Development and validation of a multimodal nomogram integrating maternal serum angiogenic markers and fetal hemodynamics for risk stratification of late-onset fetal growth restriction

Background Late-onset fetal growth restriction (L-FGR) remains a major challenge in obstetrics due to its subtle clinical presentation and high risk of adverse perinatal outcomes. This study aimed to develop and validate a multimodal nomogram integrating maternal serum biomarkers and fetal ultrasound parameters to improve the diagnostic evaluation and risk stratification of L-FGR and subsequent adverse endpoints. Methods This retrospective nested case-control study included 300 singleton pregnancies (100 L-FGR cases and 200 matched normal controls) managed at our institution. Fetal hemodynamic indices, biometry, and maternal serum levels of PlGF and the sFlt-1/PlGF ratio were retrospectively analyzed. Independent predictors were identified via multivariate logistic regression and confirmed by LASSO regression. A multimodal diagnostic-support and risk-stratification nomogram was constructed and evaluated for discrimination, calibration, and clinical utility. Results The sFlt-1/PlGF ratio, cerebroplacental ratio (CPR), abdominal circumference (AC) percentile, and PlGF were identified as independent predictors. The combined multimodal model demonstrated superior discrimination, achieving an AUC of 0.948 (95% CI: 0.918–0.978) in the training cohort ( n = 210) and maintaining robust discrimination with an AUC of 0.910 (95% CI: 0.845–0.965) in the independent internal validation cohort ( n = 90), significantly outperforming single-modality models ( P < 0.05). Calibration curves and the Hosmer-Lemeshow test ( P = 0.601) indicated robust fit. Decision curve analysis (DCA) and Net Reclassification Index (NRI = 0.452) confirmed substantial clinical net benefit. Furthermore, the high-risk group stratified by the nomogram exhibited significantly accelerated time-to-delivery (Log-rank P < 0.001) and remarkably higher incidences of severe composite adverse perinatal outcomes, specifically NICU admission, emergency cesarean section, and neonatal hypoglycemia. Benchmarking revealed that our interpretable logistic regression model achieved comparable efficacy to complex ML algorithms like XGBoost. Conclusion We successfully developed a highly robust, multimodal nomogram that synergistically combines hemodynamic and biochemical markers. This tool facilitates accurate, individualized L-FGR identification and risk stratification, effectively guiding clinical decision-making to mitigate adverse perinatal outcomes.