Machine Learning to Predict Outcomes and Dosing Frequency With Aflibercept for Macular Edema Following Central...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Macular edema following central retinal vein occlusion (MEfCRVO) requires ongoing management.
• Intravitreal aflibercept injections (IAI) are a common treatment.
• Predicting treatment response and frequency is crucial for patient care.

Purpose of the Study:

• To develop machine learning (ML) algorithms for predicting outcomes in MEfCRVO patients.
• To forecast visual acuity, anatomic changes, and treatment frequency.
• To enhance clinical decision-making and patient expectation management.

Main Methods:

• Utilized data from COPERNICUS and GALILEO trials.
• Trained ML algorithms on patient data receiving IAI 2 mg monthly, then pro re nata (PRN).
• Validated predictive accuracy for visual acuity, retinal thickness, and PRN injection frequency.

Main Results:

• ML algorithms showed strong correlation in predicting best-corrected visual acuity (BCVA) (r=0.87) and BCVA change (r=0.76).
• Accurate prediction of ≥ 15-letter gain (AUC=0.81) and central subfield thickness change (r=0.76).
• High accuracy achieved in predicting PRN injection frequency (AUC=0.83).

Conclusions:

• ML algorithms can reliably predict visual and anatomic outcomes in MEfCRVO patients.
• ML accurately forecasts treatment frequency, aiding in personalized care plans.
• These predictive tools can inform patient and clinician expectations for MEfCRVO management.