目的 构建基于支持向量机(Support Vector Machine,SVM)的妊娠期糖尿病(Gestational Diabetes Mellitus,GDM)预测模型。方法 选择2018年1~12月在福建省立医院南院定期产检及分娩的产妇115例,其中妊娠期糖尿病产妇50例为观察组,随机抽取同期正常产妇65人为对照组,收集产妇的一般资料和孕早期(8~12 W)的血常规、凝血功能和生化指标检测资料,将这些变量采用皮尔森相关系数分析,找出纳入预测模型分析的变量。结果 特征变量与GDM的皮尔森相关系数绝对值最大的前5个变量分别是甘油三酯、活化部分凝血活酶时间、抗凝血酶Ⅲ、孕前BMI和孕次,且这五个变量观察组和正常组孕妇之间的差异均有统计学意义(P值均小于0.01),使用SVM算法将这五个变量纳入预测模型分析,对GDM的预测有78.3%的准确率和84.6%的精确率。结论 使用SVM算法对预测早孕期孕妇患有GDM具有重要的临床意义。

Objective To build a prediction model of gestational diabetes mellitus based on support vector machine. Methods A total of 115 pregnant women who had regular checkups and deliveries at Fujian Provincial Hospital South Branch from January to December 2018 were selected, among which 50 cases of pregnant women with diabetes mellitus were selected as the observation group. 65 normal pregnant women in the same period were randomly selected as the control group. The general clinical data, blood routine indexes, coagulation function indexes and biochemistry analysis indexes of early pregnancy (8-12 weeks) were recorded, and these variables were analyzed by Pearson correlation coefficient to identify the variables included in the analysis of the prediction model. Results The top five variables with the largest absolute value of Pearson correlation coefficient between the characteristic variables and GDM were triglyceride, activated partial thromboplastin time, antithrombin Ⅲ, pre pregnancy BMI and pregnancy times, and the differences between the observation group of these five variables and the normal group were statistically significant (P<0.01).These five variables were included in the prediction model analysis by SVM algorithm, and the prediction accuracy of GDM was 78.3% and the precision was 84.6%. Conclusion The use of SVM algorithm had important clinical significance for predicting GDM in early pregnancy.