Study population and data sources

This prospective cohort study includes web-based survey data from ministerial employees in 14 of 17 ministries, after a car bomb attack at the Norwegian government offices in Oslo on 22 July 2011. The terror bombing caused substantial damage on buildings and infrastructure, killing eight and injuring 209 people. Previous studies of governmental employees report elevated risk of psychological and somatic health adversities.13 14 These negative health reactions were found among people who were directly and indirectly exposed.13 15 The estimated overall prevalence of post-traumatic stress disorder (PTSD) was 24% among directly exposed governmental employees and 4% among those indirectly exposed, signifying that being present at work during the bomb attack was not the only precondition for reporting PTSD symptoms. Common for the employees developing PTSD symptoms are that they all were the explicit target of the terror attack.16 It is reasonable to expect that even those who were indirectly exposed and evaluated their life as being in serious danger may develop PTSD.4 17 18 The present study therefore includes all employees with symptom-defined PTSD, whether indirectly or directly exposed.

The Norwegian Centre for Violence and Traumatic Stress Studies conducted the survey in collaboration with the National Institute of Occupational Health in Norway 10 months after the terrorist attack (T1). Eligible participants were informed about the study through their ministries and received an invitation letter containing a unique log-in code to access the web-based survey, including information on withdrawal procedures. All employees provided written consent, and strict procedures were followed to ensure confidentiality. Fifty-six per cent of the 3520 invited employees consented to participate, and 34 were excluded from the data file due to missing data, leaving us with 1974 employees. In addition, 84% consented to data linkage on sick leave, and our study population totaled 1649 participants. Data on doctor-certified sick leave was obtained from Statistics Norway and the Norwegian Labour and Welfare Administration.

For the purpose of this study, we used survey data on employees’ health and PTSD symptoms, background variables and information on psychosocial work environment from 10 months after the attack, linked with registry data on doctor-certified sick leave for the period of 12–24 months after the attack. Since the first years after the attack were most prominent for sick leave risk and PTSD, we focused on this especially sensitive period.7 In this study population (n=1649), we identified 94 participants eligible for this study, all fulfilling the PTSD symptom criteria and with relevant data on doctor-certified sick leave. Based on the personal identification number from the Norwegian Population register, Statistics Norway performed data linkage and deidentification. The research team did not have access to the key to match data nor the identity behind survey responses. Further details on design and participants have previously been described in a recent article from our research group.7 See figure 1 for more information.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Flowchart displaying participant disposition 10 months after the attack (T1). PTSD, post-traumatic stress disorder.

Patient and public involvement

There is no direct patient involvement in this study. Neither patients nor public were directly involved in the development, design or recruitment of the study.

Classification of PTSD

The Norwegian version of the Post-traumatic Stress Disorder Checklist-Specific (PCL-S) was used to assess symptoms of PTSD.19 In this version, the questions were specifically linked to the bomb attack in Oslo on 22 July 2011. All participants were asked to rate the extent to which they had experienced PTSD symptoms during the last 4 weeks and were instructed to answer with reference to the terrorist attack. The PCL-S consists of 17-items measuring the full domain of PTSD symptom severity from the D iagnostic and S tatistical M anual of Mental Disorders, Fourth Edition (DSM-IV). The validated Norwegian version has demonstrated good psychometric and diagnostic properties.20 The PTSD diagnosis criteria followed the procedure of positive items required in the three clusters in the DSM-IV system: one positive score in symptom cluster B (re-experiencing), three positive scores in cluster C (avoidance) and two positive scores in cluster D (hyperarousal).21 The A criterion (stressor exposure) was not included in the assessment. All items were scored on an ordinal 5-point scale from ‘1 = not at all’ to ‘5 = extremely’. As recommended by the National Center for Post-Traumatic Stress Disorder, an item score of 3 or above was defined as positive for PTSD.19 Chronbach’s alpha for PCL-S was 0.94.

Sick leave

Sick leave was measured in the 1-year period after the participants completed the survey 10 months after the attack. This 1-year period consisted of the third and fourth quarter of 2012, and the first and second quarter of 2013. The total number of sick leave days for each participant was weighted by the percentage of full-time employment and the percentage of full-time sick leave. The number of workdays was weighted by percentage of full-time employment, weekends, public holidays and vacations.

Psychosocial work exposures

Job demands, control at work, predictability, social interactions and organisational culture were measured by the General Nordic Questionnaire for Psychological and Social Factors at Work.22 Reliability tests of these scales have demonstrated acceptable internal consistency, with Chronbach’s alpha ranging from 0.68 to 0.87 and test–retest reliability from 0.66 to 0.83 with more than 5 weeks’ interval.23 Job demands were measured by three scales: quantitative demands comprising four items, decision demands with three items and learning demands with three items. A typical item for job demand was ‘Do you have too much to do?’ Three scales measured control at work: positive challenge comprising three items, control of decisions with five items and control of work pacing with four items. A typical item for control at work was ‘Can you set your own work pace?’ Predictability at work was measured by the scale long-term predictability (2 years) comprising three items. A typical item was ‘Do you know what is required in order for you to get a job that you consider attractive in 2 years?’ Social interactions were measured by two scales: support from superior comprising three items and support from coworkers with two items. A typical item for support was ‘If needed, can you get support and help with your work from your immediate superior?’ Finally, organisational culture was measured by three scales: social climate, innovative climate and human resource primacy, all comprising three items. A typical item was ‘Are workers well taken care of in your organisation?’ All responses were scored on an ordinal 5-point scale ranging from ‘1=very seldom or never’ to ‘5=very often or always’, and missing response to one of the items comprising each scale was allowed.

Background variables

The background variables (table 1), including sex, age, educational level, relationship status, directly and indirectly exposed, psychological distress, event centrality, life satisfaction and social support, were retrieved from the questionnaire survey completed 10 months after the terror attack. Education was measured by three levels of education: ‘less than 13 years’, ‘13–16 years’ and ‘more than 16 years’. Relationship status was dichotomised as ‘married or cohabiting’ and ‘single or widow’. Proximity to the explosion was measured by asking participants about their whereabouts at the time of the bomb attack, and responses were dichotomised as ‘being present in the government district’ and ‘not in the government district’.

Psychological distress was measured by the Hopkins Symptom Checklist-10 (SCL-10), with four response categories from ‘1=not bothered’ to ‘4=very bothered’.24 A typical item was ‘worrying too much about things’. The index was scored as the mean of the item score, where an average score above 1.85 is regarded as a valid predictor of depression. Chronbach’s alpha for SCL-10 was 0.92. The Centrality of Event Scale (CES) measures to what degree an event was integrated into an individual’s life story and identity.25 Responses were categorised from ‘1=strongly disagree’ to ‘5=strongly agree’ and were calculated as the mean of all items. A typical item was ‘I feel that the event has become part of my identity’. Chronbach’s alpha for CES was 0.92. Life satisfaction was measured with the one-item Cantril’s Ladder of Life Scale, where overall life satisfaction was rated on a 10-point scale representing steps in the ladder.26 A score of 10 represents the best possible life you can imagine, and a score of 0 is the bottom step, representing the worst possible life. Finally, social support was measured with the four item Crisis Support Scale (CSS), with response categories ranging from ‘1=never’ to ‘7=always’. A typical item was ‘How often is someone willing to listen? Chronbach’s alpha for CSS was 0.89.27

Covariates

Covariates that could potentially influence the association between the psychosocial work environment and sick leave risk were considered as potential confounders. The following covariates were included in the main analysis: sex and symptoms of PTSD. Due to a small sample size, we chose a parsimonious structure, adjusting for the most important covariates in the analysis to retain an acceptable number of cases and to be able to detect significant differences between our two groups. We did a preliminary analysis adjusting for education, but that did not change our risk estimates. Therefore, the two most significant variables shown to differ between the groups were entered as covariates in the analysis. See table 1 for information on background variables.

Statistical analysis

The outcome variable in the present study, sick leave days, is a count variable with overdispersion and an excess of zeros. We therefore used two-part hurdle models to explore sick leave according to various psychosocial work environment factors.28 In the first part, we collapsed the outcome variable into two categories: no sick leave (zero count) and sick leave (above zero count). We then estimated the odds of sick leave versus no sick leave for various predictor values and summarised this as ORs (95% CIs) per one unit increase in a given predictor using the lowest value as the reference. In the second part, we used zero-truncated negative binominal regression for the positive counts (ie, those with sick leave) to estimate the length of sick leave for various predictor values.29 We summarised this as a count or rate ratio (RR, 95% CIs) using the lowest predictor value as the reference. To account for varying workdays, as not all employees worked full time during the observation period, the hurdle models were offset by registered workdays in the relevant period.

The binomial regression analyses and the zero-truncated negative binominal analysis were computed in R (The R Foundation for Statistical Computing, Vienna Austria, 2018), with the R packages pscl for hurdle regression (R Development Core Team, 2017) and in STATA V.15.

Sensitivity analysis

In this study, sick leave prior to the attack is a covariate that can act as a confounder. If sick leave before the attack increases the risk of sick leave after the attack, then our estimated relative risks measuring the association between work factors and sick leave would most likely be attenuated. To capture uncontrolled confounding or systematic change of behaviour due to previous sick leave, we conducted sensitivity analysis by adjusting for previous sick leave prior to the attack, defined as sickness absence days from the first quarter of 2008 up until and including the second quarter of 2011 divided by the number of expected workdays for this period.