Study population and study design
The Health Workers Cohort Study (HWCS) is a cohort evaluating the relationship between lifestyles and chronic diseases among Mexican adults. Details of the study design and cohort characteristics are described elsewhere. (19) The HWCS participants include employees from three health systems and academic institutions in Cuernavaca and Toluca, Mexico. Of the 1,776 males and nonpregnant females aged 18 and older who participated in both the baseline and follow-up assessments, we excluded those who had missing adiposity (n=165) or SBS measure (n=103), participants without information or implausible values of calorie intake, inactivity, and tobacco use from the present analysis (n=491). The final sample size of 1,285 participants was used to conduct a longitudinal analysis with data from the baseline (2004) and follow-up (2010) (Figure 1). In a cross-sectional study, we used a subsample of 142 adults from the total HWCS population to correct the SBS from the self-administered questionnaire.
Calibration of self-reported SBS values with accelerometry
We used accelerometer based SBS to correct self-reported SBS answered by the participants when they returned the accelerometer. Sampling. The 142 participants were randomly split into two groups to cross-validate our corrected measure. One-half (n=71) was used to create a predictive generalized linear model (GLM) where accelerometer based SBS time was the dependent variable. The participants' second half or training sample (n=71) was used to test the correction equation obtained in the first half.
Measurements
Self-reported sedentary behaviors
We used an adapted and translated version of the Nurses' Health Study to capture self-reported SBS. (20)Study participants stated their time spent in specific activities during a typical week in the previous year. The instrument listed 21 activities with a metabolic equivalent (MET) value lower than 1.5 METs, including recreation (e.g., writing, using a computer for recreation, reading, watching TV, and seeing movies), household (activities such as sewing), and in a work setting (e.g., sitting time). Participants reported the frequency (days/week) and time spent daily on SB activities. We calculated the hours per day engaged in SBS by adding all weekly time invested in the mentioned activities and domains and dividing the result by seven. (19) We used self-reported SBS (hours/day) and categorical occupational SBS (<8 hours, 8-9 hours, <9 hours) to explore explicative variables in the correction or calibration model.
Accelerometer-based sedentary time
We used the ActiGraph GT3Xþ accelerometer to measure SBS in a subsample of the HWCS participants. The ACTi-Graph GT3Xþ is a triaxial device used as a reference method for movement assessment. (21) Accelerometers provided time spent in the different movement intensity spectra. Participants wore the accelerometer for seven days with an adjustable hip belt placed on the mid-axillary line of the dominant side. A set of accelerometry values was considered valid if it included at least 60 continuous minutes of nonzero values. We allowed a one- to two-minute series with values between 0 and 99 counts per minute (CPM), as performed by Actilife5 Software v5.7.4.12. We used counts <100 to classify SBS and included all subjects with at least four valid days of measurements with at least 10 hours. (22) Accelerometer data were processed using a MATLAB code developed by coauthors DS, UV, and other international studies.(23)
Adiposity
Dual X-ray absorptiometry (DXA) using a Lunar densitometer (model: DPX-GE 73735, serial number: 638405U77) (Lunar Radiation Corporation, Madison, WI, USA; software version.35, fast scan mode) was used for the adiposity measurements. Trained technicians carried out daily quality control checks using the manufacturer phantom.(19)
Other covariates
All participants completed a self-administered questionnaire that included sociodemographic information regarding education level and job position (retired, assistant, medical doctor, dietitian, nurse, manager, laboratory analyst, pharmacy attendant, researcher, etc.). and lifestyle (e.g., diet, smoking status, and physical activity) at the baseline and follow-up assessments. Participants visited our research center to undergo a physical examination and provide a venous fasting blood sample after 8 hours of fasting for laboratory testing. A standardized analyst measured glucose (mg/dL) and triglycerides (mg/dL) by chemiluminescence (Acces2; Beckman Coulter) with a routine enzymatic colorimetric method (Selectra XL instrument, Randox). Glucose and triglycerides were used as continuous values to calibrate self-reported SBS.
Dietary intake was assessed with a semiquantitative Food Frequency Questionnaire to record the consumption frequency and standard portion size of 116 food items during the previous year. Total calorie intake was calculated by multiplying the consumption frequency of each food by its nutrient and calorie content and used as a continuous variable in the association analysis as kcal/day. (19) Smoking status was classified as nonsmokers, ex-smokers, and smokers. Physical activity (PA) was measured through the self-administered questionnaire, which estimates the time (hours/week) and intensity (light, moderate, and vigorous) of activities carried out during a typical week in the past year. We calculated PA as hours/day by dividing the hours/week of moderate- and vigorous-intensity activities between seven. Participants were considered physically inactive if they performed < 150 min/week of moderate activity or < 75 min/week of vigorous activity as established by the World Health Organization (WHO) recommendations for adults aged 18-64. (24) Sex was defined as female=0 and male=1. We also included other variables, such as sleeping time (hours/day), back pain (yes=1, no=0), BMI (kg/height (m2)), chronic diseases (yes=1, no=0 for DMT2, hypertension, depression, overweight/obesity), age (in years as a continuous variable, or as categories of <30, 30-39, 40-49, 50-59, >60 years), and an education level variable for elementary school (where yes=1, no=0), middle school (where yes=1, no=0), high school (where yes=1, no=0), university and postgraduate (where yes=1, no=0).
Statistical analyses
Self-administered questionnaire calibration study. We performed variable selection for the correction SBS model. The prediction model had accelerometer based SBS as the dependent variable (hours/day). We used predictor variables that explained the error of self-reported SBS and its possible relationship with the SBS values obtained with accelerometry. (15,16,25–27) The following explanatory variables were included: self-reported SBS, occupation, caloric intake, sex, moderate/vigorous physical activity, sleep time, back pain, BMI, triglycerides, glucose, chronic diseases, age, and education. We evaluated the natural form of the distribution of each continuous explicative variable, possible mathematical transformations, and interactions that could improve the prediction model.
Estimation model. We graphically identified the best potential predictors of SBS measured with accelerometry and pooled them in a GLM. Each variable was withdrawn, and we tested its
contribution to the model with their beta values and significance (p<0.01) to identify the best predictor combination. We evaluated the Akaike criterion and kept the models with the lowest value.
Cross-validation and agreement. Once we generated the predictive model in the training sample of 71 subjects, we tested it in the second half of 71 subjects, our testing sample. We used the predicted SBS with the equation or corrected and noncorrected vs. accelerometry values for categorical concordance analyses of terciles. Finally, we employed Bland‒Altman analyses to evaluate the agreement between the questionnaire's SBS corrected, without correction, and accelerometry values as hours/day in the training and testing samples. We found a possible group of prediction models and calculated the SBS in the total sample of HWCS participants. We eliminated models with more than 30% implausible SB predicted values. After obtaining the calibration model, we calculated the sedentarism correction as hours per day with the values obtained in the baseline and follow-up assessments.
Association between SBS change and adiposity increment. We used a fixed-effects modeling approach to estimate the relationship between SBS and adiposity change. (28)After reviewing the literature to identify potential confounders, we obtained an acyclic diagram of causality (DAG) with age, physical inactivity, sex, tobacco, and calorie intake to avoid confounding bias from the statistical analysis (Daggity program, see Additional file 1 “DAG figure sedentary behavior”). We performed a linear regression analysis with deltas or the differences between the follow-up and baseline values of body adiposity (kg) and SBS (hrs./d) and included possible confounders. All analyses were performed using Stata version 14 (Stata Corp LLC).