This study investigates the prognostic value of cervical length (CL) measured by serial transperineal ultrasound at various gestational periods (20–23, 24–27, and 28–32 weeks) for predicting the gestational age at preterm birth in twin pregnancies. Participants underwent serial transperineal ultrasound measurements to assess CL. This retrospective cohort study included 231 cases of spontaneous preterm birth in twin pregnancies delivered at our hospital (June 2015 - June 2020), categorized into early preterm (< 28 weeks, n = 49), moderate preterm (28 to < 34 weeks, n = 45), and late preterm (34 to < 37 weeks, n = 137) groups. A control group of 231 full-term twin pregnancies was selected to achieve a 1:1 case-control ratio. Logistic regression analyzed the correlation between CL and preterm birth gestational age, adjusting for confounders. The prognostic performance of CL was evaluated by determining optimal cut-off points from receiver operating characteristic (ROC) curve analysis. Shorter CL at all measurement intervals was significantly associated with an increased risk of delivering at an earlier gestational age. All confidence intervals (CI) reported are 95%. Consistent with the study’s primary objective, CL measured at 24–27 weeks showed the strongest predictive value for early preterm birth (AUC: 0.927; CI: 0.885–0.999), with an optimal cutoff of 2.21 cm. For other subgroups, prediction models combining CL with monochorionicity or mode of conception (assisted reproductive technology) demonstrated better prognostic performance than CL alone. Specifically, CL at 28–32 weeks combined with ART showed the highest performance for predicting moderate preterm delivery (AUC: 0.970; 95% CI: 0.937–0.999). Serial transperineal CL measurements are associated with the risk of spontaneous preterm birth in twin pregnancies. Its prognostic value appears strongest for early preterm birth. For moderate and late preterm birth, combining CL with clinical risk factors such as chorionicity and conception mode may improve risk stratification. However, these findings are exploratory, as the models were not validated in an independent dataset.
Zhang et al. (Thu,) studied this question.