支持向量机算法预测腹部术后病人死亡风险模型的建立及验证

目的 通过支持向量机算法，建立预测腹部手术病人术后28 d的死亡风险模型。 方法收集2015年7月至2017年6月期间行腹部手术的病人的术前一般情况、术前访视情况、实验室检查等指标，基于支持向量机算法建立并验证预测腹部术后的死亡风险模型，并与传统logistic回归模型比较，评价支持向量机模型的工作性能。 结果共纳入手术病人1 512例，其中男911例(60.25%%)，女601例(39.75%)。训练集和测试集中，死亡组的死亡预测概率高于存活组(P < 0.01)。训练集中，支持向量机模型的ROC曲线下面积高于logistic回归模型，但差异无统计学意义(0.97 vs 0.95, P>0.05)。验证集中，支持向量机的ROC曲线下面积高于logistic回归模型(0.98 vs 0.91, P < 0.05)。支持向量机模型的敏感性(训练集68.57% vs 62.86%，验证集79.78% vs 77.78%)和阳性预测值(训练集80.00% vs 65.75%，验证集83.33% vs 77.13%)优于传统logistic回归模型。 结论支持向量机模型能够准确预测腹部手术病人28 d死亡风险，其工作性能强于传统的logistic回归模型。

Development and validation of the support vector machine model for predicting the risk of death in patients after abdominal surgery

ObjectiveTo develop a model for predicting the 28-day death risk in patients with abdominal surgery using support vector machine algorithm. MethodsThe preoperative general conditions, preoperative visits, laboratory tests and other indicators of patients treated with abdominal surgery from July 2015 to June 2017 were collected.The logistic regression model was compared to evaluate the performance of support vector machine model. ResultsA total of 1 512 surgical patients were included, including 911 males(60.25%) and 601 females(39.75%).In both of the training set and validation set, the predicted probability of death in death group was significantly higher than that in survival group(P < 0.01).In the training set, the area under ROC curve of support vector machine model was larger compared with the logistic regression model, but the difference of which was not statistically significant(0.97 vs 0.95, P>0.05).In the validation set, the area under the ROC curve of support vector machine was significantly higher than that of logistic regression model(0.98 vs 0.91, P < 0.05).The sensitivity(training set 68.57% vs 62.86%, validation set 79.78% vs 77.78%) and positive predictive value(training set 80.00% vs 65.75%, validation set 83.33% vs 77.13%) of support vector machine model were better than those of traditional logistic regression model. ConclusionsThe support vector machine model can accurately predict the risk of 28-day death in patients with abdominal surgery, and its performance is better than that of traditional logistic regression model.