EP17.26: Spontaneous preterm birth prediction using 3D ultrasound assessment of previous advanced labour Caesarean scar

To evaluate the performance of a multiparameter test including 3D ultrasound scar features to predict spontaneous preterm birth (SPTB) in women with previous advanced labour Caesarean delivery. This single-centre prospective study at University College London Hospital, UK (2019-2023) assessed cervical length (CL) and Caesarean scar characteristics at 14-24 weeks’ gestation via transvaginal ultrasound in pregnant women with previous Caesarean delivery at ≥8cm dilatation. 3D ultrasound volumes were analysed using a validated method. PTB preventive interventions (cervical cerclage and/or progesterone) were indicated on clinical history and/or short CL (<25mm). Primary and secondary outcomes were prediction of SPTB <37 weeks and development of short CL, respectively. We constructed predictive models using multivariable logistic regression analysis with calculation of the area under the receiver operating characteristics curve (AUC) and sensitivity at a fixed false-positive rate of 25%. The SPTB rate was 8% (13/166); 21% (35/166) women received preventive interventions. Caesarean scar was seen in 88% of cases with a niche (an indentation at scar site ≥2mm) present in 43%. The median scar distance from the internal cervical os (IO) was 3.5mm (range 0.7 to 9.4mm) below the IO in the SPTB group, versus 3.2mm (range 0.0 to 6.3mm) above the IO in the term birth group (p = 0.005). Prediction of SPTB and short CL were improved by including clinical history, 2D niche assessment (length, depth, width) and scar distance to IO (AUC 0.85, 95% CI 0.72, 0.96 and AUC 0.84, 95% CI 0.73, 0.93 respectively). The addition of 3D scar features (niche and cervical volume) only marginally improved the sensitivity. Pregnant women at risk of developing short CL or SPTB after Caesarean delivery could be predicted using clinical information with scar ultrasound features. Integrating 3D scar features did not appear to improve the model performance.