Related Experiment Video
Updated: May 10, 2025
A Retrograde Implantation Approach for Peritoneal Dialysis Catheter Placement in Mice
Published on: July 20, 2022
Longitudinal Changes in Peritoneal Transport and Their Impact on Dialysis Outcomes: A Machine Learning Approach
Chia-Chun Lee1,2, Jo-Yen Chao1, Kuan-Hung Liu1,2
1Division of Nephrology, Department of Internal Medicine, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan.
Introduction:
Longitudinal changes in peritoneal transport patterns and the predictive role of dialysate biomarkers remain poorly understood. This study assessed the impact of peritoneal equilibration test (PET) trajectory changes on clinical outcomes, explored biomarker contributions, and developed a machine learning-based model to predict peritoneal transport transitions.
Methods:
This prospective study enrolled peritoneal dialysis (PD) patients aged ≥18 years from 2016 to 2017, with follow-up until 2020. Patients with missing PET data or acute illness within 2 months of PET were excluded. Based on longitudinal PET changes, patients were classified into four trajectory groups: persistent high (HH), high to low (HL), low to high (LH), and persistent low (LL). Clinical outcomes included technique failure and mortality. Dialysate biomarkers were quantified using ELISA and standardized by appearance rates (ARs). A Support Vector Machine (SVM) model was developed to predict PET trajectory changes using clinical and biomarker data. Model performance was assessed using accuracy, precision, recall, F1-score, and area under the curve (AUC).
Results:
Among 132 eligible patients, cumulative risk analysis identified the HH group as having the highest risk of adverse outcomes, followed by LH, LL, and HL groups (p = 0.009). Based on prognostic trends, HH and LH were reclassified as the future high (FH) group, while HL and LL were grouped as the future low (FL) group. The FH group had a significantly higher risk of adverse outcomes than the FL group (HR: 7.87, 95% CI: 1.81-34.10, p = 0.005). Matrix metalloproteinase 2 (MMP2) AR and plasminogen activator inhibitor 1 (PAI-1) AR differed significantly across PET trajectory groups (p < 0.001 for both), with the HH group exhibiting the highest biomarker levels (MMP2 AR: 195.0 ng/min, IQR: 145.2-230.0; PAI-1 AR: 2.63 ng/min, IQR: 1.29-4.51). The SVM model integrating clinical and biomarker data outperformed models using clinical data alone, achieving a higher AUC (0.87 vs. 0.71). Risk visualization curves identified males with elevated biomarkers as particularly vulnerable to transitioning to high transporter status.
Conclusions:
Sustained or transitioning to a high transporter status significantly increases the risk of adverse outcomes in PD patients. Higher MMP2 AR and PAI-1 AR levels are associated with an increased risk of adverse PET trajectory changes, enhancing risk stratification. Integrating biomarker-based predictive models with clinical data improves prognostic accuracy, supporting early intervention strategies for high risk patients.
