Correspondence: Dr G. C. S. Smith, Laboratory for Pregnancy and Newborn Research, College of Veterinary Medicine, Cornell University, Ithaca, New York 148536401, USA.
Abstract
Objectives To establish the relation between fetal abdominal circumference and birthweight in a large population of fetuses; to identify whether the error in estimating birthweight by abdominal circumference varied with the magnitude of abdominal circumference; and to establish whether adding femur length to abdominal circumference caused a clinically important reduction of error in predicting birthweight.
Design A retrospective study.
Setting The ultrasound department of a teaching maternity hospital offering a tertiary referral service.
Sample From 3512 nondiabetic women with a normally formed singleton fetus, an abdominal circumference measurement of the infant was made within seven days of delivery; of these, 1213 had a femur length measurement performed at the same time.
Results There was a linear relation between abdominal circumference and birthweight. There was a strong inverse correlation between the proportional error in predicting birthweight from the abdominal circumference and the magnitude of the abdominal circumference. Both the Campbell and Wilkin equation (abdominal circumference alone) and the Hadlock equation (abdominal circumference and femur length) were associated with systematic errors, especially with larger birthweight infants. The median absolute errors for the two equations were not significantly different overall (6.98% and 6.86% respectively), although the Hadlock equation was significantly more accurate in predicting birthweight in infants weighing greater than 4500 g. However, no threshold value of abdominal circumference or of estimated fetal weight using the Hadlock equation had a positive predictive value in estimating infants of > 4500 g of greater than 35%.
Conclusions Prediction of birthweight should be by abdominal circumference alone. Table 1 presents robust estimates of the error of predicting birthweight using fetal abdominal circumference.
Table 1. The relation between fetal abdominal circumference (AC) and birthweight (BW).
Prediction of birthweight using ultrasonic examination of the fetus was first described by Campbell and Wilkin in 1975^{1}. The single measurement which correlates most strongly with birthweight is fetal abdominal circumference (AC), and this is widely used as a single parameter of fetal size^{2}. The definitive description relating fetal AC to birthweight is the original study of 140 fetuses which only included 11 infants < 2000 g birthweight^{1,2}. These authors hypothesised that the proportional error in predicting birthweight was the same across the range of AC. However, there were insufficient numbers of infants at the extremes of birthweight to test that hypothesis.
Some studies have used additional parameters of fetal biometry in an attempt to improve the accuracy of estimates^{3}. However, given the error inherent in all techniques of predicting birthweight^{2}, we decided to focus on quantifying the error of the prediction rather than on obtaining small improvements in accuracy. To this end we sought to quantify the variation in birthweight in infants with a given magnitude of AC in a large population with reasonable numbers of infants at the extremes of birthweight. We also sought to test the hypothesis that the proportional error in predicting birthweight from AC is unaffected by the magnitude of the AC.
Furthermore, we sought to determine whether using additional fetal measurements did indeed cause a clinically important reduction in the error of the estimate over using AC alone. The main parameters used in other equations in addition to AC are femur length, biparietal diameter and head circumference. We chose femur length as the additional parameter because it can usually be easily measured (unlike head circumference and biparietal diameter which can only be measured in about 35% of fetuses at term or during labour^{4}) and the error of the prediction from AC and femur length is only minimally greater than using all four parameters together^{3}. Of the myriad of equations available, we chose to compare the Campbell and Wilkin equation^{1} (AC alone) and the Hadlock equation^{5} (AC plus femur length) as these are the two recommended for use by the British Medical Ultrasound Society^{2}.
METHODS
Over a 10 year period the results of all ultrasound scans performed in the main department during working hours were collected in a computer database, along with comprehensive details of the given patient's medical and obstetric history and of the current pregnancy. Following delivery the database was updated with details of the labour and mode of delivery, and significant parameters of the neonate.
Measurements of AC and femur length were made on a range of ultrasound machines by trained staff in the ultrasound department using standard techniques^{2}. The AC was measured directly rather than by calculation from the abdominal diameter. The femur length was measured if requested by the clinician. A measurement of AC was obtained within seven days of delivery from the infants of 3512 nondiabetic women who had had a normally formed singleton fetus. In 1213 of these infants, a measurement of femur length was also made.
Statistics
The two equations compared were the Campbell and Wilkin equation using AC alone^{1} and the Hadlock equation, using AC and femur length^{5}. The systematic error was calculated by the signed percentage error (the difference between the birthweight and estimated birthweight expressed as a percent of birthweight). A systematic error was assumed to exist where the 95% confidence intervals of the mean percentage difference excluded zero. The absolute error (i.e. nonsigned) was summarised by the median and the interquartile range as it is, necessarily, skewed. Comparison of the absolute error was made by obtaining the difference in absolute error between the two equations for each individual patient. This was then compared with zero using a onesample Wilcoxon signed rank test and 95% CI were obtained. Statistical analysis was performed using Minitab release 8.2 on an Apple Macintosh.
RESULTS
1. Relation between AC and birthweight
The median, 10th and 90th centiles and range of birthweight for groups of infants within a given range of AC, arranged in 10 mm increments from 200 mm, are given in Table 1. There was a very strong positive correlation between the median birthweight of each group and the AC (r^{2}= 0.996) (Fig. 1). The r^{2} value was the same for a second order polynomial, and only very slightly greater for a third order polynomial curve (0.998). There was a strong inverse correlation between the proportional error in predicting birthweight using the simple regression equation and the magnitude of the AC (Fig. 2).
Equations of lines:
2. Comparison of Campbell and Wilkin and Hadlock equations
Standard existing equations for predicting birthweight were compared using the data from the 1213 fetuses that had both an AC and femur length measurement performed within one week of delivery. The Campbell and Wilkin equation (AC alone) was associated with a significant overestimation of birthweight between 500 and 999 g and 1500 and 2999 g and a significant underestimate of birthweight between 3500 and 4999 g (Fig. 3a). The Hadlock equation (AC and femur length) resulted in a significant underestimate of birthweight between 1000 and 1499 g and 4000 and 4999 g (Fig. 3b).
The median absolute error (i.e. nonsigned) was 6.98% (interquartile range 3.46 to 12.03) for the Campbell and Wilkin equation and 6.86% (interquartile range 3.48 to 11.66) for the Hadlock equation. The median difference in absolute error between the two equations was 0.18% of birthweight (95% CI −0.12 to 0.56, P= 0.314 onesample Wilcoxon signed rank). The median error was significantly less for the Hadlock equation between 500 and 1499 g, 2000 and 2499 g, and 4000 and 4999 g; significantly less for the Campbell and Wilkin equation between 3000 and 3999 g; and not significantly different between the two equations between 1500 and 1999 g and 2500 and 2999 g (Fig. 4). The only weight range where the median difference in error between the two equations exceeded 5% of birthweight was in the range 4500 and 4999 g, where the Hadlock equation was associated with a median reduction in error of 9.00% of birthweight compared with the Campbell and Wilkin equation (95% CI 7.66 to 10.44, P < 0.0001, onesample Wilcoxon signed rank).
3. Prediction of macrosomia
The sensitivity, specificity, and positive and negative predictive values in predicting a birthweight > 4500 g for threshold levels of AC or estimated fetal weight (Hadlock equation) are given in Table 2.
Table 2. Diagnostic efficacy of threshold abdominal circumference and Hadlockestimated fetal weights in predicting macrosomia (birthweight > 4500 g). Values are % (n). EFW = estimated fetal weight predicted by Hadlock equation^{5}; PPV = positive predictive value; NPV = negative predictive value.
AC (mm)
EFW(g)
Sensitivity
Specificity
PPV
NPV
^{*}When confined to infants with an AC 360 mm, the PPV for an EFW 3750 was 12% (16/139) and for 4000 was 15% (16/97). The PPV for the other thresholds were unchanged.
360

100 (16/16)
88 (1055/1197)
10 (16/158)
100 (1055/1055)
370

94 (15/16)
94 (1130/1197)
18 (15/82)
>99 (1130/1131)
380

69 (11/16)
98 (1170/1197)
29 (11/38)
>99 (1170/1175)
390

31 (5/16)
99 (1187/1197)
33 (5/15)
99 (1187/1198)
3750
100 (16/16)
86 (1033/1197)
9^{*} (16/180)
100 (1033/1033)
4000
94 (15/16)
93 (1113/1197)
15^{*} (15/99)
>99 (1113/1114)
4250
88 (14/16)
97 (1161/1197)
28 (14/50)
>99 (1161/1163)
4500
44 (7/16)
99 (1181/1197)
30 (7/23)
99 (1181/1190)
4750
6 (1/16)
>99 (1194/1197)
25 (1/4)
99 (1194/1209)
DISCUSSION
There are a number of areas of obstetric practice where prediction of birthweight is deemed important, such as contemplating vaginal breech delivery^{6} and in predicting the outcome of preterm delivery^{7}. Some would argue that the potential for large individual errors renders such estimates of little value. Clearly, where used, the error attached to the estimate is of critical importance.
It had previously been hypothesised that the proportional error in estimating birthweight from fetal AC alone was the same across the range of abdominal circumference^{1}. Our data (Fig. 2) indicate that the proportional error increases significantly with decreasing fetal AC. The underestimate of the variation in birthweight within a range of AC in smaller infants using existing data^{1} may lead to clinically important errors. It has been stated that fetuses with an AC between 210 and 219 mm would be almost 95% certain to be < 1000 g^{1}, whereas our 10th and 90th centiles for this range of AC were much wider (830 and 1370 g) and, indeed, 55% of infants within this range of fetal AC had a birthweight > 1000 g (Table 1). We propose that the data in Table 1 should replace the currently recommended tabular data relating fetal AC to birthweight^{2}. We also propose that estimates of birthweight should be accompanied by the 10th and 90th centiles of the estimate to aid the clinician in interpreting the information.
Regarding the mathematical relation between AC and birthweight, the original study^{1} proposed that it was best modelled using a third order polynomial equation of log_{e} (birthweight) versus AC. Our data (Fig. 1) illustrates that the relation between fetal AC and birthweight in the range studied is linear. The original study^{1} plotted a line through a scatter of their raw data. As most of their infants’ birthweight was > 2000 g, the equation will have been modelled to minimise error in this group. Use of the medians of each group means that our mathematical model is determined equally across the range of AC. This may account for the different relation observed.
A number of small studies have demonstrated that adding femur length to AC in equations to predict birthweight reduces the error by about 5%^{2}. However, in our group of 1213 fetuses who had both an AC and femur length measurement within seven days of delivery, we found no significant difference in the median error of the prediction between the currently recommended equations^{2} comparing AC alone^{1} with AC plus femur length^{5}.
When analysed according to eventual birthweight, there was only one group where one equation was associated with a median improvement in accuracy of > 5% of birthweight, namely, using the Hadlock equation in infants with an eventual birthweight > 4500 g. No fetus with an AC < 360 mm had a birthweight > 4500 g (Table 1). It could be argued, therefore, that any fetus with an AC > 360 mm should also have a femur length measured to improve the prediction of macrosomia. To test how useful this might be clinically, we calculated the sensitivity, specificity, and positive and negative predictive values of threshold levels of AC and estimated fetal weight in predicting a birthweight > 4500 g (Table 2). No level of AC alone or Hadlock estimated fetal weight had a positive predictive value of > 35%. The decreased error in this range of birthweight with the Hadlock equation is reflected by the slightly better sensitivities for a given positive predictive value compared with AC alone, but we would conclude from these data that neither technique adequately indicates the strong likelihood of macrosomia or detects a sufficient proportion of macrosomic infants to be clinically useful in this context.
We conclude that when attempting to predict birthweight from a single scan, it should be done by AC alone and that the prediction should be accompanied by the 10th and 90th centile of the estimate. The currently recommended tabular data relating AC to birthweight^{2} may lead to an underestimate in the error of the prediction in low birthweight infants. We present a revised table for routine clinical use which provides robust estimates of the error in predicting birthweight from a single measurement of AC (see Table 1).