Developing a Predictive Artificial Intelligence Model for Early Detection of Diabetic Nephropathy in Nigerian Type 2 Diabetes Patients

Diabetic nephropathy is the leading cause of end-stage renal disease in Nigeria, and early detection through predictive modelling could transform clinical management by identifying at-risk patients before overt proteinuria develops. Existing predictive models were developed on European or East Asian cohorts and have not been validated in Nigerian populations with a high co-burden of hypertension, sickle cell trait, and tropical infections that independently affect renal function. This study develops and validates an artificial intelligence predictive model for early diabetic nephropathy in Nigerian type 2 diabetes patients. A prospective cohort of 2,000 type 2 diabetic patients without baseline proteinuria is recruited from 10 diabetes clinics across all six geopolitical zones and followed for five years with annual renal function assessment. Incident diabetic nephropathy is defined as development of persistent microalbuminuria or a 40 percent decline in eGFR. Candidate predictor variables include glycated haemoglobin trajectory, systolic blood pressure, uric acid, serum cystatin C, urinary KIM-1, urinary NGAL, HbS carrier status, and anti-VEGF biomarkers. Random forest, XGBoost, and deep neural network models are trained on an 80:20 split with five-fold cross-validation. Shapley Additive Explanations provide model interpretability. The Diabetic Nephropathy Progression Model and Machine Learning Epidemiology Theory provide the conceptual basis. Original contributions include the first AI nephropathy prediction model validated in a Nigerian cohort. Findings will support clinical decision support tool deployment in Nigerian diabetes clinics. Keywords: diabetic nephropathy, artificial intelligence, predictive model, type 2 diabetes, Nigeria.