@Article{chd.2025.072431,
AUTHOR = {Chih-Yin Chien, Tsae-Jyy Wang, Pei-Hung Liao, Ying-Hsiang Lee, Wei-Sho Ho},
TITLE = {Artificial Neural Network-Based Risk Assessment for Cardiac Implantable Electronic Device Complications},
JOURNAL = {Structural and Congenital Heart Disease},
VOLUME = {20},
YEAR = {2025},
NUMBER = {5},
PAGES = {601--612},
URL = {http://www.techscience.com/schd/v20n5/64670},
ISSN = {3071-1738},
ABSTRACT = {<b>Background:</b> Cardiac implantable electronic devices (CIEDs) are essential for preventing sudden cardiac death in patients with cardiovascular diseases, but implantation procedures carry risks of complications such as infection, hematoma, and bleeding, with incidence rates of 3–4%. Previous studies have examined individual risk factors separately, but integrated predictive models are lacking. We compared the predictive performance and interpretability of artificial neural network (ANN) and logistic regression models to evaluate their respective strengths in clinical risk assessment. <b>Methods:</b> This retrospective study analyzed data from 180 patients who underwent cardiac implantable electronic device (CIED) implantation in Taiwan between 2017 and 2018. To address class imbalance and enhance model training, the dataset was augmented to 540 records using the Synthetic Minority Oversampling Technique (SMOTE). A total of 13 clinical risk factors were evaluated (e.g., age, body mass index (BMI), platelet count, left ventricular ejection fraction (LVEF), prothrombin time/international normalized ratio (PT/INR), hemoglobin (Hb), comorbidities, and antithrombotic use). <b>Results:</b> The most influential risk factors identified by the ANN model were platelet count, PT/INR, LVEF, Hb, and age. In the logistic regression analysis, reduced LVEF, lower hemoglobin levels, prolonged PT/INR, and lower BMI were significantly associated with an increased risk of complications. ANN model achieved a higher area under the curve (AUC = 0.952) compared to the logistic regression model (AUC = 0.802), indicating superior predictive performance. Additionally, the overall model quality was also higher for the ANN model (0.93) than for logistic regression (0.76). <b>Conclusions:</b> This study demonstrates that ANN models can effectively predict complications associated CIED procedures and identify critical preoperative risk factors. These findings support the use of ANN-based models for individualized risk stratification, enhancing procedural safety, improving patient outcomes, and potentially reducing healthcare costs associated with postoperative complications.},
DOI = {10.32604/chd.2025.072431}
}