# Optimizing Nursing Communication for Symptom Management in Hemodialysis: Development of an Artificial Intelligence–Based Web Predictive Model for Burden Classification and Evidence Navigation

## TL;DR

This study developed an AI model to classify symptom burden in hemodialysis patients, aiming to improve nursing care and resource allocation.

## Contribution

A high-accuracy XGBoost model for symptom burden classification in hemodialysis patients using machine learning.

## Key findings

- The XGBoost model achieved an AUC of 0.994 in predicting symptom burden categories.
- Key predictors included uremia toxin, electrolyte imbalance, and psychological symptoms.
- The model showed strong clinical utility and reduced nursing workload through targeted interventions.

## Abstract

Chronic kidney disease (CKD) is a global health concern, with hemodialysis (HD) being a vital life-sustaining treatment for affected patients. Symptom burden in HD patients significantly impacts quality of life and clinical outcomes. However, symptom management remains inadequate, especially in resource-constrained settings.

This study aims to develop a predictive model to categorize symptom burden and optimize nursing interventions using machine learning.

A nationwide, cross-sectional study employing nonprobability convenience and multiregional sampling.

Data were collected in five provinces in China: Liaoning (Northeast China), Fujian (Southeast China), Yunnan (Southwest China), Jiangsu (Eastern China), and Shaanxi (Central China). A total of 1866 HD patients were finally included. Machine learning algorithms, including elastic net regression, Boruta feature selection, and 8 classifiers, were used to develop and validate a predictive model for classifying symptom burden categories (mild vs. severe). The model's performance was evaluated using metrics such as accuracy, sensitivity, specificity, and AUC. Decision curve analysis (DCA) and SHapley Additive exPlanations (SHAP) values were employed for clinical utility and interpretability.

The XGBoost model demonstrated excellent predictive accuracy with an AUC of 0.994 on test data, outperforming other models. Key predictors included uremia toxin, electrolyte imbalance, and psychological symptoms. The model achieved strong calibration and high clinical utility, as confirmed by DCA. It also offered a practical tool for targeted interventions, reducing nursing workload while ensuring efficient care allocation.

This study demonstrates the potential of machine learning models to improve symptom burden classification in HD patients. The XGBoost model, utilizing a limited set of key predictors, offers high predictive accuracy and clinical utility, providing a scalable solution for resource-constrained healthcare settings. This approach supports personalized symptom management, contributing to improved patient outcomes and efficient resource use in nephrology nursing.