Predictive Models for Estimating the Probability of Successful Vaginal Birth After Cesarean Delivery: A Systematic Review.

Abstract

To systematically review all studies that developed or validated a vaginal birth after cesarean (VBAC) prediction model. MEDLINE, EMBASE, CINAHL, Cochrane Library, and ClinicalTrials.gov were searched from inception until February 2022. We included observational studies that developed or validated a multivariable VBAC prediction model in women with a singleton pregnancy and one previous lower segment cesarean delivery. A total of 3,758 articles were identified and screened. For 57 included studies, data were extracted in duplicate using a CHARMS (Critical Appraisal and Data Extraction for Systematic Review of Prediction Modelling Studies) checklist-based tool and included participants' characteristics, sample size, predictors, timing of application, and performance. PROBAST (Prediction model Risk of Bias Assessment Tool) and TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis) were used to assess risk of bias and transparency of reporting. Several studies developed or validated more than one model. There were 38 unique prediction models, 42 external validations of 10 existing prediction models, and six modifications of existing models. Of the 38 unique models, only 19 (19/38, 50%) were internally validated in the initial study. No studies externally validated their model in the initial study. Age, previous vaginal birth, and previous cesarean delivery for labor dystocia were the commonest predictors. The area under the curve in included studies ranged from 0.61 to 0.95. Models used close to delivery generally outperformed those used earlier in pregnancy. Most studies demonstrated a high risk of bias (45/57, 79%), the remainder were unclear (7/57, 12%) and low (5/57, 9%). Median TRIPOD checklist adherence was 70% (range 32-93%). Several prediction models for VBAC success exist, but many lack external validation and are at high risk of bias. Models used close to delivery outperformed those used earlier in pregnancy; however, their generalizability and applicability remain unclear. High-quality external validation and effect studies are required to guide clinical use. PROSPERO, CRD42020190930.