Objective: To compare the accuracy of 31 published formulas for estimated fetal weight (EFW) in predicting macrosomia (birthweight 4,000 gm or more) in infants of diabetic mothers.
Methods: The study population comprised 165 women with gestational or pregestational diabetes who had sonograms to estimate fetal weight after 36 weeks of gestation and within 2 weeks of delivery. Three measures of accuracy were compared: 1) area under the receiver operating characteristic (ROC) curve relating EFW to macrosomia, 2) systematic error, and 3) absolute error. For each measure, the 31 formulas were rank-ordered from 1 (best) to 31 (worst). For each formula, the three rank scores were summed to give a total score. The formula with the lowest total score was considered the "best" formula.
Results: Macrosomia occurred in 49 cases (30%). Areas under the ROC curves ranged from 0.8361-0.8978. Differences in areas were not significantly different between the 31 formulas. The 1986 formula of Ott et al. had the lowest total score. Using this "best" formula, an EFW of 4,000 gm or more had a sensitivity of 45% to predict macrosomia and a positive predictive value of 81%. To achieve 90% sensitivity with this formula would have required diagnosis of macrosomia with an EFW of 3,535 gm or more, but this would have comprised 46% of the population with a 42% false-positive rate. All 31 formulas were better at predicting macrosomia than predictions based on gestational age alone, and 28 were better than predictions based on abdominal circumference alone.
Conclusions: All 31 formulas for EFW had comparably poor accuracy for prediction of macrosomia. Delivery decisions based on EFW will often be in error. Future studies should determine whether specific sonographic measurements, ratios, or differences are better than EFW or birthweight as predictors of birth trauma.