Chao Qianwen, Pei Juhong, Wei Yuting, Yang Zhuang, Wang Xiaorui, Du Li, Han Lin
Evidence-based Nursing Center, School of Nursing, Lanzhou University, Lanzhou City, Gansu Province, 730000, China.
First Clinical School of Medicine, Lanzhou University, Lanzhou City, Gansu Province, 730000, China.
J Tissue Viability. 2025 Aug;34(3):100894. doi: 10.1016/j.jtv.2025.100894. Epub 2025 Mar 21.
Pressure injury is prevalent in clinical settings and demands precise staging for optimal care. Subjectivity and imprecision in traditional visual assessments have sparked the creation of advanced technology-based evaluation tools.
To systematically assess pressure injury staging methods, analyze their evaluation results, and provide reference for clinical practice.
Systematic review and meta-analysis.
PubMed, Embase, Cochrane Library, Web of Science, CINAHL, and manual searches of academic journals and conference proceedings were utilized.
The study conducted a systematic search of databases in April 2024, utilizing Endnote X9 to document findings. Two reviewers independently extracted data and evaluated its quality using the QUADAS-2 tool. The meta-analysis, conducted in Meta-disc, focused on metrics such as AUC, sensitivity, and specificity. Heterogeneity among the studies was assessed using Cochran's Q and I tests.
This review screened 15312 articles and ultimately included 15 studies. These studies described methods for pressure injury staging, including visual assessment, 29 machine learning models, and human-model integrated evaluation. The accuracy of traditional visual assessment was relatively low and showed significant variability. Eight studies involving 24 machine learning models were included in the meta-analysis, demonstrating significantly high accuracy, with an AUC of 0.93, and the combined sensitivity, specificity, and diagnostic odds ratio were 0.81, 0.87, and 20.48, respectively.
The review underscores the advantages of machine learning in diagnosing pressure injuries, offering higher accuracy over traditional methods. Integrating clinical expertise with machine learning enhances medical service quality and efficiency.
CRD42023462951. PROSPERO REGISTRATION LINK: crd.york.ac.uk/prospero/display_record.php?ID=CRD42023462951.
压力性损伤在临床环境中很常见，需要精确分期以实现最佳护理。传统视觉评估中的主观性和不精确性促使了基于先进技术的评估工具的产生。
系统评估压力性损伤分期方法，分析其评估结果，为临床实践提供参考。
系统评价和荟萃分析。
使用了PubMed、Embase、Cochrane图书馆、Web of Science、CINAHL，并对手学术期刊和会议论文进行了手工检索。
该研究于2024年4月对数据库进行了系统检索，使用Endnote X9记录研究结果。两名评审员独立提取数据，并使用QUADAS-2工具评估其质量。在Meta-disc中进行的荟萃分析侧重于AUC、敏感性和特异性等指标。使用Cochran's Q和I检验评估研究之间的异质性。
本综述筛选了15312篇文章，最终纳入15项研究。这些研究描述了压力性损伤分期的方法，包括视觉评估、29种机器学习模型和人机整合评估。传统视觉评估的准确性相对较低，且存在显著差异。荟萃分析纳入了8项涉及24种机器学习模型的研究，显示出显著的高准确性，AUC为0.93，综合敏感性、特异性和诊断比值比分别为0.81、0.87和20.48。
该综述强调了机器学习在诊断压力性损伤方面的优势，比传统方法具有更高的准确性。将临床专业知识与机器学习相结合可提高医疗服务质量和效率。
PROSPERO注册号：CRD42023462951。PROSPERO注册链接：crd.york.ac.uk/prospero/display_record.php?ID=CRD42023462951。
