CARTs remained essential for accurate CAN classification and staging; routinely available clinical variables provided complementary phenotypic information but were insufficient to robustly distinguish CAN categories without CART data, particularly CAN0 and CAN1.
Almost all patients with T2DM in the registry had at least one borderline or abnormal cardiovascular autonomic reflex test result.
Greater CAN severity was associated with older age, longer diabetes duration, higher systolic blood pressure, and poorer renal function.
The CART-inclusive XGBoost model achieved perfect discrimination of CAN categories in its best iteration, with all five CARTs among the leading predictors.
Excluding CART variables sharply reduced model performance in patients with T2DM only.
Adding non-diabetic controls without abnormal CART results to the CART-excluded model improved performance, most for CAN0.
Age, diabetes duration, and renal function markers were recurrent non-CART predictors, while other feature groups contributed in category- and model-specific ways.
UMAP showed substantial overlap among CAN categories, particularly for CAN1.
ML may support multidimensional phenotyping and hypothesis generation but cannot replace standardized autonomic testing.
This study used machine learning on data from 693 people with type 2 diabetes who had standardized heart-related autonomic nerve tests, to see which routine clinical features predict cardiac autonomic neuropathy (CAN), a nerve complication that raises cardiovascular risk. Nearly everyone (97.3%) had at least one borderline or abnormal test result. More severe CAN was linked to older age, longer diabetes duration, higher systolic blood pressure, and worse kidney function. When the autonomic test results themselves were given to the model, it sorted patients into categories almost perfectly, but this is expected since those tests define the categories. When the tests were removed and only routine clinical information (age, lab values, medications and so on) was used, accuracy fell substantially (weighted AUC 0.64). Adding non-diabetic control participants improved performance (0.78), mainly for identifying people without CAN. Early CAN was hardest to identify in every analysis, and visual mapping showed the categories overlapped heavily. This research suggests that routine clinical data carry some information about CAN, but not enough to replace formal autonomic testing, especially for spotting early disease. Machine learning may help researchers explore patterns and generate hypotheses. The study came from a single center, so the findings may not apply to all populations.
Farkas K, Vedelek V, Vámos M, Csajbók &, Baczkó I, Zádori J, et al.. (2026). Machine learning predictors of cardiovascular autonomic neuropathy in type 2 diabetes: insights from a prospectively collected single-center registry.. Frontiers in endocrinology. https://doi.org/10.3389/fendo.2026.1965013