- Top of page
Identifying pregnancies complicated by small-for-gestational age (SGA) fetuses has become one of the main focuses of prenatal care. This is because of the significant association between SGA and perinatal morbidity and mortality. Recent studies have established the potential advantage of using customized growth charts over population-based charts in identifying SGA fetuses at risk for adverse outcome[2-5]. For example, in a previous study from our population we showed that fetuses classified as SGA using customized charts for our singleton population were associated with a higher risk for stillbirth than were those classified as SGA by a population-based growth chart. These studies have, however, been limited to singleton pregnancies.
Twin pregnancies pose special problems with regard to the clinical use of prenatal estimates of fetal weight owing to several factors, including the influence of shared placental mass and chorionicity. There are therefore differences in growth potential between singleton and twin pregnancies that should be considered when evaluating fetal growth. There is evidence from one study that two commonly used singleton formulae have reduced sensitivity for detecting growth restriction in twin gestations.
While there are conflicting reports on the need for plurality-specific population growth charts, to our knowledge this subject has not been studied in multiple pregnancies with regard to the advantages of customized growth charts[7-10]. The aim of the present study was to compare the association between SGA and intrauterine fetal death (IUFD) in twins using customized growth charts designed for twin gestations vs those designed for singletons.
- Top of page
This was a retrospective cohort study using a database including singleton and twin pregnancies that underwent ultrasound examination at between 16 and 20 weeks' gestation. The study was approved by the Human Studies Review Board of Washington University in St Louis.
From a total population of 73 875 singleton births, we excluded preterm births at < 34 weeks, congenital anomalies and stillbirths, and identified 51 150 singleton births, which we used to derive the singleton-customized growth charts. We used the method first described by Gardosi et al. and recently validated by Larkin et al. to develop our model. Briefly, we first developed a model to predict the optimal term birth weight for each pregnancy. This was performed using multiple linear regression modeling. Coefficients for significant physiological variables affecting fetal growth were derived using backward stepwise multiple regression with a significance level of 0.05. The variables included maternal height and weight, parity and ethnicity and the sex of the baby. Data on maternal characteristics such as weight, height and smoking status were collected at the time the patients were seen prenatally. To allow comparison with previous studies[3, 5, 13-15], the birth-weight constant was calculated for a gestation length of 280 days and a ‘standard mother’, defined as of Anglo-European origin, in her first pregnancy, with height 163 cm and weight 64 kg, and the female fetal gender used as reference. Gestational age was deduced from the first day of a woman's last menstrual period. If the dating was not consistent with a first-trimester ultrasound scan or dating based on the anatomic survey (± 7 days in the first trimester or ± 10 days in the second trimester), gestational age was reassigned. The same process was repeated including only twin births in the regression models, adjusting for chorionicity to generate customized charts for twins. Cases with twin–twin transfusion syndrome and monochorionic monoamniotic twins were excluded. From a total of 2445 twin pairs, 1608 met the inclusion criteria for the regression model.
Since the distribution of birth weight across gestational age is non-linear, both quadratic and cubic terms were also examined in the multiple regression models for continuous variables. For the twin models, both twins were included. However, because twin fetuses violate the statistical assumption of independence and are likely to show intracluster correlation, all models were adjusted using a robust cluster analysis term.
After establishing the optimal weight at term for each pregnancy, we then generated the 10th and 90th percentiles at term. This was done by multiplying the coefficient of variation (SD/mean weight) by the Z-score for the 10th and 90th percentiles (−1.28 and 1.28, respectively). The mean birth weight for our singleton population was 3523.9 ± 401 g; this gave a coefficient of variation of 0.11. Therefore the final formula for calculating the 10th and 90th percentiles at term was: optimal weight at 280 days ± (optimal weight at 280 days × 1.28 × 0.11). For the twin model, we used the mean birth weight for our twin population of 3417 ± 485 g, giving a coefficient of variation of 0.14. The 10th and 90th percentiles for the twin population were calculated using the formula: optimal weight at 280 days ± (optimal weight at 280 days × 1.28 × 0.14).
The next step in modeling the customized charts was to describe how the fetus is expected to reach this predicted optimal weight at term. This was derived using the proportionality equation of Gardosi et al. based on the formula of Hadlock et al. for estimated fetal weight (EFW): EFW = exp (0.578 + (0.332 × GA) − (0.00354 × GA2)), where GA is gestational age in weeks. This is the formula currently used in our ultrasound unit for estimating fetal growth. The final proportionality equation is: % weight = 299.1 − (31.85 × GA) + (1.094 × GA2) − (0.01055 × GA3). The proportionality equation is applied to the 10th and 90th percentiles at term as well as the optimal weight in order to create customized percentiles across gestation for each fetus.
The association between the classification of SGA, defined by a birth weight < 10th percentile, and IUFD in twins was compared using the customized charts derived from singleton and twin data. IUFD was defined as fetal demise occurring after 20 weeks' gestation. Statistical analysis was performed including calculation of adjusted odds ratio (OR) for IUFD and screening accuracy using each chart. The ORs were adjusted for variables noted to be significantly associated with IUFD from historical studies, including maternal age over 35 years, Black ethnicity, chronic hypertension and pregestational diabetes. All analyses were performed using STATA version 10 (STATA Corporation, College Station, TX, USA).
- Top of page
The demographic characteristics of the twin and singleton populations before exclusions are shown in Table 1. There were significant differences in mean maternal age, gestational age at ultrasound scan, proportion of nulliparous women, ethnicity and prevalence of medical disorders such as chronic hypertension and pregestational diabetes between the singleton and twin populations.
Table 1. Demographic characteristics of the population prior to exclusions
|Characteristic||Singletons (n = 73 875)||Twins (n = 4890)||P|
|Maternal age (years)||29.9 ± 6.4||30.5 ± 6.1||< 0.01|
|Gestational age at ultrasound (weeks)||19.1 ± 1.6||18.9 ± 1.7||< 0.01|
|Nulliparous||28 322 (38.3)||2022 (41.3)||< 0.01|
|Smoker||8563 (11.6)||514 (10.6)||0.02|
|White||44 755 (60.6)||3036 (62.1)||0.04|
|Black||16 949 (22.9)||1074 (22.0)||0.11|
|Asian||1493 (2.0)||70 (1.4)||< 0.01|
|Hispanic||1195 (1.6)||80 (1.6)||0.92|
|Other||9483 (12.8)||630 (12.9)||0.92|
|Chronic hypertension||1829 (2.5)||144 (2.9)||0.04|
|Diabetes||1379 (1.9)||16 (0.3)||< 0.01|
The derived coefficients for fetal growth using twin data were different from those using singleton data, with lower constants and root mean square error or residual SD: 3422 and 288.9 in twins vs 3543 and 416 in singletons, respectively (Table 2). The standard ANOVA test yielded P < 0.0001, showing the models to be significantly different. Tests of residuals also confirmed the assumptions of normality, linearity and uniformity of variance. The coefficients for birth weight from certain variables such as parity, gender and smoking were seen to be larger in the twin-specific models than in the singleton models. Some variables known to influence fetal weight in singletons, such as some race categories, maternal weight over 64 kg and body mass index, were not significant in the twin model. In addition, the chorionicity of twin pregnancies was not significant in the regression model. An example of a twin-specific customized growth standard for a White woman with an expected birth weight of 3310 g at 40 weeks' gestation is shown in Figure 1a. The 10th percentile for the customized singleton and twin curves for the same woman are depicted in Figure 1b.
Table 2. Coefficients for birth weight derived from twin and singleton cases with complete data and excluding cases of preterm birth < 34 weeks, congenital anomalies, twin–twin transfusion syndrome and stillbirth
|Variable||Singleton model||Twin model|
|(n = 51 150; constant = 3543;||(n = 1608 pairs; constant = 3422;|
|√MSE = 416; R2 = 0.28)||√MSE = 288.9; R2 = 0.12)|
|GA (days) (from 280 days)|| || || || || || |
|Linear term||1.62||0.070||< 0.001||4.05||1.63||0.013|
|Parity|| || || || || || |
|Ethnicity|| || || || || || |
|Maternal height (from 163 cm)|| || || || || |
|Linear term||1.3||0.2||< 0.001||6.1||1.2||< 0.001|
|Maternal weight (from 64 kg)|| || || || || |
|Quadratic term||−0.088||0.021||< 0.001||−0.275||0.153||0.073|
|Cubic term||0.0009||0.0002||< 0.001||0.002||0.002||0.240|
|BMI < 20 kg/m2||−12.3||3.2||< 0.001||−14.8||25.1||0.55|
|BMI ≥ 30 kg/m2||7.8||3.2||0.015||28.7||18.9||0.19|
|Fetal gender male||19.1||2.2||< 0.001||101.6||13.6||< 0.001|
Among 3786 twin infants, 1104 were from monochorionic diamniotic gestations. IUFD was seen in 123 (3.2%) of the individual fetuses in the twin pregnancies included. The numbers of fetuses identified as SGA were 575 (15.2%) and 504 (13.3%) by the singleton and twin charts, respectively. The customized singleton charts identified as SGA 333 (66.1%) fetuses classified as SGA by the customized twin-specific charts, while the twin-specific charts classified as SGA 472 (82.1%) fetuses identified as SGA by the singleton charts.
Table 3 shows the association between SGA and IUFD and screening efficiency using both customized charts. Those identified as SGA using the customized twin charts had an adjusted OR for IUFD that was approximately twice as high as those identified using the singleton charts. The twin charts also showed a higher specificity for identifying SGA fetuses at risk for IUFD. Among 86 pregnancies classified as SGA using the customized singleton charts but not SGA by the twin-customized charts, there were three IUFD. The adjusted OR for the association between the additional cases of SGA and IUFD was, however, not statistically significant.
Table 3. Association between classification as small-for-gestational age (SGA) in twins and intrauterine fetal death (IUFD) and screening efficiency for IUFD, according to chart used to define SGA
|Chart used||IUFD (n (%)) (n = 123)||No IUFD (n (%)) (n = 3663)||Adjusted OR (95% CI)a||Sensitivity (%)||Specificity (%)||Area under ROC curve|
|Singleton-customized chart||22 (17.7)||553 (15.1)||1.2 (0.7–2.0)||17.7||84.9||0.40|
|Twin-customized chart||31 (25.0)||473 (12.9)||2.3 (1.4–3.5)||25.0||87.1||0.55|
|SGA detected only by singleton-customized chart||3 (2.4)||83 (2.3)||1.1 (0.2–3.3)||2.4||97.7||0.38|
|SGA detected using population growth chart17||37 (30.1)||883 (24.1)||1.35 (0.9–2.0)||30.1||75.9||0.53|
We also compared the association between SGA detected using the two customized charts with a non-customized chart by Alexander et al. based on data from the USA population, which is the current standard in many centers in the USA. The latter chart classified 920 (24.3%) twins as SGA, of which 37 (4.0%) had IUFD. The adjusted OR for IUFD using the non-customized chart was not statistically significant (Table 3).
- Top of page
Our study showed significant differences in maternal variables affecting optimal birth weight between twin and singleton pregnancies. In addition, we found that customized growth charts modeled with significant variables specific to twin pregnancies identified an SGA population at a higher risk for IUFD compared with those identified using singleton customized charts. The growth potential for twin fetuses as exemplified by the constants from each model was lower than that for singletons, with twin fetuses expected to be approximately 121 g lighter than singletons on average. The coefficients for certain variables such as parity, fetal gender and smoking were found to be larger (in absolute terms) for the twin-specific model. This may reflect inherent differences between twin and singleton pregnancies that should be explored in future studies. We also found that twins classified as SGA using the twin-customized charts had a higher OR for IUFD than did those classified using either the singleton-customized charts or the population-based chart. Finally, the twin-customized charts were more sensitive, with comparable specificity, for identifying at-risk pregnancies when compared with the singleton-customized standard. These findings are of clinical importance for decision-making and patient counseling.
The trade-off between a highly sensitive vs a highly specific screening chart always requires a delicate balance. While the singleton-customized charts appear to classify more pregnancies as SGA, the additional cases identified were not at significant risk for IUFD. The risk of classifying additional pregnancies as SGA is that this will lead to extra prenatal testing, with an inherent risk of iatrogenic interventions from false-positive tests. If the risk of IUFD in the additional pregnancies classified as SGA were statistically significant, then an argument could be made for using the singleton-customized charts. However, this was not the case.
This is the first study to our knowledge addressing the definition of SGA in twins using customized growth charts. Studies evaluating the use of population-based growth charts to identify complications of SGA in twins have yielded conflicting results. Grumbach et al. reported that twin and singleton fetuses show significant differences in the growth of biparietal diameter and abdominal circumference, and proposed new growth charts for these parameters specific to twins. Similarly, Joseph et al., using population data for singletons and twins from the USA National Center for Health Statistics between 1995 and 2002, showed that the optimal birth-weight range associated with serious neonatal mortality and morbidity was lower for twins than for singletons, and argued the need for plurality-specific fetal growth standards. On the other hand, Garite et al. suggested that these differences in outcome between singletons and twins/triplets had more to do with prematurity.
The strengths of our study include the use of a large database using patient-level data to address this clinically important question. This enabled us to adjust for important demographic differences between the twins and singletons in our final comparison on the association with IUFD. The study is, however, retrospective, with the associated inherent potential for missing some important confounding factors. Our study lays the foundations for future prospective studies, preferably comparing customized EFW charts designed specifically for twins with those derived from a singleton population for the identification of fetuses at risk for morbidity and mortality.
Another limitation of our study is one inherent in most twin studies, in that this population is at a higher risk for preterm delivery. Using birth weight at term to develop the customized charts may therefore have biased our results. However, since both customized charts were derived from birth-weight data at term, the bias should be non-differential.
In this study we used birth weight with respect to gestational age to classify cases as SGA or not. This approach has its limitations, since most neonates delivered prematurely also have a higher incidence of in-utero growth restriction. On the other hand, depending solely on growth charts based on ultrasound-derived EFW has the potential for including fetuses with aberrant growth as normal. The use of a hybrid chart combining birth weight at term and fetal growth biometry has been proposed. However, as such charts are yet to be validated, we believe the approach used in this study is valid. Finally, despite incorporating the proportionality growth factor based on fetal growth data in our models, fetuses were classified as SGA based on birth weight alone. This approach has been shown to have the potential of skewing the results since many infants, especially twins, deliver prematurely and our analyses do not account for missing data from pregnancies that remain in utero. Hutcheon and Platt suggested employing standard epidemiological approaches for handling missing data in these analyses. We did not use these epidemiological methods such as multiple inputting or inverse weighting in our analysis. However, if there is any error in our analysis due to this omission, it will apply to both our singleton and twin models and will therefore be non-differential to our conclusion. Future studies should, however, address these limitations prior to adoption of the customized twin models clinically.
In conclusion, charts customized specifically for twins are more effective at identifying twin pregnancies at risk for IUFD than are those derived using singleton birth data. These findings, if prospectively validated in other populations, would suggest that these charts should be adopted in the clinical evaluation of twin gestations.