Objective To construct an early-onset preeclampsia (EOSP) prediction model based on contrast-enhanced ultrasound combined with placental growth factor (PlGF) and evaluate its efficacy.

Methods Eighty-six patients with EOSP were selected as case group (strictly defined as those who first presented with symptoms and were diagnosed at a gestational age of less than 34 weeks and had not reached the state of emergency termination of pregnancy at the time of examination). Another 100 healthy late-pregnancy pregnant women during the same period were selected as control group. Pregnant women in both groups underwent placental contrast-enhanced ultrasound examination to obtain perfusion parameters such as arrival time (AT), time to peak (TTP), peak intensity (PI), and area under the curve (AUC). Enzyme-linked immunosorbent assay (ELISA) was used to detect serum PlGF levels. Univariate and multivariate logistic regression analyses were used to screen independent influencing factors, and two combined prediction models were constructed (Model 1: only included potential early indicators TTP, PI, and PlGF; Model 2: included all variables with statistically significant differences). Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, the Hosmer-Lemeshow test, and Bootstrap internal validation (1 000 repeated samplings). The clinical utility of the models was verified through decision curve analysis (DCA).

Results The median time interval from examination to diagnosis in the case group was 4.2 (2.8, 5.6) weeks. The systolic and diastolic blood pressures of pregnant women in the case group were higher than those in the control group, while the fetal biparietal diameter, femoral length, and placental thickness were smaller than those in the control group; the AT and TTP were longer, and the PI and AUC were lower in the case group compared with the control group; the serum PlGF level in the case group was also lower than that in the control group, and all differences above were statistically significant (P < 0.05). The formula for Model 1 was Logit(P)=-4.218+0.758×TTP-0.888×PI-0.954×PlGF. After Bootstrap internal validation, the AUC of Model 1 was 0.897 (95%CI, 0.852 to 0.932), and the AUC of Model 2 after validation was 0.903 (95%CI, 0.860 to 0.938), with no significant difference between the two (P=0.621). Model 1 had good calibration (Hosmer-Lemeshow test: χ²=6.325, P=0.619). Using a predicted probability of 0.35 as the cut-off value, the sensitivity was 87.2%, and the specificity was 89.0%. DCA showed that when the risk threshold was between 0.15 and 0.80, the clinical net benefit of Model 1 was significantly higher than that of the "full-intervention" or "no-intervention" strategies.

Conclusion The prediction model (Model 1) based on placental perfusion parameters (TTP, PI) from contrast-enhanced ultrasound combined with serum PlGF has good discrimination, calibration, and clinical utility. It can effectively provide early warning before the clinical symptoms of EOSP appear, with better efficacy than single indicators, and the early-warning time window is up to 4.2 weeks, providing a reliable basis for early screening, risk stratification, and clinical intervention of EOSP.