Materials and Methods
This was a retrospective cohort study of twin births, using the US National Center for Heath Statistics’s (NCHS) 1995-2000 matched multiple birth data set (the largest available linked multiple birth dataset). The NCHS-matched multiple birth data contain information on maternal and pregnancy characteristics and perinatal and infant mortality for all multiple births in years 1995-2000 in the Unites States. There were a total of 658,424 twin births.
Because the study aimed to evaluate the association of birth order with perinatal mortality, we excluded unmatched twins and births with unknown birth order (n = 77,685 births) (11.8%). Overall perinatal mortality rate was much higher for births with vs without missing data on birth order (116.3 vs 34.5 per 1000; P < .001).
Because the study’s main aim was to determine the effect of birth order on perinatal mortality by birth order–specific weight difference, we further excluded: (1) twin pairs with any reported congenital anomalies in either twin (first or second); (2) twin pairs missing data on birthweight in either twin; (3) births recorded at extreme gestational ages (<23 weeks or >42 weeks) or extreme birthweights (<500 g or >6000 g) or implausible birthweight for gestational age ; and (4) twin sets not delivered at the same gestational week.
The final study cohort included 517,600 twin births in 258,800 twin pregnancies. Research ethics approval was waived by the Shanghai Xinhua hospital Research Ethics Board because the study was based on the anonymized matched multiple birth dataset downloadable from the NCHS web site.
The NCHS multiple birth dataset contained variables on fetal sex (boy/girl), presentation (breech/malpresentation, yes/no) (missing 1.0%), mode of delivery (cesarean/vaginal) (missing 0.7%), induction or stimulation of labor (missing 0.8%), prolonged (defined as >20 hours) or dysfunction labor (yes/no) (missing 1.0%), fetal distress (missing 6.8%), gestational age (weeks), and birthweight (grams). The NCHS birth database contained an indicator variable for breech/malpresentation (impossible to distinguish between breech and other malpresentations). If the answer was a no, we classified the fetus as normal vertex presentation.
The NCHS birth database contained indicator variables for primary cesarean section and repeat cesarean section. If the answer was a yes to either question, we classified the birth as cesarean delivery; if the answer was a no to both questions, we classified the birth as vaginal delivery. The NCHS birth database contained 21 items for reporting 20 specific and other congenital anomalies. These fields were used to capture and exclude twin pairs with any reported congenital anomaly in either twin (first or second born). Fetal distress in the NCHS data is a nondescriptive term equivalent to the most contemporary use of the term nonreassuring fetal status.
The primary outcome was perinatal death because it is a more robust indicator of mortality risk than stillbirth and neonatal death that are influenced by variations in personal judgments and registration practices (whether a dead newborn is registered as a stillbirth or neonatal death), especially for births at borderline of viability. Secondary outcomes included the components of perinatal death-stillbirth (fetal deaths at a gestation of ≥20 weeks) and neonatal death (deaths during the first 4 weeks or 0-27 days of life after birth).
Cause-specific neonatal mortality were analyzed according to the classification scheme by International Collaborative Effort on Perinatal and Infant Mortality for asphyxia and injuries (may be related to delivery), immaturity-related conditions, infections, sudden infant death syndrome, and others. Causes of death are missing for all stillbirth records in the NCHS birth data.
The primary exposure of interest was birth order. Because it is very unlikely that a second twin was delivered vaginally following delivery of a first twin by cesarean section in a twin set, the frequency of such implausible records was used as an indicator of data quality in birth order. There were 782 pairs of first twin cesarean–second twin vaginal (implausible) births, and 10,959 pairs of first twin vaginal–second twin cesarean (plausible) births, indicating a birth order recording error rate of about 6.7%. Birth order was reversed to normal for these 782 pairs of first twin cesarean–second twin vaginal births, giving a total of 11,741 twin pairs of first twin vaginal–second twin cesarean births. Because it is unlikely that second twin’s cesarean delivery following first twin’s vaginal delivery is a planned event, all such births were classified as vaginal births in intention-to-treat analyses. Cesarean section deliveries of 15,731 twin pairs in cases of induction or stimulation of labor and prolonged or dysfunctional labor (an indication of failed trial of labor/vaginal delivery) were also classified as vaginal births in intention-to-treat analyses.
Stratified analyses were conducted to assess perinatal mortality risk differences between second- and firstborn twins by birth order–specific weight difference: within 5% (similar), first twins heavier by 5.0-14.9%, 15.0-24.9%, and 25.0% or greater or second twins heavier by 5.0-14.9%, 15.0-24.9%, and ≥25.0% (7 strata). To compensate for the fact that boys on average weighed about 3.6% (84 g) more than girls based on exploratory data analysis in the study cohort, birthweight of girls in different-sex twin pairs was artificially inflated by 3.6% before calculating birthweight differences in different-sex twin pairs for the purpose of comparably grouping birthweight percentage differences for all twin pairs (same sex or different sex).
To gain insight on perinatal mortality risk changes in second vs first twins by gestational age, we evaluated the risk changes in clinically important gestational age categories: extremely preterm (23-27 weeks), very preterm (28-31 weeks), mild preterm (32-36 weeks), and term (≥37 weeks). The fetuses-at-risk approach was applied to estimate gestational age stratum-specific mortality rates to avoid a potential collider effect of stratification by gestational age at birth. The fetuses-at-risk denominator is the number of all fetuses at risk of death (both born and yet unborn babies).
The unit of data analysis was the fetus. Conditional logistic regression (appropriate for paired data) was used to estimate the odds ratios (ORs) and 95% confidence intervals (CIs) of perinatal death, stillbirth, and neonatal death comparing second vs first twins. Adjusted ORs were controlled for important known fetus-specific risk factors: fetal sex, presentation (breech/malpresentation: yes/no), and birthweight for gestational age (small for gestational age [SGA]: yes/no). SGA was defined as birthweight less than the 10th percentile, according to sex- and gestational age–specific birthweight percentiles for nonmalformation births to nonsmoking mothers in the study twin cohort. Exploratory analyses observed that SGA was strongly associated with perinatal mortality (crude OR, 5.7; 95% CI, 4.7–6.8) in the study cohort.
We chose to adjust for SGA because it is strongly predictive of perinatal mortality, given the same gestational age in twin pairs, and that exploratory analyses observed similar ORs for birth order adjusting for SGA or birthweight for gestational age z score. Because twin pairs represent a perfect match in maternal characteristics, there is no need to adjust for maternal factors constant to a twin pair in comparisons of perinatal mortality between second and first twins.
In paired data analyses, only twin pairs discordant in perinatal death contributed to the relative likelihood of mortality comparing second vs first twins. However, we chose to include data for all twins (irrespective of discordance in perinatal death) to calculate and present absolute mortality rates, which are equally important in obstetric risk management decisions in addition to relative likelihood of mortality.
All data management and analyses were carried out using Statistical Analysis System, version 9.2 (SAS Institute, Cary, NC). Ad hoc power calculations indicated a power of greater than 99% to detect a risk increase or decrease of 20% or greater comparing second vs first twins in risks of perinatal mortality accounting for multiple comparisons. Two-tailed values of P < .0025 were considered statistically significant, accounting for 20 primary comparisons of interest (perinatal mortality in second vs first twins overall and in subgroups by gestational age, birth order–specific weight difference, presentation, and mode of delivery) (0.05/20 = .0025).
Results
Table 1 presents the characteristics of the twin study cohort. The majority of mothers were white (78.9%) and between 20 and 34 years of age (74.3%). Maternal smoking was reported in 10.3% of women. Comparing second- vs first-born twins, mean birthweight was significantly (mean difference, 32 g) lower, SGA (11.6% vs 8.9%), very low birthweight (<1500 g) (9.4% vs 8.8%) births were significantly more frequent, and fetal distress (4.2% vs 3.2%), breech/malpresentation (28.4% vs 20.7%), and cesarean delivery (57.6% vs 53.6%) occurred more frequently in second twins.
| Characteristic | Mothers | Twins | P value a | |
|---|---|---|---|---|
| Firstborn | Secondborn | |||
| Mothers, n, % | 258,800 | |||
| Race | ||||
| White | 204,705 (79.1) | |||
| Black | 43,873 (16.9) | |||
| Others | 10,222 (4.0) | |||
| Age, y | ||||
| <20 | 17,805 (6.8) | |||
| 20-34 | 193,584 (74.3) | |||
| ≥35 | 49,324 (18.9) | |||
| Primiparous | 106,803 (41.3) | |||
| Smoking | 21,873 (10.3) | |||
| Newborns, n | 258,800 | 258,800 | ||
| Sex, male | 129,994 (50.2) | 129,553 (50.1) | .22 | |
| Gestational age, wks | 35.6 ± 3.3 | 35.6 ± 3.3 | .99 | |
| Birthweight, g | 2401 ± 623 | 2369 ± 628 | < .001 | |
| Birthweight <1500 g | 22,645 (8.8) | 24,393 (9.4) | < .001 | |
| SGA (<10th percentile) b | 23,021 (8.9) | 29,986 (11.6) | < .001 | |
| Breech/malpresentation | 53,039 (20.7) | 72,713 (28.4) | < .001 | |
| Induction/stimulation of labor | 58,791 (22.7) | 59,200 (22.9) | .18 | |
| Prolonged/dysfunctional labor | 7324 (2.8) | 7581 (2.9) | .03 | |
| Fetal distress | 7674 (3.2) | 10,196 (4.2) | < .001 | |
| Cesarean delivery | 137,704 (53.6) | 148,223 (57.6) | < .001 | |
a P values comparing second vs first twins
b SGA is less than the 10th percentile, according to birthweight percentiles in nonmalformation births to mothers without smoking in the study twin birth cohort.
There were 3651 twin pairs discordant in perinatal death, including 1689 pairs of first twin death only (second twin survived), and 1962 pairs of second twin death only (first twin survived). There were 255,149 twin pairs concordant in outcomes; both died in 2301 pairs, and both survived in 252,848 pairs. Overall perinatal mortality rates were higher in second vs first twins (16.5 per 1000 vs 15.4 per 1000; P < .001) ( Table 2 ). The crude risk elevation was significant (OR, 1.15; 95% CI, 1.07–1.23; P < .001). However, the perinatal mortality risk difference between second and first twins became nonsignificant after controlling for fetal sex, presentation, and birthweight for gestational age (adjusted OR, 1.06; P = .09).
| Variable | All twins | Mortality, n (per 1000) | Crude a | Adjusted a | |
|---|---|---|---|---|---|
| Secondborn twins | Firstborn twins | OR (95% CI) | OR (95% CI) a | ||
| Total births | 517,600 | 258,800 | 258,800 | ||
| Total live births | 509,347 | 254,539 | 254,808 | ||
| Perinatal death b | 8253 (15.9) | 4261 (16.5) | 3992 (15.4) | 1.15 (1.07–1.23) c | 1.06 (0.99–1.14) |
| Stillbirth b | 2497 (4.8) | 1352 (5.2) | 1145 (4.2) | 1.22 (1.10–1.35) c | 1.18 (1.03–1.36) e |
| Neonatal death d | 5756 (11.2) | 2909 (11.3) | 2847 (11.0) | 1.02 (0.95–1.11) | 0.98 (0.91–1.07) |
| Asphyxia/injuries | 462 (0.9) | 243 (0.9) | 219 (0.8) | 1.13 (0.91–1.40) | 1.09 (0.88–1.36) |
| Immaturity-related | 3221 (6.3) | 1616 (6.3) | 1605 (6.2) | 1.00 (0.89–1.10) | 0.95 (0.85–1.06) |
| Infections | 1274 (2.5) | 649 (2.5) | 625 (2.4) | 1.03 (0.90–1.18) | 1.03 (0.90–1.19) |
| SIDS | 68 (0.1) | 29 (0.1) | 39 (0.2) | 0.78 (0.48–1.27) | 0.83 (0.50–1.38) |
| Others | 731 (1.4) | 372 (1.4) | 359 (1.4) | 1.02 (0.86–1.21) | 0.97 (0.81–1.16) |
a ORs from conditional logistic regression models comparing second vs first twins; adjusted ORs were controlled for fetal sex, presentation (breech: yes/no) and birthweight (SGA <10th percentile, yes/no); maternal characteristics were not included in the adjustments because they were self-matched in twin pairs
b Risk of perinatal death or stillbirth is expressed per 1000 total births
d Risk of neonatal death (0-27 days) is expressed per 1000 live births
Stepwise adjustments observed that birthweight for gestational age was the major factor accounting for the perinatal mortality risk difference, which became nonsignificant after adjusting for birthweight for gestational age alone (adjusted OR, 1.06; P = .07). The higher risk of perinatal death in second vs first twins was driven by a higher stillbirth rate (5.2 per 1000 vs 4.2 per 1000; adjusted P = .02). Overall neonatal mortality rates were similar in second vs first twins (11.3 per 1000 vs 11.1 per 1000; adjusted P = .68). Cause-specific neonatal mortality rates were all not significantly different.
The risks of perinatal death comparing second vs first twins were dependent on gestational age and birth order–specific weight difference ( Table 3 ). Higher risks of perinatal death were observed comparing second vs first twins in mild preterm (32-36 weeks) (adjusted OR, 1.22; P = .02) and term (≥37 weeks) (adjusted OR, 1.58; P < .001) births but not in very (28-31 weeks) or extreme (23-27 weeks) preterm births.
| Variable | Number of twin births n, % | Perinatal death, n (per 1000) | Adjusted | |
|---|---|---|---|---|
| Secondborn twins | Firstborn twins | OR (95% CI) a | ||
| Gestational age, wks b | ||||
| 23-27 | 17,770 (3.4) | 2605 (10.1) | 2597 (10.0) | 0.97 (0.87–1.07) |
| 28-31 | 35,820 (6.9) | 649 (2.6) | 664 (2.7) | 0.89 (0.76–1.04) |
| 32-36 | 233,424 (45.1) | 673 (2.9) | 522 (2.2) | 1.22 (1.04–1.43) c |
| ≥37 | 230,586 (44.6) | 334 (2.9) | 209 (1.8) | 1.58 (1.27–1.96) d |
| Birthweight, heavier, % e | ||||
| Larger first-born twins | ||||
| +25.0% or more | 46,928 (9.1) | 982 (41.9) | 503 (21.4) | 3.94 (3.28–4.73) d |
| +15.0-24.9% | 49,424 (9.5) | 394 (15.9) | 324 (13.1) | 1.90 (1.46–2.47) d |
| +5.0-14.9% | 100,584 (19.4) | 643 (12.8) | 580 (11.5) | 1.37 (1.14–1.66) d |
| Similar birthweight | ||||
| +4.9 to –4.9% | 151,538 (29.3) | 986 (13.0) | 999 (13.2) | 0.97 (0.84–1.12) |
| Larger second-born twins | ||||
| +5.0-14.9% | 95,788 (18.5) | 581 (12.1) | 661 (13.8) | 0.67 (0.56–0.81) d |
| +15.0-24.9% | 42,622 (8.2) | 287 (13.5) | 326 (15.3) | 0.63 (0.48–0.83) d |
| +25.0% or more | 30,716 (5.9) | 372 (24.2) | 615 (40.0) | 0.36 (0.30–0.44) d |
| Mode of delivery | ||||
| Vaginal-vaginal | 216,262 (42.1) | 2205 (20.4) | 2066 (19.1) | 1.16 (1.02–1.32) c |
| Vaginal-caesarean | 23,482 (4.6) | 174 (14.8) | 205 (17.5) | 0.98 (0.71–1.35) |
| Cesarean-cesarean | 273,844 (53.3) | 1792 (13.1) | 1663 (12.1) | 1.01 (0.92–1.11) |
| Presentation | ||||
| Vertex-vertex | 346,310 (67.7) | 2771 (16.0) | 2602 (15.0) | 1.04 (0.96–1.14) |
| Vertex-breech | 59,194 (11.6) | 503 (17.0) | 457 (15.4) | 1.10 (0.90–1.34) |
| Breech-vertex | 20,148 (3.9) | 235 (23.3) | 218 (21.6) | 1.19 (0.89–1.59) |
| Breech-breech | 85,816 (16.8) | 617 (14.4) | 598 (13.9) | 1.00 (0.84–1.20) |
a ORs comparing second vs first twins from conditional logistic regression models adjusting for fetal sex, presentation (breech: yes/no), and birthweight for gestational age (small for gestational age <10th percentile, yes/no); maternal characteristics were not included in the adjustments because they were self-matched in twin pairs
b Gestational age–specific mortality rates were based on fetuses-at-risk denominators (including unborn fetuses)
e Paired birthweight difference in percentage comparing the larger vs the smaller twins.
Comparing second vs first twins, the risks of perinatal death were similar if they had similar birthweights (within 5%) and were increasingly higher if second twins weighed progressively less (adjusted ORs were 1.37, 1.90, and 3.94 if weighed 5.0-14.9%, 15.0-24.9%, and ≥25.0% less, respectively; all P < .001) and progressively lower if they weighed increasingly more (adjusted ORs were 0.67, 0.63, and 0.36 if weighed 5.0-14.9%, 15.0-24.9%, and ≥25.0% more, respectively; all P < .001).
Similar risk patterns by birth order–specific weight difference were observed for stillbirth and neonatal death (data not shown). Among 9.1% twin pairs, first twins weighed ≥25.0% more than second twins, whereas only among 5.9% twin pairs, second twins weighed ≥25.0% more than first twins.
A slightly higher risk of perinatal death was observed in second twins if both twins delivered vaginally (adjusted OR, 1.16; P = .02), whereas there was no risk difference in perinatal mortality if both were cesarean deliveries (adjusted OR, 1.01; P = .81). Overall, there were no significant risk differences in perinatal mortality between second and first twins stratified by fetal presentation (vertex-vertex, vertex-breech, breech-vertex, breech-breech) or fetal sex (same sex, opposite sex) (data not shown). There was no significant interaction between primiparity and birth order in relation to the risk of perinatal death ( P = .92).
Intention-to-treat analyses on the risks of perinatal death in vaginal births are presented in Table 4 . Adjusting for fetal sex, presentation, and birthweight for gestational age, there was a substantially elevated risk of perinatal death in vaginal and intended vaginal deliveries at term (≥37 weeks) in second vs first twins (3.1 vs 1.8 per 1000; adjusted OR, 2.15; P < .001). However, there was no significant risk difference at preterm. Stratified analyses by presentation showed that second twins in vaginal deliveries were at a slightly elevated risk of perinatal death if the first twin was in vertex presentation, irrespective of presentation in the second twin.
