Machine Learning Models for Prediction of Diabetic Microvascular Complications

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Diabetology

Background:

• Diabetic microvascular complications, including diabetic retinopathy (DR), diabetic kidney disease (DKD), and diabetic neuropathy (DN), are major contributors to morbidity and mortality.
• Machine learning and artificial intelligence (ML/AI) offer potential for early prediction and management of these complications.

Purpose of the Study:

• To review the current landscape of ML/AI applications in predicting DR, DKD, and DN.
• To analyze the methodologies, performance, and trends in ML/AI models for diabetic microvascular complication prediction.

Main Methods:

• A comprehensive literature search of PubMed (1990-2023) was conducted to identify studies utilizing ML/AI for predicting DR, DKD, and DN.
• Key aspects analyzed included study design, patient cohorts, predictive features, ML techniques employed, prediction duration, and performance metrics (e.g., c-statistic).

Main Results:

• The review identified a growing trend in ML/AI research for microvascular complications since 2010, primarily driven by DKD studies.
• ML models demonstrated moderate predictive performance, with mean c-statistics of 0.79 (internal validation) and 0.72 (external validation). DKD models showed the highest discrimination.
• External validation was less common, particularly for DN prediction, and model performance was influenced by various factors including prediction horizon and predictor selection.

Conclusions:

• There is increasing global interest and research in applying ML/AI to predict diabetic microvascular complications.
• Diabetic kidney disease research is the most advanced, while diabetic retinopathy and neuropathy require further investigation.
• Robust external validation and adherence to established guidelines are essential for developing reliable ML/AI predictive models.