This study introduces advanced Joint Models (JM) to enhance mortality prediction in dialysis patients, addressing the inadequacies of existing models which lack clinical precision. We analyzed the relationship between albumin trajectories and mortality in 314 peritoneal dialysis patients over eight years, using training and validation datasets. We developed 12 JM, incorporating seven baseline risk factors; 10 models addressed individual and within-patient trajectory outliers, and six models included the competing events of hemodialysis transfer and kidney transplantation. These were also applied to a hemodialysis cohort of 315 patients. Findings showed a consistent albumin hazard ratio for death across all JM, ranging from 1.22 to 1.28, confirmed by two simulation studies. JM software optimizations achieved 3- to sixfold faster processing and 4.1- to 12.9-fold higher efficiency than conventional software. Longer follow-up improved all JM accuracies, and models with competing risks outperformed those considering only outliers. Across forecasting periods up to five years, all JM consistently surpassed benchmark Cox models, demonstrating significantly higher Area Under Curve (AUC) scores in both peritoneal and hemodialysis groups. In conclusion, our JM combine established risk factors with dynamic individual albumin profiles, integrating outliers and competing risks, substantially enhancing mortality predictions for dialysis patients.
Keywords: Albumin; All-cause mortality; Dynamic predictions; Haemodialysis; Peritoneal dialysis.
© 2025. The Author(s).