Ultrasound Obstet Gynecol 2002; 20: 57– 60
Prediction of fetal weight by ultrasound: the contribution of additional examiners
sBlackwell Science, Ltd
I. GULL*, G. FAIT*, J. HAR-TOOV*, M. J. KUPFERMINC*, J. B. LESSING*, A. J. JAFFA* and I. WOLMAN* *Ultrasound Unit, Department of Obstetrics and Gynecology, Lis Maternity Hospital, Tel Aviv Sourasky Medical Center, and Sackler School of Medicine, Tel Aviv University, Israel
KE YWORDS: Estimation discrepancies, Fetal weight estimation, Multiple examiners
ABSTRACT
INTRODUCTION
Objectives To assess the contribution of additional examiners to: the average discrepancy between estimated and actual fetal weights; the correlation between estimated and actual fetal weights; the reduction in major (> 10%) discrepancies between estimated and actual fetal weights.
Accurate prediction of birth weight at term can provide useful information regarding labor and mode of delivery. Many methods for fetal weight estimation have been proposed. Earlier formulae were derived solely by regression analysis, but recently far more sophisticated formulae, based on physical models, have been developed1. Despite these efforts, results are not convincing. Although accuracy in predicting fetal weight by ultrasound, expressed as mean error, is < 10%, as many as one-fifth of neonates exceed the estimated fetal weight by > 15%2. The reasons for error in predicting fetal weight are both subjective and objective. The non-medical and even medical disciplines are unfamiliar with the objective limitation in predicting fetal weight, and erroneous predictions are frequently criticised. Hence, every effort should be made to identify and reduce the subjective component in predicting fetal weight, and to recognize our limitation in accurately and consistently predicting all variations in fetal weight. This study was conducted to analyze the subjective contribution of examiners to the discrepancy in predicting fetal weight, and to evaluate the contribution of additional examiners in reducing this type of error.
Design Three experienced sonographers independently measured fetal biparietal diameter, head circumference, abdominal circumference and femur length in 39 fetuses at term. The estimated fetal weights were calculated for each examiner. Fetal biometric measurements were analyzed to obtain the source of differences in estimations among the examiners. Discrepancy, correlation and number of major (> 10%) discrepancies between the estimated and actual fetal weights were calculated for each examiner, and the contribution of additional examiners was analyzed. Results The differences in measurements of the biparietal diameter and femur length were lower than those of the head and abdominal circumferences. For each of the three examiners, the average discrepancy between the estimated and actual fetal weights was 6.1%, 5.9% and 6.3%. When the estimation was based on two examiners, the discrepancy decreased to 4.8–5.6%. The contribution of a third examiner was nil. Major (> 10%) discrepancies between estimated fetal weight and actual birth weight were found in seven, eight and nine estimations of the examiners. Estimation by two examiners decreased the number of major discrepancies, and estimation by all three examiners further decreased by approximately 50% the number of major discrepancies between the estimated and actual fetal weights. Conclusion Measurements by multiple examiners changes only slightly the average number of discrepancies between estimated and actual fetal weights. However, the reduction in major (> 10%) discrepancies is statistically and clinically significant.
METHODS The study group consisted of 39 low-risk, pregnant women at term (38–41 weeks’ gestation). Three experienced sonographers (E1, E2, E3) independently measured standard fetal biometric parameters (biparietal diameter (BPD), head circumference (HC), abdominal circumference (AC) and femur length (FL)), and estimated fetal weights were calculated for each examiner using the Hadlock formula3. The measurements were obtained as follows: (1) BPD from the outer table of the proximal skull to the inner table of the distal skull at the level of the thalamus4; (2) HC at the same level
Correspondence: Dr I. Gull, Ultrasound Unit, Department of Obstetrics and Gynecology, Lis Maternity Hospital, Tel Aviv Sourasky Medical Center, 6 Weizman Street, Tel Aviv, 64239, Israel (e-mail:
[email protected]) Accepted 6-2-02
ORIGINAL PAPER
57
Prediction of fetal weight by multiple examiners as the BPD using a built-in ellipse function5; (3) AC at the level of the stomach and the distal component of the umbilical vein using a built-in ellipse function5; (4) FL from the origin to the distal end of the shaft excluding the femoral head and distal epiphysis6. To clarify the source of differences between fetal weight estimations, fetal biometric measurements among the examiners were compared. Differences in specific measurements (absolute values) between pairs of examiners (E1–E2, E1–E3, E2–E3) were calculated and analyzed. In order to correct the estimated fetal weight at the time of estimation for the time of delivery we used the gestational adjusted projection of estimated fetal weight presented by Mongelli and Gardosi7. According to this method the normal fetal growth curves do not cross the centile cut-off lines to any significant degree. Hence, extrapolation of fetal weight is accurate, and the correction equation is: Corrected estimated fetal weight (d) = median (d) × EFW (us)/median (us), where (d) is gestational age at delivery, (us) is gestational age at examination and EFW is estimated fetal weight. The growth curves used were from an Israeli population8. This corrected fetal weight of each examiner was compared with the actual birth weight to establish the discrepancy of that individual’s measurement. The discrepancy was expressed by the average difference (in percent) between the corrected and actual weights using the equation: Discrepancy (%) = 100 × (corrected EFW − actual birth weight)/corrected EFW. The percentage discrepancy was calculated as an absolute value, since a discrepancy of +10% and one of −10% should not be averaged to 0%. We chose the corrected estimated fetal weight as the denominator, since the common clinical practice of reporting is estimated fetal weight ±10%. The correlation between corrected and actual weights was also calculated. The effect of an additional one or two examiners in reducing the discrepancy between the weights was analyzed. Furthermore, the number of major discrepancies (of > 10% between corrected estimated weight and actual weight) for each examiner was analyzed and compared to the average estimation of two and three examiners.
Statistical analysis Absolute percentage differences in specific measurements between each pair of examiners were calculated as the ratio
Gull et al. of the difference between the two examiners to the mean. These differences were then compared using an analysis of variance (anova) with repeated measurements. Maushley’s test for sphericity was used to determine whether the assumption (Huynh & Feldt’s condition for the covariance matrices) required for repeated measured analysis held. Assessment of the different methods of estimating fetal weight was performed by comparison of means by anova with repeated measurements, and association between the different estimations by calculating Pearson’s correlation coefficients. We used Student’s t-test based on Fisher’s transformation to compare correlations. In addition, we defined as a major difference a discrepancy between estimated and actual weights of > 10%. The McNemar test of symmetry was employed to examine whether the number of major differences in one estimation method was significantly different from the number of major differences in the other method. All values are expressed as mean (standard deviation (SD)).
RESULTS Of the 39 patients, 37 delivered within 10 days after the ultrasound examinations. The remaining two delivered after 17 and 18 days. The mean delivery weight was 3690 (range, 3010–4790) g. The differences in measurements of fetal biometry among the three pairs of examiners (∆E1–E2, ∆E1–E3, ∆E2–E3) are summarized in Table 1. These differences were not statistically significant. The differences in measurements of the BPD and the FL were smaller than those of the HC and AC. For each individual examiner (E1, E2, E3), the discrepancy between the estimated and actual fetal weights was 6.1 (SD, 4.2)%, 5.9 (SD, 3.8)% and 6.3 (SD, 5.1)%, respectively (Table 2). These differences were not statistically significant. The correlation between the estimated and corrected weights for E1, E2 and E3 were 0.84, 0.81, and 0.80, respectively (Table 2). For multiple examiners, the discrepancy between the estimated and actual fetal weights decreased to 4.8% for E1 + E2 (SD, 4.1; P < 0.05), 5.5% for E1 + E3 (SD, 4.3; not significant), 5.6% for E2 + E3 (SD, 4.2; not significant) and 5.1% for E1 + E2 + E3 (SD, 4.0; not significant). For multiple examiners, the correlation between estimated and corrected weights increased to 0.87 (E1 + E2; P < 0.05), 0.85 (E1 + E3; not significant), 0.83 (E2 + E3; not significant) and 0.86 (E1 + E2 + E3; P < 0.05). The number of major (> 10%) discrepancies between the estimated and actual fetal weight were 7, 8 and 9 for E1, E2
Table 1 Differences in biometric measurements of 39 term fetuses undertaken by three examiners (E1, E2 and E3)
Measurement
∆E1–E2 (%, mean (SD))
∆E1–E3 (%, mean (SD))
∆E2–E3 (%, mean (SD))
Biparietal diameter Head circumference Abdominal circumference Femur length
1.42 (1.34) 2.42 (2.2) 3.17 (2.47) 1.96 (1.99)
1.54 (1.23) 2.50 (1.75) 3.10 (2.72) 2.08 (1.95)
1.88 (1.78) 2.34 (2.01) 2.61 (2.35) 1.45 (1.54)
The percentage difference between each pair of examiners was not statistically significant.
58
Ultrasound in Obstetrics and Gynecology
Prediction of fetal weight by multiple examiners
Gull et al.
Table 2 Discrepancy and correlation of predicting fetal weight by single (E1, E2 and E3) and multiple examiners
Examiner
Discrepancy (%, mean (SD))
Correlation
E1 E2 E3 E1 + E2 E1 + E3 E2 + E3 E1 + E2 + E3
6.1 (4.2) 5.9 (3.8) 6.3 (5.1) 4.8 (4.1) 5.5 (4.3) 5.6 (4.2) 5.1 (4.0)
0.84 0.81 0.80 0.87* 0.85 0.83 0.86†
*Decrease in discrepancy was statistically significant for both examiners. †Increase in the correlation was statistically significant for examiners E2 and E3. Table 3 Major (> 10%) discrepancies in predicting fetal weight by single (E1, E2 and E3) and multiple examiners Examiner
Discrepancies (n)
E1 E1 + E2 E1 + E3 E1 + E2 + E3 E2 E2 + E1 E2 + E3 E1 + E2 + E3 E3 E3 + E1 E3 + E2 E1 + E2 + E3
7 5 6 4 8 5 7 4 9 6 7 4
and E3, respectively. When the estimation was calculated as an average of two or three examiners, the number of major discrepancies decreased (Table 3). The contribution of the additional two examiners resulted in a statistically significant reduction in major discrepancies of approximately 50% for each examiner. The advantage of the multiple examiner method over a single examiner is analyzed in Table 4 (P = 0.18 for E1, P = 0.046 for E2 and P = 0.025 for E3).
DISCUSSION Estimation of fetal weight by ultrasound has its limitations. The ultrasound estimation of fetal weight is based on two
methodological assumptions: that there is good correlation between fetal volume and weight9 and that there is good correlation between certain measurable parameters and fetal volume. Accordingly, errors in ultrasound estimation of fetal weight may arise due to different volume : weight ratios, i.e. specific gravity10, or to errors in estimating fetal volume. While fetal specific gravity cannot be predicted accurately and should be considered as an objective component of the error, measurements of fetal biometry partly depend on the skill of the examiner, and should be considered as a subjective component of the error. We analyzed discrepancies in measurements of fetal biometry that affect a person’s ability to predict fetal volume and weight. We calculated the interobserver error in measuring common parameters, i.e. BPD, HC, AC and FL, for calculating estimated fetal weight and used the anova with repeated measurements. The analysis was carried out on the differences between examiners because its aim was to compare differences and not initial values. Maushley’s test for sphericity was not significant for all four parameters (BPD, HA, AC and FL; P = 0.46, 0.79, 0.33 and 0.78, respectively). Not surprisingly, the interobserver error in linear measurement (BPD, FL) was lower than the error in the circumference measurements (HC, AC). The average interobserver error of measuring the BPD was approximately 1.5%. Mathematically, such an error should change the estimated fetal weight derived by the Hadlock equation, by approximately 0.8%3. Conversely, the interobserver error of measuring the AC was approximately 3%. Such an error may change the estimated fetal weight in the Hadlock equation by 4%. This theoretical approach emphasizes the role of proper measurement of AC. Moreover, it seems that in cases of multiple examiners it is efficient and time saving to re-measure only the AC. The discrepancy between the estimated and actual fetal weights was around 6%, with a good correlation of (r) of approximately 0.8. The contribution of additional examiners to the reduction in discrepancies and correlation between the predicted and estimated fetal weights was marginal (Table 2). The discrepancy decreased by 1%, corresponding to approximately 36 g. This statistically significant decrease in discrepancy is not, of course, clinically significant. Interestingly, even with such a low discrepancy, major discrepancies (> 10%) between estimated fetal weight and actual birth weights were found in between seven and nine estimations of each examiner. These figures are in accordance with those
Table 4 Comparison between major discrepancies of each examiner alone (E1, E2 and E3) and combination of examiners Individual examiner E1 E2 E3
Combination of examiners E1 + E2
E1 + E3
E2 + E3
E1 + E2 + E3
3* / 1† NS‡ 4/1 NS — —
3/2 NS — — 3/0 NS
— — 1/0 NS 3/1 NS
4/1 NS 4/0 0.046 5/0 0.025
*Number of major discrepancies of E1 alone (cases in which E1 exclusively had a > 10% error). †Number of major discrepancies of E1 + E2 together (but not by E1 alone). ‡Significance level of the McNemar test for symmetry.
Ultrasound in Obstetrics and Gynecology
59
Prediction of fetal weight by multiple examiners of other studies dealing with the accuracy of ultrasound examination to predict fetal weight2,11. Analysis of correlation curves explains this phenomenon. Most examinations are very close to the identity line and are responsible for the approximately 6% discrepancy. However, a few observations are remote from the identity line and are the source of the major discrepancies. As mentioned previously, such errors may be a result of different fetal specific gravity or suboptimal biometric measurements. The different specific gravities of large- and small-for-gestational age fetuses have long been recognized as a source of the inaccuracy in fetal weight estimations12. Unfortunately, no methodology can solve this limitation. However, suboptimal biometric measurements can and should be corrected. Examination by additional examiners might correct such errors. The contribution of additional examiners was assessed by using the McNemar test. This test made it possible to statistically compare the frequency that one method (single or combination of examiners) was correct and the other was not, to the frequency of the opposite situation. We found that the contribution of an additional examiner decreased the chance of suboptimal measurements and subsequently the number of major discrepancies in fetal weight estimation. It was found that this contribution was limited, since the four fetuses which had had major discrepancies in weight estimation by E1, E2, and E3 were evaluated by each of the examiners with major discrepancies. It seems that in this study the contribution of additional examiners to decrease the subjective discrepancy in measurement was approximately 50% (from 7–9 to 4). Other discrepancies in prediction may be attributable to different factors, but not to suboptimal measurement. In conclusion, we have analyzed the source of differences and the contribution of additional examiners to the sonographic evaluation of fetal weight. Measurement of AC showed a larger interobserver error and a larger effect in the Hadlock equation for weight calculation. Hence,
60
Gull et al. meticulous or even repeated measurements of AC decreased the error in prediction. The use of additional examiners reduced major discrepancies (> 10%) between the estimated fetal weight and actual birth weight. We advocate the use of additional examiners whenever the estimation of fetal weight is of major importance in planning labor and the mode of delivery.
REFERENCES 1 Dudley NJ. Selection of appropriate ultrasound methods for the estimation of fetal weight. Br J Radiol 1995; 68: 385– 8 2 Thompson HO, Casaceli C, Woods JR Jr. Ultrasonographic fetal weight estimation by an integrated computer-assisted system: Can each laboratory improve its accuracy? Am J Obstet Gynecol 1990; 163: 986– 95 3 Hadlock FP, Harrist RB, Sharman RS, Deter RL, Park SK. Estimation of fetal weight with the use of head, body, and femur measurements — A prospective study. Am J Obstet Gynecol 1985; 151: 333 –9 4 Shapard M, Filley RA. A standardized plane for biparietal diameter measurement. J Ultrasound Med 1982; 1: 145– 50 5 Deter RL, Harrist RB, Hadlock FP, Carpenter RJ. Fetal head and abdominal circumferences: II. A critical re-evaluation of the relationship to menstrual age. J Clin Ultrasound 1982; 10: 365– 72 6 Hadlock FP, Harrist RB, Deter RL, Park SK. Fetal femur length as a predictor of menstrual age: sonographically measured. AJR Am J Roentgenol 1982; 138: 875– 8 7 Mongelli M, Gardosi J. Gestation-adjusted projection of estimated fetal weight. Acta Obstet Gynecol Scand 1996; 75: 28– 31 8 Leiberman JR, Fraser D, Weitzman S, Glezerman M. Birthweight curves in southern Israel populations. Isr J Med Sci 1993; 29: 198–203 9 Baker PN, Johnson IR, Gowland PA, Hykin J, Harvey PR, Freeman A, Adams V, Worthington BS, Mansfield P. Fetal weight estimation by echo-planar magnetic resonance imaging. Lancet 1994; 343: 644– 5 10 Bernstein IM, Catalano PM. Influence of fetal fat on the ultrasound estimation of fetal weight in diabetic mothers. Obstet Gynecol 1992; 79: 561– 3 11 Watson WJ, Soisson AP, Harlass FE. Accuracy of estimated weight of the term fetus. J Reprod Med 1988; 33: 369– 71 12 Deter RL. A different way to model fetal growth. Ultrasound Obstet Gynecol 1995; 6: 307– 12
Ultrasound in Obstetrics and Gynecology
