GDM has gained increasing attention due to its hazardous outcomes and long-term adverse effects on mothers and offspring. Early detection and standardized management of GDM are essential to improve maternal and fetal outcomes [6]. Several scholars have established a GDM prediction model in early pregnancy to predict and intervene in the high-risk groups of GDM early, reducing the disease's occurrence and its complications and improving maternal and child outcomes. Sweeting et al. [7] included previous GDM medical history, family history of diabetes, age, race, parity, and BMI into the model, and the AUC was 0.88. When the model incorporated new maternal lipid markers, such as pregnancy-associated proteins, lipocalin-2, and triglycerides, the AUC was 0.91. The new model formed after the addition of new maternal lipid markers in the Sweeting model identified pregnant women at high risk of GDM more accurately than the old model, but lacks external data validation [8]. The prediction model of Teede et al. [9] includes previous GDM medical history, family history of diabetes, maternal age, pre-pregnancy BMI, and race. This model was simple and suitable for clinical application, but its predictive efficiency was low, with an AUC of 0.70. Wang et al. [10] applied four methods to establish a GDM risk prediction model in early pregnancy. The calculation of the scoring model was simple, but the AUC was 0.772, and the prediction performance was poor. The calculation formula of the logistic regression model was complicated but had a high accuracy; the AUC of training and validation sets was 0.799 and 0.834, respectively. Although the machine learning models had a high accuracy, achieving the same in clinical practice was challenging.

Some early pregnancy GDM prediction models have a good prediction performance but have not been widely used in clinical practice. The study of GDM prediction models in China started late, and a prediction model for GDM in the first trimester of pregnancy has not yet been established to provide a valuable preliminary screening tool for the early screening of pregnant women. This retrospective study analyzed the data of 6000 pregnant women. According to the clinical characteristics of pregnant women and laboratory results in the first trimester, a risk prediction model for GDM in the first trimester was established through logistic regression. The model finally included six predictors: age, pre-pregnancy BMI, HbA1c in the first trimester, UA, TG, and HDL-C. The AUC of the modeling cohort was 0.803 (95% CI: 0.788–0.817), with a sensitivity of 72% and a specificity of 73.5%. After substituting the equation into the validation cohort, the AUC was 0.782 (95% CI: 0.759–0.806), the sensitivity was 68.6%, and the specificity was 73.8%. The P values of the HL test for both the modeling and validation cohorts were > 0.05, indicating that the predictive model established in this study had a good fit.

Some studies [11] have shown that the risk of GDM increases linearly with the age of pregnant women. The prevalence of GDM increases with maternal age [12]. Li et al. [13] found that advanced age, pre-pregnancy BMI overweight, and a history of diabetes in first-degree relatives are associated with an increased risk of GDM. In early pregnancy, age and pre-pregnancy BMI are independent risk factors for GDM, and the risk of GDM in overweight/obese women aged ≥ 35 years is 2.45 times that of normal women [14]. Our results were consistent with the above findings that age and pre-pregnancy BMI are independent risk factors for the occurrence and development of GDM. However, the ability of age and pre-pregnancy BMI to predict GDM was low, the AUC was 0.583 and 0.618, respectively, and the sensitivity and specificity were low.

In early pregnancy, high concentrations of TSH and FT3 and lower concentrations of FT4 were associated with an increased risk of GDM; pregnant women with a high FT3/FT4 ratio are more likely to suffer from GDM than normal pregnant women [15, 16]. Moreover, positive anti-peroxidase antibody (TPOAb) was also associated with an increased risk of GDM [14]. A retrospective analysis of 626 subjects [17] showed that elevated UA levels in early pregnancy were positively associated with GDM risk. High UA at 13–18 weeks of gestation is a risk factor for GDM, and in pregnant women ≥ 35-years-old, serum UA has a stronger correlation with GDM [18]. Li et al. [19] showed that high UA levels during 16–18 weeks of gestation were positively and independently associated with an increased risk of GDM, and those in the highest quartile increased the risk by 55.7% compared to the lowest quartile. In the present study, no significant difference was detected in the levels of FT4 and TSH between the two groups of pregnant women. The levels of FT3 and UA in the GDM group were significantly higher than those in the non-GDM group. However, after adjusting age, pre-pregnancy BMI, HbA1c, TG, HDL-C, and other factors in the first trimester, no correlation was established between FT3 level and GDM, while UA level was correlated with GDM and was an independent risk factor for GDM. When the UA in the first trimester was > 226.55 μmol/L, the possibility of pregnant women suffering from GDM was high, and the AUC was 0.693, which had a certain predictive ability.

HbA1c showed the average blood glucose level in the past 3 months. The HbA1c of pregnant women with GDM was significantly higher than that of pregnant women with normoglycemia. Women with higher HbA1c in the first trimester had a high risk of developing GDM [20]. Kattini et al. [21] found that the risk of GDM increased when the HbA1c level was > 5.7%, and all patients with GDM could be identified when the level was > 6.0%. Fasting blood glucose (FPG), OGTT1h blood glucose level, OGTT2h blood glucose level, and HbA1c level in early pregnancy are critical predictors of GDM, among which 1 h blood glucose level has the most significant predictive value [22]. Another study found that [23], the levels of TC and TG were significantly different between the GDM and the non-GDM groups. Cao et al. [24] speculated that compared to the normal pregnant subjects, TG, TC, low-density lipoprotein (LDL) and very low-density lipoprotein (VLDL) in GDM patients were significantly higher. Conversely, the high-density lipoprotein in the GDM group (HDL) concentration was low. In this study, HbA1c, TG, and HDL-C in the first trimester were independent risk factors for GDM, but HDL-C had no independent predictive effect on GDM. The AUCs of HbA1c and TG in the first trimester were 0.722 and 0.692, respectively, and the optimal cutoff points for predicting GDM were 5.05% and 1.53 mmol/L, respectively. Thus, focusing on the glucose and lipid metabolism levels of pregnant women in the first trimester of pregnancy to prevent the occurrence of GDM is imperative.

The occurrence of GDM can be predicted based on a single index; for example, HbA1c in the first trimester, but its sensitivity and specificity are low. However, whether it could predict the occurrence of GDM alone needs to be investigated further. Moreover, the current study found that compared to individual indicators, the risk prediction model established by combining age, pre-pregnancy BMI, and laboratory indicators in the first trimester can increase the AUC from 0.583–0.722 to 0.803; also, the sensitivity and specificity have been improved.

This study mainly used factors that were easy to obtain, identify, and intervene, such as the results of early pregnancy checkups of pregnant women, as predictors, and incorporated thyroid function indicators and UA in the first trimester into the GDM risk prediction model to provide a basis for the identification of high-risk groups for GDM. Nevertheless, the present study has several deficiencies. Herein, only the pregnant women of Qinhuangdao City were included, which could not be used to infer the situation in other regions. The fitting degree of the predictive model was good, but the AUCs of the modeling cohort and the validation cohort were 0.803 (95% CI: 0.788–0.817) and 0.782 (95% CI: 0.759–0.806), respectively, and the predictive power was moderate. Also, the prediction model had not been verified externally, and needs further extrapolation.