To investigate the early pregnancy-related influencing factors of gestational diabetes mellitus (GDM), as well as the clinical value of building GDM prediction model based on the glycolipids-related biochemical indexes in early pregnancy and demographic information using four machine learning algorithms.

A total of 6 257 pregnant women with gestational age of 10 to 13^{+ 6} gestational weeks who had their first prenatal examinations in Fujian Maternity and Child Health Hospital from December 2021 to December 2022 were selected for the study. The pregnant women were categorized into the GDM group (n=1 499, GDM pregnant women) and the non-GDM group (n=4 758, non-GDM pregnant women) according to whether or not they were diagnosed with GDM at 24 to 27^{+ 6} gestational weeks by retrospective analysis. Early pregnancy-related influencing factors on the development of GDM in pregnant women were analyzed using multivariate unconditional logistic regression analysis. Based on the biochemical indexes related to glycolipids in early pregnancy and demographic information in pregnant women (8 variables), four machine learning algorithms, namely, decision tree (DT), logistic regression (LR), random forest (RF), and extreme gradient boosting (XGB) were used to build GDM prediction models, and ten-fold cross-validation was used to assess the performance of each model, and area under curve (AUC) of the receiver operating characteristic (ROC) curve among the GDM prediction models constructed by the four algorithms were compared. The study was approved by the Ethics Committee of Fujian Maternity and Child Health Hospital (Approval No. 2021KRD018). All pregnant women had signed the informed consent forms for clinical research.

①The results of multivariate unconditional logistic regression analysis showed that pregnant women with advanced age (delivery age≥35 years) (OR=1.95, 95%CI: 1.70-2.24, P<0.001), with pre-pregnancy body mass index (BMI) ≥ 18.5-24.0 kg/m² (OR=1.32, 95%CI: 1.11-1.58, P=0.002), pre-pregnancy BMI ≥ 24.0-28.0 kg/m² (OR=2.17, 95%CI: 1.73-2.73, P<0.001), pre-pregnancy BMI ≥28.0 kg/m² (OR=2.53, 95%CI: 1.70-3.78, P<0.001), elevated serum apolipoprotein (Apo) B levels during early pregnancy (OR=3.06, 95%CI: 2.14-4.37, P<0.001), and increased serum fasting glucose (FPG) concentration in early pregnancy (OR=2.08, 95%CI: 1.79-2.41, P<0.001) were all independent early pregnancy-related risk factors for the development of GDM in pregnant women. ②According to the magnitude of the eigenvalue in the classification results of the 4 classifiers, the results of GDM prediction model constructed using 8 variables of maternal age, degree of education and pre-pregnancy BMI, serum levels of total cholesterol (TC), triglyceride (TG), ApoA1, ApoB, and FPG during early pregnancy showed that the AUC of the GDM prediction models built by the 4 algorithms, namely, DT, LR, RF, and XGB, were 0.645 (95%CI: 0.591-0.698), 0.699 (95%CI: 0.641-0.749), 0.672 (95%CI: 0.621-0.772), and 0.597 (95%CI: 0.553-0.663), respectively, and the AUC of the LR algorithm was greater than that of the XGB algorithm, and the difference was statistically significant (Z=2.38, P=0.017), and there was no significant difference in AUC of pairwise comparison of the rest of the algorithms (P>0.05). ③The ten-fold cross-validation results showed that the average AUC of the GDM prediction models constructed by the four algorithms, DT, LR, RF, and XGB, were 0.586±0.025, 0.661±0.020, 0.632±0.023, and 0.576±0.019, respectively.

Based on the biochemical indexes related to glycolipids in early pregnancy and demographic data, GDM prediction models constructed with LR and RF algorithms, which has a certain predictive value for GDM, and helps to screen the high-risk group of GDM at early stage, and to provide clinical interventions when necessary to reduce GDM-related adverse pregnancy outcome of mother and fetus.