Abstract
This first systematic real-world evaluation of VO-related adverse drug reactions identifies high-risk drugs, susceptible populations, and onset patterns, thus offering guidance for preventive medication strategies. The BAG model showed higher sensitivity in real-world analysis, suggesting potential for further research in VO and floater prevention and treatment.
Introduction
Vitreous opacities (VO), commonly known as "floaters," are collagen clumps or other substances within the eye's vitreous humor that cast shadows on the retina. While often dismissed by clinicians as a benign consequence of aging, VO can significantly impair quality of life, sometimes exceeding the impact of conditions like glaucoma or diabetic retinopathy. Current treatments, such as vitrectomy or laser therapy, carry substantial risks, making prevention the most viable clinical strategy. However, the role of pharmaceuticals in inducing VO has remained largely anecdotal. This study bridges the gap between large-scale pharmacovigilance and clinical practice by identifying high-risk drugs and vulnerable patient populations.
Methods and Datasets
The researchers analyzed the FDA Adverse Event Reporting System (FAERS) database, spanning January 2004 to December 2024. The analytical pipeline involved:
- Data Mining: Disproportionality analysis using four algorithms (ROR, PRR, BCPNN, and MGPS) to identify robust signals.
- Statistical Filtering: Multivariable logistic regression to control for age, weight, and comorbidities.
- Machine Learning: Comparison of six algorithms, with the Bootstrapped Aggregating (BAG) model demonstrating superior predictive accuracy (AUC 0.824).
- Interpretability: Application of SHAP (SHapley Additive exPlanations) values to quantify the contribution of specific features to VO risk.
The final dataset focused on 3,817 subjects reporting VO-related adverse events from a cleaned pool of 18.6 million records after deduplication and standardization using DrugBank.
Key Results and Insights
The study identified 38 independent risk factors across six categories: ocular, oncologic, hormonal, antimicrobial, immunologic, and others.
High-Risk Medications
The top five high-risk drugs identified were dexamethasone, brolucizumab, triamcinolone, faricimab, and fingolimod. Notably, dexamethasone was the single most influential predictor in the machine learning model. While ocular injections (anti-VEGF) were prominent, systemic treatments like crizotinib (oncologic) and fingolimod (immunologic) also posed significant risks.
Demographic and Temporal Patterns
Risk peaks in the 61–71 age group, with a higher prevalence observed in female subjects, potentially due to estrogen withdrawal effects on ocular tissues. The onset of symptoms varies by drug class: antimicrobials (e.g., ciprofloxacin) show a rapid onset (~44 days), whereas hormonal medications (e.g., finasteride) take significantly longer (~409 days).
Discussion and Limitations
The BAG model's high performance (AUC 0.824) compared to standard logistic regression (AUC 0.698) highlights the value of advanced machine learning in pharmacovigilance. However, as a retrospective study of the FAERS database, limitations include potential reporting bias, lack of baseline vitreous state data, and the inability to definitively prove causality.
Future work should involve prospective clinical trials using standardized quantitative VO scoring to validate these findings and refine preventive strategies for patients requiring high-risk medications.
Summary updated