Dose-response associations of gestational thyroid autoantibody levels with gestational diabetes mellitus and hypertension and values for predicting pregnancy complications

We aimed to explore the dose–response associations of gestational thyroid autoantibody [thyroid peroxidase antibodies (TPOAb) and thyroglobulin antibodies (TGAb)] levels with gestational diabetes mellitus (GDM) and gestational hypertension (GHTN), and to explore their potential contribution to risk stratification of pregnancy complications in secondary machine-learning analyses. Subjects (1200 in total) were recruited from pregnant women who delivered between January 2020 and June 2024, comprising 800 cases with GDM and/or GHTN (case group) and 400 cases without GDM or GHTN (control group). Multivariate logistic regression and RCS analyses were used to evaluate associations with combined GDM + GHTN. Machine learning models were developed as secondary prediction analyses for the broader outcome of GDM and/or GHTN. Multivariate logistic regression analysis adjusted for thyroid function demonstrated that high gestational TPOAb levels were associated with a 3.31-fold increased risk of GDM combined with GHTN (OR = 3.31, 95% CI 2.06–5.17), while high TGAb levels were associated with a 2.32-fold increased risk (OR = 2.32, 95% CI 1.49–3.58), indicating that these associations were independent of thyroid function. No significant interaction effect with body mass index (BMI) or age was detected ( P >0.05). The areas under the receiver operating characteristic (ROC) curves (AUC) of eXtreme Gradient Boosting (XGBoost) and random forest models for predicting the risk of GDM or GHTN were 0.76 and 0.77, respectively, and the sensitivity (0.67) and negative predictive value (0.84) of random forest model were slightly better than those of XGBoost model. SHapley Additive exPlanations (SHAP) analysis showed that TPOAb and TGAb were main contributing features. Calibration curves and decision curve analysis (DCA) revealed acceptable calibration and potential net benefit within the analytic sample. High gestational TPOAb and TGAb levels were associated with combined GDM + GHTN and showed potential value for risk stratification of overall pregnancy complications in secondary machine learning analyses.