Construction and verification of a prediction model for gestational diabetes mellitus complicated with gestational hypertension based on a specialized disease research database

To construct and validate a model for predicting the risk of gestational hypertension disorder (HDP) in pregnant women with gestational diabetes mellitus (GDM) and to explore the nonlinear association between key metabolic indicators and the risk of HDP.

This study was a multicenter retrospective cohort study. Data were derived from the GDM-specific databases of multiple collaborating hospitals and included singleton pregnant women with GDM from May 29, 2018 to October 30, 2025. A total of 1,500 pregnant women were included in the analysis, of which 123 (8.2%) developed HDP. Using Boruta algorithm for feature screening, the optimal breakpoint determined based on restricted cubic spline (RCS) is used to convert continuous variables into binary variables, and a multivariable logistic regression is used to construct a prediction model; Apply RCS and segmented regression to analyze the nonlinear effects and thresholds of key continuous variables. Internal validation was conducted using Bootstrap resampling, and the clinical net benefit of the model was evaluated using decision curve analysis (DCA).

The final prediction model included six variables: pre pregnancy BMI ≥ 25 kg/m ² (OR = 1.92, 95% CI 1.28-2.87), assisted reproductive conception (OR = 1.80, 95% CI 1.07-3.04), history of adverse pregnancy and childbirth (OR = 1.54, 95% CI 0.97-2.47), HOMA-IR ≤ 2.6 (OR = 0.57, 95% CI 0.37-0.88), early pregnancy TG ≥ 2.867 mmol/L (OR = 1.81, 95% CI 0.91-3.59), and age ≥ 35 years (OR = 1.32, 95% CI 0.89-1.94). There is a significant non-linear correlation between HOMA-IR and triglyceride levels in early pregnancy and HDP risk, with thresholds of 2.600 and 2.867 mmol/L, respectively. When the threshold is below, the risk increases sharply with the increase of indicators. After internal validation, the final model had a corrected C-index of 0.644 (AUC = 0.665), with acceptable calibration accuracy. Decision curve analysis shows that when the decision threshold is between 2% and 18%, applying the model to guide clinical intervention can bring net benefits.

We constructed a moderately‑performing HDP risk‑prediction model for GDM‑complicating pregnancies using accessible clinical variables for auxiliary risk stratification, with HOMA‑IR and TG exploratory thresholds needing prospective validation before clinical use.