Retinopathy prediction in type 2 diabetes: Time-varying Cox proportional hazards and machine learning models