Truace
Subscribe

Birth weight estimates from machine learning beat a standard formula in a 456-woman study

Study title: Development of machine learning models for fetal birth weight prediction: A single-center study in a Turkish population

Researchers at a tertiary center in the Karaman region of Turkiye tested nine machine learning models for estimating fetal birth weight. Seven of the nine had the lowest mean absolute error, between 184.30 and 190.00 grams, and linear regression did best at 184.30 grams.

Read the study abstract

Fetal birth weight (BW) is a critical indicator of the prognosis of perinatal health. Accurate estimation of BW is essential in determining the appropriate delivery method. This single-center study aims to evaluate the performance of various machine learning (ML) algorithms in estimating BW. The study population consisted of 456 pregnant women with singleton pregnancies evaluated at a single tertiary center in the Karaman region of Turkiye between 38 and 41 weeks based on their last menstrual period, who gave bi…

Truvace record2 min read1 sourceEvidence: Moderate · 60/100 — Observational study · n = 456 · 1 independent sourcePeer-reviewedTRV-2026-2426 · cite or check it
Development of machine learning models for fetal birth weight prediction: A single-center study in a Turkish population"Medical Laboratory" by ben.dracup, CC BY 2.0.

In brief

The closest conventional comparator, the Hadlock III formula, had an error of 223.33 grams, so the best model was about 39 grams closer on average. Birth weight estimates help determine the appropriate delivery method, so smaller errors could matter in delivery planning.

The work is a single-center study of 456 women, so how the models perform in other hospitals and populations is not established.

Main points

  1. The study included 456 pregnant women with singleton pregnancies, evaluated at one tertiary center in the Karaman region of Turkiye.
  2. Women were assessed between 38 and 41 weeks based on last menstrual period.
  3. All women gave birth to a phenotypically normal live newborn within 7 days after the ultrasound examination.
  4. Nine machine learning models were developed; seven had the lowest mean absolute errors, from 184.30 to 190.00 grams.
  5. Linear regression was the most accurate model, with an error of 184.30 grams.
  6. Hadlock III, the closest formula, had an error of 223.33 grams.

The gain

In a single-center study of 456 pregnant women in Turkiye, a linear regression model estimated fetal birth weight with a mean absolute error of 184.30 grams, against 223.33 grams for the Hadlock III formula.

The full story

The study evaluated the performance of various machine learning algorithms for estimating fetal birth weight, which the authors describe as a key indicator of perinatal health prognosis and important for choosing the delivery method.

Participants were 456 women with singleton pregnancies seen at a single tertiary center in the Karaman region of Turkiye between 38 and 41 weeks. Only women who delivered a phenotypically normal live newborn within 7 days of the ultrasound were included.

Seven of nine machine learning models produced mean absolute errors between 184.30 and 190.00 grams. Linear regression performed best at 184.30 grams, compared with 223.33 grams for the Hadlock III formula.

Sources

  1. Peer-reviewedAFRICAN JOURNAL OF REPRODUCTIVE HEALTH2026-11-01

The debate