OBJECTIVES: The objective of this study was to predict ‘non-good’ response for eyes with neovascular age-related macular degeneration (nAMD) treated with anti-vascular endothelial growth factor (aVEGF) using ophthalmology electronic medical records (EMR). METHODS: This was a retrospective cohort study with data from ~40 US ophthalmology clinics. ‘Non-good’ response was defined as a best corrected visual acuity (BCVA) gain ≤5 letters. Separate models were estimated for the initiation phase (from aVEGF initiation to the visit after the third injection) and the maintenance phase (from the visit after the third injection to twelve months after aVEGF initiation). Predictors captured clinical and demographic attributes. The most recent ~20% of eyes were held back for testing, with the preceding eyes used for training. Random Forests and Logistic Regressions were estimated on the training set and predictions computed for eyes in the test set. The key performance metric was the Area Under the Curve (AUC) based on the test set. Predictor importance was measured as the reduction in AUC associated with randomly permuting values. RESULTS: The data contained 7,041 eyes for the initiation phase and 6,446 eyes for the maintenance phase. Random Forests outperformed Logistic Regressions. The AUCs for the Random Forests were 0.72 (95%CI 0.69-0.76) and 0.78 (95%CI 0.74-0.81) for the initiation and maintenance phases, respectively. For the Logistic Regressions, the AUCs were 0.70 (95% CI 0.67-0.73] and 0.75 (95%CI 0.71-0.78). The most important predictors for the initiation phase were baseline BCVA and age. For the maintenance phase, the most important predictors were related to BCVA at the start of or prior to the maintenance phase. CONCLUSIONS: This study demonstrated that ophthalmology EMR data can provide accurate predictions of clinical outcomes for eyes with nAMD. This underscores the feasibility of predictive tools to support case management, treatment decisions or physician-patient engagement at the point-of-care.