Jump to comment:

• Explaining the social gradient: The challenges of mediation
  Laust H Mortensen
  Published on: 24 January 2008

• Published on: 24 January 2008

  Explaining the social gradient: The challenges of mediation

  • Laust H Mortensen, PhD student

  Dear Editor

  In this issue of the journal, a study by Christensen et al. reported on how the social gradient in long-term sickness was explained by a number of risk factors related to work environment and health behavior in a sample of Danish employees.[1]

  I agree with Christensen et al. that it is important to examine what mediates (or explains) social gradients in health: Not only from a scholarly inter...

  Show More

  Dear Editor

  In this issue of the journal, a study by Christensen et al. reported on how the social gradient in long-term sickness was explained by a number of risk factors related to work environment and health behavior in a sample of Danish employees.[1]

  I agree with Christensen et al. that it is important to examine what mediates (or explains) social gradients in health: Not only from a scholarly interest in etiology, but also because interventions targeted at mediators (e.g. work environment) are more realistic than manipulating the determinants (e.g. socioeconomic position). Some authors seem to suggest that mediation is not a tractable problem, but I think that Christensen et al. should be applauded for this contribution.[2]

  Christensen et al report that controlling for health behavior and work environment reduced the sickness absence rate ratios by 22–57%. The percentages were arrived at by comparing the regression coefficients before and after control for health behavior and work environment. If Christensen et al had used a linear regression model with a continuous and normally distributed dependent variable the difference in coefficients could have been used to calculate the percentage explained. However, as the study by Christensen et al. uses Poisson regression the difference in coefficients does not consistently estimate the proportion of the association explained by health behavior and work environment because the two regression models that are compared are nonlinear. The bias might be small, but this depends on the actual data used by Christensen et al. as shown in simulation studies.[3][4]

  Christensen et al. use change in rate ratios - a relative measure of association - to capture the effect of intermediary variables on the social gradient in sickness absence. For example, the authors report only a small reduction of 13% (from 4.22 to 3.66) in the rate ratio between the group of executive managers/academics and the group of semiskilled and unskilled workers when controlling for health behavior. This suggests that regardless of whether we looked at employees with a ‘healthy’ health behavior or those with an ‘unhealthy’ health behavior, the semiskilled and unskilled workers would still have a 3.66 times higher rate of sickness absence. Among the ‘healthy’ employees the two rates would likely be much closer in absolute terms than among the ‘unhealthy’ even though the rate ratio is 3.66 regardless of health behavior. So, among the ‘healthy’ the (absolute) rate difference between executive managers/academics and semiskilled and unskilled workers might be quite small even though the rate ratio is high simply because the rates are low among the 'healthy' (my argument is borrowed from a study by Lynch et al. published in an earlier version of this journal[5]). I think that the authors’ analysis and interpretation is perfectly correct, but I think that it is open to debate if relative inequalities – when the risk or rates are small - are as important as a public health problem as absolute inequalities.

  I want to stress that the paper by Christensen et al. is by no means a faulty paper. The points that I have raised are minor and are not meant to discredit the work of Christensen et al.

  References

  1. Christensen C.B., Labriola M, Lund T et al. Explaining the social gradient in long-term sickness absence: a prospective study of Danish employees. J Epidemiol Community Health 2008;181-3.

  2. Kaufman JS, MacLehose RF, Kaufman S. A further critique of the analytic strategy of adjusting for covariates to identify biologic mediation. Epidemiol Perspect Innov 2004;1:4.

  3. Ditlevsen S, Christensen U, Lynch J et al. The mediation proportion: a structural equation approach for estimating the proportion of exposure effect on outcome explained by an intermediate variable. Epidemiology 2005;16:114-20.

  4. Mackinnon DP, Lockwood CM, Brown CH et al. The intermediate endpoint effect in logistic and probit regression. Clin Trials 2007;4:499- 513.

  5. Lynch J, Davey SG, Harper S et al. Explaining the social gradient in coronary heart disease: comparing relative and absolute risk approaches. J Epidemiol Community Health 2006;60:436-41.

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top