Deep Learning-Based Death Prediction Model for Chronic Kidney Disease

Area of Science:

• Nephrology
• Artificial Intelligence in Healthcare
• Public Health

Background:

• Chronic kidney disease (CKD) prevalence is increasing globally, contributing significantly to mortality.
• Traditional statistical methods are being augmented by AI for predicting mortality factors in CKD.
• Further AI research is crucial for understanding CKD patient death determinants.

Purpose of the Study:

• To identify key factors associated with mortality in CKD patients using machine learning.
• To develop a deep learning-based predictive model for assessing death risk in CKD.

Main Methods:

• Utilized data from the Korea Disease Control and Prevention Agency (2016-2021).
• Applied Least Absolute Shrinkage and Selection Operator (LASSO) regression to identify significant mortality predictors.
• Developed a deep learning model incorporating LASSO-selected variables.

Main Results:

• Identified eight factors influencing death, including length of hospital stay, emergency admission, age, severity-adjusted score, and regional differences.
• The deep learning model achieved high accuracy (96.84%) with a low loss value (0.1207).

Conclusions:

• Emergency visits and prolonged hospital stays are significant predictors of mortality in CKD.
• Regular nephrology monitoring and timely renal replacement therapy are crucial for risk mitigation.
• An AI-based predictive model using general characteristics can aid rapid prognosis assessment.