肝癌术中输血风险评分模型的建立

摘要

目的 建立肝癌术中输血风险评分(TRS)模型,预测术中输血概率.方法 回顾性分析352例肝癌手术患者的临床资料.采用x2检验对输血组和未输血组的指标进行单因素分析,将P＜0.1的因子进一步纳入多因素Logistic回归分析.用多因素分析P＜0.05的因子构建肝癌TRS模型,采用ROC曲线评价模型的性能.结果 352例肝癌患者术中输血92例,输血率为26.1％.单因素分析显示天冬氨酸转氨酶(AST)、谷氨酰基转移酶、白蛋白(Alb)、凝血酶原时间、肿瘤分期、肿瘤大小、肿瘤位置、侵犯或紧邻大血管、切肝段数和手术时长对术中输血有影响.多因素分析显示AST＞40 U/L、Alb＜ 35 g/L、肿瘤≥5 cm、肿瘤位于右叶、侵犯或紧邻大血管、手术时长＞3h是术中输血的独立危险因素.设定Alb＜ 35 g/L为2分,其他因子均为1分,计算每例患者的总分值即为TRS.TRS在0～6分范围对应的输血概率依次为0、3.8％、20.0％、35.7％、53.3％、84.6％、100％,人均输血量依次为0、0.2、0.8、1.7、3.7、5.7、8.0U,TRS与输血概率和人均输血量显著正相关(P＜0.05).TRS模型的ROC曲线下面积为0.795.以3分为是否输血的临界值,敏感度73.9％,特异度70.4％.结论 TRS模型对肝癌术中输血具有较好的预测能力,可为肝癌手术合理备血提供指导.%Objective To establish a risk scoring model to predict the likelihood of intraoperative blood transfusion in patients with hepatocellular carcinoma (HCC) undergoing liver surgery.Methods The clinical data of 352 patients with HCC who underwent liver surgery were retrospectively analyzed.The data for transfused and non-transfused patients were compared using univariate analysis.Significant variables were then entered into a multiple logistic regression model.Variables that remained in the final logistic regression analysis were used to build a transfusion risk scoring (TRS) model.The predictive value of this model was validated using ROC curves.Results Of 352 patients, 92 patients (26.1％) received intraoperative blood transfusion.Univariate analysis showed the following ten factors to be associated with blood transfusion: aspartate aminotransferase (AST), γ-glutamyl transpeptidase, albumin, prothrombin time, tumor stage,tumor size, location of tumor, great vessels invasion, number of liver segments resected and operation time.Logistic regression analysis identified six predictive factors of blood transfusion: AST ＞ 40 U/L, albumin＜35 g/L, tumor size≥5 cm, tumor in right liver, great vessels invasion and operation time ＞3 h.To calculate the total score of TRS for each patient, "albumin ＜ 35 g/L" was assigned two points, and the other factors were assigned one point.The transfusion probabilities corresponding to the different TRS between 0 to 6 were 0, 3.8％, 20.0％, 35.7％, 53.3％, 84.6％, 100％, respectively;and the corresponding amount of blood transfusion per patient was 0, 0.2, 0.8, 1.7, 3.7, 5.7, 8.0 U, respectively.TRS was significantly and positively correlated to the transfusion probability and the amount of blood transfusion per patient (P ＜ 0.05).The area under the ROC curve for the TRS model was 0.795.When a score of 3 for TRS was set as the cut-off value for transfusion, the sensitivity was 73.9％, and the specificity was 70.4％.Conclusion The TRS model has a good predictive ability for intraoperative blood transfusion, thus it can serve as an indicator for preparation of intraoperative blood transfusion.