Beyond the Glycemic Curve: Machine Learning Models for Predicting Early Microvascular Complications in Type 2 Diabetes

Matteo Romano¹, Giulia Ferrara², Luca Moretti³

Abstract

Background:
Type 2 diabetes mellitus (T2DM) remains a leading cause of microvascular complications such as retinopathy, nephropathy, and neuropathy. Early prediction of these complications could enable timely intervention and reduce long-term disability.

Objective:
To develop and validate machine learning (ML) models to predict the onset of microvascular complications within five years in patients with T2DM.

Methods:
A retrospective cohort study was conducted using electronic health records from three tertiary hospitals in northern Italy between January 2016 and December 2023. A total of 4,862 adults with newly diagnosed T2DM were included. Variables analyzed included demographics, laboratory values, comorbidities, and medication history. Random Forest, XGBoost, and Logistic Regression models were trained using an 80/20 split for training/validation. Primary outcome was the onset of at least one microvascular complication within five years.

Results:
The XGBoost model outperformed other approaches with an AUC of 0.91 (95% CI: 0.89–0.93), sensitivity of 86.4%, and specificity of 84.1%. The top predictors included baseline HbA1c >8.0%, eGFR decline >5 mL/min/year, LDL >120 mg/dL, and systolic BP >140 mmHg. Patients flagged as high-risk by the model were 3.2 times more likely to develop complications (p < 0.001).

Conclusion:
Machine learning models, particularly XGBoost, can accurately predict early microvascular complications in T2DM. Incorporating such tools into routine diabetes management could enable personalized care pathways and reduce the burden of preventable disability.