Machine Learning-Driven Investigation of Associations Between Phthalate Biomarkers and Glaucoma Using US NHANES Data (2011–2016)

Background As the leading cause of irreversible blindness globally, glaucoma involves a multifactorial etiology encompassing dysregulated intraocular pressure, optic neuropathy, and interactions between genetic predisposition and environmental determinants. Phthalates, ubiquitous endocrine-disrupting chemicals in plastics and personal care formulations, may adversely impact the central nervous and cardiovascular systems through hormonal interference, oxidative stress induction, and inflammatory pathway activation. Given their potential to target ocular structures including retinal neurofibers, microvasculature, and aqueous humor outflow pathways, research exploring the phthalate-glaucoma relationship remains nascent. Comprehensive analytical approaches are imperative for elucidating pathogenic mechanisms and characterizing associated risk profiles. Methods To evaluate associations between urinary phthalate metabolite concentrations and glaucoma susceptibility, this study integrated data from large prospective cohort studies or national health databases (2011–2016), incorporating phthalate biomarker measurements, glaucoma diagnostic status, and covariates (e.g., age, medical history). Eleven distinct machine learning algorithms were implemented for model development, with optimization conducted via stratified cross-validation. The optimal predictive model was selected guided by performance criteria such as the receiver operating characteristic curve's area under the curve (AUC). We implemented permutation feature importance analysis, evaluated accumulated local effects (ALE), and interpreted SHAP (SHapley Additive exPlanations) values to elucidate influential variables and their interaction patterns. Sensitivity analyses established the robustness of outcomes across subgroups. Conclusion This research applied explainable AI (XAI) frameworks for examining the relationship linking phthalate biomarkers to glaucoma risk. Among 11 evaluated models, the Gradient Boosting Machine (GBM) algorithm demonstrated superior predictive capability. Age constituted the most influential risk determinant. Several phthalate metabolites—specifically MEOP, MCNP, MEHP, and MECPP—were identified as significant contributors to glaucoma risk stratification. The results emphasize the vital role of integrating environmental exposure biomarkers in glaucoma prognostic models and highlight the necessity for mechanistic investigations into underlying biological pathways.