Hypertensive disorders of pregnancy, fetal biometry and umbilical artery doppler for predicting adverse perinatal outcomes in fetal growth restriction: A two-center retrospective model development and internal validation study

Maternal hypertensive disorders, fetal biometry and umbilical artery Doppler may inform risk assessment in pregnancies complicated by fetal growth restriction. We evaluated a parsimonious model combining these three predictors for a composite adverse perinatal outcome. This two-center retrospective cohort study used clinical data collected from January to December 2025 at the Department of Ultrasound Medicine, The First Affiliated Hospital of Chengdu Medical College, and the Third Clinical Medical College, Ningxia Medical University. Multivariable logistic regression combined biometric-equivalent gestational age, a recorded maternal hypertensive-disorder indicator and an umbilical artery Doppler abnormality indicator to predict a record-based composite adverse perinatal outcome. Internal validation used 20 repeats of stratified five-fold cross-validation and 2,000 bootstrap resamples. Discrimination, calibration, Brier score and exploratory decision curves were evaluated, with sensitivity analyses addressing alternative outcome definitions and analysis populations. Fixed-model validation was supplemented by a newly specified nested selection analysis and clinical gestational-age adjustment. The model-development cohort comprised 195 pregnancies, of which 73 had the composite adverse outcome (37.4%). The cross-validated area under the receiver operating characteristic curve was 0.769 (approximate 95% confidence interval, 0.694–0.837), with a Brier score of 0.179, calibration-in-the-large of 0.006 and calibration slope of 0.911. Bootstrap-corrected discrimination, Brier score and calibration slope were 0.778, 0.180 and 0.959, respectively. Discrimination was 0.748 in the temporal-consistency analysis and 0.678 after excluding participants with a positive death/induction indicator. Nested selection yielded an AUC of 0.746. After clinical gestational-age adjustment, the clinical-minus-biometric age discrepancy had an odds ratio of 1.22 per week (95% confidence interval, 0.87–1.71). The three-predictor model showed moderate discrimination for a record-based composite adverse perinatal outcome in a pooled two-center cohort. Performance varied with the analysis population and endpoint composition; growth-specific information independent of assessment timing was not established. Standardized clinical verification and independent external validation remain necessary before clinical use.