Scholarlink

<p>Acute renal failure (ARF) is one of the most common conditions encountered in the intensive care unit (ICU). ARF has a complex pathogenesis and due to the progressive weakening of the structure and function of the kidney, the incidence of ARF increases significantly in the aging group. Therefore, the development of reliable predictive model is of great importance to identify those patients in high risk for ARF, in order to provide timely and effective interventions to improve their prognosis. Extreme gradient boosting (XGBoost) is an efficient integrated learning algorithm with advantages over traditional logistic regression (LR) methods. The purpose of this study was to compare the performance of the two models in predicting 90-day mortality in elderly patients with ARF. Data of elderly patients (&gt; 60years) with ARF in ICU were extracted from MIMIC IV with 90-day mortality as end-point. The performance of the two predictive models was tested and compared by receiver operating characteristic curve and decision curve analysis (DCA). Cumulative residual distribution plot and residual box-plot were then used to determine the fit of the model. Finally, the model with better overall diagnostic value was selected and a breakdown plot was drawn. Data of 7,500 elderly ARF patients were analyzed, of whom 1,150 died within 90 days. Both models showed good discriminatory ability, but the XGBoost model had a larger area under the curve value. DCA results revealed that the net benefit of the XGBoost model had a greater range than the LR model. Moreover, the XGBoost model had the smallest sample residuals and root-mean-square residuals, indicating a better fitting of the XGBoost algorithm. Finally, a breakdown plot based on the XGBoost model was created as an individualized tool for prognosis prediction in elderly patients with ARF. Our study find that the XGBoost algorithm model was a better model for predicting 90-day mortality in elderly ICU patients with ARF compared to the LR model. The model may have clinical applications for elderly patients with ARF and may help healthcare professionals to develop detailed treatment plans as well as provide accurate care.</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

‌Application of XGBoost and logistic regression in predicting 90 days mortality for elderly severe acute renal failure patients

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月3日

查看全文
  • Jinping Zeng,
  • Yiying Zhu,
  • Feng Ye,
  • Qin Song,
  • Haiyan Ma,
  • Jun Yang,
  • Yinyin Wu

摘要

Acute renal failure (ARF) is one of the most common conditions encountered in the intensive care unit (ICU). ARF has a complex pathogenesis and due to the progressive weakening of the structure and function of the kidney, the incidence of ARF increases significantly in the aging group. Therefore, the development of reliable predictive model is of great importance to identify those patients in high risk for ARF, in order to provide timely and effective interventions to improve their prognosis. Extreme gradient boosting (XGBoost) is an efficient integrated learning algorithm with advantages over traditional logistic regression (LR) methods. The purpose of this study was to compare the performance of the two models in predicting 90-day mortality in elderly patients with ARF. Data of elderly patients (> 60years) with ARF in ICU were extracted from MIMIC IV with 90-day mortality as end-point. The performance of the two predictive models was tested and compared by receiver operating characteristic curve and decision curve analysis (DCA). Cumulative residual distribution plot and residual box-plot were then used to determine the fit of the model. Finally, the model with better overall diagnostic value was selected and a breakdown plot was drawn. Data of 7,500 elderly ARF patients were analyzed, of whom 1,150 died within 90 days. Both models showed good discriminatory ability, but the XGBoost model had a larger area under the curve value. DCA results revealed that the net benefit of the XGBoost model had a greater range than the LR model. Moreover, the XGBoost model had the smallest sample residuals and root-mean-square residuals, indicating a better fitting of the XGBoost algorithm. Finally, a breakdown plot based on the XGBoost model was created as an individualized tool for prognosis prediction in elderly patients with ARF. Our study find that the XGBoost algorithm model was a better model for predicting 90-day mortality in elderly ICU patients with ARF compared to the LR model. The model may have clinical applications for elderly patients with ARF and may help healthcare professionals to develop detailed treatment plans as well as provide accurate care.