Twin pregnancy poses a higher risk than singleton pregnancy of preterm delivery and fetal growth restriction, in addition to particular risks of weight discordancy, twin-to-twin transfusion syndrome, and others. Therefore, the accurate estimation of fetal weight in twin gestations is very important for proper management of pregnancy and selection of the optimal mode and timing of delivery. However, both the Hadlock and Shepard formulas of estimated fetal weight, which are currently in use,1,2 are based on singleton pregnancies, and even in singleton pregnancies, the standard deviation of estimated fetal weight errors is as high as 10–15%.3 Data on the accuracy of these formulas for twin pregnancies are sparse. In one of the few available studies, Roberts et al4 calculated an overall accuracy rate of 64–76% when the examination-to-delivery interval was not longer than 72 hours. Lynch et al5 found that the accuracy was similar for twin and singleton pregnancies when the fetal weight was more than 2.500 g. Jensen and Jensen6 reported a deviation of less than 10% in 72% of twin pregnancies.
Even less attention has been addressed to the subgroup of small for gestational age (SGA) fetuses in twin pregnancies. Ong et al7 reported a sensitivity of 35.7% and a specificity of 96.8% for the ultrasonographic estimated fetal weight in this setting. Others calculated the percentage error for absolute weight without correlating it with gestational age.5
The ability to predict discordancy is even less accurate than predicting individual fetal weight. Gernt et al8 calculated a sensitivity of 55% and specificity of 97% when the examination-to-delivery interval was 16 days or less. In other studies using two different statistical models, both the calculated positive predictive value and the sensitivity for discordancy were low.7,9
The aim of the present study was to compare the accuracy of fetal weight estimations between normal and growth-restricted twin and singleton pregnancies in a single tertiary center. Considering the different intrauterine environment in twin pregnancies, manifested by the different growth pattern seen in twin pregnancies during the third trimester,10 it is possible that the singleton-based formulas will be less accurate in weight estimation for twins.
MATERIALS AND METHODS
We used a case–control design, using our single-tertiary center ultrasonographic examinations database. The study was approved by the institutional review board of our medical center. All ultrasonographic fetal weight estimations at our center are performed in the ultrasound unit of the obstetrics and gynecology department. The examinations are performed by experienced ultrasound technicians (of at least 7 years of experience), or by senior physicians who are ultrasound specialists. If performed by a technician, the evaluation is reviewed and signed by a senior physician. Interrater reliability was not examined. The routine evaluation includes the standard fetal biometry measurements: abdominal circumference, femur length, biparietal diameter (BPD), and head circumference. The results for each measure are entered directly into the database.
The study group included all fetal weight estimations performed for twin pregnancies between the years 2001 and 2006. The examination-to-delivery interval did not exceed 3 days. Findings were compared with estimates made in singleton pregnancies immediately after each index twin pregnancy at a 3:1 ratio. Inclusion criteria for both groups were live fetus, birth weight more than 500 g, gestational age more than 24 weeks, and absence of fetal malformations or hydrops. Pregnancies complicated by gestational or pregestational diabetes were excluded as well. Other maternal conditions (preeclampsia, thyroid disorders, renal disease, etc) were not examined. Data on antenatal measures, gestational age at delivery, and actual birth weight were obtained from the center’s perinatal database.
In all cases, the estimated fetal weight was calculated with the Hadlock equation: [Log10 birth weight=1.335–0.0034(abdominal circumference) (femur length)+0.0316(BPD)+0.0457(abdominal circumference)+0.1623(femur length)].7 This is the formula commonly used in Israel and our center. To date, no formula has been found superior to the Hadlock equation for the Israeli population. The accuracy of the estimations was assessed with the following measures: correlation with the actual birth weight (using Pearson correlation coefficient); mean absolute error [mean absolute error=absolute value of (birth weight–estimated fetal weight)]; mean absolute percentage error [mean absolute percentage error=absolute value of (birth weight–estimated fetal weight)/birth weight]; and percentage of estimates within 10% and 15% of the actual birth weight. In addition, we calculated the sensitivity, specificity, positive predictive value, negative predictive value, positive likelihood ratio [likelihood ratio+, defined as sensitivity/(1–specificity)], negative likelihood ratio [likelihood ratio–, defined as (1–sensitivity)/specificity], and overall accuracy [defined as (true negative and true positive cases)/(all cases)] for detecting fetal growth restriction (both singleton and twin pregnancies, defined as a fetal weight below the fifth percentile) and discordancy (twin pregnancies only, defined as a weight disparity of more than 25%, using the larger twin as the index). Fetal growth restriction was defined as a fetal weight below the fifth percentile using the Dolberg curves, an Israeli population-based liveborn infants standard for birth weights commonly used in Israel.
Data were analyzed with the SPSS 13.0 (SPSS Inc., Chicago, IL). Student t test was used to compare continuous variables between groups, and χ2 test was used for categorical variables. P<.05 was considered statistically significant.
The study group consisted of 278 twin pregnancies (576 fetuses), and the control group, 834 pregnancies. Table 1 presents the demographic and obstetric characteristics of the two groups. There was a significant difference between the groups in parity, gestational age at delivery, distribution of birth weight, and rate of fetal growth restriction. Comparison within the twin group yielded a higher mean weight (±standard deviation) for the first twins than for the second twins (2,043±514 g compared with 1,942±509 g, P=.021).
The accuracy of the ultrasonographic fetal weight estimations according to the measures used is shown in Table 2. For all measures, the accuracy was significantly higher in the singleton group than in the twin group. In addition, the 8.9% mean absolute percentage error in the twin group was significantly higher than the 6.8% in the singleton group (P<.001). Within the twin group, the calculated accuracy was significantly better for the first twins by all measures, and the mean absolute percentage error was significantly lower for the first twins than for the second twins (7.5% compared with 10.3%, P<.001). These differences remained significant even when controlling for parity and gestational age at delivery.
The separate analysis of the growth-restricted gestations is presented in Table 3. The overall incidence of fetal growth restriction was higher in the twin group than the singleton group (16% compared with 5%), and higher for the second twins than for the first twins (23% compared with 9%). The twin fetal growth restriction pregnancies were also characterized by a lower positive likelihood ratio (11.4 compared with 20.5) and lower overall accuracy (88.1% compared with 95.3%). The sensitivity and specificity of the ultrasound estimated fetal weight were similar for the twin and singleton fetal growth restriction pregnancies, with an area under the receiver operating characteristics curve of 0.932 for twin pregnancies and 0.962 for singleton pregnancies. For most of the individual measures, the accuracy was better for the second twins than for the first twins; however, the overall accuracy was better for the first twins (90.6% compared with 85.6%).
The accuracy of ultrasonography in predicting discordancy is shown in Table 4. Discordancy was found in 19.4% of the twin pregnancies. The calculated sensitivity was 52%, and the specificity, 88%; the overall accuracy was 81%.
Ultrasonographic estimation of fetal weight is routinely used in the management of multiple pregnancies. The estimation of fetal weight is at times an important tool in clinical decision-making, timing of delivery, and mode of delivery. The present study demonstrates that the accuracy of ultrasonographic fetal weight estimations is greater for singleton than for twin pregnancies, overall and in the setting of fetal growth restriction, when the interval from the estimated fetal weight measurements to delivery is less than 3 days.
The significant differences between the twin and singleton groups in parity and rate of nulliparous women (Table 1) was probably attributable to the higher rate of fertility treatments preceding the twin pregnancies. We could not identify pregnancies conceived by fertility treatments.
The accuracy found here for singleton pregnancies is in accord with previous studies.11 Significant differences in the accuracy of estimated fetal weight could be seen when comparing twin gestations to singleton. These discrepancies were not observed by others.5 Regarding the twin pregnancies, the mean percentage error was higher than reported by Ong et al,7 perhaps owing to the higher proportion of twins with birth weights more than 3,000 g in their study (16%) compared with ours (3%). Ong et al7 noted a significantly reduced mean percentage error with higher birth weight. However, the proportion of fetuses with an estimated fetal weight within 10% of the actual weight was similar in the two studies. As in the present study, Ong et al calculated the estimated fetal weight with the Hadlock formula and other formulas. It is possible that the Hadlock formula is not appropriate for twin gestations, although previous attempts to formulate a specific equation did not achieve optimal results.7
The significant difference noted here in the accuracy of the ultrasonographic estimated fetal weight between the first and the second twins is noteworthy. Comparisons of the first twins and second twins with the singleton pregnancies yielded a greater difference for the second twins. Nevertheless, although the difference for the first twins was lower, it was still apparent. They might be partially explained by the higher proportion of nonvertex presentations among the second twins, causing dolichocephaly and smaller-than-anticipated BPD measurements.12,13
Fetal growth restriction is a common complication of twin gestation and a major target of surveillance during pregnancy. Our study found that the overall accuracy of the ultrasonographic estimated fetal weight for detecting fetal growth restriction is high in twin pregnancies, both for the first and second twins, and it does not differ substantially from singleton fetal growth restriction pregnancies. A low sensitivity and positive predictive value were observed both for the singleton fetuses and for the first twins, with higher values for the second twins. The latter finding might have been attributable to the high rates of fetal growth restriction in the second twins compared with the first twins and the singletons. Our detection rates for fetal growth restriction were slightly better than those of Ong et al,7 although we defined fetal growth restriction as a fetal weight below the fifth percentile, as opposed to the 10th percentile in the study of Ong et al. Furthermore, our maximal interval from examination to delivery was 3 days as opposed to 10 days. It is also possible that growth-restricted fetuses are characterized by specific abnormal growth patterns that affect the accuracy of estimated fetal weight. Therefore, they may warrant separate analysis. This concept has important clinical implications considering the high morbidity and mortality associated with fetal growth restriction and the high rate of interventions in twin pregnancies complicated by fetal growth restriction.
Discordancy is defined as an intertwin weight difference of more than 25%. In previous publications, discordancy has been related to higher rates of preterm deliveries, intrauterine fetal death and neonatal morbidity,14 especially in the presence of fetal growth restriction.15 Compared with other studies,9,16 this study observed higher sensitivity for detection of discordancy. Gernt8 observed higher specificity, positive predictive value, and negative predictive value rates with similar sensitivity rates, although the absolute number of discordant twins in the earlier study was lower (33 compared with 54). Other methods of diagnosing discordancy, such as abdominal circumference measurements, have yielded similarly disappointing results.16 Because discordancy develops over long periods of time, the interval between ultrasonographic measurements and delivery is probably not an important factor in improving accuracy rates.
The limitations of this study include its retrospective nature and our failure to control for factors that could influence the estimated fetal weight. Although rupture of the membranes4 and placental location17 have been found to have no effect on the estimated fetal weight, the role of other factors such as nonvertex presentation or chorionicity remains unclear. Another possible limitation is the significant differences between groups, mainly gestational age and birth weight. Because of the relatively small number of twin pregnancies with birth weights above 3,000 g, and singleton pregnancies below 2,500 g, we did not attempt to adjust for birth weight.
An advantage of this study is the interval between measurement of estimated fetal weight to delivery. The interval was very short, due to the strict criteria of an upper limit of three days between the estimated fetal weight and delivery. Fetal weight gain in twin gestation is approximately 200 g/wk in mid third trimester,18 correlating to approximately 10% of fetal weight. In previous studies the upper limit was 7–10 days,6,7,8 a period that could significantly lower the accuracy of fetal weight estimation. The short interval could explain the slightly higher accuracy we achieved in some of the measures compared with previous studies.
In conclusion, the present study highlights the advantages and limitations of ultrasonography for the estimation of fetal weight in twin pregnancies and in the subgroups of pregnancies complicated by fetal growth restriction and growth discordancy. We achieved better results than previous studies for some of the measures. We demonstrated that the overall accuracy of the estimated fetal weight is better for singleton than for twin pregnancies, and better for the first twin than for the second twin. Although the sensitivity of the ultrasonographic estimated fetal weight for the detection of fetal growth restriction is low, its negative predictive value is high.
1. 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;151:333–7.
2. 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;142:47–54.
3. Dudley NJ. A systematic review of the ultrasound estimation of fetal weight. Ultrasound Obstet Gynecol 2005;25:80–9.
4. Roberts WE, Gnam EC 3rd, Magann EF, Martin JN Jr, Morrison JC. Labor and membrane rupture in twin gestation. Can they affect the ability to estimate fetal weight? J Reprod Med 2001;46:462–6.
5. Lynch L, Lapinski R, Alvarez M, Lockwood CJ. Accuracy of ultrasound estimation of fetal weight in multiple pregnancies. Ultrasound Obstet Gynecol 1995;6:349–52.
6. Jensen OH, Jenssen H. Prediction of fetal weights in twins. Acta Obstet Gynecol Scand 1995;74:177–80.
7. Ong S, Smith AP, Fitzmaurice A, Campbell D. Estimation of fetal weight in twins: a new mathematical model. Br J Obstet Gynaecol 1999;106:924–8.
8. Gernt PR, Mauldin JG, Newman RB, Durkalski VL. Sonographic prediction of twin birth weight discordance. Obstet Gynecol 2001;97:53–6.
9. Blickstein I, Manor M, Levi R, Goldchmit R. Is intertwin birth weight discordance predictable? Gynecol Obstet Invest 1996;42:105–8.
10. Arbuckle TE, Wilkins R, Sherman GJ. Birthweight percentiles by gestational age in Canada. Obstet Gynecol 1993;81:39–48.
11. Colman A, Maharaj D, Hutton J, Tuohy J. Reliability of ultrasound estimation of fetal weight in term singleton pregnancies. N Z Med J 2006;119:U2146.
12. Levine D, Kilpatrick S, Damato N, Callen PW. Dolichocephaly and oligohydramnios in preterm premature rupture of the membranes. J Ultrasound Med 1996;15:375–9.
13. Kasby CB, Poll V. The breech head and its ultrasound significance. Br J Obstet Gynaecol 1982;89:106–10.
14. Hollier LM, McIntire DD, Leveno KJ. Outcome of twin pregnancies according to intrapair birth weight differences. Obstet Gynecol 1999;94:1006–10.
15. Blickstein I, Keith LG. Neonatal mortality rates among growth-discordant twins, classified according to the birth weight of the smaller twin. Am J Obstet Gynecol 2004;190:170–4.
16. Caravello JW, Chauhan SP, Morrison JC, Magann EF, Martin JN Jr, Devoe LD. Sonographic examination does not predict twin growth discordance accurately. Obstet Gynecol 1997;89:529–33.
17. Belogolovkin V, Engel SM, Ferrara L, Eddleman KA, Stone JL. Does sonographic determination of placental location predict fetal birth weight in diamniotic-dichorionic twins? J Ultrasound Med 2007;26:187–91.
18. Dollberg S, Haklai Z, Mimouni FB, Gorfein I, Gordon ES. Birthweight standards in the live born population in Israel. Isr Med Assoc J 2005;7:311–4.