A new polygenic score for refractive error improves detection of children at risk of high myopia but not the prediction of those at risk of myopic macular degeneration

Abstract

Background: High myopia (HM), defined as a spherical equivalent refractive error (SER) ≤ -6.00 diopters (D), is a leading cause of sight impairment, through myopic macular degeneration (MMD). We aimed to derive an improved polygenic score (PGS) for predicting children at risk of HM and to test if a PGS is predictive of MMD after accounting for SER.

Methods: The PGS was derived from genome-wide association studies in participants of UK Biobank, CREAM Consortium, and Genetic Epidemiology Research on Adult Health and Aging. MMD severity was quantified by a deep learning algorithm. Prediction of HM was quantified as the area under the receiver operating curve (AUROC). Prediction of severe MMD was assessed by logistic regression.

Findings: In independent samples of European, African, South Asian and East Asian ancestry, the PGS explained 19% (95% confidence interval 17-21%), 2% (1-3%), 8% (7-10%) and 6% (3-9%) of the variation in SER, respectively. The AUROC for HM in these samples was 0.78 (0.75-0.81), 0.58 (0.53-0.64), 0.71 (0.69-0.74) and 0.67 (0.62-0.72), respectively. The PGS was not associated with the risk of MMD after accounting for SER: OR = 1.07 (0.92-1.24).

Interpretation: Performance of the PGS approached the level required for clinical utility in Europeans but not in other ancestries. A PGS for refractive error was not predictive of MMD risk once SER was accounted for.

Funding: Supported by the Welsh Government and Fight for Sight (24WG201).

Keywords: ALSPAC; Generation R; Myopia; Polygenic score; UK Biobank.

Fig. 1

Accuracy of the PGS over a grid of LDpred2 parameter settings in the “tuning” sample of4000 European-ancestry participants. The model parameters varied were (a) model sparsity (full model vs. sparse model), (b) heritability (0.1, 0.2, 0.3 or 0.4). The p-value threshold from the GWAS meta-analysis summary statistics were varied across the range 1 × 10⁻⁵ to 1.0.

Fig. 2

Refractive error distribution in the “test” sample of 6000 European-ancestry particpants categorized as having a PGS above vs. below a threshold level. (a, c, e) Odds ratio for predicting high myopia of −6.00D or worse. (b, d, f) Odds ratio for predicting moderate hyperopia of +3.00 D or worse (MH). The threshold levels examined were the top and bottom 5% (a, b), 10% (c, d) and 25% (e, f).

Fig. 3

Refractive error distribution and absolute risk of high myopia in deciles of the PGS, for the “test” sample of 6000 individuals of European ancestry. (a) Distribution of refractive error by PGS decile. The white box corresponds to the interquartile range, the solid line inside the white box is the median. (b, c) Absolute risk of high myopia of at least −5.00 D (HM5; b) or of at least −6.00 D (HM6; c). Points correspond to the proportion of individuals in each decile affected by high myopia; error bars are 95% confidence intervals. Counts of affected individuals are shown above each point. The dashed horizontal line is the prevalence of high myopia in the full sample.

Fig. 4

Accuracy of the new and existing PGSs in independent samples of children and adults from the ALSPAC cohort. Three different PGSs were compared: “PGS 2018” is a PGS created from the 149 SNPs most significantly associated with avSER in UK Biobank; “PGS 2020” is a PGS created from 1.1 million SNPs associated with avSER or AOSW-inferred avSER in UK Biobank, which was previously the best-performing PGS for refractive error; “PGS 2022” is the PGS reported in the current study. (a) Prediction accuracy of PGSs in young persons from the ALSPAC cohort whose refractive error was assessed longitudinally and in a sample of their mothers (adults). Error bars are 95% confidence intervals. (b) Refractive error trajectory of ALSPAC young persons, stratified by percentile of the PGS. (c) ROC curves for detecting high myopia (HM5 or HM6) in the adults from the ALSPAC cohort. The dashed black line represents chance-level prediction accuracy.

Fig. 5

Pathway diagrams (directed acyclic graphs) illustrating potential causal relationships between the PGS, refractive error, and myopic maculopathy. (a) Refractive error is a major risk factor for MMD and it is highly plausible this relationship is causal. (b) A PGS for refractive error has a causal relationship with refractive error, therefore by logic, if refractive error is a cause of MMD then a PGS for refractive error must also be a cause of MMD. (c) While refractive error may be a mediator of the relationship between the PGS and MMD (solid arrows), it is possible that the PGS may confer an additional risk of MMD independently of the degree of refractive error (dashed arrow).