Scholarlink

Background <p>Suspected infection can progress to sepsis or septic shock, contributing to a high mortality rate among patients admitted to ICU. However, the characteristics of suspected infection remain incompletely defined. We aimed to develop and validate predictive models to identify independent risk factors for mortality and multidrug-resistant infection in patients with suspected infection upon ICU admission in mainland China.</p> Methods <p>We prospectively collected medical data from patients with suspected infection admitted to ICUs across mainland China between July 2021 and December 2022. Patients were randomly allocated to a training cohort and a validation cohort at a 7:3 ratio. Using machine learning algorithms, we identified risk factors and constructed predictive models for mortality and multidrug-resistant infection. The performance of models developed by logistic regression, random forest, extreme gradient boosting, and gradient boosting machine was evaluated using the area under the curve (AUC), Brier score, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), the Youden index, net reclassification improvement (NRI), and integrated discrimination improvement (IDI).</p> Results <p>A total of 2963 patients from 67 hospitals in mainland China were enrolled. The most common infection sites were the lung (79.28%), bloodstream (17.11%) and abdomen (16.54%). The overall mortality rate was 10.90%. The logistic regression model demonstrated the best performance for mortality prediction (AUC 0.87) and identified the following risk factors: surgery (2.39, 95CI% 1.60–3.56, <i>p</i> &lt; 0.01), APACHE II score (1.51, 95CI% 1.44–1.58, <i>p</i> &lt; 0.01), and bloodstream infection (2.08, 95CI% 1.40–3.10, <i>p</i> &lt; 0.01) were independent risk factors of mortality. For multidrug resistant infection prediction, the logistic regression model also showed the highest discriminative ability (AUC 0.86). Independent risk factors included APACHE II score (1.06, 95%CI 1.03–1.1, <i>p</i> &lt; 0.01), bloodstream infection (1.82, 95%CI 1.26–2.61, <i>p</i> &lt; 0.01), urinary infection (3.42, 95%CI 2.20–5.30, <i>p</i> &lt; 0.01), <i>Klebsiella pneumoniae</i> (11.67, 95%CI 7.91–17.21, <i>p</i> &lt; 0.01), <i>Acinetobacter baumannii</i> (85.22, 95%CI 50.03-145.18, <i>p</i> &lt; 0.01), and <i>Enterococcus faecium</i> (22.10, 95%CI 8.58–56.93, <i>p</i> &lt; 0.01).</p> Conclusions <p>Pneumonia was the most common infection among ICU patients in mainland China. Using machine learning techniques, we developed and validated predictive models that identified independent risk factors for mortality and multidrug-resistant infection upon ICU admission.</p> Clinical trial registration <p>The trial protocol was registered at ClinicalTrials.gov (Identifier: NCT04966390, Registration Date: July 14, 2021). The full record can be accessed at: <a href="https://clinicaltrials.gov/show/NCT04966390">https://clinicaltrials.gov/show/NCT04966390</a>.</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 mortality and multidrug-resistant infection risks in ICU patients with suspected infection: a prospective national multicenter cohort study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月20日

查看全文
  • Shuguang Yang,
  • Yao Sun,
  • Ting Wang,
  • Chenxiao Hao,
  • Hua Zhang,
  • Wei Sun,
  • Youzhong An,
  • Huiying Zhao

摘要

Background

Suspected infection can progress to sepsis or septic shock, contributing to a high mortality rate among patients admitted to ICU. However, the characteristics of suspected infection remain incompletely defined. We aimed to develop and validate predictive models to identify independent risk factors for mortality and multidrug-resistant infection in patients with suspected infection upon ICU admission in mainland China.

Methods

We prospectively collected medical data from patients with suspected infection admitted to ICUs across mainland China between July 2021 and December 2022. Patients were randomly allocated to a training cohort and a validation cohort at a 7:3 ratio. Using machine learning algorithms, we identified risk factors and constructed predictive models for mortality and multidrug-resistant infection. The performance of models developed by logistic regression, random forest, extreme gradient boosting, and gradient boosting machine was evaluated using the area under the curve (AUC), Brier score, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), the Youden index, net reclassification improvement (NRI), and integrated discrimination improvement (IDI).

Results

A total of 2963 patients from 67 hospitals in mainland China were enrolled. The most common infection sites were the lung (79.28%), bloodstream (17.11%) and abdomen (16.54%). The overall mortality rate was 10.90%. The logistic regression model demonstrated the best performance for mortality prediction (AUC 0.87) and identified the following risk factors: surgery (2.39, 95CI% 1.60–3.56, p < 0.01), APACHE II score (1.51, 95CI% 1.44–1.58, p < 0.01), and bloodstream infection (2.08, 95CI% 1.40–3.10, p < 0.01) were independent risk factors of mortality. For multidrug resistant infection prediction, the logistic regression model also showed the highest discriminative ability (AUC 0.86). Independent risk factors included APACHE II score (1.06, 95%CI 1.03–1.1, p < 0.01), bloodstream infection (1.82, 95%CI 1.26–2.61, p < 0.01), urinary infection (3.42, 95%CI 2.20–5.30, p < 0.01), Klebsiella pneumoniae (11.67, 95%CI 7.91–17.21, p < 0.01), Acinetobacter baumannii (85.22, 95%CI 50.03-145.18, p < 0.01), and Enterococcus faecium (22.10, 95%CI 8.58–56.93, p < 0.01).

Conclusions

Pneumonia was the most common infection among ICU patients in mainland China. Using machine learning techniques, we developed and validated predictive models that identified independent risk factors for mortality and multidrug-resistant infection upon ICU admission.

Clinical trial registration

The trial protocol was registered at ClinicalTrials.gov (Identifier: NCT04966390, Registration Date: July 14, 2021). The full record can be accessed at: https://clinicaltrials.gov/show/NCT04966390.