Abstract
Introduction: Tunnelled central venous catheters are increasingly used for vascular access in patients undergoing haemodialysis for chronic kidney disease. However, catheter dysfunction is a frequent and clinically relevant complication, impairing treatment efficacy and increasing morbidity. This study aimed to develop and internally validate predictive models for catheter dysfunction using routinely collected haemodialysis session data, with the goal of facilitating early detection and proactive clinical decision-making. Methods: We conducted a diagnostic, retrospective, cross-sectional, and analytical study based on 60,230 HD sessions recorded in 2021 across dialysis centres in Spain. A total of 743 patients with functioning catheter were included. Clinical, technical, and haemodynamic variables were analysed to identify those associated with catheter dysfunction in the subsequent session. Five logistic regression models were built; the dataset was split into training (two-thirds) and internal validation (one-third) cohorts. Model performance was evaluated using the area under the ROC curve (AUC) and the Hosmer-Lemeshow test. Results: Significant predictors included venous pressure, effective blood flow, catheter location, convective techniques, and line reversal. The bootstrapping model, selected for internal validation due to its parsimony and performance, achieved an AUC of 0.844 (95% CI: 0.824-0.863), with a sensitivity of 81.6% and a specificity of 70.9% at a 0.019 threshold. Conclusions: The bootstrapping-based predictive model is a valuable clinical tool for anticipating catheter dysfunction using routine haemodialysis data. Its implementation may enable earlier intervention, reduce reliance on reactive treatments, and enhance vascular access management in haemodialysis patients.