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

69,798 UK Biobank participants in the analytic sample (tables 1-2 and fig 2 all give 69,798; the paper's own abstract says 67,798 — an internal inconsistency in the published article, and the tables are the estimates actually reported). Instruments from 23andMe (44 myopia variants, GWAS n = 191,843) and SSGAC/Okbay (69 education variants, GWAS n = 293,723).

Adults aged 40-69 at recruitment from England, Scotland and Wales with measured refractive error and reported age of completing full-time education. Refractive error measured by NON-cycloplegic autorefraction (accommodation not paralysed), averaged over both eyes.

Bidirectional mendelian randomisation. Genotype is fixed at conception and randomised at meiosis, so it functions as a natural experiment breaking the reverse causation and confounding that make the education-myopia correlation uninterpretable. Instruments are strong (myopia allele score explains 4.32% of refractive-error variance, F = 3155; education allele score 0.71% of education variance, F = 464). Sensitivity analyses (MR-Egger, weighted median, weighted mode) all agreed in sign and magnitude and the Egger intercepts were indistinguishable from zero, so there is little evidence of directional horizontal pleiotropy. Caveats the record previously understated: (1) although labelled a "two sample" design, the main analyses computed the instrument-exposure and instrument-outcome associations IN THE SAME SAMPLE, which is why MR-Egger had to be run as a split-sample analysis; (2) the exposure is age of leaving full-time education among adults schooled 5-7 decades ago, not any specific classroom practice, and the authors state plainly that they cannot determine "which components of educational practices ... have led to increases in myopic refraction"; (3) UK Biobank is a healthier, better-educated, self-selected cohort, which the authors name as a possible source of bias in both estimates; (4) two confounders (the geographical "northing" coordinate and genetic principal component 9, which tracks being born in Wales) showed evidence of biasing the IV estimate, though adjustment left the results essentially unchanged.

DIRECTION OF THE ARROW IS THE POINT: more education CAUSES myopia, not the reverse. This is a cost of schooling, not a benefit. Observational regression gave -0.18 D per additional year of education (95% CI -0.19 to -0.17, P < 2e-16). Mendelian randomisation gave a larger point estimate: -0.27 D per year (95% CI -0.37 to -0.17, P = 4e-8) — though the Durbin-Wu-Hausman test for whether the IV estimate genuinely differs from the observational one is only P = 0.06, so "MR is stronger than observational" is a weakly evidenced claim, not an established one. In the reverse direction there was little evidence that myopia affects education (-0.008 years per dioptre, 95% CI -0.041 to 0.025, P = 0.6), and the study had 80% power to detect an effect >=0.048 y/D, so this is a powered null rather than an absence of data. The authors calculate that a UK university graduate with 17 years of education would on average be at least -1 D more myopic than someone who left school at 16 with 12 years — enough to blur a Snellen chart to 6/18 and require glasses for driving. The observational dose-response is non-linear: -0.25 D/year for those leaving between ages 15 and 18, slowing to -0.10 D/year for those leaving after 18.

Uses genetics as the identifying instrument rather than being confounded by it. The central hereditarian implication for this archive is inverted from the usual one: the observational correlation between education and myopia is not (mainly) heritable-trait confounding, it reflects a real causal effect of schooling on the eye - and the burden of that effect falls on the schools that produce the most academic success. Note the vertical-pleiotropy argument the authors rely on: if education-associated variants act on schooling THROUGH intelligence, the causal estimate is unbiased; it is only biased if intelligence causes myopia by a route that bypasses education entirely.

Not the first MR on this question. Cuellar-Partida et al. ran a bidirectional MR across three European-ancestry cohorts (combined n = 5,649) and reported -0.46 D per year of education (P = 1e-3) — same sign, larger magnitude, and independent of this team. Mountjoy et al. criticise it as under-powered and pleiotropy-prone (it used a 17,749-SNP polygenic score as the instrument) but do not dispute its direction. So the education -> myopia arrow is independently replicated; the point estimate is not pinned down (-0.23 to -0.46 D/year across studies and specifications).