Procedure and participation
Data from the Swedish Longitudinal Occupational Survey of Health (SLOSH) were used. The overall aim of SLOSH is to examine associations between work participation, work environment, social situation and health/wellbeing (Magnusson Hanson et al. 2015; Magnusson Hanson et al. 2008; Statistics Sweden 2014). SLOSH is an ongoing cohort study using biennial questionnaires, which first started in 2006. The participants are previous respondents to the Swedish Work Environment Survey, which aims to comprise a representative sample of the working population in Sweden. The response rate at baseline 2006 was 65.0% and generated an initial sample of 9214 respondents. In the data collection of 2008, 9639 new responders were recruited, representing a response rate of 61.1%. Further details of the data collections and participation in SLOSH have been published elsewhere (Magnusson Hanson et al. 2015; Magnusson Hanson et al. 2008; Statistics Sweden 2014). To be acknowledged, in the present study, only respondents that responded to the questionnaire at three data collections (baseline either 2006 or 2008), 2 years apart, and remained gainfully employed during the whole period, were included in the analytical study sample.
Analytical study sample
In the present study, SLOSH data collected in 2006, 2008, 2010 and 2012 were used. Based on those data collections, two study samples were created, study samples A (baseline at 2006) and B (baseline at 2008) (Fig. 1). To increase the size of the analytical sample in the present study, we merged samples A and B into one. Out of those, we excluded respondents who had not responded to all three measurements, were non-employees at any of the analysed measurements (e.g., retired), reported suboptimal SRH at T1 or had missing data on ICT demands at work at T1 or T2, or missing data on SRH at T1 or T3. This left 4468 gainfully employed persons (1941 [43.4%] men and 2527 [56.6%] women) as the analytical study sample (Fig. 1). The data collection of SLOSH 2014 was not included because the ICT demands at work scale was excluded in SLOSH 2012, and the latter could consequently only be used as a follow-up of SRH in the present study.
ICT demands at work
ICT demands at work were measured by a scale specifically developed for SLOSH (Stadin et al. 2016; Stenfors et al. 2013), based on previous work by Johansson-Hidén et al. (2003). The scale is introduced as follows: ‘New technology and more flexible working conditions have changed working life for many people. Technology can be very helpful but is also conducive to new types of stress. Estimate to what extent you are stressed by…’. Then follow five items in the 2006 data collection; ‘…too many calls and emails’, ‘…claims to be available on work-related issues both during work hours and leisure time’, ‘…claims to give immediate answers to emails and telephone calls that require a lot of work’, ‘…constantly being interrupted by the telephone and email’, and ‘…computers and other equipment that fail to work properly’. In the 2008 data collection, the item ‘…claims to be available on work-related issues both during work hours and leisure time’ was split into two items, separately focusing on either work hours or leisure time. Cronbach’s alpha of the ICT demands at work scale was measured at T1, and the analysis was conducted separately in group A (0.89) and group B (0.87), respectively. The response options were rated on a Likert scale from 1 (I do not have access to this at work) to 5 (very much). ICT demands at work was calculated as the mean score of the ICT demand items. The median score of ICT demands at work (3.00 at both T1 and T2) was used as the cut-off value for high and low ICT demands at work (high ICT demands was defined as strictly above the median) across all measurements. ‘Non-exposure T1, T2’ (low ICT demands at both T1 and T2), was compared with ‘exposure T1, non-exposure T2’ (high ICT demands at T1, and low ICT demands at T2), ‘non-exposure T1, exposure T2’ (low ICT demands at T1, and high ICT demands T2) and ‘exposure T1 and T2’ (high ICT demands at T1 and T2).
Self-rated health refers to the general subjective health status, and was measured by the one-item question: ‘How would you rate your general state of health?’, which originally was introduced in the SF-36 scale (Sullivan et al. 1995). The response options were rated on a Likert scale from 1 (very bad) to 5 (very good). In the analyses, suboptimal SRH (defined as responding ‘very bad’, ‘rather bad’, or ‘neither good nor bad’) was contrasted to good SRH (defined as responding ‘quite good’ or ‘very good’ to the question).
Age, sex, SEP, health behaviours, Body Mass Index (BMI), job strain, and social support were treated as potential confounding factors. These factors have been included based on previous associations with different indicators of work-related stress and SRH (Moor et al. 2016; Rydstedt et al. 2012; Stadin et al. 2016; Stenholm et al. 2017; Toivanen 2011). SEP was calculated in three categories; ‘low SEP’ (unskilled, semiskilled and skilled workers), ‘intermediate SEP’ (assistant and intermediate non-manual workers) and ‘high SEP’ (employed and self-employed professionals, higher civil servants and executives), classified in line with Statistics Sweden’s manual of the socioeconomic classification (Statistics Sweden 1982). Age was calculated in four categories, ‘20–39 years’, ‘40–49 years’, ‘50–59 years’ and ‘60–68 years’. Smoking was calculated in two categories; ‘smoking’ (daily and occasionally) and ‘non-smoking’. Physical activity during leisure time was measured by the question ‘How much do you practise physical exercise?’ The answers were rated on a four-grade ordinal scale from ‘never exercise’ to ‘exercise regularly’, which was dichotomised into the categories ‘low and occasional physical activity’ and ‘regular physical activity’. BMI was calculated by self-reported weight in kilograms/height in squared meters and classified into four categories; ‘underweight’ (< 18.50), ‘normal weight’ (18.50–24.99), ‘overweight’ (25.00–29.99) and ‘obesity’ (≥ 30.00) (World Health Organization 2015).
Job strain was calculated by the demand-control questionnaire (DCQ), which covers the dimensions ‘demands’, based on five items, e.g. ‘Does your work demand too much effort?’, and ‘control’, based on six items, e.g. ‘Do you have a choice in deciding what you do at work?’ (Karasek and Theorell 1992). The population medians of the demands (2.60) and control dimensions (3.17) at T1 were used as cut-off values for high or low scores of the dimensions. ‘Job strain’ was calculated in the categories ‘job strain’ (combination of high demands (strictly above the median), and low control (strictly below the median)) and ‘no strain’ (all other combinations of the demand and control dimensions).
Social support was measured by the social support dimension in the demand-control-support questionnaire (DCSQ), that is based on six items, e.g. ‘I get on well with my colleagues’ (Chungkham et al. 2013). The population median at T1 (1.83) was used as cut-off value for high or low score of social support. Social support was calculated in the categories ‘low social support’ (strictly below the median), and ‘high social support’ (equal to or above the median).
Sex and SEP were also treated as potential effect modifiers. This was due to previous findings of sex differences in diverse work characteristics and health-related outcomes (Swedish Work Environment Authority 2016), and findings indicating that SEP modifies the association between work-related stress and health-related outcomes (Hoven and Siegrist 2013; Toivanen 2011). Information about all potential confounding or modifying factors was included in the SLOSH questionnaire and measured at T1.
Chi-square tests were conducted for bivariate analyses to test potential differences with regard to study sample A or B, sex and SEP in the prevalence of high ICT demands at work, and other characteristics. Logistic regression analyses, calculating odds ratios (OR) with 95% confidence intervals (CI) were used to examine the association between repeated exposure to ICT demands at work and development of suboptimal SRH at follow-up. Repeated exposure to ICT demands at work in relation to suboptimal health was analysed in a crude analysis, and in four different multivariable adjusted regression models: Model 1 was adjusted for age, sex and SEP, Model 2 was adjusted for age, sex, SEP, health behaviours and BMI, Model 3 was adjusted for age, sex, SEP, health behaviours, BMI and job strain, and Model 4 was adjusted for age, sex, SEP, health behaviours, BMI, job strain and social support. All analyses were carried out in the total study sample, and stratified by sex and SEP, separately. Tests for statistical interaction between ICT demands at work and sex, and ICT demands at work and SEP, were also conducted, by including a statistical interaction term between ICT demands and sex or SEP in the respective logistic regression model. As sensitivity analyses, the main results were calculated separately in study samples A and B, and additionally calculated on a modified study sample, that excluded participants stating that they did not have access to ICT at work. The alpha was set to < 0.05. IBM SPSS Statistics 21 was used to calculate the results.