Artificial intelligence approaches for risk stratification of diabetic kidney disease

Diabetic kidney disease is a major microvascular complication of type 1 and 2 diabetes mellitus that affect the mortality of patients worldwide. Current risk stratification methods are predominated by disease-specific scoring systems. However, these models are susceptible to several limitations, including the inclusion of a limited number of risk variables, variation in results during external validation, and the dichotomized nature of scoring criteria that do not fully represent the dynamics of biological risk. Artificial intelligence (AI) can extract, stratify, and derive interrelationships between variables within complex datasets. The main approaches of AI technology in medicine are deep learning and machine learning. Recent studies have explored AI-based prediction models for the risk pattern predictions of diabetic nephropathy. With further development, this may assist physicians in clinical decision-making and improve the survival rate of diabetic patients. Moreover, early prediction of diabetic renal disease can reduce dialysis dependency, encourage lifestyle change, and reduce likelihood of end-stage renal failure. However, the implementation of AI is restricted due to the absence of an established framework for clinical use and general unfamiliarity with the technology. Therefore, this chapter aims to conduct a comprehensive review of the existing evidence of AI-based predictive models on the risk stratification of diabetic nephropathy. The current literature suggests that AI models can improve the detection of disease prognosis and identify significant predictors. Nevertheless, it is prudent to recognize that further investigation is necessary to compare the performance of AI models and conventional risk stratification methods. Furthermore, a rigorous regulatory framework must be enforced before AI is used to support clinical practice.