目的 使用最小绝对收缩与选择算子(least absolute shrinkage and selection operator,LASSO)回归筛选变量,分别构建Logistic回归模型和决策树模型,旨在探索更为简明、高效的重症急性胰腺炎(severe acute pancreatitis,SAP)早期预测模型,为识别高危人群,指导临床治疗,改善预后提供科学依据.方法 回顾分析2020年11月至2023年9月安徽医科大学第一附属医院本部及高新院区急诊科和消化内科收治的412名急性胰腺炎患者的临床资料,使用LASSO回归筛选出与SAP发生显著相关的影响因素,分别构建多因素Logistic回归模型和决策树模型,以急性胰腺炎严重程度床旁指数(bedside index for severity in acute pancreatitis,BISAP)为参考,比较和评价模型的预测效能.结果 412例研究对象中,SAP发病率为12.14％(n=50),LASSO回归筛选出7个与急性胰腺炎严重程度显著相关的变量,包括:入院呼吸频率、入院疼痛评分、胸腔积液、纤维蛋白降解物、C反应蛋白、血肌酐、血清白蛋白;Logistic回归模型纳入胸腔积液、入院时疼痛评分、血肌酐、白蛋白4项指标.训练集中,模型灵敏度=0.528,特异度=0.984,准确度(95％CI)=0.928(0.892～0.955),Kappa 值=0.606,AUC(95％CI)=0.920(0.862～0.979);测试集中,模型灵敏度=0.643,特异度=0.925,准确度(95％CI)=0.891(0.822～0.941),Kappa 值=0.519,AUC(95％CI)=0.923(0.861～0.985).决策树模型包含3个分支,4个终端节点,显示血肌酐、白蛋白和胸腔积液3个因素可以有效预测SAP发生.训练集中,模型灵敏度=0.500,特异度=0.973,准确度(95％CI)=0.914(0.876～0.944),Kappa 值=0.544,AUC(95％CI)=0.812(0.731～0.894);测试集中,模型灵敏度=0.500,特异度=0.925,准确度(95％CI)=0.875(0.802～0.928),Kappa值=0.412,AUC(95％CI)=0.709(0.565～0.853).Delong检验显示:在训练集中,Logistic回归模型AUC大于决策树模型(P＜0.01)和BISAP评分(P＜0.001),而决策树模型和BISAP评分的AUC差异无统计学意义(P=0.762);在测试集中,Logistic回归模型AUC同样大于决策树模型(P＜0.01)和BISAP评分(P=0.018),而决策树模型的AUC低于BISAP评分(P=0.017).结论 Logistic回归模型和决策树模型对SAP均有较好的预测价值,两种模型可以联合使用,对临床实践具有指导作用.
Establishment and evaluation of early prediction models for severe acute pancreatitis
Objective To explore a simplified and efficient early prediction model for severe acute pancreatitis(SAP)using the least absolute shrinkage and selection operator(LASSO)regression,and to construct both logistic regression and decision tree models.The aim is to identify high-risk individuals,guide clinical treatment,and improve patient outcomes.Methods A retrospective analysis was conducted on the clinical data of 412 patients with acute pancreatitis admitted to the Emergency and Gastroenterology Departments of the First Affiliated Hospital of Anhui Medical University and its High-tech Branch from November 2020 to September 2023.LASSO regression was employed to identify factors significantly associated with SAP,followed by the construction of a multivariate logistic regression model and a decision tree model.The predictive performance of these models was evaluated and compared to the bedside index for severity in acute pancreatitis(BISAP).Results Among the 412 patients,the incidence of SAP was 12.14％(n=50).Seven variables significantly associated with SAP severity were identified by LASSO regression,including respiratory rate at admission,pain score at admission,pleural effusion,fibrin degradation products,C-reactive protein,serum creatinine,and serum albumin.The logistic regression model incorporated four variables:pleural effusion,pain score at admission,serum creatinine,and serum albumin.In the training set,the model demonstrated a sensitivity of 0.528,specificity of 0.984,accuracy(95％CI)of 0.928(0.892-0.955),Kappa value of 0.606,and AUC(95％CI)of 0.920(0.862-0.979).In the testing set,the model showed a sensitivity of 0.643,specificity of 0.925,accuracy(95％CI)of 0.891(0.822-0.941),Kappa value of 0.519,and AUC(95％CI)of 0.923(0.861-0.985).The decision tree model comprised three branches and four terminal nodes,indicating that serum creatinine,serum albumin,and pleural effusion could effectively predict SAP occurrence.In the training set,the decision tree model had a sensitivity of 0.500,specificity of 0.973,accuracy(95％CI)of 0.914(0.876-0.944),Kappa value of 0.544,and AUC(95％CI)of 0.812(0.731-0.894).In the testing set,the model exhibited a sensitivity of 0.500,specificity of 0.925,accuracy(95％CI)of 0.875(0.802-0.928),Kappa value of 0.412,and AUC(95％CI)of 0.709(0.565-0.853).The DeLong test revealed that in the training set,the AUC of the logistic regression model was significantly greater than that of the decision tree model(P＜0.01)and the BISAP score(P＜0.001),while the AUC difference between the decision tree model and the BISAP score was not statistically significant(P=0.762).In the testing set,the AUC of the logistic regression model was again greater than that of the decision tree model(P＜0.01)and the BISAP score(P=0.018),whereas the AUC of the decision tree model was lower than that of the BISAP score(P=0.017).Conclusions Both the logistic regression and decision tree models demonstrate good predictive value for SAP,and their combined use may provide valuable guidance for clinical practice.
Acute pancreatitisLASSO regressionLogistic regressionDecision treePrediction model
万方数据
维普
