Development and Validation of Machine Learning Models: Electronic Health Record Data To Predict Visual Acuity After...

Area of Science:

• Ophthalmology
• Medical Informatics
• Machine Learning

Background:

• Cataract surgery is a common procedure to restore vision.
• Predicting postoperative visual acuity is crucial for patient management and expectations.
• Electronic health records (EHR) offer a rich data source for predictive modeling.

Purpose of the Study:

• To develop and compare machine learning models for predicting final corrected distance visual acuity (CDVA) after cataract surgery.
• To identify key predictors of visual outcomes using EHR data.

Main Methods:

• A predictive modeling study utilizing decision tree, random forest, and gradient boosting algorithms.
• Inclusion of 64,768 patients undergoing cataract surgery from Kaiser Permanente Northern California (June 2010 - May 2015).
• Evaluation of model discrimination and calibration for predicting postoperative CDVA (20/50 or worse vs. 20/40 or better).

Main Results:

• The gradient boosting model demonstrated superior discrimination ability for predicting CDVA outcomes.
• Key predictors for poor visual outcomes (CDVA 20/50 or worse) included preoperative CDVA, age, and age-related macular degeneration, contributing significantly to model accuracy.
• Other influential factors included glaucoma medication, epiretinal membrane, corneal disorders, surgery duration, surgeon experience, and socioeconomic neighborhood characteristics.

Conclusions:

• Gradient boosting is the most effective machine learning algorithm for predicting CDVA post-cataract surgery.
• Machine learning models can enhance prognostic accuracy and inform patient decision-making regarding cataract surgery.