Study design and cohort
Data for this study was obtained from a prospective population study of young adults in Sweden, aged 20–24 years, with a 1-year follow-up. The baseline cohort of 7,125 individuals, which overrepresented women and those of Swedish birth, was the basis of a project called Work Ability of Young Adults (WAYA). This cohort was derived from a questionnaire sent to an equal number of women and men in a randomly selected sample of 20,000 young adults. The aim of this project was to follow young adults over time with questionnaires focusing on contemporary exposures to work factors such as information and communication technology (ICT), environmental factors related to lifestyle, exposures at work or during studies, health, productivity, and work ability [17]. The 1-year follow-up cohort consisted of 4,163 individuals.
Study sample
The study sample consisted of 1,311 young adults, aged 21–25 years, here defined as young workers, with adult workers defined as those > 25 years. The inclusion criteria for this group from the 1-year follow-up cohort, were i) having answered the work ability score (the first dimension of the WAI) at both baseline and follow-up, and ii) having salaried work at both baseline and the 1-year follow-up. This resulted in the exclusion of 1,745 students, (Figure 1).
Drop-outs
The drop-out group of workers (not including students) consisted of 1,500 individuals (not shown in the figure). Of these, 1,490 workers did not answer the follow-up questionnaire. Another 10 individuals answered the follow-up questionnaire, but did not answer the work ability score.
A drop-out analysis showed that the lost group had similar scores to the study sample for several work factors at baseline, such as job control, social support at work, reward relative to effort, and negative influence of job demands on private life. However, the drop-out group consisted of significantly more men, (a 5% difference) than the study sample. Also, the workers in the drop-out group had a significantly lower daily use of the computer in general (a 7% difference) than workers in the study sample. For work ability, a statistically significant small difference (0.1 score levels) could be seen between the groups, but this is most likely not of clinical interest.
Data selection
Data was collected in 2007 and 2008 through two self-administrated questionnaires consisting of 78 items. The first questionnaire was sent by post and the second by web, after a one-year interval. As compensation, invited participants received a lottery ticket valued at 1 Euro with each questionnaire. The posting of the first questionnaire was followed by two reminders, and the second by three, the last including two cinema tickets. The response at baseline was 36%, and at the 1-year follow-up, 73%. This procedure has been described in detail [18].
Descriptive data were collected from the study sample at baseline, and work factors and self-reported level of work ability were collected through the questionnaires at both baseline and the 1-year follow-up needed for the analyses.
Individual characteristics
Questions about descriptive data were partly selected from previously shown associations and relationships between work ability and individual factors for both adults [9] and young adults [10]. Consequently, individual factors such as sex, civil status, educational level, main occupation, living area, country of birth, and health-related questions about smoking, body mass index (BMI), physical activity, chronic pain, symptoms of depression, and experienced health, were used.
The outcome
Work ability
The WAI is a self-report instrument consisting of seven dimensions derived from ten items, on which individuals estimate the dimensions of their own work ability [2]. This instrument has previously been used for workers as young as 16 years of age [7, 14]. The WAI has been shown to be a useful tool when investigating an entire working population, although further evaluation of the instrument is needed for workers of different ages [19].
The change in self-reported work ability
Work ability in this study was measured by the work ability score, an “age-free” item according to Ilmarinen [1]. This one item of work ability measures “current work ability compared with the lifetime best” and consists of a scale from 0 representing “cannot work at all right now” to 10 representing “my work ability is at its best right now”. A change in the work ability score has been validated to show a change in the entire WAI for women on long-term sick-leave [20].
We defined a real change in the work ability score as a decrease or increase of 2 score levels or more, based on prior analyses in a test/retest study of 29 young adults [17], in which the smallest detectable change was calculated as 1.9. Self-reported changes in work ability, in any direction, were dichotomized to 1 for changes of 2 score levels or more, and to 0 for changes of 1 score level or none.
The explanatory variable
Physical factors at work
Because of a lack of knowledge about how changes in work factors influence work ability in young workers, the selection of physical work factors was based on work factors previously found to be associated with changes in work ability in a mainly adult working population [11–13].
The time frame for all questions, except on vibration exposure during the last year, was the last 30 days.
Two questions addressed computer use. The first question asked about total daily time spent at a computer, for both work and leisure. Possible response alternatives were < 2 h/day, 2–4 h/day, and > 4 h/day. The second question concerned computer use of more than 2 hours with no breaks longer than 10 minutes, with possible answers as never, once in a while, a couple of times per month, a couple of times per week, and most days. The cut-off points for these questions were obtained from a cross-sectional study [21].
Questions about work postures had different response alternatives. “How long daily do you work with your hands above shoulder level,” could be answered by never, < 1 hour, 1–2 hours, and > 2 hours [22]. Similar responses to the question, “How long each day do you work with a flexed or extended neck” could be never, < 3 hours, 3–5 hours, and > 5 hours [23]. “How long daily do you work with a flexed back” could be answered by never, < 0.5 hour, 0.5–1 hour, and > 1 hour [24].
Questions about lifting had the answer alternatives for intensity of 5–10 kg, 11–15 kg, 16–25 kg, and > 25 kg and for frequency of 0–4, 5–15, 16–30, and > 30 times/day [25]. A question about the frequency with which they handled tools or equipment demanding a forceful grip to the equivalent of lifting 1 kg or more had the alternatives of seldom or never, several times per day, several times per hour, and several times per minute.
For the question concerning regular use of vibrating hand-held machines at work, the alternatives were yes or no.
Psychosocial factors at work
Like the physical work factors, psychosocial work factors were selected from known relationships between changes in psychosocial work factors and work ability reported mainly in adult workers [11–13].
Questions related to job demands, job control, social support at work, and reward relative to effort had the same response alternatives: corresponds very poorly, corresponds somewhat poorly, corresponds fairly well, and corresponds very well.
Job demands were defined as exposure to high demands and expectations at work, and job control as having control over and the ability to deal with situations at work. Questions about social support concerned access to support and help at the workplace in the event of problems. These one-item questions were developed from the demand–control model [26], which also included social support from co-workers in a later version [27].
From the effort–reward model [28], one-question, concerning the reward deserved in relation to the effort extended and the actual production at work was modified. There was also a question about whether demands at work negatively influenced private life (leisure, home, and family life), with the possible response alternatives very seldom, fairly seldom, sometimes, fairly often, and very often. This question, derived from the model of work–home interference [29], has been validated [30].
Two new questions, asking about the previous month, were constructed to address flexibility in work [31], with the same response categories: never, once in a while, a few times per month, a few times per week, or more or less daily. The first question asked whether work was performed outside the workplace, for example at home, and the second, whether respondents had ever to be available by mobile phone after working hours. One further question asked how often they had been working more than 12 hours in a day within the last 30 days. Possible responses were 0, 1–2, 3–8, 9–15, and > 15 times in the last month.
A question about whether they experienced noise annoyance at the workplace [32] could be answered by never, once in a while, a few times per month, a few times per week, or more or less every day.
Changes in self-reported work factors
A changed answer for any of the work factors by one step or more among the three to five response alternatives between baseline and the 1-year follow-up was defined as a change. This simplified measure of change was chosen despite diverse methods of interpreting scales [33], based partly on the test/rest study [17] in which the numbers of response alternatives were changed for some questions to increase reliability.
Statistical analyses
For all analyses in this study, SAS version 9.1 (SAS Institute, Cary, NC) was used.
Descriptive data of the sample and subgroups at baseline were first derived through frequency analyses.
Prospective analyses were performed next to assess associations between changes in work factors and changes in work ability. The Cox proportional hazard regression model was used to estimate prevalence ratios (PR) in both univariate and multivariate analyses, with time set to 1 [34]. These analyses were carried out for the sample as a whole [n = 1,311] adjusted for sex, and also stratified by gender as recommended [35]. For a more correct confident interval (CI)(95% CI) the robust variance was used [34]. In this study, prevalence refers to the proportion of individuals reporting reduced or improved work ability.
In the analyses of reduced work ability, the reference group consisted of those with either constant work ability or improved work ability at the 1-year follow-up (n = 880). The reference group in the analyses for improved work ability consisted of those with either constant work ability or reduced work ability at the 1-year follow-up (n = 1,213).
The work factor variables were coded so that a PR > 1 for reduced work ability meant that an increase in a work factor was hypothesized to have a negative effect on work ability. For improved work ability, a PR > 1 meant that an increase in a work factor was hypothesized to have a positive effect on work ability.
Finally, before the backward stepwise multivariate regression analysis, work factors with a p-value ≤ 0.2 in the univariate Cox proportional hazard regression analysis were selected for further analysis. Next, Spearman’s rank correlation was calculated amongst these selected work factors to check for multicollinearity. All paired correlations were < 0.8 and hence no multicollinearity was found. In the backward stepwise multivariate regression analysis, the variables with the highest p-values were excluded one at a time in order. When all variables had a p-value ≤ 0.05 the step-wise procedure was finished.
The study required no approval from the Regional Ethics Review Board in Gothenburg.