Scholarlink

<p>Many studies have reported increased mortality risk in patients with Parkinson’s disease (PD), but few have investigated the risk factors for PD mortality, including medical and socioeconomic factors. We applied an explainable artificial intelligence (XAI) model to predict long-term all-cause mortality in patients with PD using administrative healthcare data collected at PD diagnosis. Among seven machine learning algorithms, XGBoost achieved the best performance (10-year area under the receiver operating characteristic curve (AUROC): 0.836; 5-year AUROC: 0.894). The most important contributing feature to PD mortality was age, followed by male sex and pneumonia. Using XAI models, the nonlinear association between contributing factors and PD mortality was assessed, and an optimal target value to reduce mortality was found. In addition, prediction of individualized 10-year mortality risk for each PD participant was possible. Our XAI modeling pipeline demonstrated the feasibility to predict long-term mortality in patients with PD using preexisting healthcare data.</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

Prediction of all-cause mortality in Parkinson’s disease with explainable artificial intelligence using administrative healthcare data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月2日

查看全文
  • You Hyun Park,
  • Yong Wook Kim,
  • Dae Ryong Kang,
  • Sang Chul Lee,
  • Seo Yeon Yoon

摘要

Many studies have reported increased mortality risk in patients with Parkinson’s disease (PD), but few have investigated the risk factors for PD mortality, including medical and socioeconomic factors. We applied an explainable artificial intelligence (XAI) model to predict long-term all-cause mortality in patients with PD using administrative healthcare data collected at PD diagnosis. Among seven machine learning algorithms, XGBoost achieved the best performance (10-year area under the receiver operating characteristic curve (AUROC): 0.836; 5-year AUROC: 0.894). The most important contributing feature to PD mortality was age, followed by male sex and pneumonia. Using XAI models, the nonlinear association between contributing factors and PD mortality was assessed, and an optimal target value to reduce mortality was found. In addition, prediction of individualized 10-year mortality risk for each PD participant was possible. Our XAI modeling pipeline demonstrated the feasibility to predict long-term mortality in patients with PD using preexisting healthcare data.