Education and myopia: assessing the direction of causality by mendelian randomisation

Abstract

Objectives: To determine whether more years spent in education is a causal risk factor for myopia, or whether myopia is a causal risk factor for more years in education.

Design: Bidirectional, two sample mendelian randomisation study.

Setting: Publically available genetic data from two consortiums applied to a large, independent population cohort. Genetic variants used as proxies for myopia and years of education were derived from two large genome wide association studies: 23andMe and Social Science Genetic Association Consortium (SSGAC), respectively.

Participants: 67 798 men and women from England, Scotland, and Wales in the UK Biobank cohort with available information for years of completed education and refractive error.

Main outcome measures: Mendelian randomisation analyses were performed in two directions: the first exposure was the genetic predisposition to myopia, measured with 44 genetic variants strongly associated with myopia in 23andMe, and the outcome was years in education; and the second exposure was the genetic predisposition to higher levels of education, measured with 69 genetic variants from SSGAC, and the outcome was refractive error.

Results: Conventional regression analyses of the observational data suggested that every additional year of education was associated with a more myopic refractive error of -0.18 dioptres/y (95% confidence interval -0.19 to -0.17; P<2e-16). Mendelian randomisation analyses suggested the true causal effect was even stronger: -0.27 dioptres/y (-0.37 to -0.17; P=4e-8). By contrast, there was little evidence to suggest myopia affected education (years in education per dioptre of refractive error -0.008 y/dioptre, 95% confidence interval -0.041 to 0.025, P=0.6). Thus, the cumulative effect of more years in education on refractive error means that a university graduate from the United Kingdom with 17 years of education would, on average, be at least -1 dioptre more myopic than someone who left school at age 16 (with 12 years of education). Myopia of this magnitude would be sufficient to necessitate the use of glasses for driving. Sensitivity analyses showed minimal evidence for genetic confounding that could have biased the causal effect estimates.

Conclusions: This study shows that exposure to more years in education contributes to the rising prevalence of myopia. Increasing the length of time spent in education may inadvertently increase the prevalence of myopia and potential future visual disability.

Conflict of interest statement

Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years ; no other relationships or activities that could appear to have influenced the submitted work .

Published by the BMJ Publishing Group Limited. For permission to use (where not already granted under a licence) please go to http://group.bmj.com/group/rights-licensing/permissions.

Figures

Fig 1

Numbers of participants in UK Biobank who passed validation for mendelian randomisation study. MSE=mean spherical equivalent

Fig 2

Observational association between age full time education was completed and refractive error for 69 798 people in UK Biobank. On average, more educated people had higher levels of myopia (more negative refractive error). Whiskers represent 95% confidence intervals

Fig 3

Results of bidirectional mendelian randomisation. (Top panel) 69 variants associated with educational attainment in Okbay et al were linked to higher levels of myopia (more negative mean spherical equivalent (MSE)) in UK Biobank. (Bottom panel) 44 variants associated with myopia (more negative MSE) in Pickrell et al were not linked with more time spent in education in UK Biobank. Regression line and standard errors (shaded area) fitted using robust linear regression. Whiskers represent 95% confidence intervals

Fig 4

Confounding bias plots. Plots showing relative bias in instrumental variable estimate (blue) and standard multivariable regression estimate (white) from potential confounders including: place of birth (northing and easting coordinates), Townsend deprivation index, age, sex, breastfed, birth weight, and first 10 genetic principal components (PC), when (A) estimating the effect of time spent in education on refractive error; and (B) estimating the effect of refractive error on time spent in education. Townsend deprivation index was natural log transformed