Optimizing Predictive Models for Distance and Near Vision Outcomes in Presbyopic Surgery Using Machine Learning

Area of Science:

• Ophthalmology
• Data Science
• Biomedical Engineering

Background:

• Presbyopic corneal refractive surgery aims to restore both near and distance vision.
• Predicting visual outcomes remains challenging due to inherent trade-offs.
• Personalized surgical planning is crucial for optimizing patient satisfaction.

Purpose of the Study:

• Develop predictive models for postoperative visual acuity (near, distance, monocular, binocular).
• Create a composite outcome metric, distance-corrected visual gain (DCVG).
• Identify key factors influencing outcomes in presbyopic corneal refractive surgery.

Main Methods:

• Retrospective analysis of 914 eyes from 457 patients treated with PresbyMAX.
• Application of machine learning algorithms (Random Forest) for predicting visual acuity.
• Evaluation of feature importance using SHAP and PDPs.

Main Results:

• Binocular predictive models showed higher accuracy than monocular models.
• Corrected distance visual acuity (CDVA) was more predictable than distance-corrected near visual acuity (DCNVA).
• Optimal planned addition for balancing vision was +1.50 D to +2.25 D; patients aged 45-56 had better near vision potential.

Conclusions:

• Predictive modeling and DCVG metric enable personalized surgical planning.
• This approach aids in managing visual trade-offs in presbyopic surgery.
• Machine learning facilitates individualized decision-making for better patient counseling and optimized visual outcomes.