Development of a time-dependent random survival forest model for predicting FGR based on prenatal screening date.
Authors
Affiliations (3)
Affiliations (3)
- Center for Reproductive Medicine, Taizhou Hospital of Zhejiang Province, Wenzhou Medical College, Linhai City, China.
- Department of Obstetrics & Gynecology, Taizhou Hospital of Zhejiang Province, Wenzhou Medical College, Linhai City, China.
- Center for Reproductive Medicine, Enze OBGYN Hospital of Taizhou Hospital, Taizhou City, China.
Abstract
Fetal growth restriction (FGR) is a progressive condition that amplifies risks with advancing gestation. This study develops a time-dependent random survival forest (RSF) model integrating prenatal screening markers to predict FGR and explore temporal factors, aiding clinical decision making for targeted intervention. This was a retrospective cohort study in tertiary hospitals in Taizhou City, China (2016-2022) including 27 543 singleton pregnancies with serial prenatal records, excluding major fetal anomalies or incomplete follow-up. An RSF model was developed using gestational week as the time variable and Delphi-defined FGR as the event. The model incorporates maternal biological data, serologic markers, and ultrasound measurements as feature variables. Performance was assessed using time-dependent receiver operating characteristic (ROC) curves and decision curve analysis (DCA). Feature-based cumulative risk curves were employed for interpreting the model. Main outcomes were measured using model validation (C-index, area under the curve [AUC]) and utility (individual prediction curves, feature risk curve). The RSF model demonstrated high performance (C-index: 0.864), with peak predictive power between 28 and 36 weeks' gestation (AUC: 0.87-0.91). It identified both early- and late-onset FGR. Key predictors included ultrasound markers such as abdominal circumference and femur length. Late-onset FGR was mainly caused by a cumulative mild effect of multiple factors, while early-onset FGR was driven by maternal factors and fetal markers like AFP and E3. The RSF model outperforms traditional models in accuracy and flexibility, providing dynamic FGR risk predictions and supporting timely clinical intervention.