Scholarlink

<p>Stroke-associated pneumonia (SAP) is a frequent and severe complication following stroke. Recently, several machine learning (ML) models have been developed to predict SAP. We aimed to evaluate the predictive performance of these models in SAP prediction. We searched PubMed, Embase, Scopus, and Web of Science up to 18 June 2025, for studies developing ML, deep learning (DL), or neural network (NN) models for SAP prediction. The pooled estimates of area under the curve (AUC), accuracy (ACC), sensitivity (SEN), specificity (SPE), and diagnostic odds ratio (DOR) were calculated using the R program. A total of 27 studies were included, with a prevalence of SAP at 18.9%. Most models were ML based (77.8%), and clinical data were the most common input (77.8%). The pooled AUC was 0.84 [95% (CI): 0.80–0.87], and the pooled ACC was 0.80 (95% CI: 0.76–0.84). SEN and SPE were 0.73 (95% CI: 0.63–0.81) and 0.85 (95% CI: 0.77–0.90), respectively. The pooled DOR was 15.4 (95% CI: 10.2–23.3), and the summary receiver operating characteristic (SROC) curve showed an AUC of 0.853 with a false positive rate of 0.153 (95% CI: 0.096–0.235). No significant differences were found between ischemic and hemorrhagic subgroups. ML-based models demonstrated promising performance in predicting SAP and can help physicians through the early identification of high-risk cases. However, further external validation and integration into clinical workflows are required before widespread clinical adoption.</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 Models for Predicting Stroke-Associated Pneumonia: A Systematic Review and Meta-Analysis

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Bardia Hajikarimloo,
  • Ibrahim Mohammadzadeh,
  • Salem M. Tos,
  • Rana Hashemi,
  • Alireza Khoshrou,
  • Mohammadreza Amjadzadeh,
  • Mandana Dehghan,
  • Saba Aghajan,
  • Ehsan Goudarzi,
  • Dorsa Najari,
  • Azin Ebrahimi,
  • Mohammad Amin Habibi

摘要

Stroke-associated pneumonia (SAP) is a frequent and severe complication following stroke. Recently, several machine learning (ML) models have been developed to predict SAP. We aimed to evaluate the predictive performance of these models in SAP prediction. We searched PubMed, Embase, Scopus, and Web of Science up to 18 June 2025, for studies developing ML, deep learning (DL), or neural network (NN) models for SAP prediction. The pooled estimates of area under the curve (AUC), accuracy (ACC), sensitivity (SEN), specificity (SPE), and diagnostic odds ratio (DOR) were calculated using the R program. A total of 27 studies were included, with a prevalence of SAP at 18.9%. Most models were ML based (77.8%), and clinical data were the most common input (77.8%). The pooled AUC was 0.84 [95% (CI): 0.80–0.87], and the pooled ACC was 0.80 (95% CI: 0.76–0.84). SEN and SPE were 0.73 (95% CI: 0.63–0.81) and 0.85 (95% CI: 0.77–0.90), respectively. The pooled DOR was 15.4 (95% CI: 10.2–23.3), and the summary receiver operating characteristic (SROC) curve showed an AUC of 0.853 with a false positive rate of 0.153 (95% CI: 0.096–0.235). No significant differences were found between ischemic and hemorrhagic subgroups. ML-based models demonstrated promising performance in predicting SAP and can help physicians through the early identification of high-risk cases. However, further external validation and integration into clinical workflows are required before widespread clinical adoption.