3D Placental Bed Perfusion Ultrasound: Development and Internal Validation of Trimester-Specific Prediction Models for Early- and Late-Onset Preeclampsia.

This study aimed to develop trimester-specific and combined prediction models for early-onset preeclampsia (EOPE) and late-onset preeclampsia (LOPE) based on maternal clinical characteristics, uterine artery pulsatility index (UtA-PI), and 3D placental parameters, identify the optimal models, and construct static nomograms. A total of 1602 singleton pregnant women were prospectively enrolled from October 8, 2020 to May 31, 2025. All maternal clinical data were collected and participants were followed up until delivery. UtA-PI and 3D placental ultrasound parameters, including placental volume (PV), placental quotient (PQ), and placental blood flow vascularization index (PBVFI), were measured at 11-14 and 15-18 weeks of gestation, and the placental vascular growth rate (PVGR) was calculated. According to the gestational age at preeclampsia onset, participants were divided into the EOPE group (n = 29), LOPE group (n = 58), and normal control group (n = 1515). Univariable and multivariable Firth's logistic regression analyses were performed to identify independent predictors. Model performance was evaluated using receiver operating characteristic (ROC) curves, bootstrap-based calibration and decision curve analysis (DCA). Static nomograms were constructed based on the optimal combined models. Chronic hypertension was an independent risk factor for both EOPE and LOPE, while elevated placental perfusion parameters exerted a protetive effect. The combined model achieved optimal predictive performance, with AUCs of 0.980 (95% CI 0.968-0.992) for EOPE and 0.827 (95% CI 0.765-0.889) for LOPE. Internal bootstrap validation confirmed acceptable calibration, and DCA demonstrated positive net clinical benefit. Static nomograms established based on the optimal models enable individualized risk assessment for EOPE and LOPE. We developed and internally validated trimester-specific and combined prediction models to assess the risks of EOPE and LOPE, and constructed static nomograms based on the optimal models. The models achieved favorable apparent performance in this single-center cohort. Nevertheless, their generalizability remains unclear. External validation using independent cohorts is required before large-scale application in routine antenatal screening.