Automated machine learning models show promise for diabetic retinopathy screening, but require careful selection and external validation.