Introduction

Psychosocial work environment has been the subject of growing attention in the field of occupational health within the last decades. Psychosocial work factors have been recognised as occupational risk factors for health outcomes, especially for cardiovascular diseases and mental disorders (Belkic et al. 2004; Bonde 2008; Netterstrom et al. 2008). The most common model used to measure psychosocial work factors is the job strain model (Karasek et al. 1998), composed of two main dimensions: psychological demands and decision latitude comprising two sub-scales, skill discretion and decision authority. According to this model, the combination of high demands and low latitude may have adverse effects on health. Another model is the effort-reward imbalance (ERI) model (Siegrist et al. 2004), including effort at work and reward in terms of esteem, job promotion and job security, that postulates that the imbalance between high effort and low reward may be harmful for health. Other concepts have emerged in the literature such as job insecurity (Sverke et al. 2006).

Attributable fractions may be very informative to evaluate the burden of diseases attributable to a risk factor and also to estimate the costs of diseases imputable to a risk factor (Bejean and Sultan-Taieb 2005; Leigh 2006). Evaluations of attributable fractions have been made for a number of occupational exposures of physical or chemical nature (Driscoll et al. 2005; Fingerhut et al. 2005; Nelson et al. 2005). Although a substantial literature has been produced on the etiological role of psychosocial work factors on health outcomes, the literature is almost non-existent on the estimation of fractions of diseases attributable to these factors. To our knowledge, three studies so far were performed on this topic. The first study provided estimates of fractions of mortality attributable to occupational factors, including the fractions of cardiovascular diseases and mental disorders attributable to psychosocial work factors such as job strain according to Karasek’s model and shift work in Finland (Nurminen and Karjalainen 2001). A second study produced estimates of fractions of depression attributable to job strain in the Australian state of Victoria (Lamontagne et al. 2008). A third study estimated the fractions of cardiovascular diseases, mental disorders and musculoskeletal disorders attributable to job strain in France (Sultan-Taieb et al. 2011).

The objectives of the study were to evaluate the fractions of cardiovascular diseases and mental disorders attributable to job strain, effort-reward imbalance (ERI) and job insecurity in Europe as a whole and for each European country. Therefore, we studied not only the most widely used concept, job strain, but also other concepts, ERI and job insecurity, using harmonised European data on 31 countries to make European comparisons possible. This study is thus a first attempt to provide estimates of attributable fractions in order to contribute to a better knowledge of the burden of these factors in each country and also in Europe as a whole.

Materials and methods

Attributable fractions

Attributable fractions (AF) produce an estimate of the fraction of cases that is ‘attributable to an exposure in a population and that would not have been observed if the exposure had been non-existent’ (Nurminen and Karjalainen 2001). AFs are calculated from the estimate of the prevalence of exposure Pe (proportion of the population exposed to this risk factor), combined with estimated relative risks (RR) of disease or death due to exposure to this risk factor (Levin 1953): AF = Pe(RR−1)/[1 + Pe(RR−1)]. This formula was used for adjusted RRs.

Prevalence of exposure (Pe)

The prevalence of exposure for the whole Europe and each country was calculated from the fourth European Working Conditions Survey (EWCS) carried out by the European Foundation for the improvement of living and working conditions (Eurofound) in 2005 (Parent-Thirion et al. 2007). This periodical survey aims at providing information on working conditions in European countries and may be considered as a major source of harmonised and comparable data, the same questionnaire and protocol being used in all countries. The survey covered 25 EU countries plus four acceding and candidate countries and two members of the European Free Trade Association, making a total of 31 countries. The sample is representative of the persons in employment (employees and self-employed, according to Eurostat definition, that is, all persons aged 15 years or more who worked within the week preceding the interview). The representativeness of the sample at national level was achieved by the sampling procedure and the calculation of weighted data (see below). In each country, the EWCS sample followed a multi-stage, stratified and clustered design with a ‘random walk’ procedure for the selection of households. The selection of the worker to be interviewed (if more than one worker in the household) was the person whose birthday was the latest. All interviews were conducted face-to-face in the respondent’s own household. Details on the survey and sampling design may be found elsewhere (Parent-Thirion et al. 2007). The sample included 29,680 workers, 14,881 men and 14,799 women, with a cooperation rate (proportion of completed interviews to all eligible units contacted), often reported as a response rate, of 66 %. The data have already been used in two previous papers that compared the prevalence of exposure to eighteen psychosocial work factors between European countries (Niedhammer et al. 2012a) and that explored the association between psychosocial work factors and sickness absence (Niedhammer et al. 2012b). In the present study, we studied psychosocial work factors for which summary relative risks were available in the literature and prevalence of exposure could be measured using the EWCS data (see below).

The three exposures studied were job strain, ERI and job insecurity and were constructed from items of the EWCS questionnaire (‘Appendix’). Psychological demands were based on 5 items and decision latitude on 11 items (4 items for skill discretion, 7 items for decision authority, each sub-dimension being given the same weight in the total scale of decision latitude). The scores were dichotomised at the median of the total sample to define low and high levels of exposure following the recommendations of the original instrument (Karasek et al. 1998). People exposed to job strain were those who had both high demands and low latitude. The measures for ERI model were effort (6 items), job insecurity (1 item), job promotion (3 items) and esteem (4 items). Reward, following ERI model, was constructed using a sum of the sub-dimensions of job insecurity, job promotion and esteem, each sub-dimension being given the same weight. Finally, a measure for ERI was constructed using a weighted ratio between effort and reward, and exposure to imbalance between high effort and low reward was defined by a ratio over 1, following the recommendations of the original instrument (Siegrist et al. 2004). The item of job insecurity used for ERI was also used for measuring job insecurity alone.

The prevalence of exposure to job strain, ERI and job insecurity was calculated using weighted data to provide an estimate that was representative for the whole European working population of the 31 countries and for each country. These weights were calculated using a calibration on margins for the following calibration variables: number of workers in the household, gender, age, occupation, economic activity, region and country size (country size was used for the whole European sample only). The differences in the prevalence of exposure to job strain, ERI and job insecurity between countries were tested using the Wald test, making the assumption that the prevalence had an independent normal distribution within each country (Rothman et al. 2008).

Relative risk estimates

The relative risk (RR) estimates were obtained from the available meta-analyses on the associations between psychosocial work factors and health outcomes. Four meta-analyses using prospective studies provided summary RRs (or odds ratios that were used as estimates of RRs) on the predictive effects of psychosocial work factors on cardiovascular diseases (Kivimaki et al. 2012, 2006; Steptoe and Kivimaki 2013) and on mental disorders (Stansfeld and Candy 2006). The summary RRs used were those that were multiple adjusted in the meta-analyses. These meta-analyses were based on a systematic review of studies with high quality standard, that is, using prospective design, satisfactory instruments for measuring psychosocial work factors, adjustment for confounding and adequate statistical analysis. But such studies are scarce in the literature and for the most recent concepts for psychosocial work factors, they are still lacking. These meta-analyses (Kivimaki et al. 2006, 2012 ; Stansfeld and Candy 2006; Steptoe and Kivimaki 2013) summarised the etiological role of various concepts for psychosocial work factors on cardiovascular diseases and mental disorders. However, given the criteria used to select the studies in the meta-analyses and the available literature, only job strain, ERI, organisational justice (for cardiovascular diseases only) and job insecurity (for mental disorders only) were retained as psychosocial work factors in these meta-analyses. As no item was available in the EWCS questionnaire about organisational justice, this concept was not studied in the present study. Given that three meta-analyses (Kivimaki et al. 2012, 2006; Steptoe and Kivimaki 2013) provided information on the association between job strain and cardiovascular diseases, we retained the very recent summary RR estimate (adjusted for gender, age and socioeconomic status) produced by Kivimaki et al. (2012) that had the advantages to be multiple adjusted and based on both published and unpublished studies. The RR estimates and 95 % CI obtained from the three meta-analyses (Kivimaki et al. 2006, 2012 ; Stansfeld and Candy 2006) are presented in Table 1. For cardiovascular diseases, the RR of job strain was significant, but the RR of ERI was not. All the RRs were significant for mental disorders.

Table 1 Summary multiple adjusted relative risks and 95 % CI
Full size table

Calculation of confidence intervals for attributable fractions

The AFs were calculated for the whole Europe and for each country separately following the formula presented above.

Simulation-modelling techniques were used to obtain confidence intervals for AFs that reflect the main sources of uncertainty in the calculations. AF is in fact a simple function of two independent variables, Pe and RR, that have a normal distribution. Repeated random samples of these variables were generated to obtain 100 000 values of AF. The distribution of these values provided the mean and variance and thus the 95 % CI of the AFs.

The differences in the AFs between countries were tested using the Wald test (Rothman et al. 2008) to determine whether there was any significant departure from homogeneity of the AFs across countries.

All statistical analyses were performed using SAS software.

Results

Tables 2, 3 and 4 provide the prevalence of exposure to job strain, ERI and job insecurity and their 95 % confidence intervals for Europe as a whole and for each country. The prevalence of exposure to job strain, ERI and job insecurity was, respectively, 26.90, 20.44 and 14.11 % in Europe and varied, respectively, from 14.80 to 35.40 % for job strain, from 9.52 to 37.51 % for ERI and from 5.52 to 32.17 % for job insecurity according to country. The prevalence of exposure to job strain, ERI and job insecurity differed significantly between countries.

Table 2 Fractions of cardiovascular diseases and mental disorders attributable to job strain in Europe
Full size table
Table 3 Fractions of cardiovascular diseases and mental disorders attributable to effort-reward imbalance (ERI) in Europe
Full size table
Table 4 Fractions of mental disorders attributable to job insecurity in Europe
Full size table

Tables 2, 3 and 4 also present the fractions of cardiovascular diseases and metal disorders attributable to job strain, ERI and job insecurity for Europe and for each country. The fractions of cardiovascular diseases attributable to job strain were significantly different from zero for Europe and for each country (Table 2). The fraction was 4.46 % for Europe and varied from 2.51 % to 5.77 % according to country, without any significant difference between countries. The fraction of mental disorders attributable to job strain was 18.16 % in Europe, varied from 11.14 to 22.33 %. AFs were significantly higher than zero in most countries except Denmark, Ireland, Latvia, Netherlands, Norway, Sweden and Switzerland. The three countries with the lowest AFs were Netherlands, Sweden and Switzerland, and those with the highest AFs were Czech Republic, Greece and Slovenia. However, the differences in the fractions of mental disorders attributable to job strain were not significant between countries.

The fraction of cardiovascular diseases attributable to ERI was 18.21 % in Europe and varied from 9.78 % to 27.89 % without any significant difference between countries, and all fractions were not significantly different from zero (Table 3). The fraction of mental disorders attributable to ERI was 14.81 % and varied from 7.52 % to 24.06 %. All AFs were significantly higher than zero in Europe and each country. The fraction attributable to ERI was found to be the lowest in Bulgaria, Ireland and Latvia, and the highest in Greece, Slovenia and Turkey. The differences in the fractions of mental disorders attributable to ERI between countries were significant at p = 5 %.

The fraction of mental disorders attributable to job insecurity was 4.53 % in Europe, varied from 1.83 to 9.66 % according to country, and all AFs were significantly higher than zero (Table 4). The three countries with the lowest AFs were Denmark, Luxembourg and UK, and the three countries with the highest AFs were Czech Republic, Poland and Slovenia. However, the differences in the fractions of mental disorders attributable to job insecurity were not significant between countries.

Discussion

The fractions of cardiovascular diseases and mental disorders attributable to job strain, effort-reward imbalance and job insecurity were evaluated using three meta-analyses summarising relative risks of diseases and the fourth European working conditions survey providing estimates of prevalence of exposure. The fraction of cardiovascular diseases attributable to job strain was 4 % for Europe and significantly different from zero for Europe and all countries. The fraction of cardiovascular diseases attributable to ERI was 18 % for Europe and not significantly different from zero for Europe as well as for each country. The fractions of mental disorders attributable to job strain, ERI and job insecurity were 18 %, 15 % and 5 %, respectively, and significantly higher than zero for Europe. These fractions were also significant for each country, but there were some exceptions for job strain, as the fractions were not significant in Denmark, Ireland, Latvia, Netherlands, Norway, Sweden and Switzerland. Significant differences between countries were found for the fractions of mental disorders attributable to effort-reward imbalance, with some countries presenting lower fractions and others higher fractions.

The comparison with previous studies may be difficult as there have been almost no previous study on this topic. The study by Nurminen and Karjalainen (Nurminen and Karjalainen 2001) provided estimates of fractions of cardiovascular fatalities attributable to job strain in Finland that were 16 % for men and 19 % for women and estimated that 15 % of deaths related to depressive episodes among men and 10 % among women were attributable to job strain. As these results concerned mortality only, they may not be comparable to ours. Lamontagne et al. (Lamontagne et al. 2008) found estimates of fractions of depression attributable to job strain of 13 % for men and 17 % for women in the Australian state of Victoria. Although not from Europe, this study is very consistent with our results, something that may not be so surprising as the authors followed a very similar approach based on the available meta-analyses. Previous findings for France (Sultan-Taieb et al. 2011) showed that the fractions of cardiovascular diseases attributable to job strain were from 0 to 25 % (i.e. non-significant) and the fractions of mental disorders attributable to the same exposure were from 5 to 34 %, which is in agreement with the present results. Indeed, we found here that the fractions of mental disorders attributable to job strain may be estimated from 0.18 to 30.62 % in France (95 % CI). Finally, the recent meta-analysis by Kivimaki et al. (2012) provided a population attributable risk for job strain and coronary heart disease of 3.4 % (95 % CI 1.5–5.4) in 7 European countries (Belgium, Denmark, Finland, France, Netherlands, Sweden and UK), this estimate being very close to ours of 4.46 % (95 % CI 1.26–7.65).

Our study also explored differences between countries for attributable fractions, something that has not been done before. Significant differences were observed only for the fraction of mental disorders attributable to ERI and were explained by significant differences in ERI exposure between countries, in agreement with previous results (Niedhammer et al. 2012a). For lack of available data, we could not estimate attributable fractions on the basis of specific RRs for each European country. Differences between AFs are due to differences in exposure prevalence in our estimations. However, differences between countries in RRs associated with psychosocial work factors could be observed and explained. For example, Kim et al. pointed out that macro-level structures of welfare regimes may play a role in the association between precarious employment/job insecurity and health outcomes and in particular that the comprehensive employment policies of Scandinavian welfare states may help to moderate an array of negative health-related consequences associated with precarious employment and job insecurity (Kim et al. 2012). Another explanation might also be related to the differences between countries in terms of occupational medicine density and quality.

Strengths of the study deserve to be mentioned. Our study is the first one to provide estimates of attributable fractions for psychosocial work factors at European level. Although the causality of the associations between psychosocial work factors and health outcomes has been questioned, authors underlined that the calculation of attributable fractions may be justified even if causality has not been demonstrated to a high degree of certainty (Benichou et al. 1998; Nurminen and Karjalainen 2001; Walter 1998; Wilson et al. 1998). The study covered a large number of countries in Europe. Such an analysis has never been done before. The prevalence of exposure was calculated using representative and harmonised data (indeed, the data from the European working conditions survey use the same protocol and questionnaire in all countries) making generalisation and comparisons between countries possible. The study provided estimates of attributable fractions for Europe as a whole, but also for each country; such results may be useful at European and national levels. The study focussed on job strain that is the most well-known concept to evaluate psychosocial work factors, but not only, as ERI and job insecurity were also studied.

Limitations of the study should be mentioned. The prevalence of exposure to job strain and ERI was derived from the data of the European working conditions survey, and these exposures were calculated using proxies, as the validated instruments were not available. Such a limitation may lead to imprecision in these measures. However, additional analyses supported expected psychometric properties (Niedhammer et al. 2012b). Furthermore, other studies demonstrated the interest and validity to construct proxies, for example Karasek’s JCQ-like dimensions (Karasek et al. 2007). The RRs used in the study were extracted from three available meta-analyses (Kivimaki et al.2006, 2012; Stansfeld and Candy 2006) based on a number of prospective studies in various countries. There may be no exact comparability between the definition of exposure to job strain between studies used in these meta-analyses; in other words, the exposure to job strain may not have been defined using the median of the scores of demands and latitude in all studies. In addition, we made the assumption that the RR estimates may not be different from one country to another, which may be consistent from a theoretical point of view, but nevertheless should be verified. Furthermore, the robustness of the summary RRs may vary for job strain and the two other measures of ERI and job insecurity, as the number of prospective studies may be lower for the two last concepts. Our statistical strategy may be conservative because we used adjusted RRs to calculate the AF estimates. For example, in the meta-analysis by Kivimaki et al. (2012), age- and gender-adjusted summary RR was 1.23 for the association between job strain and cardiovascular diseases (whereas multiple adjusted RR was 1.17). The study focussed on the attributable fractions of cardiovascular diseases and mental disorders and examined neither other diseases nor addictive behaviours that may be associated with psychosocial work factors and may be risk factors for various diseases. Thus, the fractions obtained may underestimate the global burden of diseases attributable to the exposures considered. Our findings may also underestimate the burden of cardiovascular diseases and mental disorders attributable to psychosocial work factors, as other exposures were not studied, organisational justice for example. Consequently, our results are more likely to be underestimated than overestimated. Furthermore, the fractions of mental disorders obtained in our study were more related to mental health in general than to mental disorders given the outcomes used in the studies included in the meta-analysis. The analysis for men and women separately was not possible as the multiple adjusted RRs from the meta-analyses were given for both genders in combination. Moreover, available data did not allow to produce AF estimates for different occupational groups. Studies showed that there may be differences in exposure according to occupational groups, for example low-skilled occupational groups such as blue collars, clerks or service workers may be more likely to be exposed to most psychosocial work factors (Karasek et al. 1998; Niedhammer et al. 2008). The size of the EWCS sample does not allow to evaluate the prevalence of exposure according to occupational group in each country. Furthermore, very few studies produced RR estimates for specific occupational groups (Sultan-Taieb et al. 2011). Because of a lack of available data, calculation of AFs according to occupational group may be difficult. However, such AF estimations would produce very useful information for targeting prevention policies towards high-risk groups.

To conclude, the study underlines that the fractions of mental disorders attributable to job strain, effort-reward imbalance and job insecurity may be substantial in Europe and that these fractions may differ across countries especially for the fraction of mental disorders attributable to effort-reward imbalance. The calculation of such attributable fractions may be informative to estimate the burden of diseases attributable to specific occupational exposures of the psychosocial work environment. These results based on attributable fractions provide very useful data, since insurance-based data provide highly underestimated statistics on the burden of diseases resulting from occupational exposures. In the case of diseases imputable to psychosocial work factors, recognition of the occupational origin of such diseases is very seldom and workers compensations data cannot be used to estimate this burden. Therefore, these attributable fractions may be useful elements for guiding prevention policies and defining priorities at European and national levels.