Machine-Learning Prediction of Major Adverse Cardiovascular Events in Type 2 Diabetes With and Without Diabetic Foot Syndrome Using Surrogate Cardiovascular Risk Indices and Circulating Biomarkers: An Exploratory, Internally Validated Study

Abstract Background and aims. Diabetic Foot Syndrome (DFS) is a serious complication of diabetes, increasing mortality and cardiovascular risk. Although artificial intelligence (AI) is increasingly used in medicine, few studies have applied Machine Learning (ML) models combining clinical variables with cardiovascular, inflammatory, genetic and epigenetic markers in DFS. This study aimed to develop and internally validate ML models predicting major adverse cardiovascular events (MACE) in patients with type 2 diabetes with and without DFS, with hospitalisation as a secondary outcome, and to identify the variables most associated with MACE. Methods. This retrospective analysis used the database of a single-centre observational cohort of 40 patients with type 2 diabetes and DFS and 40 age- and sex-matched diabetic controls without foot ulcers, enrolled in 2014–2015 and followed for at least 8 years. We analyzed clinical, cardiovascular (PWV, RHI, LF/HF), adipo-inflammation (Omentin-1, Gremlin-1, VEGF, HIF-1α), epigenetic (miR-217-5p, miR-503-5p) and genetic (VEGF − 2578 C > A polymorphism) markers. Logistic Regression and Random Forest models were developed to predict MACE (non-fatal myocardial infarction, non-fatal ischaemic stroke and all-cause death) and internally validated with stratified 5-fold cross-validation; no external validation was performed. Results. DFS was associated with higher odds of MACE (OR 2.71, 95% CI 1.05–7.00; p = 0.039). ML models showed moderate predictive performance (Random Forest AUC: 0.78, 95% CI 0.68–0.88), achieving maximal sensitivity through decision threshold recalibration. Key predictors identified included antiplatelet therapy, LF/HF ratio, VEGF levels, PWV, and endothelial function. In exploratory univariate analyses, patients with MACE had higher PWV, LF/HF and Gremlin-1 and lower VEGF (p < 0.05). Conclusions. In patients with type 2 diabetes, DFS was associated with higher odds of MACE over at least 8 years. An exploratory model combining conventional risk factors, vascular and autonomic indices (PWV, LF/HF) and circulating biomarkers (VEGF, Gremlin-1) showed moderate discrimination at internal validation, with no advantage of Random Forest over Logistic Regression. These findings highlight the importance of comprehensive cardiovascular assessment and aggressive risk factor management in DFS patients, while supporting the potential role of ML in clinical decision-making. These hypothesis-generating findings require confirmation in larger cohorts with external validation.