Main

The prevalence of multiple births has increased in recent years (1–3), making the identification of factors that influence outcome in these infants increasingly important from a public health perspective. Infants from multiple births are known to be at higher risk of cerebral palsy (CP) than singleton infants (4–7). As the risk of CP is strongly related to low birth weight or immaturity, and twin infants are on average smaller and born earlier than singleton infants, it is necessary to separate the influence of birth weight from other factors which may independently predict unfavorable outcome.

The aim of this study is to describe the risk of CP in a large population-based cohort of over one million births, of whom more than 20,000 were twins. Our data came from five populations in Australia and the United States (9). Footnote 1 Demographic characteristics were derived from birth records and supplemented by perinatal death certifications and CP diagnoses obtained from registers with a high ascertainment of cases.

Twins are more likely than singletons to die as infants (4, 8–10), precluding a diagnosis of CP which generally cannot be made with confidence until months or years after birth. Our primary purpose was to explore factors related to CP risk in twins. It was necessary, therefore, to examine factors associated with the competing risk, which is early mortality.

METHODS

Vital statistics collected during the decade of the 1980s from five populations in the United States and Australia were combined for analysis of mortality and neurologic morbidity in multiple births. The three Australian populations consisted of all births from Western Australia, Victoria, and South Australia. The North American populations were from California (four counties in the San Francisco Bay area) and Georgia (five counties from the Metropolitan Atlanta area). Singleton and twin births only were included. Detailed descriptions of the individual datasets can be found in the following references: Western Australia (Stanley et al. 1994) (11); Victoria (Beischer et al. 1990) (12); South Australia (Chan et al. 1986, Byron-Scott et al. 1997) (13, 14); California (Grether et al. 1992) (15); and Georgia (Boyle et al. 1991, Adams et al. 1997) (16–18).

Cerebral Palsy diagnoses in the Georgia dataset were from the Metropolitan Atlanta Developmental Disabilities Surveillance Program. Cerebral Palsy was diagnosed at age 3 to 10, and was defined as “a group of nonprogressive, but often changing, motor impairment syndromes secondary to lesions or anomalies of the brain arising at any time during brain development. The impairment of motor function may result in paresis, involuntary movement, or incoordination.” Children with postnatally acquired CP diagnosed before 11 y of age were included. Excluded were motor disorders which are transient, a result of progressive disease of the brain, or which are due to spinal cord abnormalities or injuries. CP diagnoses were limited to children with known plurality. In the California CP registry (15), CP was defined as “a chronic disability of CNS origin characterized by aberrant control of movement or posture, appearing early in life and not as the result of a progressive disease.” Children were ascertained at the age of 3, and instances of CP due to infection or other adverse event after the first 28 d of life were not included. CP was ascertained at age 5 in the Australian registries and defined in a similar manner as California. CP diagnoses were available for all infants in the Victorian and Western Australian populations and twin infants only in the South Australian population.

Each entry in the database corresponded to an individual birth, not pregnancy. Thus, each twin had a separate entry in the database. The following data were available: birth weight, gestational age, gender, plurality, maternal race (except for Victoria), maternal age, and birth order (for twins). Entries with unknown birth weight or with an invalid gestational age (defined as less than 20 wk, more than 46 wk, or unknown) were excluded from further analysis except where noted. The following outcomes were recorded: stillbirth, death <28 d, death ≤1 y, and cerebral palsy. In some cases, the birth weight for gestational age data were implausible, particularly when the gestational age was very early. We chose not to exercise arbitrary rules of exclusion as these inevitably impose bias. However, our assumption is that birth weight is more likely to be correct than gestational age.

Almost all (99.6%) of the twins were matched to their co-twins. For these twins, the same demographic and outcome data were available for the co-twin, e.g. co-twin sex, co-twin birth weight, and co-twin outcome (stillbirth, neonatal death, postneonatal death <1 y, or cerebral palsy). Each matched twin was classified as being either in a same-sex pair or a different-sex pair. Four types of twins were defined: twins from male-only pairs (M:M), twins from female-only pairs (F:F), female twins from mixed-sex pairs (F:M) and male twins from mixed-sex pairs (M:F). A twin was defined as growth discordant if the absolute difference in birth weight between twin and co-twin exceeded 20% of the larger twins' birth weight. To distinguish between growth discordance due to the fetal death of one co-twin from growth discordance due to other reasons, analysis was restricted to growth discordant pairs with both twins live-born except where otherwise noted.

Crude rates of mortality and CP were calculated overall and by demographic and twin factors. Mortality rates include stillbirth and death within 28 d of birth, and were calculated as a percentage of all births (live births and still births). Rates of CP were calculated as a percentage of all infants alive at 1 y. Univariate and multivariate analyses were performed using Intercooled Stata 6.0 (Stata, Corp). Univariate comparisons were made using t-tests and χ2 tests as appropriate. The p-values considered statistically significant were modified to avoid the problem of multiple comparisons as indicated. Linear changes in study variables over the time frame of the study (1980–1989) were assessed using a nonparametric trend test (Stata, nptrend) (19). Logistic regression was used to calculate adjusted prevalence rate ratios (odds ratios) for some demographic and twin factors after adjustment for birth weight and gestational age where noted. In these logistic models, birth weight was modeled using four terms (bwt, bwt (2), bwt (3), bwt (4) and gestational age was modeled using two terms (ga, ga2) in an effort to find the best fit for the observed nonlinear association of these primary factors on the outcomes of interest. To allow for the nonindependence of data from multiple births, p-values were calculated using a robust variance calculation in comparisons using the logistic model and for other comparisons (t tests, χ2 tests), twin pairs were treated as a single sampling unit for purposes of variance computations.

RESULTS

Information on 1,142,252 births was available from the three Australian and two American populations. Of these, 1,141,351 were singletons or twins. Table 1 shows maternal and singleton/twin characteristics for the individual populations for the years covered by the study. The non-Caucasian population for the regions was as follows, excluding those with unknown race or “other” race: Western Australia: 5.1% Aboriginal; South Australia: 2.6% Asian, 1.8% Aboriginal; California Footnote 2: 10.9% Asian, 7.9% Black; Georgia; 36.8% Black. Maternal race was not available in the Victorian data set.

Table 1 Characteristics of the Individual Data Sets
Full size table

Secular trends.

The proportion of births that were twins or higher-order multiples, increased from 1.9 in 1980 to 2.5% in 1989. Footnote 3 Of three major factors associated with twinning, race, maternal age, and use of assisted reproductive therapy, it was possible to assess only the influence of maternal age, which increased slightly from 26.1 to 27.7 y over this time period. The proportion of mothers aged 35 y or more increased from 5.9 to 9.9% of all births. Compatible with increasing maternal age and increased use of assisted reproductive therapy, different-sex pairs increased from 24% to 32% of all twin pairs. All of these linear trends were statistically significant (p < 0.05) for the combined dataset. These trends were also generally apparent for each of the five study populations, although not always to a statistically significant degree.

Characteristics of twin pregnancies.

Twins or higher order multiples represented 2.3% of all births, 1.2% of all pregnancies. The prevalence of twin pair types was as follows: male-only pairs (36%), female-only pairs (35%), mixed pairs (29%). Since it is assumed that the mixed pairs represent half of the dizygotic pairs, 58% (2 times 29%) of the twins are estimated to be dizygotic and, by subtraction, 42% are monozygotic pairs. Thus 29/71 = 40.8% of the like sex pairs are dizygotic.

The average gestational age for twin pregnancies was about 3 wk less than for singleton pregnancies (Fig. 1 A). § In 16% of (live born) twins, the absolute weight difference between the infants exceeded 20% of the larger twin's birth weight. Size discordance was more common in different-sex than in same-sex pairs (18.2%versus 14.9%, p < 0.001) (including only pairs in which both infants were live-born).

Figure 1
figure 1

(A,B) Gestational age and birth weight distribution for live-born singletons and twins.

Full size image

Characteristics of twins.

On average, live-born twins were about 940 g smaller than live-born singletons (Fig. 1 B). § While only 5% of (live born) singletons weighed less than 2500 g at birth, 48% of twins weighed less than 2500 g. Similarly, only 1% of singletons but 8% of twins weighed less than 1500 g.

The weight difference between singletons and twins was apparent at all gestational ages: The higher median birth weight in singletons increased from 70 g at 24 wk to 590 g at 40 wk. Among twins, second-born live-born twins were smaller than first-born live-born twins by 43 g Footnote 4 and same-sex twins were smaller than different-sex twins by 74 g. Female twins who had a female-co-twin were about 62 g smaller than female twins who had a male co-twin. Male twins who had a male co-twin were about 86 g smaller than male twins who had a female co-twin.

Rates of early mortality and CP are shown by demographic factors for singletons and twins in Table 2 and by twin factors in Table 3, presented first for early mortality and then for CP.

Table 2 Crude early mortality (still-birth or neonatal) and crude cerebral palsy (CP) rates in singletons and twins by demographic factors
Full size table
Table 3 Crude early mortality (still-birth and neonatal), and crude cerebral palsy (CP) rates by twin factors
Full size table

Mortality Rates in Singletons and Twins

Demographic factors.

Overall, twins had approximately five times greater risk of stillbirth or neonatal death than singletons (Table 2). Singletons were at higher risk of stillbirth (0.66% [0.64%-0.68%] of all births) than neonatal mortality (0.47% [0.46%-0.48%] of all births), whereas twins were at higher risk of neonatal mortality (3.17% [2.90%-3.44%] of all births) than stillbirth (2.66% [2.42%-2.90%] of all births).

Crude rates of mortality were higher for males and for nonwhites for both singletons and twins. Nonwhites were on average smaller and less mature at birth than whites. To compare mortality rates among whites and nonwhites (and other groups with different average birth weights and gestational ages), logistic regression was used to calculate odds ratios adjusted for birth weight and gestational age (see methods). After adjustment, nonwhites had lower risk of mortality (OR = 0.81 singletons, OR = 0.61 twins, p < 0.0001) than whites. Similarly, singletons of both younger (<20 y) and older (≥35 y) mothers were at increased risk of early death. After adjustment for birth weight and gestational age, singletons of younger mothers were at reduced risk of mortality (OR = 0.89, p = 0.002), while singletons of older mothers remained at increased risk (OR = 1.20, p < 0.0001). A similar pattern was seen for twins as for singletons after adjustment, i.e. rates were lower for twins of younger mothers and increased for twins of older mothers, but was not statistically significant.

Twin factors.

While the overall rate of early mortality was higher in twins than singletons, the pattern of risk varied by birth weight and gestational age strata. Figure 2 shows the rate of mortality by birth weight and gestational age for singletons and twins. Very low birth weight twins and singletons had similar rates of mortality (Fig. 2). Above 1000 g but less than 2750 g, twins had a lower rate of mortality. Above that weight, twins had similar or higher mortality rates. When viewed by gestational age, twins had higher mortality rates below about 27 wk gestational age, lower rates from about 30 to 35 wk gestational age, and higher rates above 38 wk gestational age.

Figure 2
figure 2

(A,B) Mortality (still-birth and neonatal) by birth weight and gestational age for singletons and twins. ▪p < 0.002, difference between singletons and twins. ○p < 0.05, difference between singletons and twins. Note Birth weight is combined above 4000 grams for twins and above 5500 grams for singletons. (C, D) Log scale.

Full size image

Overall, singletons in the 3250–4500 g weight range or at 40–41 wk gestational age had the greatest chance of surviving the neonatal period (visible on a log scale –Fig. 2 C, D). For twins, the corresponding optimal birth weight and gestational age was 2500–3750 g and 37–38 wk gestational age. Even at optimal birth weight, however, twins had approximately twice the risk of early mortality compared with singletons at optimal birth weight (0.43%versus 0.20% approximately).

Table 3 shows mortality rates for twins by twin-specific factors, including birth order, twin type, gender-pair type, growth discordance, and co-twin outcome. To include both twins in each pair, twins with missing or out of range gestational age are included in these rates. Rates of mortality were increased for second-born twins and twins in same-sex pairs, and especially for male-only pairs. Adjustment for birth weight and gestational age did not change these patterns.

Mortality rates were greatly increased for twins whose co-twin had died in the neonatal period compared with twins whose co-twin survived the neonatal period. This increased risk was apparent across the birth weight range (Fig. 3 A) but appears largely confined to same-sex twins (Fig. 3 B). Overall, the risk of mortality in twins whose co-twin was a neonatal survivor was 2.2%, was 64% for different-sex twins whose co-twin died, and was 73% for same-sex twins whose co-twin died. This difference in risk persisted after adjustment for gestational age and birth weight: compared with twins whose co-twin was a neonatal survivor, same-sex twins whose co-twin died had higher risk of death (OR = 15.2[10.7–21.6], p < 0.0001) than different-sex twins whose co-twin died (OR = 3.6[2.0–6.7], p < 0.001).

Figure 3
figure 3

(A) Mortality by birth weight and co-twin mortality. ▪p < 0.004, difference between twins whose co-twin was a neonatal survivor and twins whose co-twin was not a neonatal survivor. (B) Mortality by birth weight and pair type (same-sex vs different-sex) for twins whose co-twin died. ▪p < 0.004, difference between twins from same-sex pairs whose co-twin died and twins from different-sex twin pairs whose co-twin died. ○P < 0.05, difference between twins from same-sex pairs whose co-twin died and twins from different-sex twin pairs whose co-twin died.

Full size image

Rates of mortality are also shown for growth discordant twins and are based on twins in pairs in which both twins were live-born (Table 3). Mortality was increased for both twins in the pair, but increased more in the smaller twin (p = 0.005). Not shown in this table, a similar pattern was seen when all twin pairs, including those with a stillborn co-twin, were included.

Rates of Cerebral Palsy in 1-Year Survivors

Demographic factors.

Trends suggested that the crude rates of CP varied by demographic factors in ways mostly similar to the crude rates of fetal and neonatal mortality. These associations did not always reach the level of statistical significance in the smaller twins cohort (Table 2). The crude rate of CP was higher for male infants, nonwhite singleton (but not twin) infants, singletons of both younger and older mothers, and twins of younger mothers (Table 2). Similar to mortality rates, rates of CP adjusted for birth weight and gestational age were lower in nonwhite infants for singletons (OR = 0.83, p = 0.007) and twins (OR = 0.60, p = 0.037). Singletons born to older mothers were at higher risk of CP after adjustment (OR = 1.24, p = 0.014), however twins born to older mothers were not at increased risk (OR = 1.06, p = 0.84).

Twin factors.

Overall, CP was approximately four times as common in twin 1-y survivors than singleton 1-y survivors (Table 2: 0.59%versus 0.14%, p < 0.0001). Twins appeared to be at increased risk relative to singletons above about 38 wk or 2500 g (Fig. 4). The pattern for less mature infants was less clear. It should be noted that because mortality is so high in the 20–23 week gestational age range, most of the 1-y survivors in this gestational age range probably had an incorrect gestational age – particularly so for singletons. For example, at 20 wk gestational age, the average birth weight for the 1-y surviving singletons was an implausibly high 2294 g. The corresponding birth weight for surviving twins was 1163 g.

Figure 4
figure 4

(A,B) Cerebral palsy by birth weight and gestational age for singletons and twins. ▪p < 0.004, difference between singletons and twins. ○p < 0.05, difference between singletons and twins.

Full size image

Crude and adjusted rates of CP were similar for first and second-born twins and same-sex and different-sex twins (Table 3). Rates of CP are shown for twins from growth discordant pairs in which both twins were live-born. In these pairs, surviving growth discordant twins are at increased crude risk of CP and the risk was similar for the smaller and larger twin. After adjustment for birth weight and gestational age in separate logistic models for the smaller twin and larger twin, there was no increased risk of CP for the smaller twin in a discordant pair (OR = 0. 63 [0.35–1.15]) or for the larger twin in a discordant pair (OR = 1.65, [0.95–2.88]) compared with twins from nondiscordant pairs.

Crude rates of CP were much higher in twins whose co-twin died compared with those whose co-twin survived the neonatal period (5.4%versus 0.48%, p < 0.0001). This rate was similar for same-sex (5.7%) and different-sex twins (5.3%) and for twins whose co-twin was stillborn (4.7%) versus twins whose co-twin died in the neonatal period (6.3%). The highest rate of CP (11.8%) was in twins whose co-twin had CP (latter data not shown).

The increase in risk of CP following co-twins death was apparent across the range of birth weight (Fig. 5 A), but unlike the pattern seen for mortality, the excess risk of CP in surviving twins whose co-twin died was not obviously different for same-sex versus different-sex twins by birth weight strata (Fig. 5 B). Compared with 1-y survivors whose co-twin survived, the adjusted odds of CP for twins whose co-twin died was 3.8 [2.3–6.4]. By twin type, the adjusted OR was 4.3 [2.2–8.4] for same-sex twins and 2.4 [1.0–5.8] for different-sex twins, not a statistically significant difference.

Figure 5
figure 5

(A) Cerebral palsy by birth weight and co-twin mortality. ▪p < 0.002, difference between infants whose co-twin was a neonatal survivor and infants whose co-twin was not a neonatal survivor. ○p < 0.05, difference between infants whose co-twin was a neonatal survivor and infants whose co-twin was not a neonatal survivor. (B) Cerebral palsy by birth weight and pair type (same-sex vs different-sex) for twins whose co-twin died.

Full size image

DISCUSSION

In agreement with well-documented trends in the United States, Australia, and elsewhere (2, 3), we found increases in births to older mothers and overall rates of twinning. The increasing prevalence of multiple births has become even more obvious in recent years, partly attributable to delayed childbearing but primarily fueled by the increased use of assisted reproductive techniques. Also consistent with other reports (20, 21), we found that early mortality rates (still birth and neonatal) were lower for nonwhites than whites after adjustment for birth weight and gestational age. This pattern was also observed for adjusted rates of CP in both nonwhite twin and singleton infants.

Overall, twins had a greater risk of early death or cerebral palsy compared with singletons. Singletons were at higher risk of stillbirth than neonatal mortality while twins were at higher risk of neonatal mortality than stillbirth, because very small twins (<1000 g) more often survived birth than very small singletons but were then more likely to die in the neonatal period. The greater frequency of caesarean sections in twins than singletons may play a role in this difference, as lower rates of stillbirth were seen for both singletons and twins delivered via caesarean section, and twins were more often delivered this way.

The risk of death in twins, which was lower than in singleton infants below 2500 g, increased relative to singletons with increasing birth weight or gestational age (Fig. 2). It was at more mature gestational ages that threat of mortality in twins exceeded that in singleton infants, in agreement with results reported elsewhere (4, 8, 22–24). However, even at the most favorable birth weight, twins still had approximately twice the risk of early mortality or CP than singletons at their optimal birth weight. Furthermore, the divergent growth patterns of singletons and twins were evident as early as 20 wk gestational age, in accordance with other reports (25).

Twins from growth discordant pairs were at increased risk of both early mortality and of CP. Twin-twin transfusion syndrome (TTTS), an important cause of growth discordance, occurs in about 15% to 35% of monochorionic twins (26–28) and been estimated to account for 5% to 20% of perinatal mortality in twins (28). Some case series have reported higher morbidity in the heavier preterm (29) but not term (30) twin. Size discordancy is often, but not always, found in TTTS (27). Furthermore, half or fewer of monochorionic twins who are growth discordant have TTTS (27). Other conditions that may cause size discordancy in twins include genetic differences, antenatal infection, and uteroplacental insufficiency (27). Size discordancy is thus a highly imperfect measure of TTTS.

The risk of death of a twin was greater if the co-twin died in utero or shortly after birth. While observed across the whole range of birth weight, the relative increase in risk was greatest among the largest infants due to the low risk of neonatal death in term infants (Fig. 3). Same-sex twin pairs were at greater risk than different-sex twin pairs following the death of a co-twin; this difference was statistically significant only for smaller twins, due to the relatively low mortality of larger infants.

If one twin died, the risk of CP in the survivor was at least 10-fold greater than if the co-twin survived. Death of a co-twin was thus a powerful predictor of CP in twins. Because failure to record a fetal death, especially one occurring early in a multiple gestation, is an inherent problem in vital records, there is probably under ascertainment of multiple gestations complicated by fetal death. The possible contribution of the often undocumented “vanishing twin” to neurologic morbidity in the survivor, who is often registered as a singleton birth, has been a subject of speculation (31). Presumably this would underestimate the contribution of co-twin death to childhood neurologic morbidity (32).

The higher risk of CP in the child whose co-twin died was apparent over the birth weight range examined (Fig. 5). While the excess risk of mortality associated with the death of a co-twin was more pronounced in same-sex than in different-sex twin pairs, this was not the case for CP, perhaps because of the higher rate of death in same-sex pairs presented a competing risk. One other report based on a large population-based registry found the risk of CP in a twin whose co-twin died in utero to be considerably higher in same-sex pairs (10.6%) than in different-sex pairs (2.9%) (33). As the authors of this report pointed out, an anomaly in the way the sex of infants who died in utero was registered requires cautious interpretation of this result. Similar, but nonsignificant results, were reported for rates of CP following the neonatal death of the co-twin in the same population (34) and for rates of CP following fetal or neonatal death of the co-twin in a different population (35).

A number of other investigators have stressed the relatively high risk of cerebral impairment in the surviving child whose co-twin died in utero(4–6, 36–38). The mechanism by which brain injury occurs in a twin whose co-twin died in utero is usually discussed with reference to placental anastomoses or to cord entanglement, and thus considered to be a feature of monozygotic twin pregnancies. The neuropathologic findings in such cases include massive destructive lesions that might be consequences of drastic vascular disruption, with multicystic encephalomalacia, hydranencephaly, and porencephaly. Such vascular events earlier in gestation might produce microgyria and heterotopias. Along with the cerebral lesions, consequences of vascular events in other organs systems – gastrointestinal, urogenital, and other – may be observed. Other routes to adverse outcome in both members of a twin pair also need consideration: MZ twins more often than DZ or singletons infants have malformations or chromosomal aberrations. Noxious events (i.e. coagulation disorders) that kill one twin can sub lethally injure the other, a general mechanism that can involve dizygotic as well as monozygotic pairs. It has not been demonstrated that medical intervention after one twin has died can prevent cerebral catastrophe in the other (39, 40).

This is the largest population-based study of CP to date, affording an opportunity to consider multiple factors possibly associated with this disorder. Inevitably, the statistical power and representativeness afforded by this large sample size comes with the tradeoff that clinical information is less ample and detailed than is possible in smaller clinical studies from referral centers. Two examples of the limitations are zygosity and gestational age: we were limited to using twin-pair gender type as a proxy since information on zygosity was not available. Since approximately 40% of same sex pairs are dizygotic, any risk associated with monozygosity was necessarily diluted. Gestational age as reported in vital records has known limitations (41) and because information was restricted to infants of at least 20 wk gestational age, we were not able to examine risks to a twin following the early fetal death of a co-twin. Furthermore, the populations studied differed in terms of racial composition, medical care, and rates of CP. However, our combined prevalence rates for CP in the five populations surveyed were within the range of other population-based studies using varying case and population definitions (4–6, 23, 42–58).

CONCLUSION

Although twins were at increased risk of early death or CP relative to singletons, small twins generally fared better than small singletons. Co-twin death was a strong predictor of CP in surviving twins. This risk was the same for same- and different-sex pairs, and observed across the birth weight spectrum.