Fig.6.

Patient A: true positive case study. a Input feature values for patient A. b VA changes throughout the first year of treatment, as measured at each follow-up appointment. NB: Only baseline information used to train model. c Local feature importance showing how each feature affected the AutoML model’s inference score relative to the baseline score of 0.48. d Individual conditional expectation (ICE) plot showing how the model’s inference score changes as SRF volume is hypothetically varied and other features are kept as shown in Fig. 6a. The indicated point represents patient A’s actual SRF volume at baseline, whilst the horizontal black line represents the default classification threshold of 0.5