Characterizing and Predicting Limited Early Response in Eyes With Retinal Vein Occlusion Using Machine Learning in Routine Clinical Practice.

Objective: To characterize and predict limited early responders (LER) in eyes with retinal vein occlusion (RVO) treated with anti-vascular endothelial growth factor (anti-VEGF).

Patients And Methods: This retrospective cohort study of 116 eyes with edema secondary to RVO characterized early responders (ER) and LER 3 months after initiating anti-VEGF treatment. Baseline characteristics and 12-month outcomes were compared. A machine learning (ML) algorithm was developed to predict LER.

Results: At baseline, LER had higher best-corrected visual acuity (BCVA) than ER (< 0.0001) and lower central subfield thickness (CST) than ER (< 0.01). At 12 months, change in BCVA was + 0.8 and + 21.4 letters, and change in CST was -104.4 and -187.1 μm for LER and ER, respectively ( < 0.0001, < 0.05). The ML algorithm achieved area under the receiver operating characteristic curve = 0.73.

Conclusion: ER eyes experienced greater functional and anatomical improvements at 12 months in routine clinical practice than LER. The ML algorithm achieved moderately high performance predicting LER. .