Scholarlink

Background <p>Hospital readmissions are a major burden for patients, families, and healthcare systems. Artificial intelligence (AI) and electronic medical records (EMR) offer new opportunities to improve clinical decision support by identifying patients at high risk of unplanned readmission. However, accurate prediction remains challenging in urogenital system disorder populations.</p> Methods <p>We conducted a retrospective observational study using EMR data from Taichung Tzu Chi Hospital (TTCH), Taiwan, between December 2018 and January 2022. Adult inpatients diagnosed with urogenital system disorders (ICD-10 N00–N99) were included, while planned readmissions were excluded. A total of 112 EMR features were extracted per patient. Data preprocessing and class balancing were applied. Random forest–based variable importance measures were employed for automatic feature selection, resulting in the identification of 34 key predictors. Seven machine learning models, including deep learning and ensemble approaches, were developed and evaluated using a 79.5% training and 20.5% testing split.</p> Results <p>The final dataset included 7,413 cases, with 321 unplanned 30-day readmissions. Among all models, the ensemble model achieved the best predictive performance, attaining the highest area under the receiver operating characteristic curve (AUC) of 0.827, and outperforming deep neural network, CatBoost, XGBoost, random forest, standard logistic regression, and logistic regression using only the LACE score. The selected 34 predictors contributed substantially to readmission risk identification.</p> Conclusions <p>Our AI-based ensemble framework accurately predicts unplanned 30-day hospital readmissions among patients with urogenital disorders. This approach may support early risk stratification, targeted interventions, and improved clinical decision-making, contributing to the development of smart hospitals and preventive healthcare systems.</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

Predicting unplanned 30-day hospital readmissions in patients with urogenital system disorders: a retrospective EMR-based ensemble learning study in Taiwan

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月1日

查看全文
  • Chang-Hung Lin,
  • Chuen-Horng Lin,
  • Chih-Hao Liu,
  • Chen-Yu Hsiao,
  • Yi-Ning Wang,
  • Hao-Siang Hsu,
  • Fu-Hsing Wu,
  • Yung-Kuan Chan

摘要

Background

Hospital readmissions are a major burden for patients, families, and healthcare systems. Artificial intelligence (AI) and electronic medical records (EMR) offer new opportunities to improve clinical decision support by identifying patients at high risk of unplanned readmission. However, accurate prediction remains challenging in urogenital system disorder populations.

Methods

We conducted a retrospective observational study using EMR data from Taichung Tzu Chi Hospital (TTCH), Taiwan, between December 2018 and January 2022. Adult inpatients diagnosed with urogenital system disorders (ICD-10 N00–N99) were included, while planned readmissions were excluded. A total of 112 EMR features were extracted per patient. Data preprocessing and class balancing were applied. Random forest–based variable importance measures were employed for automatic feature selection, resulting in the identification of 34 key predictors. Seven machine learning models, including deep learning and ensemble approaches, were developed and evaluated using a 79.5% training and 20.5% testing split.

Results

The final dataset included 7,413 cases, with 321 unplanned 30-day readmissions. Among all models, the ensemble model achieved the best predictive performance, attaining the highest area under the receiver operating characteristic curve (AUC) of 0.827, and outperforming deep neural network, CatBoost, XGBoost, random forest, standard logistic regression, and logistic regression using only the LACE score. The selected 34 predictors contributed substantially to readmission risk identification.

Conclusions

Our AI-based ensemble framework accurately predicts unplanned 30-day hospital readmissions among patients with urogenital disorders. This approach may support early risk stratification, targeted interventions, and improved clinical decision-making, contributing to the development of smart hospitals and preventive healthcare systems.