It is commonly accepted that twin gestations are associated with an increased risk of perinatal morbidity and mortality, largely because of a higher risk of preterm delivery. The perinatal death rate in twins is 4 to 6 times higher than that in singletons.1,2 Twins are also associated with low and very low birth weights, another risk factor for morbidity and mortality. Growth restriction in twins, as in singletons, increases the risk of morbidity and mortality.3
Unlike growth restriction, growth discordance is a complication unique to multiple gestations. Discordant growth in twins has been associated with preterm delivery, low birth weights, and intrauterine growth restriction (IUGR).4,5 Past studies have shown that discordant twins have a higher perinatal death rate, more fetal anomalies, lower Apgar scores, longer hospital stays, and delayed intellectual development in childhood, in comparison with concordant twins.6,7 Because of the strong association between low birth weights, preterm delivery, and discordance, some authors argue that discordant growth is not an independent risk factor for fetal or neonatal morbidity and mortality.8,9 Continued pregnancy may increase the risk of fetal harm, but preterm delivery is also associated with morbidity. Our objective was to estimate whether discordance is an independent risk factor for perinatal morbidity and mortality in twins.
MATERIALS AND METHODS
This was a retrospective cohort study of twin deliveries at the Mount Sinai Medical Center, New York, NY, from 1992 to 2001. Data were retrieved from a centralized database of obstetric and neonatal information for all deliveries during that time period. Only pregnancies with 2 liveborn infants born at 24 weeks or later were included. Discordance was defined as a 20% or greater difference in birth weights and was calculated by using the formula (larger twin weight − smaller twin weight)/larger twin weight × 100. Small for gestational age (SGA) was defined as less than the 10th percentile by using Brenner10 singleton norms. The following 4 groups were compared: 1) appropriate for gestational age (AGA) and concordant, 2) AGA discordant, 3) SGA concordant, and 4) SGA discordant. Inclusion in the AGA groups was limited to twin pairs in which both twins were AGA. If at least 1 twin was SGA, the twins were assigned to the SGA groups. The discordant and concordant twins were compared with regard to maternal demographics; maternal medical complications associated with poor fetal growth, such as chronic hypertension, diabetes, autoimmune disease, renal disease, and anemia, antepartum complications; and adverse neonatal outcomes. Twins were managed by their own physicians, generalists in many cases. The ultrasounds were performed by maternal fetal medicine specialists. Generalists sought maternal fetal medicine input for management at their discretion, most often to decide when delivery was indicated.
Preterm delivery in this study was characterized as delivery less than 37 weeks of gestation, with a subset less than 32 weeks, and a separate category of medically indicated preterm delivery was included. Low birth weight (LBW) was defined as less than 2,500 g at delivery, and very low birth weight (VLBW) was defined as less than 1,500 g at delivery. Chorionicity was assessed via postdelivery placental pathology and was available in 68% of cases.
Maternal demographics and medical complications were compared among the 4 groups. Then, maternal and neonatal outcomes were compared between concordant and discordant AGA twins, as well as for concordant and discordant SGA twins. Finally, multiple logistic regression analysis was performed to assess the influence of confounding variables on the association between discordance and adverse perinatal outcome. These variables included chorionicity, antenatal corticosteroid use, oligohydramnios, preeclampsia, and gestational age at delivery. Analyses of neonatal characteristics were performed by using a twin pair as the unit of analysis, and a characteristic was considered present if either twin was affected. Categorical variables were assessed by χ2 test for independence or 2-tailed Fisher exact test in cases of small-cell frequencies. P values of .05 were considered statistically significant. In addition, Maentel-Haenzel odds ratios (ORs) of discordancy adjusted for SGA status for selected adverse perinatal outcomes, and the corresponding 95% confidence intervals (CIs) were calculated. We chose to present ORs to facilitate comparison with the adjusted ORs from the logistic regression analysis. Power analysis was used as appropriate with an α of .05 and a β of .8. All statistical computations were performed with SAS (Cary, SC) statistical software. This study was approved by the Institutional Review Board of the Mount Sinai School of Medicine.
During the study period from January 1, 1992, to December 31, 2001, 1,318 twin pairs meeting inclusion criteria were born at Mount Sinai. Of these, 856 pairs were AGA and concordant, 70 pairs were AGA and discordant, 254 pairs were SGA and concordant, and 138 pairs were SGA and discordant. The concordant and discordant twins were similar with regard to maternal demographics, including maternal age, race, parity, smoking, and alcohol and drug use. There was no significant difference between concordant and discordant twins in the rates of chronic diseases associated with poor fetal growth, such as chronic hypertension, diabetes, autoimmune disease, renal disease, and anemia. The groups were also similar with regard to conception rates with in vitro fertilization, tocolysis, and the use of corticosteroids for fetal lung maturity.
Antepartum and intrapartum characteristics are as follows. Among AGA twins, discordant twins more often had medically indicated preterm delivery, cesarean delivery, suspected IUGR, and preeclampsia. Among SGA twins, discordant twins more often had delivery at 32 weeks or earlier and indicated preterm delivery, cesarean delivery, oligohydramnios, and suspected IUGR. As for AGA discordant twins, oxytocin was used less often in delivering SGA discordant twins. The rates of fetal anomalies were similar among both AGA and SGA discordant and concordant twins. When adjusted for SGA status, discordance was an independent risk factor for delivery at 32 weeks or earlier (OR 1.55; 95% CI 1.18, 1.49) and indicated preterm delivery (OR 1.84; 95% CI 1.46, 2.31), cesarean delivery (OR 1.33; 95% CI 1.18, 1.49), oligohydramnios (OR 2.02; 95% CI 1.18, 3.44), fetal anomalies (OR 1.90; 95% CI 1.07, 3.35), suspected IUGR (OR 1.84; 95% CI 1.43, 2.36), and preeclampsia. Discordance was associated a decreased risk of preterm labor (OR 0.62; 95% CI 0.45, 0.85).
Neonatal outcomes are shown in Tables 1 through 3. Among AGA twins, those with discordance were more often LBW and VLBW and had more neonatal intensive care unit (NICU) admissions, oxygen requirements, necrotizing enterocolitis, and transient tachypnea of the newborn (Table 1). Among SGA twins, discordance was more often associated with LBW and VLBW, NICU admissions, ventilator requirement, oxygen requirement, intraventricular hemorrhage, hyperbilirubinemia and transient tachypnea of the newborn (Table 2). When adjusting for SGA, discordance remained an independent risk factor for LBW, VLBW, NICU admission, ventilator requirement, oxygen requirement, intraventricular hemorrhage, necrotizing enterocolitis, hyperbilirubinemia, and transient tachypnea of the newborn (Table 3). Respiratory distress syndrome (RDS) was more common in discordant than concordant SGA twins (10.9% versus 3.2%, P = .002), but because the trend was in the opposite direction for AGA twins (7.1% versus 11.6%, P = .26), the overall risk of RDS after adjusting for SGA did not differ between discordant and concordant twin pairs (OR 1.39; 95% CI 0.80, 2.42). Similarly, among SGA twins, neonatal death was more common for discordant pairs (0.4% versus 4.4%, P = .005), but the trend was in the opposite direction for AGA twins (1.9% versus 0.0%, P = .15), resulting in a 95% CI, which includes unity for the overall risk of neonatal death after adjustment for SGA status (OR 2.49; 95% CI 0.67, 9.20).
Multiple logistic regression was used to assess whether discordance remained an independent risk factor for the above neonatal outcomes after adjusting for antepartum characteristics. These included chorionicity, antenatal steroid use, oligohydramnios, preeclampsia, and gestational age at delivery. After adjustment, discordance was still associated with a decreased risk of preterm labor (OR 0.5; 95% CI 0.3, 0.82) but an increased risk of indicated preterm delivery (OR 2.3; 95% CI 1.48, 3.59) and cesarean delivery (OR 1.87; 95% CI 1.22, 2.87). Discordance was also independently associated with LBW, VLBW, NICU admission, oxygen requirement, hyperbilirubinemia, and transient tachypnea of the newborn (Table 4). However, discordance was not independently associated with ventilator use, necrotizing enterocolitis, and RDS. A statistically nonsignificant trend toward higher rates of intraventricular hemorrhage was seen in discordant twin pairs. Power analysis revealed that 5122, 1958, 5305, and 9604 sets of twins would have been required to detect a difference in ventilator use, necrotizing enterocolitis, RDS, and intraventricular hemorrhage, respectively.
As mentioned above, neonatal death was not associated with discordance in our preliminary analysis after adjusting only for SGA. When we performed regression analysis, we found that neither discordance nor SGA were independent risk factors for neonatal death. However, monochorionicity was an independent risk factor for neonatal death, independent of discordance (OR 4.11; 95% CI 1.19, 14.24).
The association between discordant growth and neonatal complications in twins has led to the frequent practice of increased antepartum surveillance and early delivery in these pregnancies. In the recent literature, investigators have addressed two important questions regarding discordant twin growth. The first relates to the degree of discordance, and studies have attempted to determine the cutoff that best predicts adverse neonatal outcome. The lack of a standard definition for a clinically significant growth difference within a twin pair is partly a result of conflicting data on the associated adverse perinatal outcomes. Several studies suggest that clinically significant adverse outcomes are not seen until discordance reaches levels of 30% to 40%, which is obviously greater than the generally accepted 20%.4,11,12 Our study was designed to address the second question, which is whether or not discordant growth is independently associated with adverse outcomes.
Authors have used varied methods to investigate the effects of discordance. In 1988, Blickstein et al13 studied 14 pairs of term discordant twins who were all AGA and weighed more than 2,500 g. That study thus eliminated growth restriction and prematurity as possible confounders, leaving discordance as the lone risk factor. Compared with concordant twin pairs matched for gestational age, no difference in neonatal morbidity or mortality was found. The authors concluded that, provided discordant twins reach term (37 or more weeks) and weigh 2,500 g or greater, they fare well. Another study by the same group evaluated “macrosomic” twins. Fifty-six twin pregnancies were divided into deciles, and the 10th, (heaviest decile) was compared with the 9th.14 All twins were term and AGA. Although there were actually more discordant pairs in the 10th decile, there was no increase in neonatal morbidity or mortality.
Bronsteen et al8 evaluated 131 pairs of liveborn twins to determine which factors were most predictive of morbidity and mortality. In a multivariate regression analysis, IUGR, birth weight percentile, and prematurity were predictors of poor neonatal outcome, whereas discordance was not. Patterson et al9 also used multivariate regression analysis to examine 194 sets of twins. Although morbidity, mortality, and the presence of anomalies were related to small weight for gestational age and prematurity, they were not independently related to discordance.
In another study, Redman et al12 found that discordance above 31% (the 95th percentile in their population) was associated with NICU admission, low 5-minute Apgar scores, fetal acidemia, and cesarean delivery for fetal distress. After regression analysis, they concluded that growth discordance was significantly associated with fetal acidemia and cesarean delivery but not with ponderal index and chorionicity. They further concluded that 31% discordance may reflect a more appropriate definition of abnormal than 20%.
In a population-based retrospective study of 9,590 twin pairs, Hartley et al15 described a much higher rate of SGA status in discordant compared with concordant pairs (58% versus 9%). Pairs were stratified by the presence of 0, 1, or 2 SGA infants, and poor outcomes increased in direct proportion to the number present. This birth and death certificate–based study examined only perinatal death, neonatal death, 5-minute Apgar scores less than 7, or respiratory difficulty, defined as RDS or 30 minutes or more of assisted ventilation; no other neonatal morbidities were analyzed. Discordant pairs had significantly higher rates of perinatal mortality than nondiscordant pairs, even after stratification by both gestational age and SGA status. In summary, previous investigations of the significance of growth discordance have reached opposite conclusions.
Our findings suggest that 20% growth discordance in twins places them at higher risk for some adverse neonatal outcomes whether they are AGA or SGA. However, the most serious morbidities and neonatal mortality did not show an independent relationship with discordance. Analysis of our data was complicated by the closely linked relationships of variables such as gestational age at delivery and birth weight with discordance. In the reality of obstetric decision-making, it would be ideal to identify independent markers of pregnancies at high risk for neonatal death or injury. If discordance were shown to have an independent relationship with severe neonatal outcomes, such as intraventricular hemorrhage, necrotizing enterocolitis, RDS, and death, it would be reasonable to accept some additional risk from iatrogenic preterm delivery to reduce overall neonatal morbidity and mortality. Discordance of 20% or greater in of itself is not predictive in this respect. Our data indicate that discordance is an independent risk factor only for more minor adverse neonatal outcomes, and therefore, it is difficult to justify preterm delivery for the sole indication of discordant growth. However, one should not ignore a particularly high-risk group of infants, that is, the pregnancies with discordant growth and one or both infants SGA. In this group, further research is required to determine whether or not early delivery will improve outcomes.
Our study had several limitations, which may have affected our results. First, we used singleton norms to define AGA and SGA, although twins are known to demonstrate lower weights than singletons after 30 weeks.16 It is not clear whether the twin infants at the higher end of the SGA range are actually “AGA for twins” and therefore at no higher risk of morbidity. Therefore, our study may have overestimated the incidence of clinically relevant SGA pregnancies after 30 weeks of gestation. Furthermore, because the incidence of poor neonatal outcomes after 30 weeks is less frequent, it may be more difficult to detect a negative impact of discordance in this segment of the population. We chose to use singleton norms because those for twins are not universally accepted, and our goal was to maximize the applicability of our results to generalized practice.
Second, our study does not include stillbirths, clearly one of the most important adverse obstetrical outcomes. If discordance was an independent risk factor for stillbirth, early delivery certainly might be indicated to attempt to avoid this occurrence. However, the retrospective nature of our research made it impossible to differentiate between false discordance at birth, caused by growth of one twin after the other’s death, and a true difference, which predated the demise of one twin. Although a study using frequent serial ultrasounds would allow inclusion of stillbirths, it would be limited by the errors inherent in estimating fetal weight by sonogram. A recent study of 297,155 twin births from the U.S. Matched Multiple Birth File found that birth weight discordance was associated with stillborn fetus regardless of SGA status. Twins who were SGA and also discordant were at the highest risk of dying in utero; this pattern was not apparent for neonatal mortality.17 However, another group using the same data found increased rates of neonatal mortality in smaller and larger twins affected by 25% or more birth weight discordance after controlling for fetal growth.18
In conclusion, discordance does appear to increase some adverse perinatal outcomes in twin gestations, both for AGA and SGA pregnancies. In our sample, however, 20% discordance was not an independent risk factor for serious neonatal morbidities or mortality. We were limited by the small number of adverse outcomes among concordant twins, even with our large sample size. Future research should certainly be directed at SGA discordant twins because these infants may be at a particularly high risk for adverse outcomes.