Exploring prognostic factors on vascular outcomes among maintenance dialysis patients and establishing a prognosis prediction model using machine learning methods

TL;DR

This study uses machine learning to predict cardiovascular risks in dialysis patients by combining traditional and kidney disease-specific factors.

Contribution

A novel machine learning model integrating traditional and CKD-specific factors improves MACE risk prediction in dialysis patients.

Findings

The model achieved high predictive accuracy with an AUROC of 0.864.

Key predictors included age, diabetes, hypertension, and CKD-specific factors like albumin and cholesterol levels.

Patients were effectively stratified into high- and low-risk groups with significant survival differences.

Abstract

Cardiovascular diseases (CVDs) remain the leading cause of morbidity and mortality in end-stage kidney disease patients, with persistently high rates of major adverse cardiovascular events (MACEs). Traditional risk factors such as diabetes and hypertension have limited predictive value in this population, while chronic kidney disease (CKD)-specific factors including inflammation, disordered mineral metabolism, and vascular calcification play significant roles. Therefore, we developed a machine learning-based model incorporating traditional and CKD-specific variables to improve MACEs risk predictions and facilitate early intervention.

We retrospectively enrolled 412 adults undergoing maintenance hemodialysis (MHD) at a single center between October and December 2018, with follow-up until December 2021 or censoring. Enrolled patients were classified by MACEs occurrences. An elastic-net regularized Cox regression with backward selection was used to identify key MACE predictors, integrating traditional and CKD-specific risk factors. The model performance was validated via leave-one-out cross-validation and area under the receiver operating characteristic curve (AUROC). Standard statistical tests were applied for group comparisons.

The elastic-net regression identified 46 key predictors from a high-dimensional dataset. Patients who developed MACEs were older and had higher prevalences of diabetes, hypertension, coronary disease, heart failure, and lower albumin/cholesterol. A Cox regression with backward selection was used to refine the model to 13 predictors. The final model demonstrated excellent predictive performance, (AUROC = 0.864) (95% CI, 0.8131– 0.9148), effectively stratifying patients into high- and low-risk groups with significant survival differences (log-rank p < 0.001).

The machine learning-based model demonstrated high predictive accuracy for MACEs in MHD patients by integrating both traditional and CKD-specific risk factors, offering a potential tool for early identification and clinical decision-making.

Not applicable.

The online version contains supplementary material available at 10.1186/s12911-025-03302-2.

Figures