孕早期基于多因素回归分析构建妊娠期糖尿病预测模型及验证

Construction and Validation of Prediction Model of Gestational Diabetes Mellitus Based on Multivariate Regression Analysis in the First Trimester of Pregnancy

目的探讨孕早期(12~13+6周)妊娠期糖尿病(gestational diabetes mellitus,GDM)发生的危险因素,构建预测模型并验证。方法选取2020年1~12月在上海市第一人民医院定期产检并分娩的单胎孕妇433例,根据孕24~28周75 g葡萄糖耐量试验结果分为GDM组(n=188)与非GDM组(n=245)。采用电化学发光法检测孕早期血清生化指标,离子交换高效液相色谱法测定糖化血红蛋白。采用Logistic回归分析筛选GDM的危险因素并构建预测模型,绘制受试者工作特征(ROC)曲线分析模型对GDM的预测价值。选取2021年1~6月在上海市第一人民医院产检的95例孕妇对模型效能进行验证。结果与非GDM组比较,GDM组孕妇的孕前身体质量指数(body mass index,BMI)(23.41±11.17 kg/m2vs 21.18±2.88 kg/m2),γ-谷氨酰转肽酶(gamma-glutamyl transpeptidase,γ-GGT)(16.61±10.62 U/L vs 14.00±8.35 U/L),三酰甘油(triacylglycerol,TG)(1.90±0.58 mmol/L vs 1.57±0.55 mmol/L),糖化血红蛋白(glycosylated hemoglobin Alc,Hb Alc)(5.25%±0.47%vs 5.07%±0.34%),空腹血糖(fasting blood glucose,FBG)(4.68±0.47 mmol/L vs 4.36±0.36 mmol/L),LDL/HDL(1.53±0.49 vs 1.41±0.50),TG/HDL(2.93±0.59 vs 2.71±0.58)和TC/HDL(1.19±0.49 vs 0.95±0.45)水平升高,高密度脂蛋白-胆固醇(high density lipoprotein-cholesterol,HDL-C)(1.69±0.39 mmol/L vs 1.77±0.41 mmol/L)水平降低,差异均有统计学意义(t=2.613,2.818,5.874,4.582,17.701,2.458,3.815,5.310,-2.187,均P<0.05)。Logistic回归分析显示,孕前BMI,FBG,Hb Alc,TG和TC/HDL均为预测GDM的独立危险因素(Waldχ^(2)=4.48~35.549,均P<0.05),根据筛选出的危险因素建立的预测模型如下:Logit(P)=-20.562+0.085(BMI)+1.921(FBG)+1.57(Hb Alc)+2.248(TG)-2.302(TC/HDL)。该模型预测GDM的曲线下面积为0.800(95%CI:0.757~0.842),最佳截断值为0.352,敏感度和特异度分别为80.00%,66.00%。通过95例孕妇对模型进行验证,发现该模型的敏感度、特异度和准确度分别为84.50%,91.00%和85.30%。结论孕早期(12~13+6周)通过BMI联合FBG,Hb Alc,TG和TC/HDL构建的预测模型对于GDM有较高的预测价值。

Objective To explore the risk factors of gestational diabetes mellitus(GDM)in the first trimester(12~13+6 weeks)of pregnancy,build a prediction model and verify it.Methods 433 singleton pregnant women delivered in the First People’s Hospital of Shanghai from January 2020 to December 2020 were selected.They were divided into GDM group(n=188)pregnant women and non-GDM group(n=245)pregnant women according to a 75g glucose tolerance test results at 24~28 weeks of gestation.The electrochemiluminescence method measured serum biochemical indexes in early pregnancy,and glycosylated hemoglobinAlc was measured by ion exchange high-performance liquid chromatography.Using logistic regression analysis to screen the risk factors of GDM and construct a predictive model,draw the subject’s work characteristic curve to analyze the model’s predictive value.Ninety-five pregnant women who underwent prenatal examinations at Shanghai First People’s Hospital from January 2021 to June 2021 were selected to validate the model’s effectiveness.Results Compared with the non-GDM group,the level of body mass index(BMI)(23.41±11.17 kg/m2 vs 21.18±2.88 kg/m2),gamma-glutamyl transpeptidase(γ-GGT)(16.61±10.62 U/L vs 14.00±8.35 U/L),triacylglycerol(TG)(1.90±0.58 mmol/L vs 1.57±0.55 mmol/L),glycosylated hemoglobinAlc(HbAlc)(5.25%±0.47%vs 5.07±0.34%),fasting blood glucose(FBG)(4.68±0.47 mmol/L vs 4.36±0.36 mmol/L),LDL/HDL(1.53±0.49 vs 1.41±0.50),TG/HDL(2.93±0.59 vs 2.71±0.58),and TC/HDL(1.19±0.49 vs 0.95±0.45)in GDM group was increased,and the level of highdensity lipoprotein-cholesterol(HDL)(1.69±0.39 mmol/L vs 1.77±0.41 mmol/L)was decreased,the differences were statistically significant(t=2.613,2.818,5.874,4.582,17.701,2.458,3.815,5.310,-2.187,all P<0.05).Logistic regression analysis showed that pre-pregnancy BMI,FBG,HbAlc,TG,and TC/HDL were all independent risk factors for predicting gestational diabetes(Waldχ^(2)=4.48~35.549,all P<0.05).The prediction model constructed based on the selected risk factors was as follows:Logit(P)=-20.562+0.085(BMI)+1.921(FBG)+1.57(HbAlc)+2.248(TG)-2.302(TC/HDL).The model predicts that the area under the curve of GDM was 0.800(95%CI:0.757~0.842),the optimal cutoff value was 0.352,and the sensitivity and specificity were 80.00%and 66.00%,respectively.Ninety-five pregnant women validated the model,and its sensitivity,specificity,and accuracy were 84.50%,91.00%,and 85.30%,respectively.Conclusion The prediction model constructed by BMI combined with FBG,HbAlc,TG and TC/HDL in the first trimester(12~13+6 weeks)of pregnancy has a high predictive value for GDM.