Predictive models for estimating the duration of labour induction: A systematic review and critical appraisal

This systematic review aimed to identify and evaluate prediction models developed to estimate the duration of labour induction, to support individualized clinical decision-making and optimise obstetric resource planning.

A comprehensive search of MEDLINE (PubMed), Scopus, and Google Scholar identified studies published from 1 January 2000 up to 5 January 2025. Eligible studies were original research articles that developed or validated a model predicting labour induction duration, expressed either as a continuous measure or as a predefined time interval. Risk of bias was assessed using the PROBAST tool.

Nine studies, conducted primarily in high-income countries, were included. Sample sizes ranged from 204 to 8466 women undergoing labour induction.

Models used clinical and demographic predictors such as Bishop score, parity, body mass index, and gestational age. Logistic regression was the predominant modelling method, followed by linear regression and machine learning. Reported discriminative performance (area under the ROC curve) ranged from 0.73 to 0.88. The highest externally validated model reported an AUROC of 0.81 (95% CI 0.78-0.83). Calibration was assessed in only two studies. Using standard PROBAST categories, all models were judged at high or unclear risk of bias, primarily due to limitations in the analysis domain.

Despite moderate-to-good internal performance, current models have limited clinical applicability due to methodological weaknesses and lack of external validation. Future studies should prioritise calibration, transparent reporting, and usability testing to ensure reliable clinical implementation.