A machine‑learning model trained on 27,543 pregnancies now forecasts fetal growth restriction with remarkable accuracy, offering Chinese hospitals a window for earlier intervention in a condition that remains a leading cause of perinatal morbidity worldwide.
Chinese scientists have developed a time‑dependent random survival forest (RSF) model that can predict fetal growth restriction (FGR) weeks before clinical symptoms become apparent, according to a new study published in the International Journal of Gynecology & Obstetrics. The model was built and validated using retrospective data from 27,543 singleton pregnancies recorded across tertiary hospitals in Taizhou City, China, between 2016 and 2022.
FGR is a progressive condition in which a fetus fails to reach its expected growth potential, amplifying risks of stillbirth, neonatal complications, and long‑term metabolic disorders. Traditional screening relies on static risk scores and basic biometric thresholds, which often miss the condition until it is already advanced. The new approach treats gestational week as a dynamic time variable, allowing the algorithm to continuously update risk predictions as the pregnancy progresses. The model achieved a C‑index of 0.864, with peak predictive power between weeks 28 and 36—the critical window when most late‑onset FGR cases emerge.
Key predictors included ultrasound‑derived abdominal circumference and femur length, alongside maternal serologic markers such as alpha‑fetoprotein (AFP) and unconjugated estriol (E3). The team also found that early‑onset and late‑onset FGR follow distinct biological pathways. Early FGR was driven predominantly by maternal vascular factors and fetal biomarkers, while late‑onset FGR—far more common—resulted from the cumulative mild effect of multiple modest risk factors. This distinction suggests that a one‑size‑fits‑all screening schedule may be inadequate; a more flexible, risk‑stratified approach could better match interventions to the underlying cause.
China’s large, urbanising population and increasingly centralised hospital systems make it an ideal setting for implementing machine‑learning‑driven prenatal screening. The study’s RSF model outperformed conventional logistic regression and Cox proportional‑hazards models in both accuracy and flexibility, and it does not require expensive additional testing infrastructure—it works with data already collected during routine antenatal care. For healthcare systems worldwide that are grappling with rising caesarean rates and the growing burden of pregnancy‑related complications, this provides a scalable, evidence‑based tool that could be integrated into existing electronic health records.
Why it matters:
FGR is frequently diagnosed too late for effective intervention. This model offers Chinese clinicians a practical, data‑driven method for early identification that is both accurate and inexpensive to deploy. For any hospital system managing high‑volume obstetric services, the ability to predict FGR weeks in advance could meaningfully reduce neonatal intensive care admissions and improve long‑term outcomes for at‑risk infants.
ScientificChina — tracking what’s happening in Chinese science, technology, research, and industrial innovation in a way global professionals can actually use.
Follow ScientificChina for deeper insight into China’s evolving science, technology, and industrial landscape.
No products in the cart.