Main findings
This economic evaluation was set out to see if the Stress-Prevention@Work implementation strategy among employees of a large health-care organisation represents good value for money as seen from the employer’s perspective. The Stress-Prevention@Work implementation strategy requires an investment of €50 per participating employee. After the intervention, employees in the experimental condition were less often absent from their work and were in addition more productive when compared to matched health-care workers who did not receive the intervention. The greater productivity in the intervention group represented a net benefit of on average €2981 per employee per year when compared to the lesser productivity in the waitlisted control group. This is equivalent to a return-on-investment of close to €60 (rounded) per one euro invested. The outcomes are surrounded by some uncertainty. Nonetheless, there is a likelihood of 96.7% that the initial investment will at least break even after 1 year and there is an 88.2% likelihood that the net benefits will amount to €1000 per employee in a year. Hence, the overall picture is in favour of the Stress-Prevention@Work implementation strategy, which is in line with recent views that (mental) health promotion can benefit workers and employers (Fouquet et al.
2019; Sorensen et al.
2016; Thompson et al.
2018). As Stress-Prevention@Work is mainly a preventive strategy, it is interesting to see positive effects given the relatively short time horizon. This has been the case for other programmes as well (Noben et al.
2014,
2015; Oude Hengel et al.
2014); however, a short time horizon is likely to affect net benefits as worksite health promotion programmes increase per-employee costs at the short term, while the related improvements in productivity solely occur at the longer term (van Dongen et al.
2017). Nonetheless, more than a decade ago, it was concluded that employers and researchers remain largely unaware of the value of quality care and psychiatric intervention, that productivity may increase as a consequence of psychiatric intervention, and that those improvements may offset the cost of the treatment (Goetzel et al.
2002; Langlieb and Kahn
2005).
Not all programmes are equally effective, for example due to poor implementation or low compliance rates (Goetzel et al.
2014; van Dongen et al.
2016). Hence, based on the work of Goetzel et al. (
2014), employers considering implementing health promotion programmes should focus on (1) clarifying why the programme needs to be implemented (i.e., do not focus solely on financial gains); (2) ensuring that the programme fits into the culture of the organisation; and (3) ongoing outcome monitoring and evaluation to strengthen implementation (Goetzel et al.
2014).
Strengths and limitations
This study has some strengths and limitations. Among its strengths, we must mention the use of robust statistical techniques accounting for baseline imbalances, dropout, clustering of employees in their teams and the non-normality of costs. Sensitivity analyses attested to the robustness of the main findings. There are also some limitations that need to be addressed.
First, neither the individual employees nor their teams had been allocated randomly to the control and experimental conditions. Nonetheless, at baseline we did not observe between-group differences that required adjustment in the economic evaluation except costs of presenteeism (and absenteeism) at baseline. A sensitivity analysis adjusting for this possible confounder showed that the net benefits were underestimated in the main analysis and increased from €2981 to €3106, indicating that the main analysis was more conservative.
Second, loss to follow-up was substantial and could have biased the estimates at follow-up. However, we used intention-to-treat analysis by imputing missing observations. This was done with regression imputation with both predictors of outcome (to increase the accuracy of the imputed values) and by predictors of loss to follow-up (to counter the effects of selective dropout). In one sensitivity analysis, linear mixed modelling was used instead and replicated the pattern of how costs stayed much the same in the experimental group between t0 and t1 and then dropped at t2, which contrasts with the control group where costs increased between t0 and t1 and then dropped off at t2. Lastly, an analysis was performed in which missing data were imputed using multiple imputation with PMM. Given that these replications provided similar results, this attested to the robustness of our main analysis. When comparing responders with non-responders (dropouts), non-responders demonstrated higher baseline costs of absenteeism (€389 vs €321), but lower baseline costs of presenteeism (€52 vs €74). Other characteristics (e.g., job satisfaction, days of work per week, hours of work per week, female gender and age) were similar in both groups.
Third, absenteeism and presenteeism were based on self-report over the last month and this may have caused some recall bias. It should be noted that it is hard to see how the costs of presenteeism could have been measured without relying on self-report. Also, it was a conscious choice to keep the recall period short (last 4 weeks) and not, say, the last 3 months to minimise recall bias. Ideally, we would have liked to cross-validate the self-reported data of employees with company-registered sickness absence data. This was, however, not possible due to privacy regulations.
Fourth, stress reduction may also have impacted on staff turnover and the employees’ ability to continue working until the age of retirement. Now, the intervention’s impact on resignation and early retirement remains unknown, as such assessments would have required a much longer follow-up and larger sample size.
Fifth, the effect of the (additional) intervention(s) may be dependent on how these intervention activities are undertaken (i.e., during or after working hours). For example, we have assumed that more extensive interventions take place outside office hours; however, this assumption might be harmful to the level of compliance achieved in implementing the intervention and demotivate participants as the employer invites them to participate in an intervention to reduce work stress and then tells them to do it in their spare time. Hence, it is recommended that employer and employee both keep this in mind and maintain an open conversation. Further studies should examine the extent to which this issue plays an important role in the compliance and effectiveness of more extensive additional interventions.
Sixth, the waitlisted control design may have caused a more beneficial effect of Stress-Prevention@Work compared to the waitlisted control condition, as the use of a waitlisted control condition has been associated with an overestimation of effect sizes in psychotherapy (Furukawa et al.
2014; Cuijpers et al.
2019). Ideally, the control arm would consist of a care as the usual arm, but the waitlisted control was assumed to provide an important incentive to participate in the study. Moreover, Stress-Prevention@Work primarily focused on the organisation instead of individual workers. Although it is possible that organisations in the waitlisted control condition may have postponed implementation of alternative interventions, individuals were not specifically confronted with the results of the allocation process and hence negative feelings associated with the allocation process, which have been argued to be one of the causes of a biased effect size (Furukawa et al.
2014), may have played a minor role.
Lastly, the costs were restricted to the costs of offering the Stress-Prevention@Work implementation strategy, which was the sole purpose of this study. Nonetheless, it should be kept in mind that the Stress-Prevention@Work implementation strategy was designed to encourage the uptake of a stress-reduction intervention—which inevitably introduces costs of its own. These intervention costs have not been included in our analysis, but logic suggests the following: if the per-employee costs of a selected stress intervention would be as high as €2981, then the expected net benefits would cease to exist, and the employer would see a break even between the total costs (of Stress-Prevention@Work plus the stress-reducing intervention) and the benefits (stemming from lesser productivity losses). This suggests that the added per-employee cost of one or another stress intervention can be substantial well before the employer begins to see that the costs begin to exceed the benefits. It should be noted, however, that in the Netherlands, mental health care is most often covered by the health-care insurance (either basic or specialised mental health care).