A deep learning system, DeepDR Plus, was developed to predict the time to progression of diabetic retinopathy (DR) using retinal fundus images. The system was trained on 717,308 fundus images from 179,327 participants with diabetes and validated on 118,868 images from 29,868 participants. It achieved concordance indexes of 0.754–0.846 and integrated Brier scores of 0.153–0.241 for predicting DR progression within 5 years. The system was validated in real-world cohorts and showed potential to extend the mean screening interval from 12 months to 31.97 months, with a low rate of delayed detection of progression to vision-threatening DR. The system could predict individualized risk and time to DR progression, enabling personalized screening intervals. The study highlights the importance of individualized risk models for DR management and the potential of AI in improving screening efficiency and outcomes. The DeepDR Plus system demonstrated high accuracy in predicting DR progression, with performance comparable across different ethnic and clinical datasets. The system's ability to integrate into clinical workflows could significantly enhance DR screening programs by allowing more targeted and efficient interventions. The study also emphasizes the need for further research to address gaps in existing DL systems for DR prediction, including the evaluation of their impact on patient outcomes when integrated into clinical practice. The results suggest that AI-based systems can provide valuable insights for personalized DR management, potentially reducing healthcare costs and improving patient outcomes.A deep learning system, DeepDR Plus, was developed to predict the time to progression of diabetic retinopathy (DR) using retinal fundus images. The system was trained on 717,308 fundus images from 179,327 participants with diabetes and validated on 118,868 images from 29,868 participants. It achieved concordance indexes of 0.754–0.846 and integrated Brier scores of 0.153–0.241 for predicting DR progression within 5 years. The system was validated in real-world cohorts and showed potential to extend the mean screening interval from 12 months to 31.97 months, with a low rate of delayed detection of progression to vision-threatening DR. The system could predict individualized risk and time to DR progression, enabling personalized screening intervals. The study highlights the importance of individualized risk models for DR management and the potential of AI in improving screening efficiency and outcomes. The DeepDR Plus system demonstrated high accuracy in predicting DR progression, with performance comparable across different ethnic and clinical datasets. The system's ability to integrate into clinical workflows could significantly enhance DR screening programs by allowing more targeted and efficient interventions. The study also emphasizes the need for further research to address gaps in existing DL systems for DR prediction, including the evaluation of their impact on patient outcomes when integrated into clinical practice. The results suggest that AI-based systems can provide valuable insights for personalized DR management, potentially reducing healthcare costs and improving patient outcomes.