The study was a 6-month cluster RCT with 3-arms of which one was a wait-list control group. The objective of the current paper was to assess whether the multi-component interventions, incorporating individual, environmental, and organisational changes, increased physical activity or reduced sedentary behaviour after the 6-month interventions. Effects of the two interventions, either targeting physical activity (iPA) or sedentary behaviour (iSED), were compared to a wait-list control group and to each other. We hypothesised that participating in the iPA intervention would increase time spent in moderate-to-vigorous intensity physical activity (MVPA), and participating in the iSED intervention would decrease time spent sedentary.
Ethical approval was granted by The Stockholm regional ethical review board (2017/2409–31/1). All participants provided written informed consent before the first data collection. The study was conducted in accordance with the CONSORT guidelines for cluster RCTs. The trial was prospectively registered as ISRCTN92968402 on 27/02/2018, recruitment started 15/03/2018 (https://doi.org/10.1186/ISRCTN92968402). A study protocol , including a detailed description and rationale of the trial, has been published earlier, the most important details are provided below.
From two Swedish product and service producing companies (with 3 different office locations), 2033 office workers were invited to participate in the study. The target was to include 330 persons, based on sample size calculations . Inclusion criteria were ages 18–70 years and able to stand and exercise. Persons were excluded if they were very physically active, according to device-measured MVPA of more than 30 min/day in prolonged bouts (≥10 min) every day. Sedentary behaviour was not used as an exclusion criterion, as a previous cross-sectional study in a comparable sample showed that nearly all office workers reported high levels of sedentary behaviour .
Randomisation and masking
Clusters consisted of teams of office workers grouped by Human Resources (HR) personnel at each company. Minimum amount of persons per cluster was five and other considerations taken into account were: 1) having a team or line manager, 2) having regular group meetings, and 3) having limited regular meetings with other teams. Randomisation into groups was performed after finishing baseline data collection and was done on a cluster level . Groups were randomly allocated (1:1) with stratification for company and cluster size (large vs. small). Matched randomisation was applied to realise logistical capacity. The random allocation sequence was setup using a computer-generated random number list, and randomisation was performed by an independent researcher not involved in data collection. All participants were informed on allocation after baseline data collection via e-mail sent from another independent researcher not involved in data collection. Personnel involved in data collection and processing were blinded for group allocation.
Two multi-component interventions aiming to either increase physical activity or reduce sedentary behaviour were designed based on the ecological framework addressing multiple levels including individual, environmental, and organisational [18, 19]. Both interventions lasted 6 months and focused on work time as well as leisure time, to decrease the risk that participants compensate by e.g. decreasing sedentary behaviour at work but at the same time increasing sedentary behaviour during leisure time .
The individual support included motivational counselling that had a comparable design but a different focus in the two interventions. This counselling was provided by professional health coaches from a health promotion company who received additional training on CBT and on physical activity and sedentary behaviour. There were five sessions; three individual (45–60 min) and two group sessions (90 min) spread out during the 6-months intervention period. The content of the counselling sessions was standardised using manuals with checklists.
A team leader was appointed to each cluster to deliver the environmental and organisational components of the interventions, and to encourage all participants to remain in the study. HR personnel of each company asked one participant in each cluster to be the team leader, for which no formal requirements were set, but judgement of HR personnel on a suitable individual was adopted. At the start of the intervention, one of the researchers contacted all team leaders by phone to coordinate the logistics of implementing the different components of the assigned intervention.
Intervention to promote physical activity level (iPA)
The intervention aiming to promote MVPA included:
Individual: motivational counselling using CBT towards increasing time spent in MVPA, including individual feedback on physical activity (MVPA and steps/day).
Environmental: access to a commercial gym (6 months), exercise sessions and lunch walks organised by team leaders, and the provision of company bikes.
Organisational: team leaders encouraged employees to be physically active inside and outside of working hours, including commuting to work.
Intervention to reduce sedentary behaviour (iSED)
The intervention aiming to reduce sedentary behaviour, including breaking up prolonged sitting, included:
Individual: motivational counselling using CBT towards reducing time in sedentary behaviour and breaking up prolonged sitting, including individual feedback on sedentary behaviour patterns (sedentary behaviour and steps/day).
Environmental: team leaders were instructed to organise standing and walking meetings. Note that companies already provided their employees with sit-stand desks .
Organisational: team leaders encouraged employees to reduce sedentary behaviour during work, in meetings, and while sitting behind their desks as well as outside of working hours including lunch and commuting.
Control group (C)
A passive wait-list group, which received one of the described interventions after the follow-up measurement (at 6 months).
Inclusion of participants started March 2018 and ended November 2018, after all employees at the companies had been informed about the study. Data collection of the RCT finished in May 2019. At both baseline and the 6-month follow-up, all participants filled out online surveys and participated in device-measured physical activity and sedentary behaviour measurements.
Characteristics of participants were collected from the online survey, and included: company of employment, age, gender, years of education, family status (married/living together: yes vs. no), working fulltime (yes vs. no), and smoking (yes vs. no). Weight and height measured at baseline were used to calculate body mass index (BMI; kg/m2).
Device-measured physical activity and sedentary behaviour
Participants were instructed to wear an Actigraph GT3X accelerometer (Actigraph GT3X, Fort Walton Beach, Florida, USA) on the hip during wake time and on the wrist during in-bed time, along with an ActivPal inclinometer on the frontal aspect of the mid-thigh (activPAL™ 3 activity monitors, PAL technologies limited, Glasgow, UK) during 7 days. During the measurement period, participants noted in-bed times, working hours, sleep quality, and daytime sleepiness in a diary.
Accelerations from 3 axes (vector magnitude, VM) were sampled at a frequency of 30 Hz. Minimum requirement for data inclusion was 600 min of valid wear time during waking hours, based on diary information, on at least 4 days. Non-wear time was defined as at least 60 consecutive minutes with no movement (VM = 0 counts per minute, cpm), with an allowance of maximum 2 min of activity. The primary outcome was time spent in MVPA (> 2690 cpm), expressed as a percentage of wear time based on an average of all available days. Secondary outcome measures (also expressed as percentage of wear time) were light (200–2689 cpm), moderate- (2690–6166 cpm), and vigorous-intensity physical activity (> 6167 cpm) as averages over available days, as well as weekday only averages .
Inclination of the thigh was obtained to quantify time spent sedentary, standing, and walking. Sedentary was defined as the thigh in a horizontal position, i.e. sitting or lying. Recorded time was coded as wear time, non-wear time, or working time, based on diary recordings. Sleep and non-wear time were excluded. For a day to be considered valid the following rules were applied: ≥10 h of worn waking hours, < 95% of time spent in any one behaviour (sedentary, standing, walking), and ≥ 500 steps . Work time was considered valid when the device was worn for ≥80% of the time at work and ≥ 5 h of worn working hours. Data from at least 4 days were required, with at least 2 working days and 2 non-working days. The primary outcome extracted from this device was total sedentary time (% of wear time), averaged over all days. Secondary outcome measures (% of wear time) were: standing and walking, averaged over all days; and sedentary, standing, and walking for work time only.
Self-reported physical activity and sedentary behaviour (secondary outcomes)
Two validated questions for physical activity were used, one regarding exercise and one regarding daily activities (see supplementary file 1) . Self-reported physical activity was based on an index of these two questions as described previously . The index was then dichotomised into more favourable vs. more unfavourable physical activity levels using a cut-off coinciding with approximately 150 min/week of accelerometer assessed MVPA .
For sedentary behaviour, a validated question accounting for the amount of sitting time, excluding sleep, was used (see supplementary file 1) . There were 7 possible response options. A cut-off of less than 7–9 h was used to classify participants as more favourable or more unfavourable sedentary behaviour. This cut-off has shown correspondence to approximately 10 h of accelerometer assessed sedentary time .
The power calculation are presented in detail in the published protocol .
We planned to perform multilevel regression models (frequentist analyses) to evaluate differences in effects between the iPA, iSED, and control group (C) . However, when running these models issues arose regarding the singular fit. A singularity error appeared in about 25% of the models, but with no clear pattern of occurrence. We chose to perform multilevel models using Bayesian statistics for all outcomes instead, because it is a preferred method that can handle issues of singularity . The frequentist multilevel results (β, and 95% confidence interval (CI) were presented for two main outcome variables (%MVPA and %Sedentary) in the complete cases dataset, to confirm the Bayesian statistics. Note that these two models did not have a singularity error.
Bayesian regression estimates and 95% credible intervals were estimated using the packages with R statistical program language in Rstudio (R version 3.6.1, packages: brms, tidyverse and tidybayes). Bayesian analysis has a number of advantages, e.g. it can be applied to a large range of models including hierarchical multilevel models, the calculation of exact estimates without reliance on large sample sizes, and provides comprehensible answers with credible intervals that can be interpreted as probabilities .
In all models, clustering of data was taken into account using a two-level structure with participants as the second level and clusters as the first level. All groups were simultaneously regressed on the post-test value of the outcome, adjusted for the baseline value of the outcome with age, gender, education, and company as covariates. We used an uninformative prior (student t) for the coefficients. For all device-measured physical activity and sedentary behaviour, which were continuous outcomes, the Gaussian function was used (4 chains, 4 cores and 3000 iterations, 1000 warm-up). Parameter estimates and 95% credible intervals were presented. For the analyses on the dichotomised self-reported physical activity and sedentary behaviour outcome variables, the bernoulli function was used (4 chains, 4 cores and 3000 iterations, 1000 warm-up). For these outcomes, parameter estimates and credible intervals were exponentiated to provide posterior odds ratios. The achieved level of convergence was Rhat values of 1 for all models.
Furthermore, we determined posterior probabilities, which is the probability that the coefficient for a group comparison is above 0. Being above (or below) 0 with a high probability indicates that there was a difference between groups in the outcome variable. Although Bayesian models should not be interpreted like frequentist models regarding significance, we used a cut-off for indicating significant effects of these posterior probabilities as below 0.025 or above 0.975, which is comparable to the traditional p-value significance level of 0.05.
For each of the different defined outcome variables, separate models were run and repeated for each of the following four datasets:
Complete cases. The analyses performed with this dataset were considered as the primary results. Education was imputed with median values for 7 participants who had missings or impossible values but no further imputations were performed. There were no missing values for gender or age.
Intention To Treat with last observation carried forward (ITT-LOCF). The assumption in this model was that individuals with missing values at follow-up have no change from baseline. The imputation was performed in 3 steps:
Missing values for education were imputed using k Nearest Neighbour imputation using baseline variables.
Missing values for the outcomes at baseline were imputed with follow-up values if available, otherwise with k Nearest Neighbour imputation.
Missing values at follow-up were imputed with Last Observation Carried forward, which in this case is the baseline observation carried forward.
Intention To Treat with k Nearest Neighbour (ITT-kNN). All missing values were imputed with k Nearest Neighbour imputation.
Per Protocol. The inclusion criteria for this dataset were:
Individuals in iPA or iSED groups who had attended 3 to 5 counselling sessions , or individuals in C group. For those individuals that did not have this information available, it was assumed that they did not participate in at least 3 sessions.
Individuals with at least one non-missing value at follow-up, meaning that they should have at least results of either one of the device-measured or self-reported physical activity or sedentary behaviour outcomes at follow-up.
Education was imputed with median values for those who had missings or impossible values but no further imputations were performed.
Drop-outs were defined as having neither Actigraph nor ActivPAL data at follow-up. To determine whether missing data were not at random, we performed binary logistic regression analyses using generalised linear models. Differences between those who dropped-out and those who did not drop out were assessed for: group allocation, age, gender, education, as well as baseline physical activity (%MVPA, Actigraph) and baseline sedentary behaviour (%Sedentary, ActivPAL). These analyses were performed using SPSS (Statistical Package of Social Sciences, version 25).