妊娠糖尿病发病风险交互式列线图预测模型的构建和验证

Establishment and validation of interactive nomogram model for predicting the risk of gestational diabetes mellitus

目的 探讨妊娠糖尿病(gestational diabetes mellitus, GDM)发病影响因素,构建并验证GDM交互式列线图风险预测模型。方法 回顾性收集2021年1—12月入院登记的2 279例孕产妇临床数据,按4∶1随机划分为训练集和验证集。采用χ^(2)检验、Wilcoxon秩和检验分析GDM发病风险的影响因素,使用LASSO logistic回归筛选预测变量构建列线图模型,并进行模型验证,采用ROC曲线、校准曲线、临床决策曲线和临床影响曲线对模型进行评价。结果 训练集共纳入1 823例孕产妇,GDM患病率为16.84%。通过多因素LASSO logistic回归分析筛选GDM风险预测因素,其中妊娠年龄(OR=1.094,95%CI:1.055~1.135)、孕前BMI(OR=1.177,95%CI:1.126~1.230)、农村户籍(OR=0.242,95%CI:0.073~0.805)、本科及以上学历(OR=2.003,95%CI:1.291~3.106)、糖尿病家族史(OR=6.516,95%CI:4.034~10.525)、初产(OR=0.600,95%CI:0.430~0.837)、流产(OR=1.905,95%CI:1.642~2.704)与GDM存在独立关联。基于上述因素构建GDM交互式列线图预测模型,训练集和验证集ROC曲线下面积(95%CI)分别为0.747(0.717~0.777)和0.741(0.679~0.802)。校准曲线和临床影响曲线均表明模型预测值和实际值具有较高的一致性。临床决策曲线显示,GDM阈值概率界于0.10~0.65时,模型可以获得最大的净收益。结论 GDM列线图预测模型具有较高的区分度、校准度和临床适用性,有助于及早识别GDM高危人群并采取预防措施。

Objective To explore the factors affecting the onset of gestational diabetes mellitus(GDM),and to establish and validate the interactive nomogram model for the prediction of risk of GDM.Methods We retrospectively collected the clinical data about 2,279 pregnant women admitted to Zhengzhou Central Hospital Affiliated to Zhengzhou University from January to December 2021,and the dataset was randomly divided into the training set(80%)and the validation set(20%).χ^(2) test and Wilcoxon rank sum test were used to analyze the factors affecting the risk of onset of GDM.Least absolute shrinkage and selection operator(LASSO)logistic regression analysis was performed to select the predictive variables for the establishment of interactive nomogram model,and then the model was validated.Receiver operating characteristic curve(ROC),calibration curve,clinical decision curve analysis(DCA)and clinical impact curve(CIC)were employed to evaluate the performance of the model.Results A total of 1,823 pregnant womenwere enrolled into the training set,and the prevalence rate of GDM was 16.84%.Multivariate LASSO logistic regression analysis was conducted to identify the predictors of risk of GDM,among which pregnancy age(OR=1.094,95%CI:1.055-1.135),pre-pregnancy BMI(OR=1.177,95%CI:1.126-1.230),rural registered residence(OR=0.242,95%CI:0.073-0.805),bachelor degree or above(OR=2.003,95%CI:1.291-3.106),history of diabetes mellitus(OR=6.516,95%CI:4.034-10.525),primiparity(OR=0.600,95%CI:0.430-0.837)and abortion(OR=1.905,95%CI:1.642-2.704)were independently associated with GDM.The interactive nomogram model was constructed based on the above-mentioned factors,and the area under ROC(95%confidence interval)in the training set and the validation set was 0.747(0.717-0.777)and 0.741(0.679-0.802)respectively.Calibration curve and CIC revealed that the values predicted by the established model were in good agreement with the actual values.DCA indicated that the maximum net benefit value would be achieved by the nomogram modelwhen the threshold probability intervals for GDM were 0.10-0.65.Conclusion The nomogram predictive model for GDM has higher discrimination,calibration and clinical applicability,and is conducive to early identifying the high-risk population for GDM and taking preventive measures.