胰十二指肠切除术后患者胰瘘风险预测模型的系统评价

Risk prediction models for pancreatic fistula in patients after pancreaticoduodenectomy:a systematic review

目的评价胰十二指肠切除术(PD)术后患者胰瘘风险预测模型的偏倚风险与适用性。方法计算机检索Embase、PubMed、Web of Science、CINAHL数据库、Cochrane Library、中国生物医学文献数据库、中国知网、维普数据库、中华医学期刊全文数据库和万方数据库与PD术后胰瘘风险预测模型相关的文献,检索时间为建库至2023年12月31日。预测模型研究的偏倚风险评估工具对模型进行偏倚风险和适用性评价,并对模型中重复率高的预测因子采用RevMan 5.4软件进行荟萃分析。结果共纳入23篇文献,包括30个模型构建,胰瘘发生率为11.4%~41.7%。19篇文献报道了模型构建和(或)验证的区分度指标,受试者工作特征曲线下面积为0.62~0.94,其余4篇文献报告了一致性指数。23篇文献预测性能较好,但整体偏倚风险均为高风险,原因主要集中在研究对象和统计分析领域,包括样本量不足、未选择合适来源的研究对象、未合适处理连续变量、采取单因素分析筛选预测因子等。8篇文献由于纳入研究对象和结果指标与系统评价研究问题不符等原因,整体适用性被评为高风险。荟萃分析结果显示,主胰管扩张(>3 mm)(OR=0.70,95%CI:0.65~0.76)、胰腺质地软(OR=5.18,95%CI:3.90~6.88)、体质量指数≥25 kg/m^(2)(OR=1.12,95%CI:1.07~1.17)均为术后胰瘘的有效预测因子。结论现有PD术后患者胰瘘风险预测模型预测性能良好,但存在一定的偏倚风险且部分模型适用性较差,未来构建预测模型时可关注主胰管扩张、胰腺质地、体质量指数等预测因子。

ObjectiveTo evaluate the risk of bias and applicability of the risk prediction models for pancreatic fistula after pancreaticoduodenectomy(PD).MethodsThe relevant studies were systematically searched from Chinese database(Chinese medical journals database,CNKI,Wanfang,Weipu,Sinomed)and English databases(PubMed,Embase,Web of Science,Cochrane Library,CINAHL Database)and the retrieval time limit was from the establishment of the database to December 31,2023.Based on the risk of bias assessment tool of the predictive model study,the risk of bias and applicability of the model were evaluated,and the predictors with high repetition rate in the model were meta-analyzed by RevMan 5.4 software.ResultsA total of 23 studies involving 30 models were included.The incidence of pancreatic fistula was 11.4%-41.7%.nineteen studies reported the discrimination index of model construction and/or validation,and the area under the curve(AUC)was 0.62-0.94.The other four studies reported the consistency index(c-index).The predictive performance of the 23 studies is good,but there is a high risk of bias.The main reasons are that the sample size is not saturated,the research object from the appropriate source is not selected,the continuous variables are not properly processed,and the predictive factors are screened by single factor analysis.Eight studies were rated as"high risk"for applicability due to inconsistencies between the included study subjects and outcome measures and the systematic review study questions.The meta-analysis results showed that the dilation of main pancreatic duct(>3 mm)(OR=0.70,95%CI:0.65-0.76),soft pancreas texture(OR=5.18,95%CI:3.90-6.88),and body mass index≥25 kg/m^(2)(OR=1.12,95%CI:1.07-1.17)were effective predictors of postoperative pancreatic fistula(POPF).ConclusionThe existing predictive models for pancreatic fistula risk in patients undergoing PD both domestically and internationally have good predictive performance,but there is a certain risk of bias and some models have poor applicability.Predictors such as dilation of main pancreatic duct,soft pancreas texture,and body mass index≥25 kg/m^(2)should be more emphasized in future predictive models.