Truace
Subscribe

Conventional statistical and machine learning-based prediction of neonatal outcomes in late-onset fetal growth restriction

Objectives To evaluate associations between antenatal fetal biometric and Doppler parameters and neonatal outcomes in late-onset fetal growth restriction (FGR), and to evaluate machine learning models for neonatal outcome prediction. Methods This retrospective cohort included 137 singleton pregnancies. Continuous missing values were completed using a single iterative chained-equations procedure. Maternal characteristics, fetal biometric measurements, and Doppler parameters, including umbilical artery pulsatility…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1789· cite & verify
Conventional statistical and machine learning-based prediction of neonatal outcomes in late-onset fetal growth restriction
3D and 4D Ultrasound Imaging in Navi Mumbai by Goleisureintl. CC BY 4.0 · https://creativecommons.org/licenses/by/4.0

In brief

Researchers analyzed 137 singleton pregnancies complicated by late-onset fetal growth restriction. They compared conventional statistics and regression with machine learning models, using prenatal ultrasound measurements, Doppler indices and maternal characteristics to predict outcomes such as NICU admission, transient tachypnea and sepsis.

The work suggests that routinely collected prenatal data can flag newborns at higher risk, and that machine learning largely agrees with traditional methods rather than clearly outperforming them. The sample is small and retrospective, and the models have not been tested on outside populations, so clinical usefulness is still unproven.

Main points

  1. The retrospective cohort included 137 singleton pregnancies with late-onset fetal growth restriction.
  2. In final logistic models, maternal age, previous abortions and estimated fetal weight predicted NICU admission.
  3. Previous cesarean delivery predicted transient tachypnea of the newborn, while maternal age and estimated fetal weight predicted neonatal sepsis.
  4. Machine learning reached a cross-validated R2 of 0.592 for birth weight and an AUC of 0.850 for neonatal sepsis.

The gain

Machine learning models built on fetal biometry, Doppler and maternal data predicted birth weight and neonatal sepsis with moderate accuracy, supporting conventional statistical findings in late-onset FGR.

The rundown

The study assessed associations between antenatal fetal biometric and Doppler parameters, including umbilical artery and middle cerebral artery pulsatility indices and the cerebroplacental ratio, and neonatal outcomes. Missing continuous values were filled using a single iterative chained-equations procedure.

NICU admission was linked to higher maternal age and to lower gestational age, fetal biometric measurements, estimated fetal weight, birth weight, Apgar scores and umbilical cord pH after false discovery rate correction.

The birth-weight linear model had the highest explanatory power, with an adjusted R2 of 0.662. Machine learning models were evaluated with repeated 5-fold cross-validation and showed moderate predictive performance.

Sources

  1. Peer-reviewedJournal of Perinatal Medicine2026-10-07

The debate