High myopia constitutes a significant global public health concern. Although visual behaviors are well-established determinants, the relationships between systemic environmental exposures, nutritional factors, and high myopia remain incompletely characterized at the population level. We conducted a cross-sectional analysis of 3,283 adults from the National Health and Nutrition Examination Survey (NHANES) 1999–2008. Participants were categorized as healthy controls, non-high myopia, or high myopia based on non-cycloplegic autorefraction. Weighted multivariable logistic regression and restricted cubic spline models, as well as stratified and quantile analyses, were employed to assess exposure–outcome associations. Bayesian Kernel Machine Regression (BKMR) was applied to examine joint exposure patterns among correlated metals and dietary factors. Additionally, machine learning approaches (LASSO, Boruta, Elastic Net, and XGBoost) were utilized for exploratory variable selection and risk stratification. Elevated blood concentrations of cadmium and mercury were associated with increased odds of high myopia, whereas several dietary and biochemical indicators demonstrated inverse associations under specific analytical frameworks. Carbohydrate intake was positively associated with high myopia. Nonlinear relationships and joint exposure patterns were identified across models. However, these findings exhibited substantial uncertainty and variability between methods. Machine learning analyses consistently identified cadmium exposure and carbohydrate intake as influential predictors, although variable importance rankings differed across algorithms and were interpreted as exploratory. In this nationally representative adult population, environmental metal exposures and nutritional factors were cross-sectionally associated with high myopia status. Given the cross-sectional design, the limited number of high myopia cases, potential measurement constraints, and residual confounding from unmeasured behavioral factors, these findings should be considered hypothesis-generating rather than indicative of causal or preventive relationships. Further longitudinal studies incorporating life-course exposure assessment and comprehensive behavioral data are warranted to clarify temporal dynamics and underlying biological mechanisms. Not applicable.
Hui et al. (Fri,) studied this question.