Estimation of fetal weight: accuracy of regression models versus accuracy of ultrasound data

To determine the respective importance of equation accuracy and variability in interexaminer measurements in estimation of fetal weight (EFW).

The study included 3 samples from three different French maternity hospitals. Sample 1 (6,508 fetuses) was used to compute a new linear regression model estimating fetal weight (FW) from ultrasound measurements. Sample 2, with 705 fetuses, was used to compare the accuracy of the new equation with Hadlock's equation. Sample 3 (1,461 fetuses) was used to assess - from our equation and from Hadlock's - the estimation errors due to the variability of ultrasound measurements recorded by 11 distinct examiners. Accuracy was determined by the signed percent difference (%Diff). Statistical analysis included the F test for correlated variances and comparisons of correlated variances.

The random error of our model is 6.8%, significantly lower (p < 0.01) than Hadlock's which is around 7.4% (8.9% for low FW and 7.4% for high FW). The variability of ultrasound measurements among the different examiners in sample 3 generated a significant variation (p < 0.01) in the random error, ranging from 6.7 to 12.5%. This significant variation was also observed using Hadlock's equation for the same sample (from 7.2 to 12%).

The accuracy of EFW depends much more on the quality of ultrasound measurements than on the choice of equation. Nevertheless, optimizing the accuracy of EFW (about 6.7-6.8%) imperatively requires standardized data collecting.