Predictive models for adverse pregnancy outcomes in fetal growth restriction: a systematic review and meta-analysis

We aim to conduct a systematic review and meta-analysis to evaluate the performance of existing predictive models for adverse pregnancy outcomes in pregnancies with fetal growth restriction from multiple perspectives. A systematic literature search was conducted in PubMed, the Cochrane Library, Web of Science, and CNKI up to January 2025 to identify studies on the development, validation, or updating of clinical prediction models for adverse pregnancy outcomes in fetal growth restriction. Risk of bias was assessed using the Prediction Model Study Risk of Bias Assessment Tool. Subgroup analyses were based on study design, country income level, predictor types, and outcome categories. A random-effects model was used for meta-analysis. A systematic literature search identified 40 studies describing predictive models for adverse pregnancy outcomes in the context of fetal growth restriction. Of these, 26 studies were ultimately included in the meta-analysis based on diagnostic criteria for FGR and risk assessment methodologies. Meta-analysis showed pooled AUCs of 0.823 (95% CI: 0.767–0.868; I² = 52.52%) for early-onset FGR, 0.749 (95% CI: 0.671–0.813; I² = 92.78%) for late-onset FGR, and 0.791 (95% CI: 0.714–0.852; I² = 84.09%) for studies without onset differentiation. Models using multiple predictors consistently outperformed single-predictor models (AUC: 0.758 vs. 0.743; 0.845 vs. 0.746) in the late-onset FGR and FGR groups. We also demonstrated models with different predictor sets for composite adverse pregnancy outcomes. Incorporating Doppler indices, biochemical markers, or maternal clinical factors into models based on estimated fetal weight may moderately enhance overall predictive performance. Our study demonstrates that integrating Doppler parameters, biochemical markers, or maternal clinical factors into models based on estimated fetal weight may moderately improve overall predictive performance for composite adverse pregnancy outcomes with FGR. However, few have undergone rigorous internal or external validation; therefore, their clinical application warrants caution. Future research should focus on multi-center studies, improve existing thresholds, and explore new indicators. It is also necessary to combine machine learning and deep learning techniques to enhance model performance. PROSPERO registration number: CRD42025643903.