Deep Learning-Based Death Prediction Model for Chronic Kidney Disease

Abstract

Objectives: The prevalence of chronic kidney disease (CKD) continues to rise, making it one of the leading causes of deathworldwide. Recent advances in medical and health research have progressed beyond traditional statistical methodologies,increasingly leveraging artificial intelligence to identify and predict factors influencing mortality. Further AI-based researchis therefore essential to deepen understanding of the determinants of death among CKD patients. Methods: This studyused data from the Korea Disease Control and Prevention Agency’s in-depth survey of patients discharged between 2016and 2021. Least absolute shrinkage and selection operator (LASSO) regression, a machine learning technique, was appliedto identify significant factors associated with death in CKD patients. These selected variables were then incorporated into adeep learning-based predictive model. Results: Eight factors influencing death were identified, including length of hospitalstay (coefficient = 0.023), emergency admission (0.016), age (0.013), severity-adjusted score (0.008), and regional differences(0.003). The developed deep learning model achieved a loss value of 0.1207 and an accuracy of 96.84%. Conclusions: Thisstudy identified emergency visits and prolonged hospital stays as key predictors of death in CKD patients. To mitigate theserisks, regular monitoring by nephrology specialists and timely initiation of renal replacement therapy are essential. Age alsoemerged as a critical determinant, emphasizing the importance of age-stratified clinical guidelines amid global aging trends. The high-performing, simplified predictive model based on general characteristics offers a valuable tool for rapid prognosisassessment in primary and secondary healthcare settings.