Study design and setting
This study was carried out as a quasi-experimental trial. The outcomes were measured at baseline, three and twelve months. A process evaluation was performed at worker level at three and nine months. The Medical Ethics Committee of the University Medical Center Groningen approved the study design, the protocols and the procedures.
Study population and recruitment
The study was performed at the University Medical Center Groningen (UMCG) and the University of Groningen. The participating UMCG departments were the paediatric and intensive care units (nurses). The University departments which participated were technical services (maintenance and repair), secretarial administrative services (administration), financial economic affairs (policy development, consulting and implementation), human resource advice (policy staff), facility services (cleaning maintenance) and the University Library (management and services). The mean age of the workers was 52.4 (SD 4.9), or 51.7 (SD 4.8) and 52.9 (SD 5.1) from the intervention group and the business as usual group, respectively. Most frequently self-reported chronic health conditions were musculoskeletal diseases (62%), followed by neurological or sensory disease (28%) and mental health conditions (24%).
Participants in this study were selected in a two-step procedure. First, eligible supervisors were invited to participate in the trial. Human Resource Professionals (HRPs) from the two organizations selected the supervisors and thus their workers. The inclusion criteria for potentially participating departments in this study were: a) a higher proportion of workers aged 45 and over than other departments; b) no other intervention studies being performed simultaneously; and c) no planned reorganizations. Supervisors at the departments recommended by the HRPs were invited and informed about the set-up of the intervention by the researcher prior to their decision to participate in this study. Those first to agree to participate in this study were allocated to the intervention group until half the estimated available workers (n = 131) had been allocated. The remaining supervisors and their workers formed the business as usual group. Supervisors willing to participate in the study received information about the process. Recruitment of the supervisors started in June 2009.
In the second step, the eligible workers of supervisors who consented to participate were asked to participate in this study. Workers were informed about this study by their supervisor in November 2009. The workers then received a letter inviting them to participate in the study, describing its aim, content and set-up. Workers on long-term sick leave with no prospect of recovery, or workers who were certain of retiring within a year were excluded from the study. Participation was voluntary and workers were free to leave the study at any time without further consequences. All the workers with the same supervisor followed the same treatment regime, i.e. treatment allocation was at the level of the supervisor. This was done to minimize the probability of contamination [28].
Twenty-eight supervisors, thus 28 departments, were approached to participate in this study. Nine departments declined for reasons such as upcoming reorganization and time pressure. The remaining 19 departments and their supervisors were allocated to either the intervention group (n = 12) or business as usual group (n = 7). Following the allocation of the supervisors, the total of 236 workers were divided between the intervention group (n = 129) or business as usual group (n = 107). An overview of the recruitment flow is presented in Figure 1.
Non-participation and loss to follow-up
Workers who did not meet the inclusion criteria were excluded from this study (n = 16). The most common reasons for exclusion were sick leave for longer than 1 year (n = 10) and upcoming retirement (n = 6). In the intervention group, three workers refused to participate, and seven did not participate for unknown reasons. In the business as usual group, 36 workers refused to participate (n = 5) or did not participate for unknown reasons (n = 31). There was no data on the demographics or work-related outcomes of these workers to compare them to the workers participating in this study.
During the intervention, two supervisors from the intervention group were unable to implement the intervention: because of an unforeseen departmental reorganization for one and long-term sickness of the other. Therefore, we lost 26 workers in the intervention group after the baseline measure. For the remaining workers, data for both follow-up measures was missing for 18 and 5 workers in the intervention and business as usual groups, respectively.
The ‘staying healthy at work’ intervention
The goal of the intervention is to make workers aware of their own responsibility and behaviour in creating a healthy and motivating work environment (e.g. life-long learning) by encouraging them to go through a problem-solving process (performing a problem orientation) to find and implement effective actions and solutions to achieve improvement, life-long learning or to tackle problems which obstruct and deteriorate a sustainable working life (to acquire different problem-solving styles). At the worker level, the intervention comprised three stages: 1) an inventory of work-related problems, needs and the workers’ career and personal development opportunities, including an assessment of the degree to which each issue was amenable to change; 2) a dialogue between worker and supervisor to discuss solutions following a brainstorm format; and 3) making an action plan to plan and implement solutions for a follow-up period next year. The intervention was incorporated into the annual appraisal within the organizations.
The intervention includes a booklet for the workers (stage 1) to prepare for their dialogue (stage 2) and complete their action plan based on solutions chosen and recorded by the workers themselves (stage 3). The supervisors are responsible for recognizing and rewarding excellent performance and providing coaching and for feedback where needed to improve performance shortcomings. Supervisors were also expected to ensure that employees had the tools, resources and training required to carry out their responsibilities successfully. Therefore, they were trained in challenging the workers to reflect on the feasibility of solutions and how to present themselves as a source of support for the worker: not by taking over responsibilities but by strengthening the workers’ autonomy. The first two-hour training course focused on knowledge regarding sustainable employability and on problem-solving techniques. After two weeks the second training event (five hours) was held, which consisted of an active training module in which the problem-solving techniques were practised by role-play with an actor.
Business as usual
Workers in the business as usual group received their regular annual appraisal.
Outcome measures
Primary outcome variables
Work ability was assessed using the Work Ability Index (WAI), a self-administered questionnaire comprising seven items: (i) subjective estimation of current work ability compared with lifetime best (0–100 points); (ii) subjective work ability in relation to both the physical and mental demands of the work (2–10 points); (iii) number of diagnosed diseases (1–7 points); (iv) subjective estimate of work impairment due to diseases (1–6 points); (v) sickness absenteeism during the past year (1–5 points); (vi) own prognosis of work ability after two years (1 or 4 of 7 points); and (vii) psychological resources (enjoyment of daily tasks, activity and personal energy, optimism about the future) (1–4 points) [29]. The WAI is a reliable and valid standardized measure of work ability [30]. Scores range from 7 to 49: higher scores indicate better work ability.
Vitality was measured using the single-item vitality scale of the 12-Item Short Form Health Survey (SF12) and its reliability and validity have been documented [31,32]. The item scores were ‘never’, ‘seldom’, ‘sometimes’, ‘mostly’ and ‘always’.
Productivity was measured using the quantity scale of the Quality and Quantity (QQ) method measuring productivity loss at work [33]. The QQ provides a reliable and valid tool for measuring quantity and quality of work on a daily basis. Workers were asked to indicate how much work they had done during normal working hours on their last regular working day when compared to normal. The quantity of productivity was measured on a scale from 0 (nothing) to 10 (normal amount) [7]. The outcome was dichotomized into productivity loss (scoring 0–9) and no productivity loss (scoring 10). The quality of productivity was not measured because the quality and quantity question are highly correlated [7].
Secondary outcome variables
Secondary outcomes were assessed at baseline and at twelve months follow-up. Fatigue was assessed with the eight-item subscale ‘the subjective feeling of fatigue’ from the Checklist of Individual Strength (CIS) [34]. The items were scored on a seven-point Likert scale ranging from ‘Yes, that is true’ to ‘No, that is not true’. Scores were summed to yield possible scores ranging from between 8 and 56; higher scores indicate higher degrees of fatigue. Psychosocial work characteristics were measured using the Job Content Questionnaire (JCQ) [35]. The JCQ consists of five subscales: job demands (5 items, range 12–48), decision authority (3 items, range 12–48), skill discretion (6 items, range 12–48), social support from supervisors (4 items, range 4–16) and coworker support (4 items, range 4–16). Each item was rated on a four-point scale from ‘strongly disagree’ to ‘strongly agree’.
Perceived work attitude was measured using the Dutch Language version of the Work Involvement Scale (WIS-DLV), which covers six items on a four-point scale (strongly disagree to strongly agree). Higher scores (range 0–100) indicate more positive attitudes towards work [36]. Self-efficacy was measured using the standardized Dutch version of the General Self-Efficacy Scale (ALCOS-16) [37], which assesses the subjects’ expectations of their general capacities [38]. The instrument consists of sixteen items assessing the subscales ‘willingness to exert effort in completing the behaviour’ (range 6–30), ‘persistence in the face of adversity’ (range 6–30) and ‘willingness to initiate behaviour’ (range 4–24). The five response categories range from ‘strongly disagree’ to ‘strongly agree’: a higher score on this questionnaire indicates higher self-efficacy. Work engagement was measured using mean scores from the short Dutch version of the Utrecht Work Engagement Scale (UWES-9) [39]. The nine items reflect three subscales covering the underlying dimensions of engagement. These subscales – vigour, dedication and absorption (each subscale 3 items) – are scored on a seven-point frequency rating scale ranging from ‘never (0)’ to ‘always/ever’ (6). High mean scores on all scales indicative for work engagement are.
Potential confounders
At baseline, data on potential confounders at the worker level was assessed using a questionnaire including age, gender (male/female), education (low = lower vocational education/medium = intermediate secondary or vocational education/high = higher vocational education and university), occupation (executive/secretarial or administrative/policy/management), sector (healthcare/education), shift work (yes/no), duration of current position and years of paid work (0-10/11-20/21-30/31-40/>40 years).
Process evaluation at the worker level
The process evaluation was carried out using a questionnaire at worker level to measure whether the intervention was carried out as intended. The extent to which the intervention was delivered as planned, the exposure and the engagement of the workers with the intervention and the workers’ attitude towards the intervention were included based on the process elements as described by Steckler and Linnan [40].
The extent to which the intervention activities were executed as planned was measured by the intervention dose delivered (representing how much of the intervention was delivered to the worker in question), operationalized by variables such as whether the information about the intervention was sent to the workers, whether the planned dialogues between the worker and supervisor occurred, and whether the workers received the booklet to prepare for this dialogue.
The extent to which the workers were exposed to the intervention was operationalized by measuring reach and dose received. Reach was defined as the attendance rate. Dose received represents the extent to which the workers actively engaged with the intervention components (e.g. preparing for the dialogue using the first part of the booklet or drawing up an action plan).
Workers in the intervention group were asked (yes/no) to evaluate the content and the relevance of the information leaflets they received to prepare for the dialogue, to what extent they experienced support from the booklet and whether the focus on the amenability of situations to change motivated them to plan and implement solutions. The quality of the dialogue (rated on a five-point scale from excellent to very poor), the duration of the dialogue (in minutes) and the support they experienced from the supervisor were assessed.
Statistical analysis
All statistical analyses were carried out at the worker level and according to the intention-to-treat principle. The chi-square test (ordinal and nominal variables) or t-test (mean scores) were used to compare differences on baseline characteristics between the intervention and business as usual group. For the primary outcomes, we performed linear (work ability), ordinal (vitality) and logistic (productivity) multilevel analyses. We had planned to incorporate three levels (supervisor, worker and observation (time)) in all the models. However, random coefficients at the supervisor level did not improve model fit and resulted in the same effect estimates. Therefore, two random effects (worker and observation) were incorporated in the final models. We tested for interactions between the intervention and time to follow-up by incorporating interaction terms in all the multilevel analyses. For the secondary outcomes we performed linear regression analyses. All analyses included adjustment for baseline levels of the outcome, baseline levels of the other primary and secondary outcomes, and for the potential confounders. A detailed description of the sample size analyses is provided elsewhere [41]. The multilevel analyses were performed with MLWin version 2.24. Linear regression analyses were carried out using the statistical package SPSS version 18.0 (SPSS Inc. Released 2009, Chicago: SPSS Inc). A two-tailed p-level of < .05 was considered to indicate statistical significance for all analyses.