预测微通道经皮肾镜碎石取石术治疗鹿角形肾结石术后严重出血的贝叶斯网络模型的建立与验证
Establishment and validation of Bayesian network model:for predicting the risk of severe bleeding after microchannel percutaneous nephrolithotripsy in the treatment of staghorn renal calculi
宋伟航 1李泽宇 1张春锋 1吴春磊1
作者信息
- 1. 新乡医学院第一附属医院泌尿外科,河南新乡 453100
- 折叠
摘要
目的 探讨微通道经皮肾镜碎石取石术(M-PCNL)治疗鹿角形肾结石术后严重出血的危险因素,以此构建预测术后严重出血的贝叶斯网络模型.方法 回顾性分析2020年1月-2022年1月于新乡医学院第一附属医院由同等资历术者行M-PCNL的160例鹿角形肾结石患者,采用计算机产生的随机数法以3∶1的比例将患者分为建模组(120例)和验证组(40例),将建模组患者按照术后出血情况分为严重出血组(38例)和非严重出血组(82例),比较两组患者一般资料,分析患者术后严重出血的独立危险因素,采用R软件构建贝叶斯网络模型,Netica软件进行贝叶斯网络模型推理预测;采用受试者工作特征(ROC)曲线对模型进行评价.结果 对建模组进行多因素logistic回归分析,初步筛选出肾功能不全(OR:2.845,95％CI:1.563～6.515)、结石最大径≥2 cm(OR:2.063,95％CI:1.824～4.555)、手术时间≥90 min(OR:3.632,95％CI:2.365～7.114)、一期手术(OR:2.321,95％CI:1.874～6.332)、多通道取石(OR:1.842,95％CI:1.366～3.687)是影响患者术后严重出血的独立危险因素(P＜0.05).结合多因素logistic回归分析的结果,以肾功能不全、结石最大径、通道数目、手术时间、手术分期和术后严重出血建立贝叶斯网络模型.利用建模组和验证组对贝叶斯网络模型进行内、外部评价.建模组的AUC为0.879(95％CI:0.804～0.931,P＜0.001),灵敏度、特异度分别为87.68％和89.63％;验证组的AUC为0.875(95％CI:0.818～0.908,P＜0.001),灵敏度、特异度分别为87.55％和89.40％.模型区分度良好.结论 肾功能不全、结石最大径≥ 2 cm、手术时间≥ 90 min、一期手术、多通道取石是影响患者M-PCNL后严重出血的危险因素.采用贝叶斯网络构建的鹿角形肾结石术后严重出血的预测模型具有较好的预测能力,并且能够更加直观地描述疾病与各因素间复杂的网络风险机制.
Abstract
Objective To explore the risk factors of severe postoperative hemorrhage in patients with staghorn renal calculi treated with mini-percutaneous nephrolithotomy(M-PCNL),and to construct a Bayesian network model to predict postoperative hemorrhage.Methods A retrospective analysis was conducted on 160 patients with staghorn renal calculi who were treated with M-PCNL by surgeons with equivalent qualifications at the First Affiliated Hospital of Xinxiang Medical College during Jan.2020 and Jan.2022.A computer-generated random number method was used to divide them into a modeling group(120 cases)and a validation group(40 cases).Patients in the modeling group were divided into severe bleeding group(38 cases)and non-severe bleeding group(82 cases).The general information of the two groups was compared,and the independent risk factors of severe postoperative hemorrhage were analyzed.A Bayesian network model was constructed using R software,the inference prediction was conducted using Netica software,and the performance of the model was evaluated with receiver operating characteristic(ROC)curve.Results Multivariate logistic regression analysis showed that renal insufficiency(OR:2.845,95％CI:1.563-6.515),mixmum diameter of stones ≥2 cm(OR:2.063,95％CI:1.824-4.555),operation time ≥ 90 minutes(OR:3.632,95％CI:2.365-7.11),one-stage operation(OR:2.321,95％CI:1.874-6.332),and multi-channel stone removal(OR:1.842,95％CI:1.366-3.687)were independent risk factors of postoperative severe hemorrhage(P＜0.05).Based on the above parameters,a Bayesian network model was established,which was then evaluated with the modeling and validation groups internally and externally.The AUC of the modeling group was 0.879(95％CI:0.804-0.931,P＜0.001),with sensitivity and specificity being 87.68％and 89.63％,respectively.The AUC of the validation group was 0.875(95％CI:0.818-0.908,P＜0.001),with sensitivity and specificity being 87.55％and 89.40％,respectively.The model showed good discrimination.Conclusion Renal dysfunction,mixmum diameter of stones ≥2 cm,operation time ≥90 minutes,one-stage operation,and multi-channel stone removal are risk factors of severe hemorrhage in patients after M-PCNL.The prediction model has good predictive ability and can effectively describe the complex mechanism between diseases and risk factors.
关键词
微通道经皮肾镜碎石取石术/鹿角形肾结石/贝叶斯网络模型/严重出血/肾功能Key words
mini-percutaneous nephrolithotomy/staghorn renal calculi/Bayesian network model/severe hemorrhage/renal function引用本文复制引用
出版年
2024
NETL
NSTL
万方数据