Birth weight discordance and adverse perinatal outcomes

Introduction

Studies from Canada [1], the United States [2], Australia [3], Norway [4], England [5], and Korea [6] have shown an increase of over 50% in the twinning rate in the last decade. The increased rate of twinning across the globe has affected the clinical perspective pertaining to specific perinatal complications of twin pregnancies. Associated complications such as increased perinatal mortality and morbidities are established in twin pregnancies compared with singleton gestations [7], [8], [9], [10]. The addition of discordant fetal growth exacerbates the potential for future complications. The literature is rich with studies retrospective and prospective [11], [12], [13], hospital-based [14], [15], [16] and population-based [17], and small [13], [14], [18], [19] and large [20], producing widely disputed rates of perinatal mortalities and morbidities. Some studies used a birth weight discordance (BWD) of 20%–30% to define growth discordance associated with higher rates of adverse perinatal outcomes [21], while other studies suggest that growth discordance is common and the majority of twins do well regardless of their weight difference [14], [15], [20]. These studies adjusted for variables most commonly thought to be confounders of the relationship between BWD and perinatal outcomes such as gestational age [15] intra-uterine growth retardation (IUGR) [12], twin size [20], infant sex [22], [23], birth order [20], sex pairing [24], maternal age [25], chorionicity [13], [26], [27], [28], [29], prenatal care [30], preterm birth [15], and pregnancy due to assisted reproductive technology [31], [32]. However, a few studies to date have adjusted for confounding factors related to maternal complications at pregnancy [e.g. diabetes, pregnancy-induced hypertension (PIH) and preeclampsia], prior obstetric history, and maternal weight gain during pregnancy [33], [34].

While the majority of studies have focused on presenting the association between BWD and perinatal morbidities such as respiratory distress syndrome (RDS), congenital anomalies, hyperbilirubinemia, hypoglycemia, and anemia, a few studies have investigated the outcomes related to management of newborn viability at birth and neonatal respiratory support in infancy. Our study primarily aimed at investigating the association between BWD and perinatal mortality. The secondary objective of this study was to assess the difference in short-term neonatal outcomes across discordant twins at birth and in the first 28 days of life. This information will convey more accurate prognostic information to parents and caregivers as the twinning rate increases in British Columbia. Our purpose was to examine the association between discordant growth and perinatal mortality and morbidities among twins with careful adjustment for maternal confounding factors and chorionicity as well as to adjust for the inter-dependent manifestation of outcomes among twin pairs by using general estimating equation (GEE) modeling which has not been used extensively in twin studies.

Patients and methods

The study was approved by Ethics Review Board of the University of British Columbia, Vancouver (H11-03281). We conducted a retrospective population-based cohort study of twins delivered during the period of 2000 through 2010 in BC, Canada. Data were obtained from perinatal services BC’s perinatal data registry (PSBC) [14]. The registry contains information on approximately 99% of births in the province of BC. Information in the registry is collected from standardized forms completed by clinicians. The validity of the data is insured by quality control measures including built-in warnings in the data entry software and periodic checking of the data. It should be noted that not all the variables were available for all the twins for a number of reasons a) failure of clinicians to record information or clearly mis-entered data and b) some variables were added to the PSBC data set after 2004 (e.g. maternal education) or after 2008 (e.g. IVF).

Data were accessed via Population Data BC’s Secure Research Environment [35]. This environment is a central server accessible only via an encryptical virtual private network through the firewall and use of a Yubikey^® (Yubico, Stockholm, Sweden) for two-factor authentication. It provides researchers with secure storage and backup, centralized location for access and processing of research data, SPSS software (SPSS, Statistical Package for the Social Sciences, IBM, NY, USA) and security standards that meet the Canadian Standards Association’s ten privacy principles. These principles protect personal information as per detailed requirements of the BC Freedom of Information and protection of Privacy Act. None of the authors have any competing interests.

Cases with major structural congenital anomalies (n=42), twin-to-twin transfusion syndrome (TTTs, n=234), those with a weight <500 g (n=325) and single stillbirths (n=49), missing birth weight (n=10), sex (n=14) were excluded. The exclusion of single stillbirths and birthweight under 500 g was an attempt to remove cases where a stillbirth of one twin occurs remote from the birth of the survivor or where there was intentional selective feticide, both of which would lead to excessive birth weight discordance. After excluding these cases, 12,656 twins remained in the dataset for analyses.

A variable was created by taking the difference between the birth weights of the larger twin and the smaller twin, divided by the larger twin weight and multiplied by 100; see Equation 1:

We adopted the definition of growth discordance for twins as larger than 30% BWD.

Gestational age was established based on an algorithm considering the last menstrual period and ultrasound estimate of gestation. When the two estimates were concordant, the last menstrual period-based estimate was used, whereas in cases of discordancies of more than 2 weeks between the estimates, the ultrasound estimate was used. The algorithm was adapted from the Society of Obstetrician and Gynecologists guidelines [15].

Outcome variables related to perinatal mortality were early and late neonatal mortalities. Early neonatal death was defined as the death of the baby within 7 days of birth, whereas late neonatal death was described as babies’ death within 28 days of birth. Both of these variables included inpatient death and death identified through linkage with vital statistics.

Perinatal morbidity was divided into short-term and long-term morbidities. Short-term morbidity was inclusive of 1- and 5-min Apgar scores of <7; use of antibiotics, surfactant, steroids for lung maturation, or resuscitation drugs; whether or not the newborn underwent chest compression; and intermittent positive pressure ventilation with an endotracheal tube (IPPV-ETT). Diagnostic tests comparing growth-discordant and -concordant twin pairs were streptococcus culture test, cord arterial gases base and cord arterial gases pH using blood from cord. The number of days spent under a ventilator and the number of days receiving continuous positive airway pressure (CPAP) were also analyzed. Long-term morbidity included newborn septicemia, hypoglycemia, neonatal diabetes, anemia, apnea, intraventricular hemorrhage (IVH), pneumonia, RDS, and retinopathy.

Fetal-to-placenta weight ratio was calculated by dividing “fetal weight” over “placenta weight”. For monochorionic (MC) twins, one placenta is used for this calculation, while for dichorionic (DC) twins, the combined weight of both placentas are used.

Confounding variables considered for adjustment in the analysis were inclusive of clinical factors [chorionicity, gestational age, twin size, fetal sex, sex discordance, fetal growth (small for gestational age, appropriate for gestational age, large for gestational age), parity, number of prenatal visits and preterm birth, history of stillbirth, congenital anomalies, preterm birth, or low birth weight in the association between birth weight/sex discordance and perinatal outcomes, history of cesarean section, maternal age, body mass index (BMI), maternal weight gain during pregnancy, and chronic conditions such as diabetes and hypertension, preeclampsia, and PIH] and maternal age. Variables with no association with BWD (P>0.05) were not included in the models. We were unable to adjust for some variables such as IVF (collected after April 1, 2008), maternal education (collected after April 1, 2004), and as such there is a large proportion of missing data for some variables, which is a limitation of our analysis.

The threshold level for BWD≥30% was assumed based on our receiver operating characteristic analysis as the optimal cut-off point associated with perinatal adverse outcomes, results of which are published elsewhere [36]. The difference in outcome between discordant and concordant twin pairs was determined employing the sample t-tests or Mann-Whitney tests for quantitative variables (e.g. gestational age) and χ² tests (e.g. diabetes) for qualitative variables. A P-value of <0.05 was considered significant. To account for the dependent manifestation of outcomes among twin pairs, the strength of association between BWD and perinatal outcomes was estimated using general estimating equation (GEE) modeling. The parsimonious model was selected using the least quasi-likelihood under an independent model criterion estimate.

Results

The study population comprised 12,656 twins, of whom 452 (3.57%) were discordant by the percentage definition of ≥30%. The growth-concordant and -discordant twins were similar with regard to maternal demographics including maternal age, education, smoking habits, BMI and weight gain during pregnancy (all P>0.05, see Table 1). Limitations in the availability and the percentages of missing data is noted in Table 1. There were no significant differences between concordant and discordant twin pairs in the odds of prior congenital anomalies, low birth weight, stillbirth, and neonatal death (all P>0.05). There were, however, statistically significant differences between growth-concordant and -discordant twins for the following variables: gestational age, parity, number of antenatal visits, PIH, preeclampsia (all P<0.01) and gestational diabetes (P=0.02, see Table 1).

Perinatal outcomes are shown in Table 2. Perinatal mortality was more frequent among BWD twins compared with concordant twins (P<0.01). Growth-discordant twins had higher frequencies of 1- and 5-min Apgar scores <7, antibiotic and steroid prescriptions, IPPV-ETT use and a longer average of ventilator days compared with growth-concordant twins (all P<0.01). The two groups were similar in terms of surfactant use, chest compression, and use of resuscitation drugs (all P>0.05). Comparable distributions of cord arterial base gases and cord arterial pH were also found between the two groups (all P>0.05). Frequencies of streptococcus positive tests were also alike in BWD and concordant twins (P>0.05).

For the long-term perinatal morbidities under study, apnea, RDS, and retinopathy were found to be higher in frequency among growth-discordant twins (≥30%) in comparison with the growth-concordant twins (all P<0.01).

Lighter discordant vs. heavier concordant twins

A lighter twin was defined as a twin who was lighter than the co-twin within a growth-discordant pair. The results for the lighter discordant twins (n=236) were compared with the combined group of heavier co-twins and twins with concordant birth weight (n=1214). Fetal weight to placenta weight (F/P) ratio was estimated for these two categories. Lighter twins have significantly lower F/P compared with the comparison group (Figure 1).

Figure 1:

A comparison of fetal-to-placenta weight ratio between lighter discordant twins (n=236) and combined group of heavier co-twins and twins with concordant birth weight (n=1214, P<0.01).

Table 7 shows an overview of adjusted odds of perinatal mortalities and morbidities in both the BC population and twins born in C&W hospital. Perinatal mortality, and 5-min Apgar score <7 were found to be similarly associated with BWD in both cohorts.

Discussion

The association between BWD and adverse perinatal outcomes has been established in the literature [37], [38], [39], [40], [41], [42]. However, the literature is inconsistent in terms of adjusting the predictive models for chorionicity. Our finding is consistent with the literature in that we found higher odds of perinatal mortality and morbidity in BWD twins compared with growth concordant twins. These findings were observed in both cohorts, with and without chorionicity data. We also found increased odds of a 5-min Apgar score <7 in growth-discordant twins compared with the reference category.

Previous studies have demonstrated that the risk associated with discordant twins is related to the increased incidence of prematurity and the growth restriction of the twin pairs. To reduce mortality, resuscitation interventions and respiratory supports are used to manage complications associated with prematurity complications. The literature lacks the frequency with which twins with BWD undergo these interventions both at birth (e.g. resuscitation medication, cord blood testing for gases) and after birth (e.g. respiratory support, described here as the need for a ventilator, CPAP, or oxygen during hospitalization). Our findings suggest that at birth, the incidence of resuscitation medication used and the result of cord arterial gases testing are similar between growth-discordant and -concordant twins. However, there is a higher frequency of IPPV-ETT use among growth-discordant twins compared to those with <30% growth discordance. With regard to infection, we found that BWD infants have a higher rate of antibiotic prescription (26.7% vs. 17.4%) compared to growth-concordant infants even though the strep test results were similar between growth-discordant and -concordant twins (26.9% vs. 24.6%, P=0.70, Table 2) and newborn septicemia rates were similar. Our findings demonstrate that the BWD twins required early respiratory support (more ventilator days than growth-concordant twins); this finding is consistent with those of prior studies that identified the same difference [43], [44].

As chorionicity is not readily available in many large population based datasets, analysis by sex discordance can serve as a proxy. Information on adverse neonatal outcomes of growth discordant opposite-sex twins within the early hospitalization period is scarce. Our stratified analysis of sex-discordant twins was conducted as a substitute for chorionicity. This type of analysis has been adopted by other studies where no access to chorionicity data was possible [45], [46]. The adjusted odds (for sex discordance and other confounders) of early and late neonatal mortality and Apgar score of 5 min <7 were higher in BWD twins compared to concordant twins (Table 3). These findings are also consistent with the literature [13], [15], [47], [48], [49], [50]. If we assume that the differences in chorionicity confound the association between BWD and adverse perinatal outcomes, we would expect similar or lower odds among sex-discordant twin pairs who have DC placentas. Two other studies that used sex discordance as a surrogate variable for chorionicity found that BWD and neonatal mortality were associated in both same- and opposite-sex twins [15], [51].

When we stratified the data according to sex discordance, the odds of early and late neonatal mortalities in opposite-sex twins with growth discordance were higher than the corresponding odds in the sex-concordant data (Table 4). This result is in contrast with a large study of 4091 unlike-sexed and 10,875 like-sexed twins with a gestational age of 28 weeks and longer. This inconsistency might be due to several factors. Pregnancy loss, in the publication, was twice as high in like-sexed compared with unlike-sexed pairs, and only in like-sexed pairs was pregnancy loss strongly correlated with BWD. This study only included twins with a birth weight of 500 g or more. Furthermore, twins with TTTs and congenital anomalies or single stillbirths were not excluded [52].

Regarding perinatal morbidity, the findings from our subsample data analysis, taking chorionicity into consideration, were consistent with the results from the literature. A prospective multicentre study of 1028 twin pairs over a period of 2 years used a composite measure of perinatal morbidity inclusive of RDS, hypoxic ischemic encephalopathy, periventricular leukomalacia, necrotizing enterocolitis and sepsis. Adjusting for gestational age at delivery, perinatal mortality and composite perinatal morbidity were associated with BWD in both MC and DC twins [53]. Another report from the same study found that a composite measure of adverse perinatal outcomes, which included any of the morbidity measures described above or perinatal death, was more frequent in MC twins [54]. The literature lacks a validated composite index from perinatal morbidity that would be easily available in most databases. We therefore decided not to pool the variables together to create a composite index that could not be validated.

Information obtained from our study is important because it confirms many of the increased early mortality and morbidities that accompany discordant twins after adjustment for confounders and using a method of analysis, which adjusts, for the interdependence of outcomes associated with twin pairs. As the rate of twinning, prematurity and associated complications increase, this knowledge is necessary for both healthcare professionals and parents to better comprehend and anticipate outcomes in this group of twins who are “higher risk”. Future research to assess the role of chorionicity is beneficial to further clarify the most specific group of twins in need of neonatal intensive care unit services.

It is worth noting that we used “actual birth weight” rather than “estimated birth weight” for this study. Actual birth weight is research-oriented data and, therefore, accurate as we actually measure the birth weight of the baby on a calibrated scale. Our study results are relevant given that we assume that the estimated birth weight is equal or close to/equal to the actual birth weight data. A limitation of our data was that we did not have access to ultrasound data to identify twins with IUGR. Large studies using ultrasound defined estimated fetal weight discordance are required to examine if these associations can be extrapolated to the ultrasound based, antenatal realm which can then be applied to prenatal care. We also do not have information to stratify MC twins into different types of selective IUGR, which may affect the resulting findings [55]. Other limitations in our study involve missing data which does not allow for us to adjust for potential confounders. For instance, information on education was not available until after April 2004 nor was IVF data available until after 2008. Other variables with significant amounts missing information in the PSBC data set are BMI, weight gain, and number of antenatal visits (see footnote Table 1). For most part, the percentage of missing data is not large for the variables We utilized in our analysis and small amounts of missing data should not affect our findings.

In conclusion, twin pregnancies with BWD are at high risk of prenatal mortality, and lower Apgar score, irrespective of chorionicity. With the increasing twin pregnancy, prematurity, and higher utilization of intensive care unit resources both at delivery time and during infancy of discordant twin growth becomes more important. This knowledge is necessary for both healthcare providers and parents of such twins to be aware of the increased adverse outcomes in this highly specific population. We should be careful interpreting the result of this study as apart from BWD, many other factors determine the adverse perinatal mortality and morbidity.