胰十二指肠切除术后胰瘘风险预测模型的系统评价与Meta分析

Risk prediction models for pancreatic fistula after pancreaticoduodenectomy:A systematic review and a Meta-analysis

目的系统评价胰十二指肠切除术后胰瘘(POPF)风险预测模型,为临床筛选应用POPF相关风险模型提供参考。方法本研究根据PRISMA指南完成,PROSPERO注册号:CRD42023437672。计算机检索PubMed、Scopus、Embase、Web of Science、Cochrane Library、中国知网、维普网、万方、中华医学期刊全文数据库和中国生物医学文献数据库公开发表的胰十二指肠切除POPF风险预测模型构建的研究文献,检索时限为建库至2024年4月26日。采用PROBAST工具评价文献质量,RevMan 5.4、MedCalc软件进行Meta分析。结果共纳入36篇文献、20119例患者,胰十二指肠切除POPF发生率为7.4%~47.8%。36篇文献中,共构建55个风险预测模型,受试者工作特征曲线下面积(AUC)为0.690~0.952,其中52个模型AUC>0.7。文献质量评价结果均为高偏倚风险和适用性好。采用MedCalc软件对模型预测性能AUC进行统计学分析,合并的AUC为0.833(95%CI:0.808~0.857)。Meta分析显示:BMI、术后第1天引流液淀粉酶、术前血清白蛋白、胰管直径、胰腺质地、脂肪评分、肿瘤位置、失血量、性别、手术时间、主胰管指数、胰腺CT值是POPF的预测因子(P值均<0.05)。结论目前胰十二指肠切除POPF风险预测模型仍处于探索阶段,大部分预测模型的校准方法缺失,缺少外部验证,仅仅采用单因素分析筛选变量,偏倚风险较高,未来还需完善模型构建方法,以开发出预测准确度更高的风险预测模型。

Objective To systematically review the risk prediction models for postoperative pancreatic fistula(POPF)after pancreaticoduodenectomy(PD),and to provide a reference for the clinical screening and application of POPF-related risk models.Methods This study was conducted according to the PRISMA guidelines,with a PROSPERO registration number of CRD42023437672.PubMed,Scopus,Embase,Web of Science,the Cochrane Library,CNKI,VIP,Wanfang Data,China Medical Journal Full-text Database,and CBM were searched for studies on establishing risk prediction models for POPF after PD published up to April 26,2024.The PROBAST tool was used to assess the quality of articles,and RevMan 5.4 and MedCalc were used to perform the Meta-analysis.Results A total of 36 studies were included,involving 20119 in total,and the incidence rate of POPF after PD was 7.4%-47.8%.A total of 55 risk prediction models were established in the 36 articles,with an area under the receiver operating characteristic curve(AUC)of 0.690-0.952,among which 52 models had an AUC of>0.7.The quality assessment of the articles showed high risk of bias and good applicability.MedCalc was used to perform a statistical analysis of AUC values,and the results showed a pooled AUC of 0.833(95%confidence interval:0.808-0.857).The Meta-analysis showed that body mass index,amylase in drainage fluid on the first day after surgery,preoperative serum albumin,pancreatic duct diameter,pancreatic texture,fat score,tumor location,blood loss,sex,time of operation,main pancreatic duct index,and pancreatic CT value were predictive factors for POPF(all P<0.05).Conclusion The risk prediction models for POPF after PD is still in the exploratory stage.There is a lack of calibration methods and internal validation for most prediction models,and only the univariate analysis is used to for the screening of variables,which leads to the high risk of bias.In the future,it is necessary to improve the methods for model establishment,so as to develop risk prediction models with a higher prediction accuracy.