Prediction models for caesarean section following induction of labour: A systematic review of methodology and reporting quality

Abstract Background: Model development and validation constitute the critical initial steps in the clinical prediction modelling pipeline, where methodological rigour and reporting transparency determine the validity of all subsequent stages. Our systematic review aimed to appraise the methodology and reporting quality of recently published studies that developed or externally validated prediction models for caesarean section at term following induction of labour. Methods: MEDLINE, Scopus, Embase, IEEE Xplore and CINAHL Complete databases were searched to identify original research studies published since 2017. Studies that did not report any performance measurement of the prediction model were excluded. No restrictions were applied to study designs, populations, modelling algorithms, type and timing of predictors, and methods of induction. Descriptive analysis was performed to assess heterogeneity across eligible studies. Study risk of bias and model applicability were assessed using the Prediction model Risk Of Bias ASsessment Tool (PROBAST). Study reporting quality was evaluated using the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis + Artificial Intelligence (TRIPOD+AI) Statement. Results: Among the 14 included studies, 12 developed a single model using logistic or Bayesian regression, and two externally validated included models. All studies reported discrimination performance using the area under the receiver operating characteristic (AUROC) curve, and 11 reported calibration performance. However, only one validation-only study assessed clinical utility using decision-curve analysis, and none evaluated model fairness. Overall, 12 studies had unclear or high risk of bias, mainly attributable to the selection of predictors through univariate analysis. Three studies raised applicability concerns due to the inclusion of predictors not available before induction. The handling of missing data was not reported in five studies. Conclusions: More external validation is needed to assess the clinical utility and fairness of existing models before they can be recommended for the subsequent stages in the clinical prediction pipeline and ultimately support clinical decision-making and improve maternal and neonatal outcomes. Future research should prioritise the inclusion of comprehensive performance measures and strengthen methodological rigour and reporting transparency to improve model reliability and clinical applicability, as emphasised in existing reporting guidelines.