoka, Japan. All participants provided written informed consent.

A total of 823 full-time workers aged 20 to 64 years from the two enterprise groups were included as the target population in the present study. Of the initial sample, 13 individuals with missing data on self-reported sitting time, 14 individuals without valid accelerometer data, and 7 with missing values for biomarkers were excluded. Individuals with triglycerides ≥400 mg/dL were further excluded due to a known limitation of Friedewald’s estimation as described below (n = 6). In addition, 77 individuals with missing data on other covariates were excluded. Finally, 661 participants (145 women) were included in the analyses.

Daily sedentary time was assessed using a tri-axial accelerometer device (Active Style Pro HJA 350-IT, Omron Healthcare Co. Ltd., Kyoto). Since participants may have voluntarily modified their lifestyle behaviors after participating in the health examination, we sent accelerometers to participants prior to the health examination. Participants were given instructions to wear the device during waking hours for 10 consecutive days, except while bathing or sleeping. Data were recorded in 60-second epochs. In this study, we defined objective sedentary behavior according to activity intensity (≤1.5 metabolic equivalents [METs]) [14] and calculated daily sedentary time as minutes/day. Intensity of minute-by-minute activity was estimated by built-in algorithms containing a specific equation for sedentary activities [15]. The accuracy of the intensity estimation has been validated with the Douglas bag method [15, 16]. Non-wear time was defined as time periods of at least 60 consecutive minutes of no activity (i.e., estimated activity intensity < 1.0 MET) with allowance for up to two consecutive minutes of activities with intensity equal to 1.0 MET. With respect to intensity threshold of non-wear time, previous studies employed the same threshold with sedentary behavior (i.e., typically 100 counts per minute or 1.5 METs). Since the tri-axial accelerometer we used is highly sensitive and valid for estimating low-intensity activities, it enabled us to employ 1.0 MET as the threshold for non-wear-time. We used the SAS macro program provided by the National Institute of Cancer to compute daily wearing time, with modifications based on our accelerometer [17]. Days with at least 600 minutes of wearing time were considered valid [18]. Participants had to have at least four valid days to be included in the analyses [7].

Subjective sedentary time was assessed with the Japan Arteriosclerosis Longitudinal Study Physical Activity Questionnaire (JALSPAQ). This measure was developed to assess daily amount of physical activity for the general Japanese population and has been validated against total energy expenditure, using the doubly labeled water technique [19]. The JALSPAQ consists of 14 detailed questions in five domains (sleep, occupational activity, transportation, housework, and leisure time activity) [19]. Two domains (occupational and leisure time activity) include questions about sedentary behaviors. Overall sedentary time was calculated as the sum of leisure time and occupational sedentary time. Time spent in sedentary activity during leisure time was assessed by a single item from the JALSPAQ: “How many hours do you usually stay sedentary every day, such as television viewing, reading, listening to music, playing board games, and using the computer, in leisure time?” Similar questions on leisure time sitting have been used in other population studies [20, 21]. Occupational sedentary time was estimated based on the standard procedure for the JALSPAQ [19], from responses to two questions regarding working hours and proportion of time spent in sitting during work. Information on working hours was obtained through the following question asking: “How many hours per week do you usually work?” Proportion of time spent in sitting during work was rated as follows: 1) almost all of the time, 2) more than half of working hours, 3) approximately half of working hours, 4) less than half of working hours, or 5) almost none of the time. A proportion coefficient was given corresponding to the chosen answer as 1.0, 0.75, 0.5, 0.25, and 0, respectively. Participants were also asked to report their working hours. Occupational sedentary time was then calculated by working hours multiplied by the given proportion coefficient. Since this estimation method of occupational sedentary time has not been validated against a criterion measurement, we also coded the original Likert scale as a continuous variable in the analyses.

All cardio-metabolic risk factor data were obtained from the annual health examinations. Specialized health-examination nurses measured height, weight, waist circumference, and blood pressure using standard protocols. Body mass index (BMI) was calculated as weight divided by height squared (kg/m²). Waist circumference was measured to the nearest 0.1 cm. Systolic and diastolic blood pressure was measured at rest by an automated sphygmomanometer. Blood samples were analyzed for triglycerides, total and high-density lipoprotein (HDL) cholesterol, and blood glucose by enzymatic methods and glycosylated hemoglobin (HbA1c) by latex agglutination methods or high performance liquid chromatography. Total:HDL cholesterol ratio was subsequently calculated. Low density lipoprotein (LDL) cholesterol was calculated using the Friedewald formula [22]. All participants were asked to fast overnight before the blood test.

Information about educational attainment (more or less than college education), current smoking and drinking habits (yes/no), and marital status (married, not married) were obtained from a self-report questionnaire. Occupation was classified into the following categories: managers, professionals, clerks, sales, and others. Total calorie intake and saturated fat consumption in the past month were assessed using the brief-type self-administered diet history questionnaire, which was validated and commonly used in the Japanese adult population to assess dietary habits and nutrition intake [23, 24]. Use of antihypertensive, anti-diabetic, and lipid lowering medications were confirmed by physicians in the health examination. Depressive symptoms were measured by the Center for Epidemiological Studies Depression Scale (CES-D) [25]. Volume of MVPA was measured using the accelerometer and defined as activity intensity ≥ 3 METs. Daily amount of MVPA was calculated as MET multiplied by number of hours in a day. Self-reported MVPA was also assessed and quantified as MET · hours/day using a set of items from the JALSPAQ, which included type, frequency (number of days in a month), and duration (minutes/day) of habitual exercise and other leisure time activities. A MET value was assigned to each self-reported activity according to the 2011 Compendium of Physical Activities by Ainsworth and colleagues [26]. We did not include MVPA at work because it is highly correlated with sedentary time at work, and thus it is possible to over-adjust as previously suggested [27]. We conducted sensitivity analyses where models were adjusted for MVPA during leisure time plus other domains (transportation and housework), and results changed little (data not shown).

All statistical analyses were performed with the SAS software version 9.3 (SAS Institute, Cary, NC, USA) with a significance level of α = 0.05. Data were expressed as mean ± standard deviation (SD), or frequency and percentage. To show participant characteristics by levels of sedentary time, descriptive statistics were computed by tertiles of both accelerometer-derived and self-reported sedentary time, where the cut-off values were 7.99 and 9.76 hours/day for accelerometer and 7.14 and 9.28 hours/day for self-report, respectively.

To assess associations of subjective (self-report) and objective (accelerometer) indicators of sedentary time with cardio-metabolic risk factors, multiple linear regression analyses were performed. Triglycerides, blood glucose, and HbA1c were logarithmically transformed in order to normalize their skewed distributions. The first model adjusted for sex and age. In the second model, we additionally adjusted for educational attainment, marital status, smoking and drinking status, total calorie intake, saturated fat consumption, use of medication, and occupation. The third model further adjusted for accelerometer-assessed MVPA or self-reported MVPA, respectively, to examine whether the associations were independent of MVPA. Accelerometer wearing time was also adjusted in models where accelerometer-derived sedentary time was included as an independent variable. Further, we repeated multiple linear regression analyses with specific domains of self-reported sedentary time (leisure time and occupational) entered as independent variables. Multicollinearity for variables in each model was tested using the variance inflation factor (VIF) test. In our models, acceptable VIF value was no greater than 3 for all covariates. For the markers which were associated with both measures, we further conducted the Wald test to compare regression coefficients of accelerometer-derived and total self-reported sedentary time to determine whether magnitude of associations differed. The null hypothesis was that there was no difference between those two standardized regression coefficients. Furthermore, an interaction term was fitted to assess whether the effects of accelerometer-derived sedentary time on cardio-metabolic factors varied across self-perceived levels of sedentary behavior, and vice versa. Both independent variables were mean centered and an interaction term between two measures was included in the same model. Agreement between accelerometer-derived and self-reported sedentary time was examined using the method outlined by Bland and Altman [28]. Systematic and random error between two measurements was assessed using a linear regression model. Spearman’s correlation coefficient rho was calculated for each variable of sedentary time.

Additional file 1: Table S1 presents the characteristics of the participants by tertiles of sedentary time derived from the accelerometer and the self-report questionnaire (see Additional file 1: Table S1). The mean age of the participants was 43 ± 9 years and 78% were men. The participants reported sedentary time of 8.4 ± 3.4 hours/day on average in the questionnaire, and the mean sedentary time recorded by the accelerometer was 8.8 ± 2.2 hours/day with wearing time of 14.2 ± 1.6 hours/day. There was an increasing trend of objective sedentary time with older age, higher proportion of men, being married, higher education level, increased level of total and occupational sedentary time and MVPA measured by self-report, and decreased levels of MVPA measured by accelerometry. Total subjective sedentary time showed a significant trend with greater saturated fat consumption, longer leisure time and occupational sedentary time, and less objective MVPA.

Additional file 1: Table S2 presents the distribution of the cardio-metabolic risk factors according to the levels of sedentary time derived from two measurements (see Additional file 1: Table S2). Both self-reported and accelerometer derived sedentary time showed a significant trend with increased levels of triglycerides, blood glucose, higher total-HDL ratio, and decreased levels of HDL-cholesterol. Additionally, accelerometer-determined sedentary time showed a significant trend with BMI and HbA1c.

As shown in Tables 1 and 2, both self-reported and accelerometer-derived sedentary time showed detrimental associations with triglycerides, HDL-cholesterol, total:HDL ratio, and HbA1c, after adjusting for potential confounders including MVPA (Model 3). Although the regression coefficients for accelerometry were greater than those of self-report, the Wald test showed no difference between the coefficients except for regression on triglycerides (p = 0.012), and no significant interactions were found between two measures across the outcomes. In addition, significant associations were observed between objective sedentary time and BMI and between subjective sedentary time and blood glucose.

Table 3 shows different associations of sedentary time during occupation and leisure time with risk factors. Sedentary time during leisure time was only associated with total:HDL cholesterol ratio, while occupational sedentary time was associated with increasing levels of triglycerides, glucose, and HbA1c, and decreasing levels of HDL-cholesterol, after adjustment for all covariates. Similar results were found when occupational sedentary time was included in the model as a continuous variable (results not presented). Mutual adjustment for leisure time and occupational sedentary behavior did not alter the significance of the associations.

As shown in the Bland-Altman plot (Figure 1), the 95% limits of agreement ranges were wide (mean difference ± 6.3 hours). In addition, the difference in the two indicators increased as the average of two measures increased (beta = 0.602, p < 0.001), indicating that there was proportional bias between two measures. The correlation between accelerometer-determined and self-reported sedentary time was moderate (Spearman’s rho = 0.401, p < 0.001). In the domain-specific analysis, accelerometer-derived sedentary time was significantly correlated with occupational sedentary time (rho = 0.486, p < 0.001) but not with leisure sedentary time (rho = 0.036, p = 0.351), while self-reported total sedentary time was significantly correlated with sedentary time in both domains (occupational, rho = 0.711, p < 0.001; leisure, rho = 0.601, p < 0.001).