文献速递重症综合管理与镇静镇痛消化、代谢与肾脏急症

危重症患者侵袭性念珠菌病预测模型的研究进展

[Research progress on predictive modeling of invasive candidiasis in critically ill patients].
PubMed 文献 · Zhonghua Wei Zhong Bing Ji Jiu Yi Xue · 2026-01-01
一句话结论综述指出传统念珠菌病预测模型阳性预测值与泛化性有限,生物标志物与人工智能整合模型有望提升早期识别,但多缺乏大规模多中心前瞻性验证。

该综述总结ICU侵袭性念珠菌病预测模型研究现状。传统模型基于临床危险因素、念珠菌定植状态和微生物学标志物,在风险分层和经验性抗真菌决策中有作用,但阳性预测值与泛化性有限。随着精准医学发展,BDG、炎症与营养指标、免疫细胞亚群、宿主反应检测及分子诊断技术被纳入风险评估,推动模型从单一临床指标向多维整合演变。机器学习、深度学习和联邦学习可挖掘电子健康记录复杂信息,在提高预测准确性、动态风险评估和临床决策支持方面显示潜力,但多数模型缺乏大规模多中心前瞻性验证,可解释性、泛化性和临床效用待明确。

要点

  • 传统预测模型基于临床危险因素、念珠菌定植状态和微生物学标志物,可用于ICU侵袭性念珠菌病风险分层和经验性抗真菌治疗决策。
  • 传统模型阳性预测值和泛化性仍有限,单靠其指导经验性抗真菌策略存在局限。
  • BDG、炎症与营养指标、免疫细胞亚群、宿主反应检测及分子诊断技术正被纳入风险评估框架,推动预测模型向多维整合演变。
  • 机器学习、深度学习和联邦学习可挖掘电子健康记录复杂信息,在提高预测准确性、动态风险评估和临床决策支持方面显示潜力。
  • 多数现有模型缺乏大规模多中心前瞻性验证,可解释性、泛化性和临床效用需进一步明确。

适用人群与场景

  • 重症监护病房(ICU)患者,用于侵袭性念珠菌病风险评估与早期识别。

诊断标准与关键阈值

  • 原文未提及具体诊断标准、评分界值或启动/停止阈值。

处置流程与药物剂量

  • 原文未提及

红旗征与不能漏的情况

  • 侵袭性念珠菌病临床表现缺乏特异性,常规诊断方法耗时长且阳性率低,常导致治疗延迟和死亡风险增加。

研究设计与关键结果

  • 综述汇总:传统预测模型基于临床危险因素、念珠菌定植状态和微生物学标志物,在风险分层和经验性抗真菌治疗决策中起重要作用,但阳性预测值和泛化性有限。
  • 生物标志物如(1,3)-β-D-葡聚糖(BDG)、炎症与营养指标、免疫细胞亚群、宿主反应检测及分子诊断技术正被纳入风险评估框架,推动模型从单一临床指标向多维整合演变。
  • 人工智能方法包括机器学习、深度学习和联邦学习,可有效挖掘电子健康记录中的复杂信息,在提高预测准确性、实现动态风险评估和支持临床决策方面显示可观潜力。
  • 多数现有模型缺乏大规模多中心前瞻性验证,其可解释性、泛化性和临床效用仍需进一步明确。

证据类型、等级与局限

  • 证据类型:综述。
  • 局限:多数模型缺乏大规模多中心前瞻性验证,可解释性、泛化性和临床效用待明确。

对床旁实践的改变

  • 传统临床预测模型可用于ICU侵袭性念珠菌病风险分层和经验性抗真菌治疗决策,但需认识其阳性预测值和泛化性有限。
  • 生物标志物整合策略和人工智能模型有望提高预测准确性并支持动态风险评估,但原文强调多数模型缺乏大规模多中心前瞻性验证,临床效用待明确,尚不能据此改变床旁实践。
  • 原文未提及具体适用与不适用人群及常见陷阱。

指南或共识出处

  • 原文未提及

以上提炼依据论文摘要(全文未开放获取,或许可不允许翻译转载);全文请点「阅读原文」。

英文原文(摘要)· 点击展开

Invasive candidiasis is one of the common invasive fungal infections in the intensive care unit (ICU), characterized by high incidence, high mortality, and diagnostic difficulty. The clinical presentation of invasive candidiasis lacks specificity, and conventional diagnostic methods are time-consuming with low positivity rates, often leading to delayed treatment and increased death risk. Therefore, early and precise identification of high-risk patients, along with the development of sensitive and specific risk assessment tools to guide individualized antifungal strategies, remains a critical clinical challenge. In recent years, traditional prediction models based on clinical risk factors, Candida colonization status, and microbiological markers have been progressively refined and have played an important role in risk stratification and empirical antifungal therapy decisions. However, their positive predictive value and generalizability remain limited. With advances in precision medicine, biomarkers such as (1,3)-β-D-glucan (BDG), inflammatory and nutritional indicators, immune cell subsets, host response assays, and molecular diagnostic techniques are increasingly being incorporated into risk assessment frameworks, driving the evolution of predictive models from single clinical indicators to multidimensional integration. Meanwhile, artificial intelligence approaches, including machine learning, deep learning, and federated learning, can effectively mine complex information from electronic health records and have demonstrated considerable promise in improving predictive accuracy, enabling dynamic risk assessment, and supporting clinical decision-making. Nevertheless, most current models lack large-scale, multicenter prospective validation, and their interpretability, generalizability, and clinical utility require further clarification. This review summarizes the current state of research on invasive candidiasis prediction models in critically ill patients, covering traditional clinical predictive models, biomarker integration strategies, and artificial intelligence-based models, and discusses future research directions to facilitate early identification and precision diagnosis and treatment of invasive candidiasis in critically ill settings.

来源:Zhonghua Wei Zhong Bing Ji Jiu Yi Xue,查看原文。版权归原作者所有。

参考来源

  1. Li M, Meng S, Li X, et al. [Research progress on predictive modeling of invasive candidiasis in critically ill patients]. Zhonghua Wei Zhong Bing Ji Jiu Yi Xue. 2026. PMID: 42563683. doi:10.3760/cma.j.cn121430-20260214-00092

供执业医师抢救与诊疗参考,不替代临床判断;剂量请结合体重、肝肾功能和本院药品规格复核,以指南原文为准。