广西医学

目的基于重症监护医学信息数据库（MIMIC-Ⅳ）构建重症胰腺炎（SP）患者院内死亡风险的预测模型。方法在MIMIC-Ⅳ中筛选SP患者，提取其临床资料。根据患者住院期间是否死亡分为存活组和死亡组，比较存活组和死亡组患者的临床资料。按照8 ∶2的比例将SP患者随机分成训练集和测试集，基于训练集的资料，采用Logistic回归模型和支持向量机（SVM）算法构建SP患者院内死亡风险的预测模型，再基于测试集的资料，绘制受试者工作特征（ROC）曲线评价两个模型的预测效能。结果死亡组和存活组患者的年龄、红细胞分布宽度（RDW）、阴离子间隙、HDL-C差异有统计学意义（P＜0.05）。Logistic回归分析结果显示，年龄、RDW、阴离子间隙是SP患者院内死亡的独立危险因素，HDL-C是SP患者院内死亡的独立保护因素（P＜0.05）。SP患者院内死亡风险预测模型的回归方程为P=ea1+ea，其中P为SP患者院内死亡的概率，e为自然常数，a=-6.264+0.033×年龄+0.061×RDW+0.091×阴离子间隙-0.019×HDL-C。当核函数为Linear函数、松弛变量为0.1、容错率为0.42时，SVM模型预测SP患者院内死亡的准确率最高。两个模型预测SP患者院内死亡效能的评价结果显示，SVM模型预测SP患者院内死亡的准确率、灵敏度和ROC曲线下面积高于或大于Logistic回归模型，但差异无统计学意义（P＞0.05），特异度与Logistic回归模型相当。结论年龄、RDW、阴离子间隙是SP患者院内死亡的独立危险因素，HDL-C是其独立保护因素。基于上述因素构建的SVM模型和Logistic回归模型均可有效预测SP患者院内死亡风险，而SVM模型的准确率和灵敏度稍高，预测效能更优。

ObjectiveTo establish a prediction model of in-hospital mortality risk for patients with severe pancreatitis (SP) based on the Medical Information Mart for Intensive Care (MIMIC)-Ⅳ. MethodsSP patients were screened from the MIMIC-Ⅳ,and their clinical data were extracted. Patients were assigned to survival group or death group according to whether patients were dead or not during hospitalization, and the clinical data were compared between the survival group and the death group. SP patients were randomly assigned to training set or validation set according to the ratio of 8 ∶2. On the basis of data in the training set, the Logistic regression model and support vector machine (SVM) algorithm were used to establish a prediction model of in-hospital mortality risk for SP patients; furthermore, based on data in the validation set, the receiver operating characteristic (ROC) curve was drawn to evaluate predictive efficiency of the two models. ResultsThere were statistically significant differences in age, red blood cell distribution width (RDW),anion gap,and HDL-C between the death group and the survival group (P＜0.05). The results of Logistic regression analysis revealed that age, RDW, and anion gap were the independent risk factors for in-hospital mortality in SP patients, and HDL-C was the independent protective factor for in-hospital mortality in SP patients (P＜0.05). The regression equation for in-hospital mortality risk prediction model in SP patients was P=ea1+ea, among which P was the probability of in-hospital mortality in SP patients, and e was the natural constant,a=-6.264+0.033×age+0.061×RDW+0.091×anion gap－0.019×HDL-C.When the kernel function was Linear function, the slack variable was 0.1, and the error-tolerant rate was 0.42, SVM model exerted the highest accuracy for predicting SP patients′ in-hospital mortality. The evaluation results of efficiency of the two models for predicting patients′ in-hospital mortality revealed that accuracy rate, sensitivity, and area under the ROC curve of SVM model for predicting SP patients′ in-hospital mortality were higher or larger than those of Logistic regression model, but no statistically significant difference was found (P＞0.05), whereas specificity was comparable to that of Logistic regression model. ConclusionAge, RDW,and anion gap are the independent risk factors for in-hospital mortality in SP patients,and HDL-C is the independent protective factor for SP patients′ in-hospital mortality. The SVM model and Logistic regression model established based on the aforementioned factors both can effectively predict in-hospital mortality risk for SP patients, whereas the SVM model exhibits slightly higher accuracy rate and sensitivity, and superior predictive efficiency.