Original Article
Ultrasonographic Fetal Weight Estimation: Should Macrosomia-Specific Formulas Be Utilized? Blake Porter, MD1
Cherry Neely, RDMS2
Jeff Szychowski, PhD2,3
1 Division of Maternal-Fetal Medicine, Department of Obstetrics and
Gynecology, University of Missouri, Kansas City, Missouri 2 Division of Maternal-Fetal Medicine, Department of Obstetrics and Gynecology, University of Alabama at Birmingham, Alabama 3 Department of Biostatistics, University of Alabama at Birmingham, Alabama
John Owen, MD, MSPH2
Address for correspondence John Owen, MD, MSPH, Division of Maternal-Fetal Medicine, Department of Obstetrics and Gynecology, University of Alabama, 176F 10270L 619 19th St S, Birmingham, AL 35249-7333 (e-mail: [email protected]).
Am J Perinatol 2015;32:968–972.
Abstract
Keywords
► macrosomia ► estimated fetal weight ► macrosomia formula ► ultrasound ► ultrasound estimate
Objective This study aims to derive an estimated fetal weight (EFW) formula in macrosomic fetuses, compare its accuracy to the 1986 Hadlock IV formula, and assess whether including maternal diabetes (MDM) improves estimation. Study Design Retrospective review of nonanomalous live-born singletons with birth weight (BWT) 4 kg and biometry within 14 days of birth. Formula accuracy included: (1) mean error (ME ¼ EFW BWT), (2) absolute mean error (AME ¼ absolute value of [1]), and (3) mean percent error (MPE, [1]/BWT 100%). Using loge BWT as the dependent variable, multivariable linear regression produced a macrosomic-specific formula in a “training” dataset which was verified by “validation” data. Formulas specific for MDM were also developed. Results Out of the 403 pregnancies, birth gestational age was 39.5 1.4 weeks, and median BWT was 4,240 g. The macrosomic formula from the training data (n ¼ 201) had associated ME ¼ 54 284 g, AME ¼ 234 167 g, and MPE ¼ 1.6 6.2%; evaluation in the validation dataset (n ¼ 202) showed similar errors. The Hadlock formula had associated ME ¼ 369 422 g, AME ¼ 451 332 g, MPE ¼ 8.3 9.3% (all p < 0.0001). Diabetes-specific formula errors were similar to the macrosomic formula errors (all p ¼ NS). Conclusions With BWT 4 kg, the macrosomic formula was significantly more accurate than Hadlock IV, which systematically underestimates fetal/BWT. Diabetesspecific formulas did not improve accuracy. A specific formula should be considered when macrosomia is suspected.
Macrosomia is commonly defined as a birth weight (BWT) 4kg.1,2 Macrosomia significantly increases morbidities for both the mother (cesarean birth, genital tract lacerations, and hemorrhage) and the neonate (brachial plexus injury and asphyxia).3–7 While the optimal intrapartum management of
the pregnancies complicated by suspected macrosomia is still unclear,8–11 better estimates of fetal weight may allow for improved obstetrical management. While sonography has been proven to be superior to fundal height measurement12–14 in determining BWT, its accuracy for determining
received August 7, 2014 accepted after revision January 8, 2015 published online March 2, 2015
Copyright © 2015 by Thieme Medical Publishers, Inc., 333 Seventh Avenue, New York, NY 10001, USA. Tel: +1(212) 584-4662.
DOI http://dx.doi.org/ 10.1055/s-0035-1545664. ISSN 0735-1631.
Downloaded by: NYU. Copyrighted material.
968
BWT is still imperfect. Previous studies have found that commonly used formulas have sensitivities as low as 53 to 59% for diagnosing BWT 4 kg.15 Currently, most formulas for estimated fetal weight (EFW) are based on ultrasonographic measurements of fetal biometry. Various formulas have been developed to calculate EFW based on these measurements with each formula preferentially utilizing different fetal measurements, product terms, and regression coefficients. A plethora of formulas have been developed, and a few have been assimilated into common clinical use. In the United States, the formulas derived by Hadlock et al are some of the most commonly used.16 The Hadlock formulae are commonly preferred due to their accuracy calculating EFW across a wide range of weights.17 However, internationally, formulas such as those published by Campbell and Wilkin,18 Shepard et al,19 and Merz et al20 are commonly used to estimate fetal weight. The accuracy of predicting BWT by a variety of different formulas has been studied extensively.15,21,22 However, the studies performed thus far are contradictory at best, with most coming to the conclusion that of existing formulas, none are optimal for the estimation of fetal weight and prediction of macrosomia.15,21,22 More recently, investigators have shown that maternal characteristics, including maternal diabetes mellitus (MDM), and first trimester serum analytes can predict macrosomia.23–25 EFW formulas that incorporate selected maternal characteristics other than fetal biometry and derived specifically in a population of macrosomic fetuses might enhance the accuracy of prenatal weight estimation in cases of fetal overgrowth. The objective of this study was to derive an EFW formula in a population of macrosomic fetuses and compare its accuracy to the commonly used Hadlock IV formula.16 Further, we sought to estimate whether inclusion of other maternal characteristics in the formula, specifically diabetes, might further improve the accuracy of fetal weight estimation.
Methods After receiving approval from the Institutional Review Board approval at the University of Alabama at Birmingham, we used our electronic perinatal and linked ultrasound databases to identify all singleton infants born at our institution between January 1, 2005 and December 31, 2012 with a BWT 4.0 kg and complete sonographic biometry. All biometric parameters were ascertained by American Registry of Diagnostic Medical Sonographers (ARDMS)-certified sonographers in our American Institute of Ultrasound in Medicine (AIUM)-certified facility, using GE Voluson 730 Expert or E8 units (GE Healthcare, Milwaukee, WI) equipped with 3.5 to 5.0 MHz curvilinear transducers. The fetal head measurements were made in the axial plane at the level where the continuous midline echo is broken by the cavum septum pellucidum in the anterior third. The head circumference (HC) was measured around the perimeter using an electronic ellipse.26 The abdominal circumference (AC) was measured in a transverse plane that included the umbilical vein and the fetal stomach.26 The fetal femur length (FL) was measured in a view where the full femoral diaphysis was seen in a plane as close as possible to a right angle to the ultrasound beam; measurements were taken from
Porter et al.
one end of the diaphysis to the other, not including the distal femoral epiphysis.26 Measurements were made with calibrated calipers on the machine on frozen images. All images were reviewed and signed by faculty sonologists, board-certified in maternal-fetal medicine. Exclusion criteria included fetal biometric assessment performed > 14 days before birth, fetal death or a prenatally recognized, clinically significant fetal anomaly. Maternal variables included maternal age, self-reported race, parity, diabetes (gestational or pregestational), body mass index, and gestational age at delivery. Ancillary sonographic data included amniotic fluid assessment, placental position, and fetal lie. Using a natural logarithm-transformed BWT as the dependent variable, we selected a multivariable linear regression model to derive regression coefficients for the four biometric parameters considered: biparietal diameter (BPD), HC, AC, and FL. We also included all second order (e.g., AC2) and interaction terms (e.g., AC FL). The most parsimonious model was selected using a forward selection process with an α-value of 0.5 to enter the model and 0.1 to remain. The selected model was derived from the “training” dataset, comprised of a randomly selected subset (50%) of the entire available sample. The formula’s robustness was then evaluated in a “validation” dataset, comprised of the remaining observations. We defined formula validation as a nonsignificant difference between the errors generated in the training dataset from which the formula was derived and the errors observed when the same formula was applied to the validation dataset. If the new formula was validated, we planned to utilize the validated new formula in the entire dataset to compare the performance of the new formula to the Hadlock IV formula. To assess the effect of maternal diabetes on the prediction formula, we included interaction terms of MDM (yes/ no) with the other first-order biometric parameters in a multivariable linear regression model. If the interaction terms were observed to be statistically significant, indicating that MDM was a potential effect modifier, our plan was to stratify diabetic and nondiabetic patients and derive and evaluate MDM-specific formulas. Following assessment of the macrosomic-specific formula derived in the training data in the validation data, the new macrosomic-specific formula was compared with Hadlock IV and also to the MDM-specific formulas in the entire study population. To account for the possibility of unmeasured fetal growth in the scan-to-birth interval, we derived a correction factor (i.e., g/d) using a regression model that included the observed scan-to-birth interval and that also considered the fetal biometric terms as covariates. The coefficient for this term from the model was then applied uniformly to the formula-derived fetal weight estimates (as the product of the coefficient and scan-to-birth interval) for all patients’ with nonzero scan-to-birth intervals. Accuracy of sonographic formulas was assessed by the following: (1) mean error (ME ¼ EFW BWT), (2) absolute mean error (AME ¼ absolute value of [1]), and (3) mean percent error (MPE ¼ [1]/BWT 100%). Categorical variables were compared using the chi-square test of association, while the mean error values were compared using a t-test. American Journal of Perinatology
Vol. 32
No. 10/2015
969
Downloaded by: NYU. Copyrighted material.
Ultrasonographic EFW: Should Macrosomia-Specific Formulas Be Used
Ultrasonographic EFW: Should Macrosomia-Specific Formulas Be Used Our chosen α-level to represent statistical significance was 0.05. All statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC).
Results A total of 403 studies met the inclusion/exclusion criteria. Selected population characteristics are shown in ►Table 1. The mean scan-to-birth interval was 6.7 4.3 days. Notably, 29% of our patients had either gestational or pregestational MDM. The mean BWT was 4,236 g ( 306 g) with a range of 4,000 to 5,770 g. The formula derived from the linear regression model in the training data (N ¼ 201) was: Loge EFW ¼ (11.36717–0.18023 AC) þ (0.00258 AC2) þ (0.00186 BPD FL). It was associated with ME ¼ 54 284 g, AME ¼ 234 167 g, and MPE ¼ 1.62 6.2%. Further evaluation of this formula in the validation data (n ¼ 202) confirmed all comparisons to be statistically nonsignificant; p ¼ 0.54–0.99 (►Table 2) confirm-
Porter et al.
ing the internal validity of the formula. The derived correction factor for the scan-to-birth interval was 9.5 g/d. After verifying that all interaction terms between MDM and each biometric parameter were statistically significant, the stratified diabetes-specific formulas were: Loge EFW ¼ 8.11472 þ (0.00022062 HC AC) if MDM is present, and Loge EFW ¼ 11.36717 þ (0.00018421 AC2) þ (0.00125 BPD FL) if MDM is absent. Having validated the macrosomic formula, we compared its accuracy to both Hadlock IV and the MDM-specific formulas in the entire population (►Table 3). Hadlock IV systematically underestimated BWT (mean error 369 g), and compared with the macrosomic formula, the associated errors were all significantly larger; p < 0.0001. Using a MPE of 10% as a clinically relevant benchmark for accuracy, we observed that, using Hadlock IV, 45% of the Hadlock IV Downloaded by: NYU. Copyrighted material.
970
Table 1 Selected characteristics of the study population, N ¼ 403 n
Characteristics Race African American
169 (42%)
Caucasian
124 (31%)
Hispanic
103 (26%)
Asian
7 (2%)
Preterm birth < 37 wks
21 (5.2%)
Multiparous
284 (70%)
Male Infant
237 (59%)
Amniotic fluid volume Normal
346 (86%)
Oligohydramnios
9 (2.2%)
Polyhydramnios
48 (12%)
Maternal diabetes
118 (29%)
Gestational age at delivery (wk)
39.5 ( 1.4)
Birth weight (g)
4,236 ( 306)
Ultrasound to delivery interval (d)
6.7 ( 4.3)
Note: Values listed as n (%) except where listed as mean standard deviation.
Table 2 Performance of the new formula derived from the general linear model in both the training dataset and the validation dataset
a
Error terms
Training data (n ¼ 201)
Validation data (n ¼ 202)
p-Valuea
Mean error
54 284 g
50 295 g
0.96
Absolute mean error
234 167 g
223 199 g
0.54
Mean percent error
1.6 6.2%
1.6 6.2%
0.99
p-Values reflect comparison of the formula errors from the training data to the errors from the same formula when applied to the validation data.
American Journal of Perinatology
Vol. 32
No. 10/2015
Ultrasonographic EFW: Should Macrosomia-Specific Formulas Be Used
Porter et al.
971
Table 3 Comparison of Hadlock IV, macrosomic, and diabetes-specific formulas Error term
p-Valuea
Hadlock IV Mean error
369 422 g
Absolute mean error
451 333 g
Mean percent error
8.3 9.3%
Macrosomic formula Mean error
53 289 g
< 0.0001
Absolute mean error
229 184 g
< 0.0001
Mean percent error
1.6 6.2%
< 0.0001
Mean error
39 280 g
0.5
Absolute mean error
217 182 g
0.35
Mean percent error
1.3 .6%
0.45
Diabetes-specific formulas
Hadlock IV is the referent for the macrosomic formula; macrosomic formula is the referent for the diabetes-specific formulas.
estimates fell outside this range, compared with 8% of estimates using the macrosomic formula (p < 0.001). However, the MDM-specific formulas, when applied to the diabetic and nondiabetic populations, respectively, performed no better than the macrosomic formula with regards to the intergroup errors; p ¼ 0.35 to 0.5 (►Table 3).
Discussion In cases of BWT 4 kg, our macrosomic-specific formula was significantly more accurate than Hadlock IV. Importantly, we observed that Hadlock IV systematically underestimates fetal BWT in a population of macrosomic infants. Much work has been done previously regarding the proposed clinical interventions for suspected macrosomia.1,7–9 However, all previous work was based on the use of formulas which were not validated in macrosomic populations. To determine the appropriate clinical management of macrosomia we must first become more accurate at predicting its true presence or absence during pregnancy. In our study, the use of our new formula resulted in only 8% of weight estimates occurring outside the error range of 10% compared with 45% of weight estimates occurring outside 10% using the Hadlock formula. Recent studies have suggested the use of parameters other than traditional ultrasound-generated fetal biometrics to estimate fetal weight and better predict macrosomia.23–25 Given that infants of diabetic mothers have a different body composition than infants of nondiabetics, we assessed whether including a parameter for the presence or absence of maternal diabetes would significantly improve fetal weight estimation. As we had hypothesized, the interaction terms between the presence or absence of diabetes and the biometric parameters were all statistically significant, suggesting that body composition might affect formula derivation (and thus performance). Indeed, the formulas derived for the diabetic and nondiabetic gravidas were noticeably different: the diabetic formula included only an HC–AC product term while the nondiabetic
formula included AC2 and BPD–FL product terms. Nevertheless, the formulas derived specifically for the presence and absence of MDM did not significantly improve prediction accuracy when compared with our new formula; the difference was neither clinically nor statistically different. However, our study is not without limitations. First, the retrospective nature of our study has specific inherent issues. The indications for performing the ultrasounds were inconsistent and may affect the accuracy of the ultrasound biometry. For example, if more women were having ultrasounds for obesity and the inability to measure fundal heights, this might increase the chance of measurement error. Second, our population may not be representative of other populations. Further studies would need to be performed to verify the robustness of our formula in a different patient population; it is not surprising that a formula derived in a study population would have a performance advantage. Finally, given that this was a retrospective study, there was not a rigorous study protocol with defined measurement standards for fetal biometry, nor could we assess the accuracy of the individual ultrasonographers. We believe that the strengths of our study outweigh the limitations. As previously mentioned, the strength of our study is its internal validity, based on our ARDMS-certified sonographers, high-quality ultrasound units, and the rigorous nature of our sonographic measurements at our AIUMcertified center. However, this is potentially a limitation to the clinical generalizability of our results. The formula would need to be tested in other clinical settings and confirmed to be an improvement over existing formulas before its widespread clinical use. We were able to develop a new formula and add to the existing tools available to estimate fetal weight; our findings were clinically significant. Although this formula was developed in a single population, it was then validated in a separate cohort with excellent reproducibility before comparing it to another published formula. Given the widespread use of fetal weight estimation and the ability of ultrasound units and reporting software to utilize different formulas, it American Journal of Perinatology
Vol. 32
No. 10/2015
Downloaded by: NYU. Copyrighted material.
a
Ultrasonographic EFW: Should Macrosomia-Specific Formulas Be Used would be relatively easy to clinically implement macrosomicspecific formulas. As mentioned above as a limitation, while we currently have developed and validated the formula in a single population, our results should be generalizable to other populations, and we welcome further testing of its performance at other centers. In conclusion, our findings support the use of a macrosomic-specific formula in cases of suspected macrosomia based on either clinical examination or the Hadlock IV formula. Work still needs to be done for improving the accuracy of a macrosomia-specific formula. One potential area of future research will be to prospectively test our formula in other populations to examine if it is sufficiently robust to be widely applicable. Optimal estimation of fetal weight may allow us to make better recommendations regarding clinical interventions for suspected macrosomia.
10 Chauhan SP, Gherman R, Hendrix NW, Bingham JM, Hayes E.
11
12
13
14 15
16
Note This research was presented in abstract form at the 34th Annual Meeting of the Society of Maternal-Fetal Medicine; February 2014; New Orleans, LA.
17
18
Note The authors have no disclosures of any financial support received for the work being published.
19
20
References 1 Boyd ME, Usher RH, McLean FH. Fetal macrosomia: prediction, risks,
proposed management. Obstet Gynecol 1983;61(6):715–722 2 American College of Obstetricians and Gynecologists. ACOG Prac-
3
4
5
6 7 8
9
tice Bulletin: Number 22, November 2000: Macrosomia. Obstet Gynecol 2000;96(5): Gyurkovits Z, Kálló K, Bakki J, et al. Neonatal outcome of macrosomic infants: an analysis of a two-year period. Eur J Obstet Gynecol Reprod Biol 2011;159(2):289–292 Spellacy WN, Miller S, Winegar A, Peterson PQ. Macrosomia— maternal characteristics and infant complications. Obstet Gynecol 1985;66(2):158–161 Ezegwui HU, Ikeako LC, Egbuji C. Fetal macrosomia: obstetric outcome of 311 cases in UNTH, Enugu, Nigeria. Niger J Clin Pract 2011;14(3):322–326 Handa VL, Danielsen BH, Gilbert WM. Obstetric anal sphincter lacerations. Obstet Gynecol 2001;98(2):225–230 Henriksen T. The macrosomic fetus: a challenge in current obstetrics. Acta Obstet Gynecol Scand 2008;87(2):134–145 Caughey AB. Obstetric ultrasound for estimated fetal weight: is the information more harm than benefit? Am J Obstet Gynecol 2012; 207(4):239–240 Rouse DJ, Owen J, Goldenberg RL, Cliver SP. The effectiveness and costs of elective cesarean delivery for fetal macrosomia diagnosed by ultrasound. JAMA 1996;276(18):1480–1486
American Journal of Perinatology
Vol. 32
No. 10/2015
Porter et al.
21
22
23
24
25
26
Shoulder dystocia: comparison of the ACOG practice bulletin with another national guideline. Am J Perinatol 2010;27(2):129–136 Zhang X, Decker A, Platt RW, Kramer MS. How big is too big? The perinatal consequences of fetal macrosomia. Am J Obstet Gynecol 2008;198(5):517.e1–517.e6 Okonofua FE, Ayangade SO, Chan RC, O’Brien PM. A prospective comparison of clinical and ultrasonic methods of predicting normal and abnormal fetal growth. Int J Gynaecol Obstet 1986; 24(6):447–451 Kayem G, Grangé G, Bréart G, Goffinet F. Comparison of fundal height measurement and sonographically measured fetal abdominal circumference in the prediction of high and low birth weight at term. Ultrasound Obstet Gynecol 2009;34(5):566–571 Farrell T, Holmes R, Stone P. The effect of body mass index on three methods of fetal weight estimation. BJOG 2002;109(6):651–657 Hoopmann M, Abele H, Wagner N, Wallwiener D, Kagan KO. Performance of 36 different weight estimation formulae in fetuses with macrosomia. Fetal Diagn Ther 2010;27(4):204–213 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(3): 333–337 Kurmanavicius J, Burkhardt T, Wisser J, Huch R. Ultrasonographic fetal weight estimation: accuracy of formulas and accuracy of examiners by birth weight from 500 to 5000 g. J Perinat Med 2004; 32(2):155–161 Campbell S, Wilkin D. Ultrasonic measurement of fetal abdomen circumference in the estimation of fetal weight. Br J Obstet Gynaecol 1975;82(9):689–697 Shepard MJ, Richards VA, Berkowitz RL, Warsof SL, Hobbins JC. An evaluation of two equations for predicting fetal weight by ultrasound. Am J Obstet Gynecol 1982;142(1):47–54 Merz E, Lieser H, Schicketanz KH, Härle J. [Intrauterine fetal weight assessment using ultrasound. A comparison of several weight assessment methods and development of a new formula for the determination of fetal weight]. Ultraschall Med 1988;9(1): 15–24 Melamed N, Yogev Y, Meizner I, Mashiach R, Pardo J, Ben-Haroush A. Prediction of fetal macrosomia: effect of sonographic fetal weight-estimation model and threshold used. Ultrasound Obstet Gynecol 2011;38(1):74–81 Combs CA, Rosenn B, Miodovnik M, Siddiqi TA. Sonographic EFW and macrosomia: is there an optimum formula to predict diabetic fetal macrosomia? J Matern Fetal Med 2000;9(1):55–61 Poon LC, Karagiannis G, Stratieva V, Syngelaki A, Nicolaides KH. First-trimester prediction of macrosomia. Fetal Diagn Ther 2011; 29(2):139–147 Boisvert MR, Koski KG, Burns DH, Skinner CD. Early prediction of macrosomia based on an analysis of second trimester amniotic fluid by capillary electrophoresis. Biomarkers Med 2012;6(5): 655–662 Kuc S, Wortelboer EJ, Koster MP, de Valk HW, Schielen PC, Visser GH. Prediction of macrosomia at birth in type-1 and 2 diabetic pregnancies with biomarkers of early placentation. BJOG 2011; 118(6):748–754 Papageorghiou AT, Sarris I, Ioannou C, et al; International Fetal and Newborn Growth Consortium for the 21st Century. Ultrasound methodology used to construct the fetal growth standards in the INTERGROWTH-21st Project. BJOG 2013;120(Suppl 2):27–32, v
Downloaded by: NYU. Copyrighted material.
972
Ultrasonographic Fetal Weight Estimation: Should Macrosomia-Specific Formulas Be Utilized? Blake Porter, MD1
Cherry Neely, RDMS2
Jeff Szychowski, PhD2,3
1 Division of Maternal-Fetal Medicine, Department of Obstetrics and
Gynecology, University of Missouri, Kansas City, Missouri 2 Division of Maternal-Fetal Medicine, Department of Obstetrics and Gynecology, University of Alabama at Birmingham, Alabama 3 Department of Biostatistics, University of Alabama at Birmingham, Alabama
John Owen, MD, MSPH2
Address for correspondence John Owen, MD, MSPH, Division of Maternal-Fetal Medicine, Department of Obstetrics and Gynecology, University of Alabama, 176F 10270L 619 19th St S, Birmingham, AL 35249-7333 (e-mail: [email protected]).
Am J Perinatol 2015;32:968–972.
Abstract
Keywords
► macrosomia ► estimated fetal weight ► macrosomia formula ► ultrasound ► ultrasound estimate
Objective This study aims to derive an estimated fetal weight (EFW) formula in macrosomic fetuses, compare its accuracy to the 1986 Hadlock IV formula, and assess whether including maternal diabetes (MDM) improves estimation. Study Design Retrospective review of nonanomalous live-born singletons with birth weight (BWT) 4 kg and biometry within 14 days of birth. Formula accuracy included: (1) mean error (ME ¼ EFW BWT), (2) absolute mean error (AME ¼ absolute value of [1]), and (3) mean percent error (MPE, [1]/BWT 100%). Using loge BWT as the dependent variable, multivariable linear regression produced a macrosomic-specific formula in a “training” dataset which was verified by “validation” data. Formulas specific for MDM were also developed. Results Out of the 403 pregnancies, birth gestational age was 39.5 1.4 weeks, and median BWT was 4,240 g. The macrosomic formula from the training data (n ¼ 201) had associated ME ¼ 54 284 g, AME ¼ 234 167 g, and MPE ¼ 1.6 6.2%; evaluation in the validation dataset (n ¼ 202) showed similar errors. The Hadlock formula had associated ME ¼ 369 422 g, AME ¼ 451 332 g, MPE ¼ 8.3 9.3% (all p < 0.0001). Diabetes-specific formula errors were similar to the macrosomic formula errors (all p ¼ NS). Conclusions With BWT 4 kg, the macrosomic formula was significantly more accurate than Hadlock IV, which systematically underestimates fetal/BWT. Diabetesspecific formulas did not improve accuracy. A specific formula should be considered when macrosomia is suspected.
Macrosomia is commonly defined as a birth weight (BWT) 4kg.1,2 Macrosomia significantly increases morbidities for both the mother (cesarean birth, genital tract lacerations, and hemorrhage) and the neonate (brachial plexus injury and asphyxia).3–7 While the optimal intrapartum management of
the pregnancies complicated by suspected macrosomia is still unclear,8–11 better estimates of fetal weight may allow for improved obstetrical management. While sonography has been proven to be superior to fundal height measurement12–14 in determining BWT, its accuracy for determining
received August 7, 2014 accepted after revision January 8, 2015 published online March 2, 2015
Copyright © 2015 by Thieme Medical Publishers, Inc., 333 Seventh Avenue, New York, NY 10001, USA. Tel: +1(212) 584-4662.
DOI http://dx.doi.org/ 10.1055/s-0035-1545664. ISSN 0735-1631.
Downloaded by: NYU. Copyrighted material.
968
BWT is still imperfect. Previous studies have found that commonly used formulas have sensitivities as low as 53 to 59% for diagnosing BWT 4 kg.15 Currently, most formulas for estimated fetal weight (EFW) are based on ultrasonographic measurements of fetal biometry. Various formulas have been developed to calculate EFW based on these measurements with each formula preferentially utilizing different fetal measurements, product terms, and regression coefficients. A plethora of formulas have been developed, and a few have been assimilated into common clinical use. In the United States, the formulas derived by Hadlock et al are some of the most commonly used.16 The Hadlock formulae are commonly preferred due to their accuracy calculating EFW across a wide range of weights.17 However, internationally, formulas such as those published by Campbell and Wilkin,18 Shepard et al,19 and Merz et al20 are commonly used to estimate fetal weight. The accuracy of predicting BWT by a variety of different formulas has been studied extensively.15,21,22 However, the studies performed thus far are contradictory at best, with most coming to the conclusion that of existing formulas, none are optimal for the estimation of fetal weight and prediction of macrosomia.15,21,22 More recently, investigators have shown that maternal characteristics, including maternal diabetes mellitus (MDM), and first trimester serum analytes can predict macrosomia.23–25 EFW formulas that incorporate selected maternal characteristics other than fetal biometry and derived specifically in a population of macrosomic fetuses might enhance the accuracy of prenatal weight estimation in cases of fetal overgrowth. The objective of this study was to derive an EFW formula in a population of macrosomic fetuses and compare its accuracy to the commonly used Hadlock IV formula.16 Further, we sought to estimate whether inclusion of other maternal characteristics in the formula, specifically diabetes, might further improve the accuracy of fetal weight estimation.
Methods After receiving approval from the Institutional Review Board approval at the University of Alabama at Birmingham, we used our electronic perinatal and linked ultrasound databases to identify all singleton infants born at our institution between January 1, 2005 and December 31, 2012 with a BWT 4.0 kg and complete sonographic biometry. All biometric parameters were ascertained by American Registry of Diagnostic Medical Sonographers (ARDMS)-certified sonographers in our American Institute of Ultrasound in Medicine (AIUM)-certified facility, using GE Voluson 730 Expert or E8 units (GE Healthcare, Milwaukee, WI) equipped with 3.5 to 5.0 MHz curvilinear transducers. The fetal head measurements were made in the axial plane at the level where the continuous midline echo is broken by the cavum septum pellucidum in the anterior third. The head circumference (HC) was measured around the perimeter using an electronic ellipse.26 The abdominal circumference (AC) was measured in a transverse plane that included the umbilical vein and the fetal stomach.26 The fetal femur length (FL) was measured in a view where the full femoral diaphysis was seen in a plane as close as possible to a right angle to the ultrasound beam; measurements were taken from
Porter et al.
one end of the diaphysis to the other, not including the distal femoral epiphysis.26 Measurements were made with calibrated calipers on the machine on frozen images. All images were reviewed and signed by faculty sonologists, board-certified in maternal-fetal medicine. Exclusion criteria included fetal biometric assessment performed > 14 days before birth, fetal death or a prenatally recognized, clinically significant fetal anomaly. Maternal variables included maternal age, self-reported race, parity, diabetes (gestational or pregestational), body mass index, and gestational age at delivery. Ancillary sonographic data included amniotic fluid assessment, placental position, and fetal lie. Using a natural logarithm-transformed BWT as the dependent variable, we selected a multivariable linear regression model to derive regression coefficients for the four biometric parameters considered: biparietal diameter (BPD), HC, AC, and FL. We also included all second order (e.g., AC2) and interaction terms (e.g., AC FL). The most parsimonious model was selected using a forward selection process with an α-value of 0.5 to enter the model and 0.1 to remain. The selected model was derived from the “training” dataset, comprised of a randomly selected subset (50%) of the entire available sample. The formula’s robustness was then evaluated in a “validation” dataset, comprised of the remaining observations. We defined formula validation as a nonsignificant difference between the errors generated in the training dataset from which the formula was derived and the errors observed when the same formula was applied to the validation dataset. If the new formula was validated, we planned to utilize the validated new formula in the entire dataset to compare the performance of the new formula to the Hadlock IV formula. To assess the effect of maternal diabetes on the prediction formula, we included interaction terms of MDM (yes/ no) with the other first-order biometric parameters in a multivariable linear regression model. If the interaction terms were observed to be statistically significant, indicating that MDM was a potential effect modifier, our plan was to stratify diabetic and nondiabetic patients and derive and evaluate MDM-specific formulas. Following assessment of the macrosomic-specific formula derived in the training data in the validation data, the new macrosomic-specific formula was compared with Hadlock IV and also to the MDM-specific formulas in the entire study population. To account for the possibility of unmeasured fetal growth in the scan-to-birth interval, we derived a correction factor (i.e., g/d) using a regression model that included the observed scan-to-birth interval and that also considered the fetal biometric terms as covariates. The coefficient for this term from the model was then applied uniformly to the formula-derived fetal weight estimates (as the product of the coefficient and scan-to-birth interval) for all patients’ with nonzero scan-to-birth intervals. Accuracy of sonographic formulas was assessed by the following: (1) mean error (ME ¼ EFW BWT), (2) absolute mean error (AME ¼ absolute value of [1]), and (3) mean percent error (MPE ¼ [1]/BWT 100%). Categorical variables were compared using the chi-square test of association, while the mean error values were compared using a t-test. American Journal of Perinatology
Vol. 32
No. 10/2015
969
Downloaded by: NYU. Copyrighted material.
Ultrasonographic EFW: Should Macrosomia-Specific Formulas Be Used
Ultrasonographic EFW: Should Macrosomia-Specific Formulas Be Used Our chosen α-level to represent statistical significance was 0.05. All statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC).
Results A total of 403 studies met the inclusion/exclusion criteria. Selected population characteristics are shown in ►Table 1. The mean scan-to-birth interval was 6.7 4.3 days. Notably, 29% of our patients had either gestational or pregestational MDM. The mean BWT was 4,236 g ( 306 g) with a range of 4,000 to 5,770 g. The formula derived from the linear regression model in the training data (N ¼ 201) was: Loge EFW ¼ (11.36717–0.18023 AC) þ (0.00258 AC2) þ (0.00186 BPD FL). It was associated with ME ¼ 54 284 g, AME ¼ 234 167 g, and MPE ¼ 1.62 6.2%. Further evaluation of this formula in the validation data (n ¼ 202) confirmed all comparisons to be statistically nonsignificant; p ¼ 0.54–0.99 (►Table 2) confirm-
Porter et al.
ing the internal validity of the formula. The derived correction factor for the scan-to-birth interval was 9.5 g/d. After verifying that all interaction terms between MDM and each biometric parameter were statistically significant, the stratified diabetes-specific formulas were: Loge EFW ¼ 8.11472 þ (0.00022062 HC AC) if MDM is present, and Loge EFW ¼ 11.36717 þ (0.00018421 AC2) þ (0.00125 BPD FL) if MDM is absent. Having validated the macrosomic formula, we compared its accuracy to both Hadlock IV and the MDM-specific formulas in the entire population (►Table 3). Hadlock IV systematically underestimated BWT (mean error 369 g), and compared with the macrosomic formula, the associated errors were all significantly larger; p < 0.0001. Using a MPE of 10% as a clinically relevant benchmark for accuracy, we observed that, using Hadlock IV, 45% of the Hadlock IV Downloaded by: NYU. Copyrighted material.
970
Table 1 Selected characteristics of the study population, N ¼ 403 n
Characteristics Race African American
169 (42%)
Caucasian
124 (31%)
Hispanic
103 (26%)
Asian
7 (2%)
Preterm birth < 37 wks
21 (5.2%)
Multiparous
284 (70%)
Male Infant
237 (59%)
Amniotic fluid volume Normal
346 (86%)
Oligohydramnios
9 (2.2%)
Polyhydramnios
48 (12%)
Maternal diabetes
118 (29%)
Gestational age at delivery (wk)
39.5 ( 1.4)
Birth weight (g)
4,236 ( 306)
Ultrasound to delivery interval (d)
6.7 ( 4.3)
Note: Values listed as n (%) except where listed as mean standard deviation.
Table 2 Performance of the new formula derived from the general linear model in both the training dataset and the validation dataset
a
Error terms
Training data (n ¼ 201)
Validation data (n ¼ 202)
p-Valuea
Mean error
54 284 g
50 295 g
0.96
Absolute mean error
234 167 g
223 199 g
0.54
Mean percent error
1.6 6.2%
1.6 6.2%
0.99
p-Values reflect comparison of the formula errors from the training data to the errors from the same formula when applied to the validation data.
American Journal of Perinatology
Vol. 32
No. 10/2015
Ultrasonographic EFW: Should Macrosomia-Specific Formulas Be Used
Porter et al.
971
Table 3 Comparison of Hadlock IV, macrosomic, and diabetes-specific formulas Error term
p-Valuea
Hadlock IV Mean error
369 422 g
Absolute mean error
451 333 g
Mean percent error
8.3 9.3%
Macrosomic formula Mean error
53 289 g
< 0.0001
Absolute mean error
229 184 g
< 0.0001
Mean percent error
1.6 6.2%
< 0.0001
Mean error
39 280 g
0.5
Absolute mean error
217 182 g
0.35
Mean percent error
1.3 .6%
0.45
Diabetes-specific formulas
Hadlock IV is the referent for the macrosomic formula; macrosomic formula is the referent for the diabetes-specific formulas.
estimates fell outside this range, compared with 8% of estimates using the macrosomic formula (p < 0.001). However, the MDM-specific formulas, when applied to the diabetic and nondiabetic populations, respectively, performed no better than the macrosomic formula with regards to the intergroup errors; p ¼ 0.35 to 0.5 (►Table 3).
Discussion In cases of BWT 4 kg, our macrosomic-specific formula was significantly more accurate than Hadlock IV. Importantly, we observed that Hadlock IV systematically underestimates fetal BWT in a population of macrosomic infants. Much work has been done previously regarding the proposed clinical interventions for suspected macrosomia.1,7–9 However, all previous work was based on the use of formulas which were not validated in macrosomic populations. To determine the appropriate clinical management of macrosomia we must first become more accurate at predicting its true presence or absence during pregnancy. In our study, the use of our new formula resulted in only 8% of weight estimates occurring outside the error range of 10% compared with 45% of weight estimates occurring outside 10% using the Hadlock formula. Recent studies have suggested the use of parameters other than traditional ultrasound-generated fetal biometrics to estimate fetal weight and better predict macrosomia.23–25 Given that infants of diabetic mothers have a different body composition than infants of nondiabetics, we assessed whether including a parameter for the presence or absence of maternal diabetes would significantly improve fetal weight estimation. As we had hypothesized, the interaction terms between the presence or absence of diabetes and the biometric parameters were all statistically significant, suggesting that body composition might affect formula derivation (and thus performance). Indeed, the formulas derived for the diabetic and nondiabetic gravidas were noticeably different: the diabetic formula included only an HC–AC product term while the nondiabetic
formula included AC2 and BPD–FL product terms. Nevertheless, the formulas derived specifically for the presence and absence of MDM did not significantly improve prediction accuracy when compared with our new formula; the difference was neither clinically nor statistically different. However, our study is not without limitations. First, the retrospective nature of our study has specific inherent issues. The indications for performing the ultrasounds were inconsistent and may affect the accuracy of the ultrasound biometry. For example, if more women were having ultrasounds for obesity and the inability to measure fundal heights, this might increase the chance of measurement error. Second, our population may not be representative of other populations. Further studies would need to be performed to verify the robustness of our formula in a different patient population; it is not surprising that a formula derived in a study population would have a performance advantage. Finally, given that this was a retrospective study, there was not a rigorous study protocol with defined measurement standards for fetal biometry, nor could we assess the accuracy of the individual ultrasonographers. We believe that the strengths of our study outweigh the limitations. As previously mentioned, the strength of our study is its internal validity, based on our ARDMS-certified sonographers, high-quality ultrasound units, and the rigorous nature of our sonographic measurements at our AIUMcertified center. However, this is potentially a limitation to the clinical generalizability of our results. The formula would need to be tested in other clinical settings and confirmed to be an improvement over existing formulas before its widespread clinical use. We were able to develop a new formula and add to the existing tools available to estimate fetal weight; our findings were clinically significant. Although this formula was developed in a single population, it was then validated in a separate cohort with excellent reproducibility before comparing it to another published formula. Given the widespread use of fetal weight estimation and the ability of ultrasound units and reporting software to utilize different formulas, it American Journal of Perinatology
Vol. 32
No. 10/2015
Downloaded by: NYU. Copyrighted material.
a
Ultrasonographic EFW: Should Macrosomia-Specific Formulas Be Used would be relatively easy to clinically implement macrosomicspecific formulas. As mentioned above as a limitation, while we currently have developed and validated the formula in a single population, our results should be generalizable to other populations, and we welcome further testing of its performance at other centers. In conclusion, our findings support the use of a macrosomic-specific formula in cases of suspected macrosomia based on either clinical examination or the Hadlock IV formula. Work still needs to be done for improving the accuracy of a macrosomia-specific formula. One potential area of future research will be to prospectively test our formula in other populations to examine if it is sufficiently robust to be widely applicable. Optimal estimation of fetal weight may allow us to make better recommendations regarding clinical interventions for suspected macrosomia.
10 Chauhan SP, Gherman R, Hendrix NW, Bingham JM, Hayes E.
11
12
13
14 15
16
Note This research was presented in abstract form at the 34th Annual Meeting of the Society of Maternal-Fetal Medicine; February 2014; New Orleans, LA.
17
18
Note The authors have no disclosures of any financial support received for the work being published.
19
20
References 1 Boyd ME, Usher RH, McLean FH. Fetal macrosomia: prediction, risks,
proposed management. Obstet Gynecol 1983;61(6):715–722 2 American College of Obstetricians and Gynecologists. ACOG Prac-
3
4
5
6 7 8
9
tice Bulletin: Number 22, November 2000: Macrosomia. Obstet Gynecol 2000;96(5): Gyurkovits Z, Kálló K, Bakki J, et al. Neonatal outcome of macrosomic infants: an analysis of a two-year period. Eur J Obstet Gynecol Reprod Biol 2011;159(2):289–292 Spellacy WN, Miller S, Winegar A, Peterson PQ. Macrosomia— maternal characteristics and infant complications. Obstet Gynecol 1985;66(2):158–161 Ezegwui HU, Ikeako LC, Egbuji C. Fetal macrosomia: obstetric outcome of 311 cases in UNTH, Enugu, Nigeria. Niger J Clin Pract 2011;14(3):322–326 Handa VL, Danielsen BH, Gilbert WM. Obstetric anal sphincter lacerations. Obstet Gynecol 2001;98(2):225–230 Henriksen T. The macrosomic fetus: a challenge in current obstetrics. Acta Obstet Gynecol Scand 2008;87(2):134–145 Caughey AB. Obstetric ultrasound for estimated fetal weight: is the information more harm than benefit? Am J Obstet Gynecol 2012; 207(4):239–240 Rouse DJ, Owen J, Goldenberg RL, Cliver SP. The effectiveness and costs of elective cesarean delivery for fetal macrosomia diagnosed by ultrasound. JAMA 1996;276(18):1480–1486
American Journal of Perinatology
Vol. 32
No. 10/2015
Porter et al.
21
22
23
24
25
26
Shoulder dystocia: comparison of the ACOG practice bulletin with another national guideline. Am J Perinatol 2010;27(2):129–136 Zhang X, Decker A, Platt RW, Kramer MS. How big is too big? The perinatal consequences of fetal macrosomia. Am J Obstet Gynecol 2008;198(5):517.e1–517.e6 Okonofua FE, Ayangade SO, Chan RC, O’Brien PM. A prospective comparison of clinical and ultrasonic methods of predicting normal and abnormal fetal growth. Int J Gynaecol Obstet 1986; 24(6):447–451 Kayem G, Grangé G, Bréart G, Goffinet F. Comparison of fundal height measurement and sonographically measured fetal abdominal circumference in the prediction of high and low birth weight at term. Ultrasound Obstet Gynecol 2009;34(5):566–571 Farrell T, Holmes R, Stone P. The effect of body mass index on three methods of fetal weight estimation. BJOG 2002;109(6):651–657 Hoopmann M, Abele H, Wagner N, Wallwiener D, Kagan KO. Performance of 36 different weight estimation formulae in fetuses with macrosomia. Fetal Diagn Ther 2010;27(4):204–213 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(3): 333–337 Kurmanavicius J, Burkhardt T, Wisser J, Huch R. Ultrasonographic fetal weight estimation: accuracy of formulas and accuracy of examiners by birth weight from 500 to 5000 g. J Perinat Med 2004; 32(2):155–161 Campbell S, Wilkin D. Ultrasonic measurement of fetal abdomen circumference in the estimation of fetal weight. Br J Obstet Gynaecol 1975;82(9):689–697 Shepard MJ, Richards VA, Berkowitz RL, Warsof SL, Hobbins JC. An evaluation of two equations for predicting fetal weight by ultrasound. Am J Obstet Gynecol 1982;142(1):47–54 Merz E, Lieser H, Schicketanz KH, Härle J. [Intrauterine fetal weight assessment using ultrasound. A comparison of several weight assessment methods and development of a new formula for the determination of fetal weight]. Ultraschall Med 1988;9(1): 15–24 Melamed N, Yogev Y, Meizner I, Mashiach R, Pardo J, Ben-Haroush A. Prediction of fetal macrosomia: effect of sonographic fetal weight-estimation model and threshold used. Ultrasound Obstet Gynecol 2011;38(1):74–81 Combs CA, Rosenn B, Miodovnik M, Siddiqi TA. Sonographic EFW and macrosomia: is there an optimum formula to predict diabetic fetal macrosomia? J Matern Fetal Med 2000;9(1):55–61 Poon LC, Karagiannis G, Stratieva V, Syngelaki A, Nicolaides KH. First-trimester prediction of macrosomia. Fetal Diagn Ther 2011; 29(2):139–147 Boisvert MR, Koski KG, Burns DH, Skinner CD. Early prediction of macrosomia based on an analysis of second trimester amniotic fluid by capillary electrophoresis. Biomarkers Med 2012;6(5): 655–662 Kuc S, Wortelboer EJ, Koster MP, de Valk HW, Schielen PC, Visser GH. Prediction of macrosomia at birth in type-1 and 2 diabetic pregnancies with biomarkers of early placentation. BJOG 2011; 118(6):748–754 Papageorghiou AT, Sarris I, Ioannou C, et al; International Fetal and Newborn Growth Consortium for the 21st Century. Ultrasound methodology used to construct the fetal growth standards in the INTERGROWTH-21st Project. BJOG 2013;120(Suppl 2):27–32, v
Downloaded by: NYU. Copyrighted material.
972
