A preliminary investigation of diabetic nephropathy biomarkers through machine learning approaches

Diabetic nephropathy (DN) is a serious complication of type 2 diabetes (T2D), necessitating early risk assessment and monitoring to guide intervention strategies. This study explores an integrated machine learning approach to identify potential risk predictors of this complication. Using a clinical dataset of diabetic patients, we developed a model for predicting the risk of occurrence of diabetic nephropathy in patients with T2D from electronic medical records (EMR). The results show that the support vector machine (SVM) algorithm with Gaussian kernel can predict the risk of developing DN from 3 to 8 years in patients with T2D with an accuracy of about 75% and that a patient with high levels of body mass index (BMI), white blood cells (WBC) count, creatininemia, triglyceride-glucose index (TyG), glycated hemoglobin (HbA1c), potential advanced diabetic condition with comorbidities (such as hypertension and cardiac diseases), low levels of hemoglobin, estimated glomerular filtration rate (eGFR), indirect bilirubin at baseline, with longer duration of diabetes and with past or current smoking habits represent potential risk factors in the onset of this pathology at a distance of 3 to 8 years from the visit. This work provides the basis for studying the dataset under investigation and then enriching it with features produced by diabetes-specific mathematical models (MMs) to obtain a hybrid model (AI/MM) that can improve the predictive model.