Risk of caesarean delivery in labour induction: a systematic review and external validation of predictive models

The CASPAR study protocol. Can cervical stiffness predict successful vaginal delivery after induction of labour? a feasibility, cohort study

BACKGROUND

Induction of labour (IOL) is a common obstetric intervention in the UK, affecting up to 33% of deliveries. IOL aims to achieve a vaginal delivery prior to spontaneous onset of labour to prevent harm from ongoing pregnancy complications and is known to prevent stillbirths and reduce neonatal intensive care unit admissions. However, IOL doesn't come without risk and overall, 20% of mothers having an induction will still require a caesarean section birth and in primiparous mothers this rate is even higher. There is no reliable predictive bedside tool available in clinical practice to predict which patient's undergoing the IOL process will result in a vaginal birth; the fundamental aim of the IOL process. The Bishop's Score (BS) remains in routine clinical practice as the examination tool to assess the cervix prior to IOL, despite it being proven to be ineffective as a predictive tool and largely subjective. This study will assess the use of the Pregnolia System, a new objective antenatal test of cervical stiffness. This study will explore its' potential for pre-induction cervical assessment and indication of delivery outcome following IOL.

METHODS

CASPAR is a feasibility study of term, primiparous women with singleton pregnancies undergoing IOL. Cervical stiffness will be assessed using the Pregnolia System; a novel, non-invasive, licensed, CE-marked, aspiration-based device proven to provide objective, quantitative cervical stiffness measurements represented as the Cervical Stiffness Index (CSI, in mbar). A measurement is obtained by applying the sterile single-use Pregnolia Probe directly to the anterior lip of the cervix, visualised via placement of a speculum. Following informed consent, CASPAR study participants will undergo the Pregnolia System cervical stiffness assessment prior to their IOL process commencing. Participant questionnaires will evaluate the acceptability of this assessment tool in this population. This study will directly compare this novel antenatal test to the current BS for both patient experience of the different cervical assessment tools and for IOL outcome prediction.

DISCUSSION

This feasibility study will explore the use of this novel device in clinical practice for pre-induction cervical assessment and delivery outcome prediction. Our findings will provide novel data that could be instrumental in transforming clinical practice surrounding IOL. Determining recruitment rates and acceptability of this new assessment tool in this population will inform design of a further powered study using the Pregnolia System as the point-of-care, bedside cervical assessment tool within an IOL prediction model.

STUDY REGISTRATION

This study is sponsored by The University of Liverpool and registered at ClinicalTrials.gov, identifier NCT05981469, date of registration 7th July 2023.

Predicting vaginal delivery after labor induction using machine learning: Development of a multivariable prediction model

INTRODUCTION

Induction of labor, often used for pregnancy termination, has globally rising rates, especially in high-income countries where pregnant women present with more comorbidities. Consequently, concerns on a potential rise in cesarean section (CS) rates after induction of labor (IOL) demand for improved counseling on delivery mode within this context.

MATERIAL AND METHODS

We aim to develop a prognostic model for predicting vaginal delivery after labor induction using computational learning. Secondary aims include elaborating a prognostic model for CS due to abnormal fetal heart rate and labor dystocia, and evaluation of these models' feature importance, using maternal clinical predictors at IOL admission. The best performing model was assessed in an independent validation data using the area under the receiver operating curve (AUROC). Internal model validation was performed using 10-fold cross-validation. Feature importance was calculated using SHAP (SHapley Additive exPlanation) values to interpret the importance of influential features. Our main outcome measures were mode of delivery after induction of labor, dichotomized as vaginal or cesarean delivery and CS indications, dichotomized as abnormal fetal heart rate and labor dystocia.

RESULTS

Our sample comprised singleton term pregnant women (n = 2434) referred for IOL to a tertiary Obstetrics center between January 2018 and December 2021. Prediction of vaginal delivery obtained good discrimination in the independent validation data (AUROC = 0.794, 95% CI 0.783-0.805), showing high positive and negative predictive values (PPV and NPV) of 0.752 and 0.793, respectively, high specificity (0.910) and sensitivity (0.766). The CS model showed an AUROC of 0.590 (95% CI 0.565-0.615) and high specificity (0.893). Sensitivity, PPV and NVP values were 0.665, 0.617, and 0.7, respectively. Labor features associated with vaginal delivery were by order of importance: Bishop score, number of previous term deliveries, maternal height, interpregnancy time interval, and previous eutocic delivery.

CONCLUSIONS

This prognostic model produced a 0.794 AUROC for predicting vaginal delivery. This, coupled with knowing the features influencing this outcome, may aid providers in assessing an individual's risk of CS after IOL and provide personalized counseling.

The value of ultrasonographic factors in predicting cesarean following induction

This study aimed to develop and validate a prediction model of cesarean following induction of labor (IOL). A nomogram for the prediction of cesarean following IOL for singleton, cephalic term deliveries was created by comparing combinations of ultrasonographic and nonultrasonographic factors in a retrospective manner using patient data collected from a Chinese hospital between July, 2017 and December, 2023. Model discrimination and calibration were evaluated using the area under the receiver operating characteristic curve (AUROC) and a calibration curve. Subsequently, decision curve analysis (DCA) was conducted to pinpoint the optimal probability threshold for the predictive model to exhibit practical significance for clinical decision-making. A total of 738 women were included. The inclusion of ultrasound factors yielded a higher AUC when combined with nonultrasonographic factors. Of the three ultrasonographic factors analyzed, the most predictive factor for cesarean following IOL was fetal head circumference. After generating a nomogram with eight validated factors, including maternal age, gestational age, height, prior caesarean delivery, previous vaginal delivery, modified Bishop score, body mass index at delivery, and fetal head circumference by ultrasound, the trained and validated AUC values were 0.826 (95% confidence interval 0.786-0.867) and 0.883 (95% confidence interval 0.839-0.926), respectively. Decision curve analysis indicated that the model provided net benefits of between 0% and 80% of the probability threshold, indicating the benefits of using the model to make decisions concerning patients who fall within the identified range of the probability threshold. Our nomogram based on obstetric factors and fetal head circumference as obtained by ultrasound could be used to help counsel women who are considering IOL. The model demonstrates favorable net benefits within a probability threshold range of 0 to 80%.

External validation and updating of the Rossi nomogram for predicting cesarean delivery following induction: is the Bishop score valuable?

OBJECTIVE

This study sought to validate the Rossi nomogram in a Chinese population and then to include the Bishop score to see if it has an effect on the accuracy of the nomogram.

MATERIALS AND METHODS

The Rossi predictive model was applied and externally validated in a retrospective cohort from August 2017 and July 2023 in a Chinese tertiary-level medical center. For the revision and updating of the models, the regression coefficients of all the predictors (except race) were re-estimated and then the cervical Bishop score at the time of induction was added. Each model's performance was measured using the receiver-operating characteristic and calibration plots. Decision curve analysis determined the range of the probability threshold for each prediction model that would be of clinical value.

RESULTS

A total of 721 women met the inclusion criteria, of whom 183 (25.4%) underwent a cesarean delivery. The calibration demonstrated the underestimation of the original model, with an area under the curve (AUC) of 0.789 (95% confidence interval [CI] 0.753-0.825, p < 0.001). After recalibrating the original model, the discriminative performance was improved from 0.789 to 0.803. Moreover, the discriminatory power of the updated model was further improved when the Bishop score at the time of induction was added to the recalibrated multivariable model. Indeed, the updated model demonstrated good calibration and discriminatory power, with an AUC of 0.811. The decision curve analysis indicated that all the models (original, recalibrated, and updated) provided higher net benefits of between 0 and 60% of the probability threshold, which indicates the benefits of using the models to make decisions concerning patients who fall within the identified range of the probability threshold. The net benefits of the updated model were higher than those of the original model and the recalibrated model.

CONCLUSION

The nomogram used to predict cesarean delivery following induction developed by Rossi et al. has been validated in a Chinese population in this study. More specifically, adaptation to a Chinese population by excluding ethnicity and including the Bishop score prior to induction gave rise to better performance. The three models (original, recalibrated, and updated) offer higher net benefits when the probability threshold is between 0 and 60%.

Ensemble learning for fetal ultrasound and maternal-fetal data to predict mode of delivery after labor induction

Providing adequate counseling on mode of delivery after induction of labor (IOL) is of utmost importance. Various AI algorithms have been developed for this purpose, but rely on maternal-fetal data, not including ultrasound (US) imaging. We used retrospectively collected clinical data from 808 subjects submitted to IOL, totaling 2024 US images, to train AI models to predict vaginal delivery (VD) and cesarean section (CS) outcomes after IOL. The best overall model used only clinical data (F1-score: 0.736; positive predictive value (PPV): 0.734). The imaging models employed fetal head, abdomen and femur US images, showing limited discriminative results. The best model used femur images (F1-score: 0.594; PPV: 0.580). Consequently, we constructed ensemble models to test whether US imaging could enhance the clinical data model. The best ensemble model included clinical data and US femur images (F1-score: 0.689; PPV: 0.693), presenting a false positive and false negative interesting trade-off. The model accurately predicted CS on 4 additional cases, despite misclassifying 20 additional VD, resulting in a 6.0% decrease in average accuracy compared to the clinical data model. Hence, integrating US imaging into the latter model can be a new development in assisting mode of delivery counseling.