Scholarlink

Background <p>Despite advances in patient safety, adverse transfusion reactions (ATRs) continue to occur in clinical settings and remain a primary focus of hospital hemovigilance committees. Artificial intelligence (AI) has emerged as a promising tool for detecting and preventing these complications. The objective of this study is to synthesize the evidence on the applications of AI in identifying and predicting ATRs, and to examine the existing evidence regarding the feasibility and effectiveness of employing these tools in active clinical management.</p> Methods <p>This systematic review (SR) was conducted according to the PRISMA 2020 guidelines. English-language articles published within the last decade, focusing on the application of AI in the Identification and Prediction of ATRs and Implications for Clinical Management, were retrieved from the PubMed, Scopus, and Web of Science databases and subsequently analyzed. The quality of the included studies was assessed using the QUADAS-AI tool, and the findings are presented descriptively.</p> Results <p>This SR showed that in the 24 included studies, AI models were primarily applied across four main focal areas: transfusion risks and outcomes, risk and moderating factors, transfusion volume and intensity, and classification and extraction of ATRs. In the included studies, the most essential model evaluation metrics were AUROC and Sensitivity, each reported in nine studies, followed by Accuracy and F1-Score, each reported in five studies. Among the studies, the Random Forest (RF) model was used more frequently than other models. Moreover, none explicitly addressed the development, implementation, or clinical evaluation of an active management system based on AI. Clinically, most studies focused on transfusion-related complications such as mortality, bleeding, and morbidity. The majority of the studies were conducted in the field of Hematology, followed by cardiology, surgery, and ICU units.</p> Conclusions <p>Based on the interpretation of results, individual patient factors and transfusion volume play a pivotal role in the occurrence of ATRs. Implementing safe transfusion strategies, including the use of clinical decision support systems (CDSS) integrated with electronic health records (EHR) and personalized medicine approaches, alongside adherence to ethical considerations and patient privacy protection, is essential in future research. This study also identified two significant research gaps: first, the lack of research on the implementation or clinical evaluation of AI-based active management systems for ATRs; and second, the analysis of population groups revealed that research has been predominantly focused on adults, highlighting a gap concerning vulnerable populations, particularly pediatric patients.</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

Artificial intelligence in the identification and prediction of adverse transfusion reactions(ATRs) and implications for clinical management: a systematic review of models and applications

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月28日

查看全文
  • Mahdie ShojaeiBaghini,
  • Mohammad Mehdi Ghaemi,
  • Alihasan Ahmadipour

摘要

Background

Despite advances in patient safety, adverse transfusion reactions (ATRs) continue to occur in clinical settings and remain a primary focus of hospital hemovigilance committees. Artificial intelligence (AI) has emerged as a promising tool for detecting and preventing these complications. The objective of this study is to synthesize the evidence on the applications of AI in identifying and predicting ATRs, and to examine the existing evidence regarding the feasibility and effectiveness of employing these tools in active clinical management.

Methods

This systematic review (SR) was conducted according to the PRISMA 2020 guidelines. English-language articles published within the last decade, focusing on the application of AI in the Identification and Prediction of ATRs and Implications for Clinical Management, were retrieved from the PubMed, Scopus, and Web of Science databases and subsequently analyzed. The quality of the included studies was assessed using the QUADAS-AI tool, and the findings are presented descriptively.

Results

This SR showed that in the 24 included studies, AI models were primarily applied across four main focal areas: transfusion risks and outcomes, risk and moderating factors, transfusion volume and intensity, and classification and extraction of ATRs. In the included studies, the most essential model evaluation metrics were AUROC and Sensitivity, each reported in nine studies, followed by Accuracy and F1-Score, each reported in five studies. Among the studies, the Random Forest (RF) model was used more frequently than other models. Moreover, none explicitly addressed the development, implementation, or clinical evaluation of an active management system based on AI. Clinically, most studies focused on transfusion-related complications such as mortality, bleeding, and morbidity. The majority of the studies were conducted in the field of Hematology, followed by cardiology, surgery, and ICU units.

Conclusions

Based on the interpretation of results, individual patient factors and transfusion volume play a pivotal role in the occurrence of ATRs. Implementing safe transfusion strategies, including the use of clinical decision support systems (CDSS) integrated with electronic health records (EHR) and personalized medicine approaches, alongside adherence to ethical considerations and patient privacy protection, is essential in future research. This study also identified two significant research gaps: first, the lack of research on the implementation or clinical evaluation of AI-based active management systems for ATRs; and second, the analysis of population groups revealed that research has been predominantly focused on adults, highlighting a gap concerning vulnerable populations, particularly pediatric patients.