Study design and population

The Netherlands Epidemiology of Obesity (NEO) study is a population-based, prospective cohort study designed to investigate pathways that lead to obesity-related diseases. Between 2008 and 2012, 6671 individuals aged 45–65 years were included, with an oversampling of individuals living with overweight or obesity. The study design and population is described in detail elsewhere [27]. Men and women living in the greater area of Leiden (in the west of the Netherlands) were invited to participate if they were aged between 45 and 65 years and had a self-reported BMI of 27 kg/m2 or higher. In addition, all inhabitants aged between 45 and 65 years from one municipality (Leiderdorp) were invited to participate, irrespective of their BMI. The Medical Ethical Committee of the Leiden University Medical Center (LUMC) approved the design of the study. All participants gave their written informed consent.

Participants came to the NEO study center of the LUMC for a study visit after an overnight fast of ≥10 h. Prior to this study visit, participants completed a questionnaire at home to report demographic, lifestyle and clinical information. All participants underwent a physical examination, including anthropometry and blood sampling, and completed a screening for potential contraindications for MRI. Approximately 35% of the participants without MRI contraindications were randomly selected to undergo assessment of liver fat content by proton magnetic resonance spectroscopy (1H-MRS). This study is a cross-sectional analysis of baseline measurements of the random subset of participants who carried an accelerometer.

Data collection Objective assessment of physical activity and sedentary time and breaks

Daily levels of activity were objectively assessed in a random subsample of NEO study participants (n=955) with a combined uniaxial acceleration and heart rate monitor (Actiheart; CamNtech, UK). Participants were instructed to wear the monitor continuously for four consecutive days and nights, and to carry on with all normal activities during this time.

Details of assessment have been described previously [28]. Using a branched equation algorithm, the acceleration and heart rate information (recorded in 15 s epochs) was summarised into estimates of physical activity energy expenditure (PAEE, in kJ kg−1 day−1) and time spent at different activity intensities was expressed as metabolic equivalents of task (MET) [29, 30].

Sedentary time was defined as time spent in activities with an intensity ≤1.5 MET, excluding sleep time, which was assumed as time between 23:00 hours and 07:30 hours on weekdays and between 23:30 hours and 08:30 hours on weekend days. A break in sedentary time was defined as a period of activity with an acceleration >0.75 m/s2 following a sedentary bout (uninterrupted period). Light physical activity (LPA) was defined as activity with an intensity >1.5 MET and ≤3 MET, and MVPA was defined as activity >3 MET. We differentiated the total amount of MVPA from MVPA accumulated in bouts ≥5 min.

Further, we divided the day into three 6 h blocks: 06:00–12:00 hours (morning); 12:00–18:00 hours (afternoon); and 18:00–24:00 hours (evening). For each time block we calculated the proportion of total daily MVPA that was spent in each 6 h time block. Based on these proportions, we categorised participants as most active in the morning, afternoon or evening, or as having an even distribution of MVPA throughout the day. For this, we used a minimum difference of 5% between time blocks. For example, an individual with 40%, 30% and 30% of total MVPA in the morning, afternoon and evening, respectively, was classified as most active in the morning, whereas a person with respective values of 35%, 33% and 32% was classified as having an even distribution of MVPA throughout the day. Participants were excluded from analyses if valid total wear time <24 h or if wear time within any single hour of the day was <30 min.

Blood sampling

Fasting blood samples were drawn from the antecubital vein after the participant had rested for 5 min in a seated position. Within 5 min after the first blood sample, participants drank a liquid mixed meal (total 400 ml, containing 2510 kJ (600 kcal)), with 16% of energy derived from protein, 50% from carbohydrates and 34% from fat. Two postprandial blood samples were drawn at 30 and 150 min after the mixed meal. Fasting and postprandial plasma glucose and serum insulin were determined as well as fasting HbA1c concentrations [27]. We calculated the updated HOMA-IR from fasting glucose and insulin using the Oxford University online calculator (https://www.dtu.ox.ac.uk/homacalculator) and the Matsuda Insulin Sensitivity Index (Matsuda ISI).

Liver fat content

In a random subgroup of participants without MRI contraindications, liver fat content was quantified by 1H-MRS [31] on a 1.5 Tesla MR system (Philips Medical Systems, Best, the Netherlands). An 8 ml voxel was positioned in the right lobe of the liver, avoiding gross vascular structures and adipose tissue depots. Sixty-four averages were collected with water suppression. Spectra were obtained with an echo time of 26 ms and a repetition time of 3000 ms. Data points (1024) were collected using a 1000 Hz spectral line. Without changing any parameters, spectra without water suppression, with a repetition time of 10 s and with four averages, were obtained as an internal reference. 1H-MRS data were fitted using Java-based magnetic resonance user interface software (jMRUI version 2.2, Leuven, Belgium) [32]. Hepatic triacylglycerol content relative to water was calculated as the sum of signal amplitudes of methyl and methylene divided by the signal amplitude of water and then multiplied by 100. Non-alcoholic fatty liver disease (NAFLD) was defined as liver fat content ≥5.56% [33].

Covariates

On a questionnaire, participants reported ethnicity, which we grouped into White (reference category) and Other. Highest level of education was reported in ten categories according to the Dutch education system and grouped into high (including higher vocational school, and university) vs low education (reference). Tobacco smoking was reported in three categories: current; former; and never smoking (reference). Body weight and per cent body fat were estimated by the Tanita bio impedance balance (TBF-310; Tanita International Division, UK) with the participant not wearing shoes; 1 kg was subtracted from the body weight to account for clothing. BMI was calculated by dividing the body weight (kg) by the height squared (m2). Participants reported their habitual dietary intake (including alcohol intake) using a semi-quantitative self-administered 125-item food-frequency questionnaire [34, 35]. Dietary intake of nutrients and total energy was estimated using the Dutch Food Composition Table (NEVO-2011). We calculated the Dutch healthy diet index that indicated adherence to the Dutch Guidelines for a Healthy Diet of 2015 [36]. The context of physical activity (leisure or occupational) was derived from the Short Questionnaire to Assess Health-enhancing physical activity (SQUASH) [37].

Statistical analyses

In the NEO study, individuals with a BMI of 27 kg/m2 or higher were oversampled. To correctly represent baseline associations in the general population, adjustments were made for this oversampling [38]. This was done by weighting all participants towards the BMI distribution of participants from the Leiderdorp municipality, the BMI distribution of whom was similar to the BMI distribution in the general Dutch population [39, 40]. Consequently, all results apply to a population-based study without oversampling of individuals with a high BMI.

Population characteristics were summarised as mean (SD), median (25th, 75th percentile) or proportion, stratified by timing of physical activity. As a consequence of the weighted analyses, no absolute numbers could be given, only proportions.

We examined associations between daily total sedentary time, the number of breaks in sedentary time and different intensities of physical activity (i.e. total PAEE, LPA, MVPA and MVPA in 5 min bouts) with liver fat content and insulin resistance using linear regression analyses.

For this, we constructed three models with potential confounding factors: model 1 was adjusted for age, sex, ethnicity and level of educational background; model 2 was additionally adjusted for lifestyle variables (alcohol consumption, smoking and Dutch healthy diet index); and model 3 was additionally adjusted for total body fat. As body fat may both confound and mediate the relationship between physical activity and liver fat content and insulin resistance, this model allows one to observe whether physical activity is associated with liver fat content and insulin resistance beyond effects via total body fat. In addition, associations with breaks in sedentary time were adjusted for total sedentary time and total volume of MVPA. Since liver fat content and insulin resistance were non-normally distributed, these were transformed using the natural logarithm. For interpretation of the results, the linear regression coefficients were back-transformed into a relative change (with 95% CIs). As an example, a relative change of 0.8 can be interpreted as 0.8-fold decrease in liver fat content for each hour of MVPA per day, reflecting a difference in liver fat content from, for example, 5% to 4%. Additionally, these relative changes were expressed as percentage change in the text: [exp(β)−1]×100 if β>0 and: −[1/exp(−β)−1]×100 if β<0, with 95% CI.

Next, using linear regression analyses, we examined liver fat content and insulin resistance in participants who were most active in morning, afternoon or evening compared with those with an even MVPA distribution. From the regression coefficients we calculated relative changes with 95% CIs. These analyses were adjusted for the same covariates as mentioned above as well as for total MVPA. To investigate whether associations were different between men and women, we tested for interaction by including product terms of sex and all physical activity variables in the adjusted models.

We repeated all analyses with timing of LPA and total PAEE. Further, we repeated all analyses for the following additional outcomes: insulin sensitivity; and fasting glucose, insulin and HbA1c. In addition, we added the amount of occupational activity and activity during leisure time as covariates to the models with timing of MVPA. Further, we restricted the analyses of breaks in sedentary behaviour to those with a similar amount of sedentary time (mean±1 SD).

Data was analysed using STATA 16.1 (StataCorp, College Station, TX, USA).