Between 8.8 and 10.2% of CHD morbidity was attributable to job strain, and between 9.4 and 11.2% of CHD mortality was attributable to this exposure for men. Between 15.2 and 19.8% of MD (depression and anxiety) was attributable to job strain for men, between 14.3 and 27.1% for women. As a whole, between 450 000 and 590 000 cases of diseases and 910–1130 deaths were attributable to job strain for men. From 730 000 to 1 380 000 cases of diseases and from 150 to 280 deaths were attributable to job strain for women. The number of deaths attributable to job strain is approximately twice the number of deaths due work fatalities declared in France in 2003. The total number of sick leave days for the year 2003 amounted from 5 to 6.6 million days for men, and from 8.5 to 16 million days for women. The total costs of CHD and MD attributable to job strain exposure ranged from approximately 1.8 to 3 billion euros for the year 2003 in France (0.12-0.19% GDP). Medical costs accounted for 11% of the total costs, value of life costs accounted for 13-15% and sick leave costs for 74-77%. The cost of CHD was estimated at 113–133 million euros and the cost of MD was between 1.7 and 2.8 billion euros. Thus MD accounted for approximately 94-96% of the total costs attributable to job strain exposure. This is due to a higher prevalence of the disease among the population of working age, higher levels of attributable fractions, and also a high amount of sick leave days due to MD. The sensitivity analysis conducted on a more restrictive definition of mental disorders including depression only, produced much lower values in terms of number of cases and costs. This calculation can be considered as very conservative. Indeed, the definition of mental disorders was restricted to clinical depression, and thus excluded anxiety but also symptomatic cases of depression and anxiety. Furthermore, the RR estimates were based on only one available prospective study that produced RR estimates using a standardized diagnostic interview (CIDI) . This study yielded non-significant multi-adjusted RR estimates for women, which did not allow to calculate a low range value of AF estimates.
Some limitations of the study must be pointed out. This estimate of the cost of diseases attributable to job strain exposure may be more under-estimated than over-estimated. We did not produce AF estimates of CHD morbidity and mortality for women since available data were very scarce and the rare included studies yielded non-significant RRs. Therefore the estimation of the cost of CHD attributable to job strain was provided for men only. Another way of interpreting the results may be that our estimates of costs may be null for women, something that may be a bit premature to conclude given the scarcity of the literature. Indeed, recent studies show an increase in CVD prevalence among women in the French population [73, 74], which underlines the need for more investigations on the etiological role of job strain on CVD among women. The estimates did not take account of potential delayed effects of job strain exposure on cardiovascular or mental health. We included only CHD and MD in our study. The number of studies selected in our literature review was very small for musculoskeletal disorders (MSD) and concerned different locations (back, low back, neck, shoulder, upper extremity, elbow, hand, and wrist) . Therefore we could not compute a summary OR and AF estimates for a specific MSD location which could be used for cost of illness estimates. Our study grasped only a part of the total amount of costs attributable to work stress since job strain is only one aspect of stress among other concepts, such as Effort-Reward Imbalance developed by Siegrist . Cost data had also some limitations. Costs did not include production losses due to presenteeism and to early retirement for lack of data. Other cost categories could not be included in our estimates, such as the costs of informal care (provided by families and friends), intangible costs (cost of suffering, pain and discomfort) and the cost of job turnover for lack of data. We also limited the value of lost years of life to the value of lost production. We did not take account of the total losses of social welfare due to deaths through an estimation of willingness to pay or value of a statistical life for instance, for lack of available data. For all these reasons, our estimates can be considered as conservative. In addition, we made a number of assumptions in our calculations and these assumptions may not be verified easily. For example, we assumed that the RR of mental disorders associated with job strain was the same for morbidity and mortality, something that may be consistent for cardiovascular diseases but was not checked for mental disorders, the literature being rare on this topic . Another example, the proportion of suicides due to depression was estimated to be 54%-64% of the total number of suicides in the general population but it was not possible to check whether these figures may be used for the working population.
This study has also several strengths. Our meta-analyses for RR estimates showed a satisfactory homogeneity between studies, except for age-adjusted ORs of MD for women which yielded a significant level of heterogeneity. Overall, the risk of heterogeneity between studies was mitigated by the fact that we included only high quality studies in our systematic review of the literature, with similar exposure and with statistical analyses allowing the estimation of RRs. The data we used are highly consistent: RR, prevalence of exposure, prevalence of disease and cost data are based on the same definition of disease and exposure. To ensure such consistency, we conducted a systematic review of the literature for disease prevalence and costs data in France and we produced data for each gender separately. The only exception may be the use of mental disorders medical costs based on the medical cost of depression only that may underestimate the costs of mental disorders attributable to job strain. We performed a sensibility analysis including depression only to provide a very conservative estimate of mental disorders costs attributable to job strain. The medical costs used in our study included out-of pocket payments by patients along with medical expenses paid by insurances, which is well appropriate for estimations from a societal perspective. We produced two range values of AF estimates with confidence intervals, based on a meta-analysis of RRs. Our calculation method had several strengths: we took account of multi-adjusted and age-adjusted estimates in our calculations, which produced the AF range values. And we combined this approach to meta-analyses to get more precise estimates of AFs based on available data in the literature. This method allowed us to encompass various high-quality RR estimates in our summary RRs, and at the same time to take account of a certain level of uncertainty regarding RRs, since age-adjusted and multi-adjusted estimates yielded different values.
Our summary RRs for CVD are consistent with those from the meta-analysis by Kivimaki et al.  providing a summary age and gender-adjusted RR of 1.45 (95% CI: 1.15-1.84) and a multi-adjusted RR of 1.16 (95% CI: 0.94-1.43). Our RR estimates regarding men are higher, but this difference could be explained by the fact that Kivimaki et al’s estimates included both men and women. Our results for MD are also consistent with those summarized in the meta-analysis by Stansfeld and Candy  (summary OR: 1.82, 95% CI: 1.06-3.10). Our estimates of attributable fractions for CVD mortality are more conservative than those reported by Nurminen & Karjalainen , who found estimates of 16% for men for the fractions of cardiovascular deaths attributable to job strain in Finland. Our AF estimates for MD are in line with those of LaMontagne et al.  who reported fractions of 13.2% for men and 17.2% for women attributable to job strain in Australia. Their fractions resulted from the summary OR estimates from Stansfeld and Candy’s meta-analysis . Our results are higher than the costs of diseases attributable to job strain exposure in a previous study by Béjean et Sultan-Taïeb . For the year 2000 in France, total costs amounted for 1.2-1.6 billion euros but included also musculoskeletal disorders. Differences can be explained by the fact that the fractions of MD for women were underestimated (4.8%) since they were based on a limited selection of OR estimates in the literature. The cost of depression attributable to job strain in Australia was estimated by LaMontagne et al.  at 730 million dollars AUD (approx.. 510 million euros) in 2007, given that 1.54 million persons suffer from depression in the Australian workforce. It is however difficult to compare these results with ours since categories of costs included in the estimations are different: job turnover and presenteeism costs are included in LaMontagne et al.’s study, while indirect costs related to premature death (suicide) are excluded.
Our results provide an evaluation at one point of time, allowing projections of the cost of job strain according to the evolution of working environments and the trend of prevalence of exposure to job strain. It also allows comparisons with other countries. Our summary estimates of RRs for CHD and MD could be used for AF calculations in other countries where the prevalence of exposure to job strain has been measured in the working population.