冠状动脉造影/经皮冠状动脉介入治疗中对比剂肾病风险预测模型的文献分析

The risk prediction models of contrast-induced nephropathy in the coronary angiography/percutaneous coronary intervention population:a literature analysis

目的分析目前国内外针对冠状动脉造影(CAG)或经皮冠状动脉介入治疗(PCI)患者建立的对比剂肾病(CIN)风险预测模型的特征,为早期识别CIN高危人群提供参考。方法检索PubMed、Embase、Cochrane Library、中国知网、万方和维普数据库,检索时间为建库至2020年8月1日。收集有关针对CAG/PCI患者建立的CIN风险预测模型的文献,提取相关数据进行描述性统计分析。结果共检索出488篇文献,最终纳入31个CIN风险预测模型,4个为多中心研究。17个模型由中国研究者建立,14个模型由国外研究者建立。29项研究为logistic回归建模,2项研究通过随机森林法建模。31个模型包含的预测因子主要包括肾功能不全(27个,占87%)、年龄(23个,占74%)、心功能不全(20个,占65%)、糖尿病(15个,占48%)、主动脉内球囊泵(12个,占39%)、对比剂用量(9个,占29%)、低血压(8个,占26%)、贫血(5个,占16%)和高血压(4个,占13%)等。在模型性能方面,仅3个模型(10%)完整报告了训练集和验证集的区分度和校准度,其他模型均缺失部分指标。结论目前CAG/PCI患者CIN风险预测模型数量较多,但模型性能,尤其是校准度信息报告不完整,且各模型的预测因子差异较大,有必要对现有模型进行临床验证,进一步考察其临床适用性。

Objective To summarize the risk prediction models of contrast-induced nephropathy(CIN)in the coronary angiography(CAG)/percutaneous coronary intervention(PCI)population,and provide a reference for early identification of high-risk group. Methods PubMed,Embase,Cochrane Library,CNKI,Wanfang and VIP databases were searched from the date of establishment to August 1,2020. The literatures about risk prediction models of contrast-induced nephropathy in the CAG/PCI population were collected,and the relevant data were extracted for descriptive statistical analysis. Results A totally of 488 articles were retrieved. Finally,31 valid articles were included,of which 4 were multicenter studies. Seventeen models were developed by Chinese researchers and 14 by foreign researchers. Twenty-nine models were established by logistic regression,and 2 were modeled by random forest method. Predictors of 31 models mainly included renal insufficiency(27,87%),age(23,74%),cardiac insufficiency(20,65%),diabetes(15,48%),intra-aortic balloon pump(12,39%),contrast medium volume(9,29%),hypotension(8,26%),anemia(5,16%)and hypertension(4,13%),etc. In terms of model performance,only 3 models(10%)fully reported the discrimination and calibration degree of training set and validation set,while other models missed some indicators. Conclusion There were a relatively large number of CIN risk prediction models for CAG/PCI patients. However,the reporting of model performance,especially the calibration,was incomplete,and the predictors of each model were quite different. It is necessary to verify the existing models and further investigate their clinical applicability.