Prediction of complications of diabetes mellitus utilising novel retinal image analysis, genetics, and linked electronic health records data (360G-Wellcome-224390_Z_21_Z)

Early targeted intervention reduces risk of diabetic complications. We will compare the effectiveness of artificial intelligence (AI) based retinal image analysis, genotyping and linked health record data from people with diabetes mellitus to predict diabetes related complications - incident diabetic eye disease (particularly progression and sight-threatening retinopathy) and systemic (nephropathy, neuropathy, cardiovascular disease). We will apply novel end-to-end AI/machine learning approaches to retinal feature analysis and validated automated AI-enabled retinal vasculometry measurement to one of the largest and most ethnically diverse NHS Diabetic Eye Screening Programmes from North-East London (105,000 patients seen annually, 90,000 with 5 years follow-up, 40,000 with 10 years). We will develop new prognostic models, using retinal image analysis, genetics and health record data (alone and in combination) to predict diabetic complications (particularly retinopathy progression); model performance will be compared with published risk scores/algorithms. Models will be externally validated using the US Early Treatment Diabetic Retinopathy Study, cohort with gold standard retinopathy grading and high complication rates (20% mortality at 5 years). Findings will provide an early warning system of complications, which could be exploited within existing healthcare pathways to trigger early intervention.