血液透析患者导管相关性血流感染风险预测模型的系统评价

Risk predictive models for catheter-associated bloodstream infection in hemodialysis patients:a systematic review

目的 系统评价血液透析患者导管相关性血流感染风险预测模型,为临床选择合适的预测模型或开发新模型提供参考依据。方法 检索CINAHL、PubMed、Web of Science、Cochrane Library、Embase、万方数据库、中国知网、维普网、中国生物医学文献数据库中有关血液透析患者导管相关性感染风险预测模型的文献,检索时限均为建库至2023年10月1日。由2名研究者独立筛选文献和提取数据,并采用预测模型研究的偏倚风险评估工具(prediction model risk of bias assessment tool,PROBAST)分析纳入文献的偏倚风险和适用性。结果 共纳入7篇文献。4篇关于血液透析患者长期导管相关性感染的风险预测模型,1篇血液透析患者临时导管相关性感染的风险预测模型,2篇关于血液透析患者中心静脉导管相关性感染的风险预测模型。共5篇文献将血清白蛋白水平、导管留滞时间和合并糖尿病都纳为血液透析患者导管相关性感染的主要预测因素,3篇文献将患者的年龄当做血液透析患者导管相关性感染的主要预测因子,置管部位和手卫生预测因素在2篇预测模型中被纳为预测因子。预测模型的受试者工作特征曲线下面积范围为0.734~0.889。7篇文献中共有4个模型进行了Hosmer-Lemeshow校准度分析,模型的呈现形式也大多采用风险评分公式。所有研究适用性较好,但存在方法学缺陷(如缺失数据处理不当、未采用恰当方法选择变量、未提及盲法等)导致较高的偏倚风险。结论 血液透析患者导管相关性血流感染风险预测模型预测性能较好,但存在较高的偏倚风险,不能直接应用于临床实践。未来应选择已有模型进行深入广泛的验证或开展大样本、多人群的前瞻性研究以开发预测性能优良、使用简便的预测模型。

Objective To systematically evaluate the risk prediction model for catheter-related bloodstream infection in hemodialysis patients and to provide references for selecting appropriate prediction models or developing new models.Methods We searched related literature in CINAHL,PubMed,Web of Science,Cochrane Library,Embase,Wanfang database,CNKI,VIP website,China and Chinese biomedical literature database from establishing the database to October 1,2023.Two investigators independently screened the literature,extracted data,and analyzed the risk of bias and applicability of included literature using the prediction model risk of bias assessment tool(PROBAST).Results A total of seven articles were included.Age,combined diabetes,serum albumin level,hand hygiene and catheter retention time were the major predictors for catheter-related bloodstream infection.The area under the curve was 0.734 to 0.889,from which four models were calibrated.All studies were well applicable but had a risk of bias.Conclusion The risk prediction models of catheter-related bloodstream infection in hemodialysis patients have better prediction performance,but with higher risks of bias caused by methodological defects,such as improper processing of missing data,variable selection without appropriate methods,no mention of blindness,and others.Therefore,they cannot be directly applied in clinical practice yet.In the future,the existed models should be verified in-depth and extensively by prospective studies using large sample and multiple population to develop prediction models with excellent predictive performance and simple use.