Diagnostic Accuracy of Ultrasound-Estimated Fetal Weight versus Actual Birth Weight: A Prospective Cohort Study on Predictive Performance, Systematic Bias, and Clinical Agreement Near Term

Abstract

Fetal weight prediction is used to aid decision making, timing of delivery, and counselling for possible macrosomia and fetal growth restriction (FGR). The clinical standard, most commonly the Hadlock formulas, ultrasound estimated fetal weight (EFW) has also been shown to agree with actual birth weight (BW) to a varying degree depending on gestational age, extremes of fetal weight, maternal body habitus, and operator experience. To assess the accuracy, systematic bias, and clinical agreement of ultrasound-derived EFW with BW recorded within 7 days of delivery, as well as to establish factors associated with decreased estimation accuracy. N = 420 (SIMULATED) singleton pregnancies were studied using ultrasound biometry within 7 days prior to delivery. The accuracy was evaluated using mean percentage error (MPE), mean absolute percentage error (MAPE), proportion of estimates within +-10% of birth weight, Bland-Altman limits of agreement, and Lin's concordance correlation coefficient (CCC) for each birth weight category. In this synthetic cohort, EFW showed a mean percentage error of - 0.8% (SD 5.9%), mean absolute percentage error of 4.8% (SD 3.6%), and 90.2% of the estimates were within +-10% of the true birth weight. Bland-Altman analysis showed a mean bias of -25.9 g (95% limits of agreement: -412.5 g to +360.6 g). Pearson r was 0.868 and Lin's CCC was 0.853, suggesting high correlation and agreement. The AUC for EFW to predict macrosomia (birth weight >= 4000 g) and predicted small-for-gestational-age status were 0.976 (71.4% sensitivity and 96.1% specificity) and 0.963 (64.3% sensitivity and 96.0% specificity, respectively). None of the maternal, fetal, or sonographic covariates were statistically significant when explored as a separate predictor of absolute percentage error in adjusted regression. For this simulated dataset, ultrasound EFW was in good overall agreement with actual birth weight with small systematic underestimation and wide individual (personal) limits of agreement, consistent in direction with published datasets this exercise is based on. The findings presented in these tables are intended as hypothetical examples of the analytic process that the study protocol calls for, but cannot be considered evidence of clinical effectiveness since a true cohort of participants with sufficient power would be needed before any of these numbers could support clinical practice.