Effects of depression on employment and social outcomes: a Mendelian randomisation study

Abstract

Background: Depression is associated with socioeconomic disadvantage. However, whether and how depression exerts a causal effect on employment remains unclear. We used Mendelian randomisation (MR) to investigate whether depression affects employment and related outcomes in the UK Biobank dataset.

Methods: We selected 227 242 working-age participants (40-64 in men, 40-59 years for women) of white British ethnicity/ancestry with suitable genetic data in the UK Biobank study. We used 30 independent genetic variants associated with depression as instruments. We conducted observational and two-sample MR analyses. Outcomes were employment status (employed vs not, and employed vs sickness/disability, unemployment, retirement or caring for home/family); weekly hours worked (among employed); Townsend Deprivation Index; highest educational attainment; and household income.

Results: People who had experienced depression had higher odds of non-employment, sickness/disability, unemployment, caring for home/family and early retirement. Depression was associated with reduced weekly hours worked, lower household income and lower educational attainment, and increased deprivation. MR analyses suggested depression liability caused increased non-employment (OR 1.16, 95% CI 1.06 to 1.26) and sickness/disability (OR 1.56, 95% CI 1.34 to 1.82), but was not causal for caring for home/family, early retirement or unemployment. There was little evidence from MR that depression affected weekly hours worked, educational attainment, household income or deprivation.

Conclusions: Depression liability appears to cause increased non-employment, particularly by increasing disability. There was little evidence of depression affecting early retirement, hours worked or household income, but power was low. Effective treatment of depression might have important economic benefits to individuals and society.

Keywords: Keywords.

Figure 1

Scatter plot of sick/disabled–SNP associations versus exposure–SNP associations. X axis includes depression–SNP regression coefficient estimates from Howard and colleagues; Y axis includes sick/disabled–SNP log odds from UK Biobank regressions. Also plotted are the fits for several causal effect estimation methods. MR, Mendelian randomisation; RAPS, Robust Adjusted Profile Score; SNP, single-nucleotide polymorphism.

Figure 2

Forest plots of causal effect estimates of depression on employment outcome. Causal effect estimates for change in depression affection status from unaffected to affected on (A) not in paid employment, (B) sick/disabled, (C) caring for home/family, (D) early retirement and (E) unemployment. The association estimates transformed onto the same scale as the Mendelian randomisation estimates are presented in rows prefixed ‘assoc. with depression’, for example, ‘assoc. with depression, prev=0.1’, where prev=0.1 indicates a baseline depression prevalence of 10%. RAPS, Robust Adjusted Profile Score.

Figure 3

Forest plots of causal effect estimates of depression on other outcomes. Causal effect estimates for change in depression affection status from unaffected to affected on (A) Townsend Deprivation Index, (B) hours worked, (C) highest educational attainment and (D) household income level. The association estimates transformed onto the same scale as the Mendelian randomisation estimates are presented in rows prefixed ‘assoc. with depression’, for example, ‘assoc. with depression, prev=0.1’, where prev=0.1 indicates a baseline depression prevalence of 10%. RAPS, Robust Adjusted Profile Score.