Investigating the impact of poverty on mental illness in the UK Biobank using Mendelian randomization

Abstract

It is unclear whether poverty and mental illness are causally related. Using UK Biobank and Psychiatric Genomic Consortium data, we examined evidence of causal links between poverty and nine mental illnesses (attention deficit and hyperactivity disorder (ADHD), anorexia nervosa, anxiety disorder, autism spectrum disorder, bipolar disorder, major depressive disorder, obsessive-compulsive disorder, post-traumatic stress disorder and schizophrenia). We applied genomic structural equation modelling to derive a poverty common factor from household income, occupational income and social deprivation. Then, using Mendelian randomization, we found evidence that schizophrenia and ADHD causally contribute to poverty, while poverty contributes to major depressive disorder and schizophrenia but decreases the risk of anorexia nervosa. Poverty may also contribute to ADHD, albeit with uncertainty due to unbalanced pleiotropy. The effects of poverty were reduced by approximately 30% when we adjusted for cognitive ability. Further investigations of the bidirectional relationships between poverty and mental illness are warranted, as they may inform efforts to improve mental health for all.

Fig. 1. Research design.

β₁ is the association between the genetic variants and the exposure of interest (MR relevance assumption), i is the potential violation of the MR independence assumption, p is the potential pleiotropic effect of the genetic variants on the outcome (that is, violation of the MR exclusion restriction assumption) and β₂ is the causal association of interest. Created with BioRender.com.

Fig. 2. Manhattan plot of the latent poverty factor.

The x axis indicates the chromosomal position, and the y axis indicates the significance of the association (−log₁₀(P)). The red line represents the genome-wide significance level (5 × 10⁻⁸). SNP effects are estimated with a linear regression model. The P values are two-sided and not adjusted for multiple testing.

Fig. 3. Results of univariable bidirectional MR analysis of poverty against mental illness.

a, Forward analysis. b, Backward analysis. Poverty is a latent variable built using HI as the unit identification, so that an increase in the indicator’s load stands for increased income; the regression coefficients have therefore been reversed to facilitate interpretation of the effect of poverty. The effect estimates on the x axis are log-odds for binary traits (that is, for mental illnesses) and unstandardized linear regression coefficients for continuous traits (that is, for poverty); the error bars represent 95% CIs. Poverty factor, N = 453,688; ADHD, N = 225,534; AN, N = 72,517; ANX, N = 21,761; ASD, N = 46,350; BD, N = 413,466; MDD, N = 138,884; OCD, N = 9,725; PTSD, N = 206,655; SZ, N = 320,404.

Fig. 4. Results of univariable bidirectional MR analysis for each poverty measure against mental illness.

a, Forward analysis. b, Backward analysis. To have a consistent direction of the effect, the SD effect has been reversed. The effect estimates on the x axis are log-odds for binary traits (that is, for mental illnesses) and unstandardized linear regression coefficients for continuous traits (that is, for the poverty indicators); the error bars represent 95% CIs. Missing results are due to an insufficient number of SNPs selected for the MR analysis. HI, N = 379,598; OI, N = 282,963; SD, N = 440,350; ADHD, N = 225,534; AN, N = 72,517; ANX, N = 21,761; ASD, N = 46,350; BD, N = 413,466; MDD, N = 138,884; OCD, N = 9,725; PTSD, N = 206,655; SZ, N = 320,404.

Fig. 5. Results of univariable bidirectional MR analysis of HI levels against mental illnesses.

a, Forward analysis. b, Backward analysis. For LHI, cases <£18,000 and controls ≥£18,000; for LMHI, cases <£29,999 and controls ≥£29,999; for MHHI, cases >£52,000 and controls ≤£52,000; and for HHI, cases >£100,000 and controls ≤£100,000. The effect estimates on the x axis are log-odds for both the forward and backward analyses given that all traits are binary; the error bars represent 95% CIs. Missing results are due to an insufficient number of SNPs selected for the MR analysis. HI, N = 379,598; ADHD, N = 225,534; AN, N = 72,517; ANX, N = 21,761; ASD, N = 46,350; BD, N = 413,466; MDD, N = 138,884; OCD, N = 9,725; PTSD, N = 206,655; SZ, N = 320,404.