Investigating causal effects of income on health using two-sample Mendelian randomisation

Income is associated with many health outcomes, but it is unclear how far this reflects a causal relationship. Mendelian randomisation (MR) uses genetic variation between individuals to investigate causal effects and may overcome some of the confounding issues inherent in many observational study designs.

We used two-sample MR using data from unrelated individuals to estimate the effect of log occupational income on indicators of mental health, physical health, and health-related behaviours. We investigated pleiotropy (direct effects of genotype on the outcome) using robust MR estimators, CAUSE, and multivariable MR including education as a co-exposure. We also investigated demographic factors and dynastic effects using within-family analyses, and misspecification of the primary phenotype using bidirectional MR and Steiger filtering.

We found that a 10% increase in income lowered the odds of depression (OR 0.92 [95% CI 0.86-0.98]), death (0.91 [0.86-0.96]), and ever-smoking (OR 0.91 [0.86-0.96]), and reduced BMI (- 0.06 SD [- 0.11, - 0.003]). We found little evidence of an effect on alcohol consumption (- 0.02 SD [- 0.01, 0.05]) or subjective wellbeing (0.02 SD [- 0.003, 0.04]), or on two negative control outcomes, childhood asthma (OR 0.99 [0.87, 1.13]) and birth weight (- 0.02 SD, [- 0.01, 0.05]). Within-family analysis and multivariable MR including education and income were imprecise, and there was substantial overlap between the genotypes associated with income and education: out of 36 genetic variants significantly associated with income, 29 were also significantly associated with education.

MR evidence provides some limited support for causal effects of income on some mental health outcomes and health behaviours, but the lack of reliable evidence from approaches accounting for family-level confounding and potential pleiotropic effects of education places considerable caveats on this conclusion. MR may nevertheless be a useful complement to other observational study designs since its assumptions and limitations are radically different. Further research is needed using larger family-based genetic cohorts, and investigating the overlap between income and other socioeconomic measures.