重症监护病房患者多重耐药菌感染风险预测模型的系统评价

Prediction models for multidrug-resistant organism infection in patients in the intensive care unit: a systematic review

摘要:

目的 系统评价重症监护病房（ICU）患者多重耐药菌（MDRO）感染风险预测模型。 方法 检索PubMed、Embase、Web of Science、Cochrane Library、CINAHL、CBM、万方和中国知网等数据库建库至2022年6月ICU MDRO感染风险预测模型相关文献。由2名研究者独立筛选文献、提取资料，并评价偏倚风险和适用性。 结果 共纳入17篇文献，16个模型受试者工作特征曲线下面积均>0.7（0.64~0.94），偏倚风险评估显示纳入模型均存在高偏倚风险，模型适用性较好。模型中出现最多的预测因子包括抗菌药物、机械通气、ICU住院时间、留置导尿管、性别、基础疾病、共病。 结论 现有关于ICU患者MDRO感染风险的预测模型不理想，模型在开发设计、统计分析及报道方面存在一定偏倚。未来应重点关注研究设计的方法学细节和报告的规范性，并通过多中心、大样本量的研究以及进行模型验证与更新，提高模型性能。

Abstract:

Objective To systematically evaluate the prediction models of multidrug-resistant organism (MDRO) infection in patients in the intensive care unit (ICU). Methods Literatures related to the prediction models of MDRO infection in ICU patients were retrieved from PubMed, Embase, Web of Science, Cochrane Library, CINAHL, CBM, Wanfang, and China National Knowledge Infrastructure (CNKI) from the establishment of the databases up to June 2022. Two researchers independently screened the literatures, extracted data, and evaluated the risk of bias and applicability. Results A total of 17 literatures were included, and the area under the receiver operating characteristic curve of all 16 models were >0.7 (0.64-0.94). Risk of bias assessment showed high risk of bias in all included models, but the models were all applicable. The most common identified predictive factors in the models included antimicrobial drugs, mechanical ventilation, length of ICU stay, indwelling urinary catheter, gender, underlying diseases, and comorbidities. Conclusion The existing predictive models for MDRO infection in ICU patients are not ideal, as they exhibit bias in the development, design, statistical analysis, and reporting of the models. In the future, attention should be focused on the methodological details of research design and the standardization of reports. Additionally, larger-scale, multicenter studies and model validation and updates should be con-ducted to improve model performance.