Results
(1) Development cohort
During the study period, a total of 1,356 twin pregnant women who met the inclusion criteria and delivered at SNUH, were assigned as the development cohort. The indications for IOL were maternal request (n=517), prolonged pregnancy (n=484), suspected intrauterine growth retardation (n=147), rupture of membrane (n=110), preeclampsia (n=60), gestational diabetes (n=13), suspected large for gestational age fetus (n=9), chronic hypertension (n=6), oligohydramnios (n=5) and other reasons for maternal medical condition (n=5).
Of these twin pregnant women, 17.0% (n=230) underwent cesarean delivery and 83.0% (n=1126) underwent vaginal delivery. The reasons for cesarean delivery were failure to progress (n=81), failed induction (n=64), maternal condition (such as medical reason or request, n=54), and non-reassuring fetal status (n=31). Failed induction was defined as failure to give birth after more than three days of serial induction without rupture of the membrane.23,24
Table 1 compares the clinical variables according to the final mode of delivery. The gestational age at labor induction and presentation of the second twin did not differ between the two groups. However, patients who underwent cesarean delivery were older, had shorter height, higher pregestational body mass index, less effaced and dilated cervix, and heavier birth weight of twins. In addition, patients who underwent cesarean delivery were more likely to become pregnant after assisted reproductive techniques and had a higher frequency of nulliparous and dichorionic twins.
To find the best prediction model for cesarean delivery with these clinical variables, we conducted a three-fold CV with 100 repetitions. The study population in the development cohort was randomly divided into a training set and a test set with a ratio of 2:1, and the prediction model was developed using logistic regression analysis in the training set, and the AUROC was calculated in the test set. Table S1 shows the mean AUROC for each prediction model. Among the possible models, the prediction model including maternal age, parity, maternal height, cervical effacement, and total birth weight of twins, had the highest average AUROC value in the test set and was selected as the best prediction model [AUROC, 0.742 (95% CI 0.700-0.785) in the training set and 0.733 (95% CI, 0.671-0.794) in the test set]. Table 2 summarizes the odds ratios of each variable in the best prediction model in the SNUH development cohort. In addition, a nomogram for predicting the risk of cesarean delivery after IOL in twin pregnancy (Figure 1) and a web-based predictive calculator (Figure 2) was developed.
(2) Validation cohort.
In SNUBH, a total of 347 twin pregnant women who met the inclusion criteria and delivered between 2005 and 2018 were assigned as the external validation cohort. In this validation cohort, 26.5% of women (n=92) underwent cesarean delivery. External validation of the prediction model for cesarean delivery derived from the SNUH cohort was performed on this cohort. The AUROC in this cohort was 0.714 (95% CI, 0.650-0.777), which was similar to that of the development cohort (Figure 3).