Development and validation of an ultrasound-based estimated fetal weight reference for Chinese twin pregnancy: a retrospective cohort study Authors: Jing Yang, Hongzhao Yu, Yangyu Zhao, Zhihao Cheng, Yufeng Du, Pengbo Yuan, Aiqing Zhang, Yang Liu, Xiaoli Gong, Yang Yu, Yixin Li, Xiaoyan You, Nana Huang, Youzhen Zhang, Xiya Sun, Hongbo Qi, Hong-tian Li, Yuan Wei Published: 2024 ABSTRACT AbstractBackgroundFetal growth monitoring is important for twin pregnancies. However, there has been no clinically validated tool for monitoring fetal growth of twin pregnancies in China. This study aims to develop and validate a chorionicity-specific growth chart of ultrasound estimated fetal weight (EFW) for Chinese twin pregnancies.MethodsThis retrospective cohort study included all twin pregnancies who delivered two live fetuses with gestational age ≥ 34 weeks without severe obstetric complications at a tertiary hospital from January 2007 to March 2021. The participants were divided into a development set (delivered in or before December 2017) and a validation set (delivered in or after January 2018). Chorionicity-specific growth charts were created using the generalized additive models for location, scale, and shape (GAMLSS) based on the development set. The fetuses from the validation set were classified into three groups based on the last EFW: small-for-gestational-age (SGA) indicated by both the newly established twin charts and the Hadlock singleton chart currently used for twin pregnancies in China, suspected SGA indicated by only the singleton chart, and no SGA indicated by either chart. The incidence of neonatal outcomes among the three groups was then compared accordingly, including intensive care unit (NICU) stay length, respiratory diseases, and neurological disorders.ResultsThe development set included 883 twin pregnancies and a total of 6374 EFW measurements between 16 and 38 weeks of gestation, and the validation set included 801 twin pregnancies and 7630 EFW measurements. In the development set, monochorionic diamniotic (MCDA) twins had a significantly lower EFW compared to dichorionic diamniotic (DCDA) twins beginning at 26 weeks, with the difference gradually increasing thereafter, supporting the establishment of chorionicity-specific growth charts. Of the 1,602 twin neonates in the validation set, 103 (6.4%) were classified into the SGA group, 164 (10.2%) into the suspected SGA group, and 1335 (83.3%) into the no SGA group. The incidence of respiratory diseases and neurological disorders was comparable between the suspected SGA group and the no SGA group, but apparently higher in the SGA group. Meanwhile, NICU stay lengths were consistently longer for twins in the SGA group compared to the no SGA group (difference: 0.57, 95% CI: 0.31–0.83), with no significant differences observed between the suspected SGA and no SGA groups.ConclusionsThe fetal growth trajectories differed by chorionicity, with a lower EFW for MCDA twins beginning at 26 weeks. The establishment of chorionicity-specific growth chart could reduce overdiagnosis of SGA and improve fetal growth monitoring of twin pregnancies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12884-024-06924-y.Keywords: Chorionicity, Estimated fetal weight, Growth charts, Twin pregnancy, Fetal growth restriction, Small for gestational age, Adverse outcomes Background Over the past three decades, the incidence of twin pregnancy has increased by more than 70% globally, which could be largely attributed to the increased use of assisted reproductive technology and growing maternal age at conception [1, 2]. The risk of perinatal death is about six times higher in twin fetuses compared to singleton fetuses [3], primarily due to preterm delivery and fetal growth restriction (FGR) [4–7]. Therefore, meticulous prenatal surveillance of fetal growth is vital for twin pregnancies. Currently, many countries use singleton fetal growth charts to monitor the growth of twin fetuses [8]. However, accumulating studies have shown that fetal growth patterns differ significantly between twin and singleton pregnancies, especially during the third trimester [9–11]. Clinically, small-for-gestational-age (SGA) indicated by a growth chart is often interpreted as pathological growth restriction, so using singleton growth charts for twin fetuses could lead to the overdiagnosis of SGA as well as unnecessary interventions. More recently, considerable efforts have been made to establish twin-specific growth charts of ultrasound estimated fetal weight (EFW) [12–16]. Although some of these charts were developed based on completely uncomplicated healthy twin pregnancies (i.e., using strict exclusion criteria) [15, 16], including the one from China [15], most charts adopted less strict exclusion criteria since the charts derived from healthy twin pregnancies could also lead to overdiagnosis of SGA. Moreover, the performance of most of the twin-specific growth charts remains to be validated [15]. By adopting well-acknowledged less strict exclusion criteria [17, 18], this study aimed to establish and validate twin-specific reference charts of EFW for pregnant women in China. Methods Study design and participants This retrospective longitudinal cohort study included twin pregnancies who received prenatal health care and delivered at the Peking University Third Hospital, a tertiary hospital in Beijing, China, between January 2007 and March 2021. Participant selection criteria were made following the recommendations by Altman DG [17] and Silverwood RJ [18]. Specifically, the inclusion criteria were as follows: (1) a twin pregnancy with known chorionicity; (2) the delivery of two live fetuses; (3) having at least two ultrasound examinations at the second and last trimester; and (4) no missing information on gestational age, maternal height, weight, parity, and ethnicity. The exclusion criteria were as follows: (1) delivery at less than 34 weeks of gestation; (2) severe obstetric complications such as early-onset preeclampsia, gestational diabetes mellitus (GDM) treated with insulin, and autoimmune disease; GDM was defined when any of the glucose measurements reached 92 mg/dL (5.1 mmol/L), 180 mg/dL (10.0 mmol/L), or 153 mg/dL (8.5 mmol/L) for fasting, 1-hour, and 2-hour levels, respectively, according to the 75-g “one-step” approach OGTT taken between 24 and 32 weeks. The definition of early-onset preeclampsia required disease occurring at ≥ 20 and < 34 weeks’ gestation comprised of hypertension and proteinuria and/or maternal end-organ dysfunction, indicated by headache and visual symptoms and laboratory abnormalities of platelets, creatinine, or liver enzymes [19]. (3) major congenital fetal anomalies or aneuploidy in at least one fetus; (4) monochorionic-monoamniotic twins; (5) complicated monochorionic twin pregnancies; (6) incomplete or abnormal ultrasound data; or (7) dichorionic twins with an intrapair birthweight difference ≥ 25%. We grouped the participants into a development set and a validation set. The development set, consisting of individuals who delivered between January 2007 and December 2017, was used to establish reference growth charts of ultrasound EFW. The validation set comprised those who delivered between January 2018 and March 2021 and was used to assess the performance of the newly established growth charts. All the data were extracted from the electronic medical information system and the obstetrical ultrasound system of the study hospital. The records of the newborns in the validation set were extracted by the study team and independently checked by an experienced neonatologist. Gestational age and chorionicity We performed color Doppler ultrasound using GE Healthcare Voluson E8 machines (Cincinnati, OH, USA) with an abdominal probe, C1-6, at a frequency of 2–5 MHz. Gestational age was calculated according to the last menstrual period and confirmed by or revised according to a measurement of the crown-rump length of the larger twin in the first trimester, before being rounded to the nearest completed week. The ultrasound determination of chorionicity was based on sonographic findings obtained during the first trimester (a single placental mass with a T sign for a monochorionic pregnancy and two distinct placental masses or a lambda sign with a single placental site for a dichorionic pregnancy). Chorionicity was further confirmed by placental examination after delivery. Biometric measurements Serial ultrasonographic measurements, including biparietal diameter (BPD), head circumference (HC), abdominal circumference (AC), and femur length (FL), were carried out following the guidelines of the International Society of Ultrasound in Obstetrics and Gynecology [20]. EFWs were calculated using the Hadlock III formula [21] containing HC, AC, and FL. EFWs exceeding the mean ± 4 times standard deviation for gestational weeks were excluded from the analysis. These ultrasonographic fetal biometric measurements were longitudinally collected in pregnant women. Transabdominal ultrasound scans of fetal biometric measurements were conducted by two experienced sonographers over the entire study period. To make a post-hoc evaluation of the inter- and intra-operator agreement, 20 twin pregnancies independent of the present study were enrolled between 32 and 36 weeks of gestation, and all their fetuses were measured by both sonographers twice independently. We found that both inter- and intra-operator agreement on EFW measurements were high, with the intraclass correlation coefficient (ICC) for inter-operator agreement was 0.997 (95% CI: 0.994, 0.999), and those for the two sonographers were 0.996 (95% CI: 0.989, 0.999) and 0.997 (95% CI: 0.993–0.999), respectively. Statistical analysis The baseline characteristics were described using mean ± standard deviation or median (interquartile range) for continuous variables depending on the normality of the variables and using the frequency (%) for categorical variables. T-test or Wilcoxon rank-sum test or Chi-square test was used to compare the differences in these characteristics between DCDA and MCDA pregnant women as appropriate. To account for the autocorrelation of the same pregnancy, a linear mixed model or generalized linear mixed model was used to compare the differences in the baseline characteristics between DCDA and MCDA fetuses, and a multiple linear mixed model was used to examine the association of maternal and fetal characteristics with birth weight. The distribution of EFWs of DCDA and MCDA twins was presented as mean ± standard deviation by gestational weeks, and a multiple linear mixed model was used to compare the differences in EFWs between the two groups. Given that a single power transformation may not adequately normalize EFW distributions across all gestational ages, the generalized additive models for location, scale, and shape (GAMLSS), a widely accepted approach for modeling growth curves developed by Rigby and Stasinopoulos [22], was used to construct fetal growth charts. Various candidate distributions were compared using the generalized Akaike Information Criterion (G-AIC) to determine the best-fitting model (sTable 1). Finally, a Box–Cox t (BCT) distribution, a generalization of the Box–Cox normal distribution, was considered in order to model both skewness and kurtosis. Briefly, the location, scale, skewness, and kurtosis of the EFW data were smoothed across gestational ages using P-splines and the smoothed parameters were then used to obtain percentiles for specific gestational ages. Given the dramatic influence of chorionicity on EFW, separate growth charts were established for DCDA and MCDA twins. We assessed the performance of the newly established twin charts compared to the currently used Hadlock singleton chart for twin pregnancies in clinical settings in China [23]. With SGA defined as EFW < 10th percentile for gestational age, we categorized the fetuses in the validation set into three groups: growth restriction indicated by both the singleton and twin charts for the last measurement (SGA group), growth restriction indicated only by the singleton chart (suspected SGA group), and no growth restriction indicated by either chart (no SGA group). The incidences of SGA-associated neonatal outcomes were then compared among the three groups, including neonatal intensive care unit (NICU) stay length, respiratory diseases, and neurological disorders, including severe cerebral white matter injury, intraventricular hemorrhage (IVH), posthaemorrhagic ventricular dilatation and hydrocephalus. Given the right-skewed distribution and excessive zeros in the length of NICU stay data, the zero-inflated negative binomial (ZINB) model was utilized for analysis. Additionally, we assessed the predictive accuracy of the newly established twin charts compared to the Hadlock singleton curves for birth weight using the root mean square error (RMSE) as the evaluation metric. Statistical analyses were performed using R software (version 4.0), and the GAMLSS model was developed using the gamlss package [24]. A two-sided P value of less than 0.05 was considered statistically significant in all analyses. Results Establishing chorionicity-specific twin growth charts In total, 883 (69.7%) out of the 1267 twin pregnancies who delivered between January 2007 and December 2017 were included in the development set, including 179 MCDA and 704 DCDA twins (Fig. 1). Compared with DCDA pregnancies, MCDA pregnancies had lower maternal age (31.5 ± 4.5 vs. 32.7 ± 3.8), a lower proportion of conception through assisted reproductive technology (11.2% vs. 41.2%), lower neonatal birth weight (2350 ± 403 g vs. 2565 ± 366 g), and higher incidence of preterm delivery (69.3% vs. 54.4%) (P < 0.01 for all comparisons) (Table 1). 883 pregnancies underwent a total of 6374 EFW measurements. The total and median numbers of EFW measurements by gestational weeks for DCDA pregnancies were 4509 and 139.5 (interquartile range: 77.0-289.8), and for MCDA pregnancies were 1865 and 80.5 (interquartile range: 72.0-106.0), respectively. Notably, both DCDA and MCDA pregnancies had more than 50 EFW measurements across nearly all gestational weeks (sFigure 1). Flow chart for the participants in the development set Table 1Characteristics of DCDA and MCDA pregnancies in the development setCharacteristicsAll(n = 883)Dichorionic(n = 704)Monochorionic (n = 179)P valueMaternal age, year32.4 ± 4.032.7 ± 3.831.5 ± 4.5< 0.01Han Chinese819 (92.8)654 (92.9)165 (92.2)0.74Pre-pregnancy BMI, kg/m223.2 ± 4.423.4 ± 4.522.8 ± 3.90.09Gestational age, week36.0 (35.0 to 37.0)36.0 (36.0 to 37.0)36.0 (35.0 to 37.0)< 0.01Preterm delivery (< 37 weeks)507 (57.4)383 (54.4)124 (69.3)< 0.01Gestational diabetes mellitus193 (21.9)159 (22.6)34 (19.0)0.30Gestational hypertension disorders160 (18.1)120 (17.0)40 (22.3)0.10Conception through ART310 (35.1)290 (41.2)20 (11.2)< 0.01Female fetus870 (49.3)689 (48.9)181 (50.6)0.58Birth weight, g2521.5 ± 384.02565.0 ± 366.02350.0 ± 403.0< 0.01Data was presented as mean ± sd or median (25th percentile to 75th percentile) for continuous variables and n (%) for categorical variablesART, assisted reproductive technology Characteristics of DCDA and MCDA pregnancies in the development set Data was presented as mean ± sd or median (25th percentile to 75th percentile) for continuous variables and n (%) for categorical variables ART, assisted reproductive technology The EFW distributions for DCDA and MCDA twins according to gestational weeks are shown in sTable 1. The MCDA twins exhibited a significantly lower EFW compared to DCDA twins beginning at 26 weeks, with the differences gradually increasing with gestational weeks (P < 0.01 for the linear trend test). Figure 2 depicts the fetal growth charts for DCDA and MCDA twins. The corresponding model had a high R2 value of 0.95 and 0.93 for DCDA and MCDA twins, respectively. Fetal growth charts for dichorionic and monochorionic twins. Notes: The blue points were the actual data points used in the modelling The proportions of EFW values below the 10th percentile were 10.3% in MCDA twins and 10.0% in DCDA twins, while proportions above the 90th percentile were 9.3% and 9.9%, respectively. Gestation-specific parameter values and EFW percentiles (3rd, 5th, 10th, 25th, 50th, 75th, 90th, 95th, and 97th) for DCDA and MCDA twins are presented in sTable 2 and sTable 3, respectively. Performance of the newly established twin growth charts In total, 801 (60.3%) of the 1329 twin pregnancies who delivered between January 2018 and March 2021 were included in the validation set, including 593 DCDA twins and 208 MCDA twins (sFigure 3). Among the 1602 neonates, 103 (6.4%) were classified into the SGA group, 164 (10.2%) into the suspected SGA group, and 1335 (83.3%) into the no SGA group. Detailed characteristics of the three groups of neonates are shown in sTable 4. The incidence of adverse outcomes was comparable between the suspected SGA group and the no SGA group, but apparently higher in the SGA group (Tables 2 and 3). Specifically, when compared to the no SGA group, the SGA group had numerically higher risks of respiratory diseases (RR:1.88 vs. 1.18) and neurological disorders (RR:5.67 vs. 0.51) than the suspected SGA group. ZINB models for NICU stay length further indicated that, compared to the no SGA group, the log odds of being an excessive zero decreased by 0.99 (95% CI: 0.55–1.43) for the SGA group, indicating a higher likelihood of NICU admission. Additionally, NICU stay lengths were consistently longer for twins in the SGA group compared to the no SGA group (0.57, 95% CI: 0.31–0.83), with no significant differences between the suspected SGA and no SGA groups. These patterns were generally consistent for both MCDA and DCDA twins. Table 2Respiratory diseases and neurological disorders among three groupsOutcomesNo SGAn (%)Suspected SGAn (%)SGAn (%)Suspected SGA vs. No SGA aSGA vs. No SGA aOverallRespiratory disease62 (4.64)9 (5.49)9 (8.74)1.18 (0.60–2.33)1.88 (0.96–3.68)Neurological disease16 (1.20)1 (0.61)7 (6.80)0.51 (0.07–3.81)5.67 (2.39–13.47)DCDA twinsRespiratory disease36 (3.54)3 (3.53)5 (5.88)1.00 (0.31–3.17)1.66 (0.67–4.12)Neurological disease8 (0.79)-3 (3.53)-4.48 (1.21–16.58)MCDA twinsRespiratory disease26 (8.15)6 (7.59)4 (22.22)0.93 (0.40–2.19)2.73 (1.07–6.98)Neurological disease8 (2.51)1 (1.27)4 (22.22)0.50 (0.06–3.98)8.86 (2.94–26.68)a Risk ratio (95% confidence interval) Respiratory diseases and neurological disorders among three groups a Risk ratio (95% confidence interval) Table 3Coefficients of ZINB models for NICU stay length among three groupsGroupsLogitNegative binomialOverallNo SGA Reference Reference Suspected SGA-0.54 (-0.93 - -0.14)-0.04 (-0.31–0.22)SGA-0.99 (-1.43 - -0.55)0.57 (0.31–0.83)DCDA twinsNo SGA Reference Reference Suspected SGA-0.28 (-0.91–0.35)-0.30 (-0.76–0.17)SGA-0.88 (-1.41 - -0.36)0.39 (0.05–0.72)MCDA twinsNo SGA Reference Reference Suspected SGA-0.39 (-0.94–0.15)0.03 (-0.29–0.36)SGA-1.83 (-2.85 - -0.81)0.82 (0.41–1.23) Coefficients of ZINB models for NICU stay length among three groups Discussion Principle findings In this retrospective cohort study of Chinese twin pregnancies, we found that MCDA twins exhibited a significantly lower weight than DCDA twins beginning at 26 weeks of gestation and developed separate EFW reference charts for MCDA and DCDA twins accordingly. In addition, we found that the newly established twin- and chorionicity-specific charts performed significantly better than the currently used singleton growth chart in clinical settings in China. Consistent with most studies in the literature [9, 11, 13, 25], significantly lower EFWs were observed in MCDA twins compared to DCDA twins in our study, starting from 26 weeks, with the differences progressively widening with increasing gestational age, thereby highlighting the importance of the establishment of chorionicity-specific growth charts. Relatively small differences in EFWs between MCDA and DCDA twins, even in the third trimester, have also been documented in some other studies; however, these studies only included uncomplicated healthy twin pregnancies [15, 16]. Given that MCDA twins are at a higher risk of moderately compromised growth than DCDA twins [26], chorionicity-specific EFW references are therefore more suitable for fetal growth monitoring of twin pregnancies in clinical settings. In our study, the chorionicity-specific charts were developed using appropriate statistical methods. The variabilities in skewness and kurtosis across gestational ages underscored the inadequacy of applying a single power transformation for all gestational ages, emphasizing the necessity and benefits of using the GAMLSS method. The R2 value of the model for DCDA twins was 0.95 and that for MCDA twins was 0.93, showing an excellent fit to the data. Additionally, the proportions of data points that fell below the 10th percentile and above the 90th percentile were each close to 10%. The growth chart for MCDA twins was consistently lower than that for DCDA twins from 28 weeks until delivery, and the difference in the 50th percentile of EFWs at 37 weeks between MCDA and DCDA twins was about 170 g, which is consistent with Shivkumar et al. [13]. Moreover, our charts are significantly lower compared to another chorionicity-specific charts from uncomplicated Chinese twins [15, 27]. Although many studies have developed chorionicity-specific EFW references in twin pregnancies [12, 17, 18, 20, 28], few have simultaneously assessed the clinical performance of the references. In addition to the development of the references, we also assessed their performance against the Hadlock singleton reference currently used for twin pregnancies in China. We found that the incidence of adverse outcomes in the suspected SGA group (i.e., identified purely based on the singleton reference) was comparable to that in the no SGA group, while that in the SGA group (i.e., identified based on the newly established twin references) was significantly higher than that in both suspected SGA and no SGA groups. Therefore, we believe that our charts, which remarkably decreased SGA diagnosis from 16.6 to 6.4%, have the capacity to identify twins that would be improperly classified as SGA by the singleton chart, thereby reducing the overdiagnosis of SGA and potential iatrogenic interventions. Moreover, the combined use of our charts and the singleton chart enables clinicians to stratify twins into low-risk (no SGA), medium-risk (suspected SGA), and high-risk (SGA) categories. The frequency of subsequent ultrasound monitoring might be appropriately reduced in low- and medium-risk twins with normal fetal biometry, amniotic fluid volume, and umbilical artery Doppler. Such stratification may facilitate targeted monitoring and interventions, promoting personalized prenatal care while avoiding unwarranted interventions for twins exhibiting normal growth, thereby optimizing clinical outcomes. Strengths and limitations The selection of participants of this study, as aforementioned, adhered to relevant recommendations [17, 18], ensuring that it aligns with clinical practice. High-quality ultrasound measurements and advanced statistical modelling strategies enabled us to reliably estimate the growth trajectories of twins. Moreover, the performance of the twin- and chorionicity-specific charts as observed in the validation set indicated a strong potential for clinical application. The study also has limitations. The distribution of ultrasound examinations was uneven across weeks of gestation. However, the relatively large sample size allowed for a certain number of ultrasound measurements at each gestational week. Besides, the development and validation sets of patients were from the same study hospital, so the external validity of the newly established reference charts remains to be evaluated. In an independent external validation study, Shea et al. [29] found that the twin-specific reference chart performed significantly better than the commonly used Hadlock singleton reference chart in identifying SGA-associated adverse outcomes. Although this study only involved validation of the reference chart for DCDA twins, similar performance might be achieved for that of MCDA twins in view of the comparable performance of both reference charts observed in our study. Additionally, the retrospective nature of our study may limit the accuracy and completeness of the recorded data. Moreover, our charts were not customized to account for other factors affecting fetal growth or outcomes, such as parental factors [30], umbilical artery Doppler flow pattern in MCDA twins [31], and EFW discordance [32]. As there is ongoing debate about which growth charts best identify fetal growth restriction [33], future research could optimize and validate various curves through prospective studies, potentially enhancing their reliability and applicability. Conclusions This study developed twin- and chorionicity-specific growth reference charts in line with the findings that MCDA twins had lower EFW than DCDA twins beginning at 26 weeks. These charts had the potential to substantially reduce overdiagnosis of SGA as compared with the currently used singleton chart in clinical settings in China. Future studies are encouraged to validate and improve the performance of these charts. Electronic Supplementary Material Below is the link to the electronic supplementary material. Supplementary Material 1 Author contributions Yuan Wei, Hong-tian Li, Yangyu Zhao, Hongbo Qi and Jing Yang conceived and designed the study. Pengbo Yuan, Aiqing Zhang, Xiaoli Gong, Yang Yu, Yixin Li, Nana Huang, Youzhen Zhang, Xiya Sun, and Xiaoyan You contributed the conduct of the study and collection of data. Hongzhao Yu, Zhihao Cheng, Hong-tian Li, Yang Liu and Yufeng Du analysed the data. Jing Yang and Hongzhao Yu drafted the manuscript. Hong-tian Li and Yuan Wei critically revised the manuscript. All authors reviewed and approved submission of the manuscript. Funding This work was in part supported by the National Key R&D Program of China (2022YFA1303900), the Emerging Engineering Interdisciplinary Young Scholars Project of Peking University (PKU2023XGK015), the Clinical Medicine Plus X-Young Scholars Project of Peking University (PKU2021LCXQ009) and National Key Research and Development Program of China (2023YFC2705900). Data availability The data used and analyzed during the study are available from the corresponding author if the request is reasonable. Declarations Ethical approval and consent to participate This study was approved by The Human Research Ethics Committee of Peking University Third Hospital (IRB00006761-2016145). Due to the retrospective nature of the study, the need for informed consent was waived by The Human Research Ethics Committee of Peking University Third Hospital. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Jing Yang and Hongzhao Yu are joint first authors. Contributor Information Hongbo Qi, Email: qihongbo728@163.com. Hong-tian Li, Email: lihongtian@pku.edu.cn. Yuan Wei, Email: weiyuanbysy@163.com. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data used and analyzed during the study are available from the corresponding author if the request is reasonable.