Scholarlink

Background <p>Lung transplantation is an important therapeutic option for patients with end-stage chronic obstructive pulmonary disease. However, post-transplantation hospitalizations are associated with substantial cost variation. Accurately predicting high in-hospital expenditures is essential for optimizing clinical management and healthcare resource allocation.</p> Methods <p>This study utilized the Nationwide Inpatient Sample database (2005–2014) to identify adult patients who underwent lung transplantation. After rigorous data preprocessing, seven post-discharge or manually coded variables were excluded to prevent potential information leakage. High-cost hospitalizations were defined as those with total hospital charges falling within the top 20% (≥ 80th percentile) of the overall cost distribution. Feature selection was performed using a two-stage process combining correlation analysis and Recursive Feature Elimination. Six machine learning algorithms were trained through ten-fold cross-validation with optimized hyperparameters, and model performance was compared across seven evaluation methods. The optimal model was further interpreted to identify the key features associated with high-cost hospitalizations.</p> Results <p>The Gradient Boosting Machine model achieved the most robust predictive performance across all evaluation metrics, with consistently high accuracy (0.816), F1-score (0.6773), and area under the curve (AUC = 0.8078). To further interpret the model’s decision mechanism, Shapley Additive Explanations analysis was applied to quantify each feature’s contribution to high-cost prediction. The results revealed that hospital region, mechanical ventilation, acute kidney injury, lung transplant type, and complications of the transplanted lung were the most influential predictors of high-cost hospitalizations. Notably, hospitalizations in the northeast region were associated with the highest predicted costs, indicating substantial regional variation in post-transplant expenditure patterns.</p> Conclusions <p>This study developed a transparent machine learning method for predicting high-cost hospitalizations after lung transplantation. The Gradient Boosting Machine model demonstrated optimal predictive performance, and Shapley analysis identified key clinical and regional factors driving hospital expenditures, providing evidence to guide cost management and clinical decision-making.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning–based prediction of high hospital charges after lung transplantation in chronic obstructive pulmonary disease patients

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月12日

查看全文
  • Qunzhe Ding,
  • Tianyu Sun,
  • Zihao Zhang,
  • Yuhong Liu,
  • Mengli Zheng,
  • Pengtao Bao,
  • Xinxin Zhang

摘要

Background

Lung transplantation is an important therapeutic option for patients with end-stage chronic obstructive pulmonary disease. However, post-transplantation hospitalizations are associated with substantial cost variation. Accurately predicting high in-hospital expenditures is essential for optimizing clinical management and healthcare resource allocation.

Methods

This study utilized the Nationwide Inpatient Sample database (2005–2014) to identify adult patients who underwent lung transplantation. After rigorous data preprocessing, seven post-discharge or manually coded variables were excluded to prevent potential information leakage. High-cost hospitalizations were defined as those with total hospital charges falling within the top 20% (≥ 80th percentile) of the overall cost distribution. Feature selection was performed using a two-stage process combining correlation analysis and Recursive Feature Elimination. Six machine learning algorithms were trained through ten-fold cross-validation with optimized hyperparameters, and model performance was compared across seven evaluation methods. The optimal model was further interpreted to identify the key features associated with high-cost hospitalizations.

Results

The Gradient Boosting Machine model achieved the most robust predictive performance across all evaluation metrics, with consistently high accuracy (0.816), F1-score (0.6773), and area under the curve (AUC = 0.8078). To further interpret the model’s decision mechanism, Shapley Additive Explanations analysis was applied to quantify each feature’s contribution to high-cost prediction. The results revealed that hospital region, mechanical ventilation, acute kidney injury, lung transplant type, and complications of the transplanted lung were the most influential predictors of high-cost hospitalizations. Notably, hospitalizations in the northeast region were associated with the highest predicted costs, indicating substantial regional variation in post-transplant expenditure patterns.

Conclusions

This study developed a transparent machine learning method for predicting high-cost hospitalizations after lung transplantation. The Gradient Boosting Machine model demonstrated optimal predictive performance, and Shapley analysis identified key clinical and regional factors driving hospital expenditures, providing evidence to guide cost management and clinical decision-making.