Birth weight percentiles by sex and gestational age for twins born in southern China

Mean birth weight of twins is known to be lower than that of singletons, however, southern China lacks a twin-specific birth weight reference. In this paper, we use data from the Birth Certificate System in southern China, collected between January 1st 2014 and December 31st 2017 and including 161,076 twins, to calculate sex- and gestational week-specific birth weight percentiles (the 3rd, 10th, 25th, 50th, 75th, 90th, and 97th). We applied generalized additive models for location, scale and shape (GAMLSS) when calculating the birth weight percentiles, and calculated percentiles for monochorionic and dichorionic twins separately. We next used data collected between Jan 1st 2018 and Apr 30th 2018, encompassing 12,371 live births, to calculate the SGA and LGA ratios using birth weight references in Australia, South Korea and China (based on birth defects surveillance system) and birth weight percentiles calculated in this study. Compared to dichorionic twins, monochorionic twins had lower birth weights at 25 to 42 weeks of gestation. The calculated SGA and LGA ratios were relatively stable compared to the other references.

The current study aims to construct the sex-and gestational age (week)-specific birth weight reference percentiles for twins born in southern China, stratified by placental chorionicity (monochorionic and dichorionic placentation).

Materials and Methods
All birth data were obtained from the Guangdong Provincial Birth Certificate System between Jan 1st, 2014 and Dec 31st, 2017. The system covers more than 1900 medical institutions and collects all information about mothers and infants. After birth, maternity medical workers place newborn infants on electronic scales to obtain stable weight data (weighing accuracy is within 1 g). In some cases, health care attendants or midwives fill in the newborns' information in the regional maternal and child information system. The system sets logic correction to ensure that the entered birth weight falls within a feasible range. Finally, regional maternal and child information are uploaded to the Guangdong Provincial Birth Certificate System. The Chief of Midwives and the Chief of Physicians in hospitals then confirm the information entered into the data system. Before the birth certificate is issued, the Department of Medical Administration and parents are also asked to confirm the birth information. All of the information is verified by medical professionals. The birth registry database includes the child's date of birth, gestational age (week) at birth, birth weight, infant sex, parents' ages, registered residence, method of delivery and placenta chorionicity, etc. From the database, we obtained 161,134 cases of twins. We excluded stillbirths (48 cases) and deaths within seven days (10 cases), which together accounted for about 0.04% (58 cases) in all twins. The final analytical sample included 161,076 twin births. Because this study is based on administrative data collected from a large population, it was not possible to obtain informed consents; however, the study was reviewed and approved by the Ethics Committee of Guangdong Women and Children Hospital.
We analyzed the raw data of all twin newborns (40,090 in 2014, 38,285 in 2015, 42,241 in 2016 and 40,460 in 2017). The gestational age (week) was determined by combining mother-reported last menstrual period, ultrasound examination, and postnatal gestational age (week) assessment. The chorionicity of the placenta was judged by ultrasound data collected during the first trimester (about 6~7 weeks of gestation) and confirmed by data collected during examination of the placenta after birth. Birth weight percentiles were created by using the Lambda Mu Sigma (LMS) method, which were fit using the GAMLSS package, based on the assumption that birth weight had a Box-Cox Cole and Green (BCCG) distribution 14,15 . The GAMLSS method allows modeling of various kurtosis asymmetric distribution and the estimation of smooth percentiles to establish birth weight percentile curves for newborns of both genders. According to Cole's reports 16 , a sample size of n >1000 is needed to use the GAMLSS technique to fit a curve. The Schwarz Bayesian criterion, which entails stricter curve smoothing, can be used to judge the pros and cons of the model, as well as to ensure the smoothness and accuracy of the model. GAMLSS is based on the LMS method with a specific distribution of (μ,σ,υ,τ). We used Box-Cox t (BCT) to model birth weight, a method that combines Box-Cox-Cole-Green (BCCG) with the Box-Cox-power-exponential (BCPE) distribution. Note that we take into account the skewness and kurtosis of the data to express the value of the predictor. In addition, we made the model residuals better  N  C3  C10  C25  C50  C75  C90  C97  Mean  SD  N  C3  C10  C25  C50  C75  C90  C97  Mean  SD   25  34  670  761  837  926  1011 1077 1150 924  214  35  632  689  745  805  875  926  975  813  231   26  117  730  832  918  1017 1112 1184 1262 1021  238  75  696  765  833  904  986  1046 1105 901  241   27  221  794  897  1004 1114 1208 1298 1372 1123  256  130  763  847  927  1012 1106 1177 1247 1008  248   28  421  878  993  1099 1219 1322 1410 1524 1232  273  332  833  933  1027 1125 1232 1314 1395 1124  293   29  597  962  1087 1203 1333 1443 1551 1692 1340  292  433  908  1026 1135 1247 1367 1462 1554    modified and the shape of the curve tends to be smoother. Model selection was based on the generalized Akaike Information Standard (G-AIC). That is, we selected the model with the smallest GAIC value. The smoothed data were represented by birth weight percentile curves. The curves appeared in intervals of one gestational week. We estimated mean birth weights and corresponding standard deviations for twins at the 3rd, 5th, 10th, 25th, 50th, 75th, 90th, 95th, and 97th percentiles from 25 to 42 completed weeks based on the smoothed, estimated curves. The percentiles were estimated separately by infant sex (male and female) and by chorionicity. SGA and LGA were defined as birth weights below the 10 th or above the 90th percentile values at a given sex-and gestational week, respectively. Next, we used twin birth weight data collected between Jan 1st, 2018 and Apr 30st, 2018, encompassing 12,371 twin births, to verify the reliability of the four standards. We accomplished this by calculating the SGA ratio and the LGA ratio according to the standards' 10th and 90th percentile values. If standards are reliable, the gestational age (week)-specific SGA and LGA ratios should fluctuate around 10%. We also compared the SGA and LGA ratios we generated to those generated using birth weight references from Australia, South Korea and China (established based on a birth defects surveillance system). Since birth weight may differ by race and ethnicity, the birth weight standards from other countries may differ from those we produced. Moreover, given that birth weights in China may have changed since the implementation of the two-child policy in China in 2016, previously produced birth weight standards in China may be outdated. In both cases, this could result in inaccuracies in the classification of infants as SGA or LGA.
The GAMLSS package (version 5.0.6) for R statistical software (version 3.4.2) was used for analysis.

Results
As showed in Table 1 16,720 were dichorionic male twins and 15,486 were dichorionic female twins. The mean birth weights and associated standard deviations (SD) for male twins with monochorionic and dichorionic placentation were (2436 ± 453) g and (2506 ± 480) g, respectively. While the mean birth weights and associated standard deviations (SD) of female twins with monochorionic and dichorionic placentation were (2361 ± 423) g and (2400 ± 459) g, respectively. Premature twins born at 28-36 weeks and term twins born at ≥37 weeks accounted for 45.9% and 53.7% of all twins, respectively. Low birth weight twin births (birth weight <2500 g) and normal birth weight twin births (birth weight ≥2500 g) accounted for 52.2% and 47.8% of all twins, respectively. Table 2 displays smoothed percentiles for birth weights by gestational age (week) for male and female twins. We next grouped all monochorionic twins based on gestational age (week) and present the resulting data at the 3 rd ,10 th ,25 th ,50 th ,75 th ,90 th , and 97 th percentiles in Table 3. Dichorionic twins were plotted in the same way, with Table 4 displaying smoothed percentiles for birth weights (in grams) of dichorionic male twins and dichorionic female twins. As the gestational age (week) increases, the growth curves for various percentiles become smoother and increasingly steadily. In the 10th, 50th, and 90th percentile graphs of monochorionic twins and dichorionic twins, male twins showed higher BWs than females in the total infant graphs at each GA. Twins showed the most  25  15  730  770  839  918  998  1072 1197 920  192  16  683  734  803  855  923  986  1052 850  196   26  26  802  856  931  1016 1102 1181 1300 1018  221  17  762  815  883  965  1026 1114 1192 967  211   27  35  880  951  1033 1125 1218 1302 1426 1122  261  25  847  900  975  1064 1133 1224 1341 1071  217   28  87  969  1057 1147 1246 1345 1435 1554 1256  254  70  916  1003 1096 1192 1265 1356 1495 1193  279   29  100  1055 1173 1271 1378 1485 1582 1690 1374  277  86  1008 1110 1208 1309 1386 1480 1658   weight gain at 34-35 weeks, with growth slowing after 38 weeks (Fig. 1). Table 4 provides the sex-specific proportions of births at 25-42 gestational weeks. Table 5 provides the SGA and LGA ratios of four standards. The curves showing the incidence of SGA at different gestational ages were used to produce criteria, which were then compared to the previous criteria in China, as well as the criteria from Australia and South Korea (Fig. 2). Since Australia and South Korea's standards only cover gestational ages ranging from 25-40 weeks, we only use these references to calculate SGA and LGA at 25-40 weeks. Moreover, the China birth defects surveillance system standards only cover gestational ages between 28-42 weeks. As a result, we only use this reference to calculate SGA and LGA at 28-42 weeks. As expected, the thresholds derived from Australia standards captured a greater proportion of SGA births (45.9%) in 40 gestational age (week), while included only 4.1% in 28 gestational age (week) among the gestation ranges in their research dataset. On the other hand, the thresholds derived from South Korea standards below the 10th and above the 90th percentile across all gestational age (week) categories were from 3.3% to 37.9%. The thresholds derived from China birth defects surveillance system standards captured a greater proportion of LGA births (46.7% in 41 gestational age (week)), while included only 6.7% (40 gestational age (week)) within the gestation ranges in their research dataset. In our research, the 10th and 90th-percentile proportions of birth weight for gestational week which got by Birth weight percentiles of southern China were relatively stable. The maximum value was found in SGA of 27 and 41 gestational age (week) (13.3%), while the minimum value is found in LGA of 27 gestational age (week) (6.7%).

Discussion
In this study, we constructed new birth weight percentage curves for twins born in southern China. We have estimated percentage curves separately by chorionicity in order to account for chorionic membranes during twin births. Our comparison of percentile curves by chorionicity showed that birth weights of dichorionic twins were higher than monochorionic twins at 25 to 42 weeks of gestation. This finding is consistent with research conducted in south India and the US 7,17 . The low birth weight of monochorinic twins can be attributed to a reduction in weight due to a shared placenta, as well as to reduce effectiveness of the placenta 18 . The average birth weight of male infants is greater than that of female infants for both monochorionic twins and dichorionic twins. The overall pattern of change in birth weight over gestational age is characterized by a rapid increase in birth weight up until week 37, followed by a reduced rate of change afterward. Both male and female infants grew at the fastest rate between 34 and 35 weeks, gaining an average of 192 g and 182 g per week, respectively. Studies of twin pregnancies in the US have found that twin infants have the fastest weight gain between 32 weeks and 34 weeks 17 , while the East Flanders Prospective Twin Survey (EFPTS) found that the most rapid period of infant weight gain occurred between 32 weeks and 34 weeks, with 156 g gained per week 19 .
Because of improvement in medical care facilities and nutrition in China, the proportion of twins with fetal growth restriction has declined, while perinatal survival has improved. Furthermore, due to the large sample size used in our analysis, the birth weight standard we have produced can shed new light on the current situation of twins in southern China.
According to the twin birth weight standard we have constructed for southern China, the highest prevalence (13.3%) of SGA was observed at 41 weeks of gestation, while the lowest (9.1%) was observed at 26 weeks of gestation. Relative to the other three standards, our standard led to a more stable estimate of the prevalence of LGA, which ranged from 6.7% to 11.5%. If we were to instead use one of the other three standards, we would likely misclassify SGA and LGA across all gestational age groups. In particular, the other standards lead to very different estimates of the SGA rate between 39 and 40 weeks of gestation. Compared to southern China's twin birth weight standard, a smaller number of twins were classified as LGA by the Australian and Korean standards. However, a larger number of twins were defined as LGA if we were to use the Chinese standard (based on birth defects surveillance system). This suggests that twin growth standards for healthy twins developed in other countries are not applicable to the population of southern China. Moreover, it is important to regularly update the reference, in order to identify changes in birth weight distributions of twins over time. In addition, previous research has suggested that chrionicity should be taken into account when assessing twin fetal development 17 . In particular, fetal growth appears to differ for twins with monochorionic and dichorionic placentation. Until now, classification of chorionicity was not established for twin birth weight standards in southern China.
Due to the lack of appropriate reference tools, birth weight percentiles for singletons are commonly used in clinical practice in China. In this study, the use of a large, nationally representative population-based sample of twins ensures a more representative and accurate estimate of percentiles.
Unfortunately, we did not collect data on environmental factors that may have affected the pregnant women and fetuses in the study, including socio-economic conditions, diet or nutritional status. Therefore, we cannot directly analyze the relationship between environmental factors and birth weight distributions. Secondly, as with other population-based studies, our data are based on birth registry data, rather than longitudinal measurement of the development of the same fetuses over the course of pregnancy. That is, we have not measured in utero fetal growth. Birth weight percentiles are not the same as intrauterine growth percentiles in that birth weight percentiles do not reflect fetal growth but rather size at birth. The birth weight of premature babies may be affected by the pathological process leading to premature birth and the developmental status during the period of extrauterine growth to full term may be different from that of intrauterine growth until full term 20,21 . It has been suggested that preterm births should be assessed using estimated utero fetal growth trajectories rather than birth weight percentile, given that preterm neonates are likely affected by fetal growth restriction 20 . However, it is difficult to estimate utero fetal growth weight, due to challenges in obtaining accurate measurements, including estimates for fetal weight calculations and the formulas needed for the calculations 21 .