Birth Characteristics, Maternal Reproductive History, and the Risk of Infant Leukemia: A Report from the Children's Oncology Group

Leukemia diagnosed in infants (<1 year of age) differs from that in older children. Most notably, the leukemic cells of ∼70% of infant cases harbor an MLL gene rearrangement compared with only 5% of cases in children ages 1 to 10 years (1, 2). The 5-year survival rate for infant leukemia of ∼45% is considerably lower than that for leukemia at older ages (3). Presence of an MLL rearrangement is an important predictor of survival for infants with acute lymphoblastic leukemia (ALL); in a recent review, event-free survival was <15% and >50% among MLL⁺ and MLL⁻ cases, respectively (4). The prognostic importance of MLL rearrangements in infant acute myelogenous leukemia (AML) is less clear (5).

There is compelling evidence that the MLL rearrangement originates in utero and may be sufficient to cause leukemia (6). In addition, a few initial studies have found that associations of DNA-damaging drugs (7), dietary DNA topoisomerase II inhibitor consumption (8), and NAD(P)H quinone oxidoreductase gene polymorphisms (9-13) with infant leukemia differ according to MLL status. These data suggest that MLL⁺ infant leukemia may have an etiology distinct from that of other leukemias.

Many investigations of childhood leukemia have reported on birth characteristics and maternal reproductive history. A recent meta-analysis of leukemia risk in children of all ages and birth weight included 18 studies and found a significantly increased risk of ALL among high birth weight children (14). Fewer studies have focused specifically on infants. These have reported, with varying consistency, that high birth weight (15-26), higher birth order (15-17, 19, 21, 26-28), and prior fetal loss (15-17, 19, 27, 29) increase the risk of infant leukemia. Maternal age (15-17, 19, 23, 28) and gestational age (16) have not been notably associated with infant leukemia, whereas maternal pre-pregnancy anthropometrics and weight gain during pregnancy have not been examined. The definition of infant leukemia has varied between studies; cases of both ALL and AML, diagnosed at up to 2 years of age, have been included. However, the inclusion of cases >1 year, who are unlikely to have MLL rearrangements, has questionable validity. Here, we present for the first time an analysis that reports results for infant leukemia with respect to the MLL gene rearrangement.

Details of subject enrollment and data collection for this case-control study have been reported elsewhere (8). Any child with leukemia diagnosed at a Children's Oncology Group institution was eligible to be included in a database. We identified cases of infant leukemia diagnosed at <1 year of age between January 1, 1996 and August 20, 2002 in the Children's Oncology Group database and contacted their mothers with physician approval. Controls were identified using a standardized random digit dialing methodology (30). Informed consent was obtained from the mothers of all children enrolled in the study. Children were eligible for the study if they resided in a home with a telephone at the time of diagnosis, if they did not have Down syndrome, and if their biological mother was available for an interview in English. Controls were frequency matched to cases on year of birth. Data were collected through structured, computer-assisted telephone interviews of mothers. Interviews were completed by the mothers of 240 of 348 (69%) of eligible cases diagnosed at 126 participation Children's Oncology Group institutions. We calculated random digit dialing response rates using the method of Slattery et al. (31); the fielding response rate was 67% (7,822 of 11,713 phone numbers successfully screened) and the random digit dialing field rate was 59% (255 of 430 mothers of eligible controls interviewed; ref. 8).

Pathology and cytogenetic reports of cases were also obtained and reviewed by two cytogeneticists as described in our previous publication (8). We considered a case to be MLL⁺ if molecular (Southern blot and PCR), fluorescence in situ hybridization, or cytogenetic methods indicated the rearrangement. When reviewing karyotypes or cytogenetic reports, cases with balanced translocations with 11q23 breakpoints were considered to be MLL⁺. Patients with deletions of 11q or other recurrent aberrations were classified as MLL⁻. Normal chromosomes were considered indeterminate unless molecular testing had been done.

Birth weight (in grams), birth order, gestational age (in weeks), maternal age at birth of child (in years), maternal history of fetal loss, maternal pre-pregnancy body mass index (BMI = weight in kilograms/height in meters squared), and maternal weight gain during pregnancy were the variables of interest. We defined categorical variables for quartiles of birth weight based on the distribution in all cases and controls combined (≤3,203 g, 3,204-3,515 g, 3,516-3,854 g, and ≥3,855 g). For comparison with the literature, we also defined a dichotomous variable for high birth weight using the common cutoff value of 4,000 g (14). We defined categorical variables for birth order (first, second, and third or higher), gestational age (<37 weeks, ≥37 weeks), maternal age (≥35 years, <35 years), and prior fetal loss (any, none) also based on prior literature (32). Pre-pregnancy BMI was classified according to WHO guidelines into normal (BMI, <25), overweight (BMI, 25-29.9), and obese (BMI, ≥30; ref. 33). Overweight and obese were also collapsed into a single category for further analysis. Mothers reported weight gain during pregnancy in pounds and exhibited end-digit preference. The modes of weight gain at 20, 30, and 40 pounds divided the data roughly into quartiles and thus were used as cut points. The four categories of maternal weight gain reported as kilograms were ≤9.07, 9.53 to 13.61, 13.61 to 18.14, and >18.14. All variables were examined in their original form (i.e., as continuous or discrete variables) to assess linear trend. Other covariates were sex, race of the child (White versus non-White), and maternal education (<high school, high school graduate, and college graduate).

We first compared values of the study variables between cases and controls using the t test for two means for continuous variables and the χ² test for discrete variables. We also compared values of the study variables between cases for whom MLL status could be determined and those with indeterminate status. In the main analyses, we compared controls to combined infant leukemia as well as to the case subgroups defined by leukemia type (ALL and AML) and MLL status separately and in combination. Unconditional logistic regression was used to calculate odds ratios (OR) and 95% confidence intervals (95% CI) for combined leukemia, ALL, and AML; polytomous regression was used to calculate ORs and 95% CIs for MLL⁺ and MLL⁻ leukemia. We adjusted for sex, race, and maternal education in all analyses. In addition, we mutually adjusted birth weight, birth order, and gestational age because these factors are interrelated (34). Similarly, we adjusted weight gain during pregnancy for pre-pregnancy BMI (35). All analyses were conducted using SAS version 9.1 (SAS Institute, Cary, NC).

Table 1 describes the distribution of covariates of case and control subjects. Age at birth of child, education, smoking, drinking during pregnancy, or morning sickness did not differ significantly between mothers of cases and controls. Cases were more often male compared with controls, although the difference was not statistically significant. Differences between cases and controls in birth weight and maternal education also were not statistically significant. There was significant dissimilarity of cases and controls on race (79.5% of cases were White versus 85.5% of controls; P = 0.003) and household income (14.4% of cases >$75,000/year versus 24.0% of controls; P = 0.03).

There were 149 cases classified as ALL and 91 as AML. Among cases of ALL, 84 (56.4%) were classified as MLL⁺ (67 by molecular methods), 43 (28.9%) were MLL⁻ (30 by molecular methods), and 22 (14.8%) had indeterminate MLL status. Among cases of AML, 29 (31.9%) were MLL⁺ (11 by molecular methods), 37 (40.7%) were MLL⁻ (9 by molecular methods), and 25 (27.5%) had indeterminate MLL status. Mean birth weight (3,536 g versus 3,541 g; P = 0.95), mean gestational age (39.5 weeks versus 39.1 weeks; P = 0.14), mean birth order (1.82 versus 2.02; P = 0.18), mean number of prior fetal losses (0.36 versus 0.38; P = 0.80), and the percent male (51.8 versus 53.2; P = 0.87) did not differ significantly between cases with determinable MLL status and indeterminate cases. Mean age at delivery, however, was higher among mothers of cases with indeterminate MLL status (30.5 years versus 28.7 years; P = 0.04). Similar results were obtained when comparing cases with and without determinable MLL status by leukemia type (data not shown).

Table 2 presents adjusted ORs for variables of interest by MLL status for combined acute leukemia. Selected results, including those for ALL and AML by MLL status, are presented in the text below with respect to each variable.

Birth weight as a categorical variable was not significantly associated with acute leukemia overall or in any subgroup of cases. However, there was a borderline significant linear trend of increasing birth weight with MLL⁺ acute leukemia (P = 0.06). ORs comparing the second with fourth quartile of birth weight were 1.10 (95% CI, 0.56-2.19), 1.31 (95% CI, 0.66-2.61), and 1.62 (95% CI, 0.82-3.19), respectively, in the MLL⁺ subgroup and 1.42 (95% CI, 0.78-2.57) comparing birth weight of ≥4,000 g with <4,000 g. ORs for MLL⁺ cases by leukemia type were imprecise but suggested that AML(MLL⁺) drove the overall results. ORs comparing the fourth with first quartile of birth weight were 1.45 (95% CI, 0.70-3.02; P for trend = 0.20) for ALL(MLL⁺) and 2.55 (95% CI, 0.61-10.7; P for trend = 0.09) for AML(MLL⁺). We examined interaction between birth weight and sex in ad hoc analyses of MLL⁺ acute leukemia. Interaction was not significant (P = 0.11), but inspection suggested that the association between birth weight and MLL⁺ acute leukemia was limited to females. ORs comparing the second to fourth quartile of birth weight with the first were 1.57 (95% CI, 0.63-3.91), 1.68 (95% CI, 0.62-4.53), and 2.16 (95% CI, 0.82-5.70) among females (P for trend = 0.03), respectively, and 0.57 (95% CI, 0.19-1.74), 0.93 (95% CI, 0.34-2.52), and 1.06 (95% CI, 0.40-2.79) among males (P for trend = 0.63).

Birth order was significantly inversely associated with combined leukemia, MLL⁺ leukemia, ALL, and ALL(MLL⁺), adjusting for birth weight. Comparing children born third or later with firstborn children, the respective ORs were 0.70 (95% CI, 0.43-1.13), 0.52 (95% CI, 0.28-0.99), 0.56 (95% CI, 0.32-0.98), and 0.50 (95% CI, 0.25-1.01). Ps for linear trend were 0.01 for each respective group of cases. There was no association of birth order with MLL⁻ leukemia; Ps for linear trend were 0.18, 0.28, and 0.35 for combined MLL⁻ leukemia, ALL(MLL⁻), and AML(MLL⁻). No significant interaction was detected between birth order and sex in MLL⁺ leukemia in ad hoc analyses (data not shown).

Maternal history of fetal loss was not associated with any group of infant leukemia. With the exception of a significant inverse trend in risk of MLL⁺ acute leukemia (P = 0.04), no significant associations of maternal age with infant leukemia were noted. Gestational age was not significantly associated with any group of infant leukemia. Finally, no observation was observed between infant leukemia and maternal pre-pregnancy BMI or weight gain during pregnancy.

We examined birth characteristics, maternal reproductive history, and risk of infant leukemia with respect to leukemia subtype and the presence of the MLL rearrangement. There was a borderline significant linear trend of birth weight with MLL⁺, but not MLL⁻, infant leukemia, which further seemed by inspection to be confined to females. An inverse association between birth order and infant leukemia, especially MLL⁺ infant leukemia, which was independent of birth weight, was also apparent. By contrast, there was little evidence of associations between infant leukemia and gestational age, maternal age, maternal history of fetal loss, pre-pregnancy BMI, and weight gain during pregnancy with any grouping of infant leukemia.

Several strengths and limitations of this study should be recognized. The study features the largest case series of infant leukemia and, importantly, was able to distinguish between cases with and without MLL gene rearrangements. The 18% of cases for whom MLL status could not be determined did not seem to differ greatly from other cases. However, sample size was small, and power therefore was limited, for some case subgroup analyses. Any apparent differences in association by MLL status must therefore be regarded as tentative.

Although recall bias is often a concern with case-control studies, other investigators have shown that most of the factors examined in this study are recalled by mothers with great reliability and without respect to disease status (15, 36, 37). Maternal anthropometrics are reported less reliably (37, 38), although there is little reason to suspect differential recall. Additionally, the mean intervals between birth and interview were relatively short at 141 weeks (SD, 81) and 194 weeks (SD, 95) for mothers of cases and controls, respectively.

As is common with case-control studies (39-41), our control group exhibited higher socioeconomic status than did cases. This fact and the relatively low rate of participation among controls arouse concern about selection bias. However, selection bias would not explain associations limited to particular leukemia subgroups, such as we observed. Moreover, some characteristics are available from national vital statistics (42), which should accurately reflect the distribution among our theoretical study base (i.e., all children born in the United States and Canada during 1995-2002; ref. 43). Comparison of these figures with our control group may indicate the extent of any selection bias.

Neither the vital status at the time of recruitment nor the presence of MLL translocations, which is a significant prognostic factor for ALL (4), was known for the 108 cases not included in the study. Our results could have been affected by the preferential participation of surviving cases if the factors under study affect prognosis. However, none of the variables investigated here has been associated with survival (4, 5). In addition, because proxy interview was the object of data collection, deceased cases were eligible for the study; 27% of cases were not living at the time of maternal interview. Although these facts suggest a minimal role for survival bias in our results, because of limitations in the data we cannot rule out the possibility. With these study attributes in mind, we compare our results with the body of previous literature below.

Table 3 presents select characteristics, the variables examined, and a summary of results for the 15 studies of birth factors or maternal reproductive history that reported results for infant leukemia (<2 years of age) as a separate entity. It should be noted, however, that previous studies of infant leukemia have not differentiated cases by MLL status and thus do not directly compare with the present results.

Among the birth characteristics, birth weight has been most often studied. In a meta-analysis of 18 studies, Hjalgrim et al. (14) reported a modest but significant association of birth weight ≥4,000 g with childhood ALL (OR, 1.26; 95% CI, 1.17-1.37) as well as a significant linear trend in risk of ALL with rising birth weight (OR, 1.14/1 kg increase in birth weight; 95% CI, 1.08-1.20). High birth weight seemed to raise the risk of AML, but the association was not significant. Although not singled out in the meta-analysis, infant leukemia has been analyzed separately in 12 studies. Eight studies found significant associations of high birth weight with infant leukemia (17-19, 22-26), one reported a suggestive but nonsignificant association (21), and three found no associations (15, 16, 20). The nine studies with significant or suggestive findings (17-19, 21-26) reported ORs ranging from 1.62 to 8.14 comparing high birth weight, variously defined, with the referent category. In our study, the OR comparing the top with bottom quartile of birth weight for MLL⁺ cases was 1.62. The association of MLL⁺ infant leukemia with birth weight ≥4,000 g was elevated but not significant in our data. However, 9.9% of children born in the United States during 1995 to 2002 had birth weights ≥4,000 g (42) compared with 16% of controls and 20.4% of MLL⁺ infant leukemias in this study, suggesting that we may have underestimated the OR. One possible explanation for our findings is that elevated birth weight is associated with childhood leukemia diagnosed at any age, with the association being stronger for MLL⁺ leukemia specifically. There may not have been sufficient MLL⁻ cases to detect a smaller OR in our analysis. We previously have hypothesized that the association between leukemia and high birth weight, especially among infants, is mediated by the insulin-like growth factor-I (44). In light of our results, it would be of interest to know if MLL⁺ lymphoblasts are particularly sensitive to growth stimulation by insulin-like growth factor-I.

Four studies have examined birth order and infant leukemia. Two have suggested an increased risk of infant leukemia with higher birth order (16, 17), whereas two did not (26, 28). By contrast, the present study found a significant inverse association of higher birth order with infant leukemia, which seemed to be driven primarily by ALL(MLL⁺). The percentage of first, second, and third or higher births were 40.6%, 32.5%, and 26.9%, respectively, among children born in the United States during 1995 to 2002 (42). First births were underrepresented and second births were overrepresented in the control series of this study compared with national vital statistics data; 33.7%, 43.5%, and 22.7% of controls were first, second, and third or later-born children, respectively. The excess of second children among controls is not easily explained by selection bias because parity decreases with increasing socioeconomic status (45). These observations notwithstanding, the percentage of MLL⁺ cases that were third or later born (17.7%) were substantially less than is found in the general population.

Many studies have examined birth order and leukemia diagnosed past infancy (46). The results, although inconsistent, have been interpreted with respect to infectious hypotheses of childhood leukemia etiology, as the presence of older siblings is thought to increase the likelihood of postnatal exposure to infectious agents. Postnatal factors, however, are likely not involved in causing infant leukemia. Rather, these findings may signify that prenatal correlates of birth order are important in the etiology of MLL⁺ infant leukemia. Insulin-like growth factor-I could not, however, explain these findings because levels increase in later pregnancies (47). However, maternal estrogen levels are known to decline as parity increases (48, 49). High birth weight and female pregnancies have also been associated with higher maternal estrogen levels (48, 49). Although we did not observe an association with sex per se, and did not detect significant interaction, the association of high birth weight with MLL⁺ leukemia seemed confined to females. Thus, several observations in this study seem consistent with a role for endogenous estrogen exposure in MLL leukemogenesis.

Yeazel et al. (29) found a strikingly large and significant increase in risk of infant leukemia in the offspring of mothers with prior fetal loss. Three other studies offered evidence of a more modest increases in risk (17, 19, 27), whereas two others found null associations (15, 16). Although study design might explain these discrepancies, the registry-based studies by Cnattingius et al. (19) and Reynolds et al. (16), which would not be susceptible to recall or selection biases, presented conflicting results. The null results of the current study do little to clarify the literature about fetal loss and infant leukemia.

Only one other study has reported on gestational age and infant leukemia. Reynolds et al. (16), as in the current study, found no association. Similarly, none of the six studies to have examined maternal age and infant leukemia has found significant associations (15-17, 19, 23, 28).

No studies, to our knowledge, have examined pre-pregnancy BMI and weight gain during pregnancy in relation to infant leukemia. However, it was recently reported that risk of ALL, particularly that diagnosed at 0 to 4 years of age, was increased in mothers with high pre-pregnancy weight and high weight gain during pregnancy; interaction between maternal pre-pregnancy weight and birth weight of the child was also noted (50). This study did not reveal like findings for infant leukemia.

The literature about infant leukemia and birth characteristics is small and has not, to date, included analyses stratified by MLL gene rearrangement status. We observed what seemed to be independent, opposing associations of birth weight and birth order with MLL⁺, but not MLL⁻, infant leukemia. However, that these associations varied by MLL status cannot be stated firmly because confidence intervals overlapped. Maternal history of fetal loss and other birth characteristics were not related to infant leukemia. Because of the lack of comparable data, these results are most useful for guiding future studies. This investigation also further supports the contention that molecularly defined subtypes of infant leukemia have separate etiologies (7, 8, 12).