Development and validation of a nomogram model for predicting cardiac autonomic neuropathy in patients with diabetes

Diabetic Cardiac Autonomic Neuropathy (DCAN), a critical yet frequently underdiagnosed microvascular complication, is associated with increased mortality. Standard cardiovascular autonomic reflex tests (CARTs) are complex and time-consuming, hindering their widespread use in routine screening in clinical settings. This study aimed to develop and validate a predictive nomogram for DCAN in patients with diabetes using readily available clinical variables.

We retrospectively analyzed the clinical data of 453 patients with type 1 or type 2 diabetes hospitalized at Shenzhen People's Hospital between February 2022 and December 2025. The dataset was randomly divided into training (70%) and validation (30%) cohorts. Key predictors were identified using a rigorous selection strategy that combined univariate analysis, least absolute shrinkage and selection operator (LASSO) regression, and multivariate logistic regression. Four candidate prediction models (Logistic Regression (LR), Random Forest, Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) were constructed and evaluated for discrimination, calibration, and clinical utility. The optimal model was visualized as a nomogram and interactive web calculator.

The prevalence of DCAN in the study population was 45.0% (204/453). The following seven independent predictors were identified: a history of diabetic retinopathy (DR) or diabetic kidney disease (DKD), diabetes duration, age, heart rate (HR), fasting plasma glucose (FPG), and HbA1c. Among the algorithms tested, the LR model exhibited the most balanced performance in the validation cohort (area under the curve (AUC) = 0.838) with the highest sensitivity (77.0%) and was thus selected as the optimal prediction tool. Consequently, the LR model was transformed into a predictive nomogram. This nomogram demonstrated good calibration and potential clinical utility for individualized risk assessment.

We successfully developed and validated a high-sensitivity prediction model for DCAN applicable to type 1 and type 2 diabetes. The developed visual nomogram and interactive web-based tool are cost-effective and user-friendly instruments that can facilitate early risk assessment and personalized clinical management.