대한신장학회

논문분류	춘계학술대회 초록집
제목	Prediction of Peritoneal Equilibration Test Results and Nutritional Status Using ShareSource Data and Machine Learning in Automated Peritoneal Dialysis Patients
저자	Hyung Woo Kim
출판정보	2026; 2026(1):
키워드	Peritoneal equilibration test, Machine learning, Remote patient monitoring, ShareSource, automated peritoneal dialysis
초록	Objectives: Remote patient monitoring through ShareSource™ enables real-time collection of automated peritoneal dialysis (APD) session data. We aimed to develop machine learning-based prediction models for peritoneal equilibration test (PET) results and nutritional status using ShareSource™ data combined with clinical parameters in APD patients. Methods: This multicenter study included APD patients from Severance Hospital and Korea University Guro Hospital. ShareSource™ data including therapy time, fill volume, ultrafiltration, pre-dialysis weight, and blood pressure were collected along with clinical data. PET results were classified into four categories (high, high-average, low-average, and low). Nutritional status was evaluated using serum albumin, total cholesterol, and C-reactive protein (CRP) levels. Prediction models were developed using XGBoost with external validation (Severance as training set, Korea University as test set) and 5-fold cross-validation on the combined dataset. Feature importance was assessed for each model. Results: A total of 48 APD patients (26 from Severance, 22 from Korea University) with 5,236 APD sessions were analyzed for PET prediction. The mean age was 53.6 ± 12.1 years, and 40.5% were female. For PET prediction, the training set achieved an overall AUC of 0.867, while the external validation AUC was 0.649, with the highest performance for the high category (AUC 0.922). The top predictive features were hemoglobin, serum creatinine, serum albumin, weight, and diastolic blood pressure. For nutritional status prediction, the external validation AUCs were 0.853 for serum albumin, 0.576 for total cholesterol, and 0.672 for CRP. In 5-fold cross-validation, the AUCs improved to 0.887, 0.614, and 0.744, respectively. Conclusion: Machine learning models utilizing ShareSource™ and clinical data demonstrated promising predictive performance for PET results and serum albumin levels in APD patients. These findings suggest the potential of remote monitoring data for early prediction of clinically relevant outcomes, warranting further validation with larger cohorts and standardized data collection.
원문(PDF)	PDF 원문보기

간행물 검색

대한신장학회 회원관리 규정