Antenatal Prediction of Shoulder Dystocia and Birth Trauma Using Routine Maternal and Ultrasound Variables: Retrospective Cohort Study

ABSTRACT Objective To develop antenatal prediction models for shoulder dystocia and birth trauma using routinely collected maternal and sonographic variables. Design Retrospective cohort study. Setting Single tertiary referral centre in the UK. Population or Sample All singleton term liveborn pregnancies delivered between January 2016 and November 2024 with a third‐trimester ultrasound performed at or beyond 36 weeks' gestation. Methods Multivariable logistic regression was used to develop antenatal prediction models for shoulder dystocia and birth trauma, incorporating maternal characteristics and fetal biometry including abdominal circumference (AC; centile or mm) and estimated fetal weight (EFW; grams or centile). Model performance was assessed using tests for multicollinearity, discrimination (area under the ROC curve, AUC) and calibration. Main Outcome Measures Shoulder dystocia and birth trauma, the latter defined as a composite of shoulder dystocia, postpartum haemorrhage requiring blood transfusion, caesarean delivery at full dilatation, or hypoxic–ischaemic encephalopathy (HIE ≥ 1). Results A total of 24 334 singleton term pregnancies were included; 432 (1.8%) were complicated by shoulder dystocia and 1210 (5.0%) by birth trauma. The model including maternal characteristics and AC centile demonstrated the best discrimination. For shoulder dystocia, the apparent AUC was 0.706 (95% CI 0.682–0.730); the optimism‐corrected AUC after bootstrap validation was 0.699. For birth trauma, the apparent AUC was 0.669 (95% CI 0.654–0.685); the optimism‐corrected AUC was 0.665. At a 10% false‐positive rate, sensitivity was 31.5% for shoulder dystocia and 22.8% for birth trauma, compared with 20.4% and 14.0%, respectively, using EFW ≥ 90th centile. Conclusions Antenatal models combining fetal AC centile with maternal risk factors outperform EFW‐based thresholds currently used in clinical practice. Although discrimination was modest, the model may be useful for antenatal risk stratification and counselling, rather than as a stand‐alone clinical test. Such models may help identify pregnancies at increased risk of delivery‐related complications associated with fetal overgrowth and inform future studies evaluating targeted interventions.