Cesarean Delivery and Risk of Infant Leukemia: A Report from the Children's Oncology Group

This article is featured in Highlights of This Issue, p. 359

Leukemia that occurs in infancy (<12 months of age) is a rare disease, and its etiology is poorly understood. In contrast with leukemia diagnosed in older children, infants with leukemia have a nearly equal distribution of acute lymphoblastic (ALL) to acute myelogenous leukemia (AML) diagnosis (1) and have a high frequency of somatic MLL gene rearrangements, present in 50% to 80% of infant ALL and 34% to 50% of infant AML (2–4). There is strong evidence that infant leukemia is initiated in utero (5) and events during pregnancy and childbirth may impact disease development.

In recent years, several studies have suggested that birth by cesarean delivery (CD) affects both short- and long-term outcomes for the neonate (6), including development of the immune system (7). CD may impede normal immune development through altered colonization of the microbiome or lack of a stress response during labor and delivery, and these effects may differ according to whether CD occurred before the onset of labor (7).

Meta-analyses have reported associations between birth by CD and subsequent risk of immune-related disorders, including asthma (8) and type 1 diabetes mellitus (9). Furthermore, several recent studies have reported an association between CD and childhood leukemia overall (10–14). One of these, a pooled analysis (10) from the Childhood Leukemia International Consortium (15), reported an association between pre-labor CD (PLCD) and ALL among children age 0 to 14 years [OR, 1.23; 95% confidence interval (CI), 1.04–1.47]. Within analyses stratified by age at diagnosis, a strong estimated effect between PLCD and infant leukemia was observed, although this result was based on very few exposed cases and controls and did not achieve statistical significance (OR, 2.62 and 95% CI, 0.96–7.19). An association between CD and infant leukemia has not been established, and many previous studies have not examined infants separately or may be underpowered to detect a modest association.

To examine this potential association more closely, we carried out an analysis of CD and infant leukemia using data from a population-based case–control study.

This analysis was carried out using data from the Children's Oncology Group (COG) AE24 study. This study was conducted in accordance with the U.S. Common Rule. Institutional review boards at the University of Minnesota and each case's COG institution approved the study protocol. We obtained written informed consent from participating mothers. Infants diagnosed with acute leukemia, including both ALL and AML, at less than 1 year of age were identified at COG institutions in two phases: between 1 January 1996 and 13 October 2002 (Phase I) and between 1 January 2003 and 31 December 2006 (Phase II). Cases were eligible if they did not have Down syndrome, if their biological mother was available for a telephone interview in English or Spanish, and if they were diagnosed and treated at a participating COG institution in the United States or Canada.

During Phase I, controls were selected from the United States and Canada using a random digit dialing procedure and were frequency matched to cases by year of birth. Phase II controls were randomly selected from 15 U.S. state birth registries and were frequency matched to cases on year of birth and region of residence based on the geographical and annual birth distribution of Phase I cases. Controls were eligible if they did not have Down syndrome and had a biological mother who was available for a telephone interview in English or Spanish.

Mothers of cases and controls completed a structured, computer-assisted telephone interview. Interview questions included items on demographics, reproductive history, family history of disease, and exposures during pregnancy with the index child. Pathology and cytogenetic reports from diagnosis were acquired for cases, and MLL gene rearrangement status (MLL⁺, MLL⁻, indeterminate) was ascertained through central review by three independent reviewers who specialize in MLL rearrangements and cytogenetics.

Data on labor and delivery, including birth characteristics, mode of birth, and, where applicable, indication for CD were collected in the interview. For Phase I cases and controls, we also requested consent to obtain all medical records beginning 6 months before conception of the index child until 1 week after the birth or discharge from hospital (whichever occurred later). Records were requested from all providers seen during pregnancy and delivery. Data were abstracted by trained nurse abstractors and included indication for CD, presence/absence of labor, labor onset time, and delivery time.

From interview data, we classified CD as pre-labor (PLCD), emergency (ECD), or indeterminate based on the indication provided. We considered the following indications as likely to have resulted in a scheduled pre-labor delivery: Baby was due or overdue; placenta previa; congestive heart failure in mother; breech presentation; multiple gestation; multiple gestation with malpresentation of one fetus or more; heart-shaped uterus; previous CD; mother developed illness; and issues of convenience. We considered the following indications as likely to have resulted in an emergency delivery: labor was slow or stopped, long labor, or did not dilate; baby was sick or in danger, fetal distress, intrauterine growth retardation, placenta not viable; baby's heart rate dropped; cord around baby's neck; cord complications (cord prolapse, cord wrapped around wrists or ankles); and cephalopelvic disproportion. All other indications were considered indeterminate.

Within the subset of subjects for whom we had medical record data available, we used presence or absence of labor, as recorded in the labor/delivery medical record, to determine whether a CD was pre-labor or emergency. If a CD occurred after the onset of labor, it was considered an ECD. Otherwise, if the CD occurred before onset of labor, we classified it as PLCD. Finally, for all mothers who experienced some labor before the birth, whether the ultimate delivery method was vaginal or cesarean, we used data on labor onset time and delivery time abstracted from medical records to calculate duration of labor. We created a categorical variable for length of labor (0 to <3 hours, 3 to <6 hours, 6 to <10 hours, ≥10 hours).

We used unconditional logistic regression (SAS version 9.4, SAS Institute) to estimate odds ratios (ORs) and 95% CIs for the association of infant leukemia with CD due to all indications, PLCD, and ECD. We first evaluated PLCD and ECD as classified according to interview data only, then separately examined the associations for PLCD and ECD with infant leukemia among the Phase I participants with medical record data available. We did not collect medical records from Phase II participants and therefore these participants are excluded from all analyses based on medical record data. Covariates selected a priori for inclusion based on established associations with childhood cancers included infant sex, birth weight, gestational age, plurality, maternal age, maternal education, maternal race, household income, maternal smoking, and maternal alcohol consumption. Variables were retained in the final multivariate models if they changed the ln(OR) estimate substantially (e.g., ≥10%). Final models included maternal race (black, white, Hispanic, other), birth weight as a continuous variable, and the matching variable, year of birth. In addition, we stratified by leukemia subtype (ALL and AML), and by MLL status (MLL⁺, MLL⁻, indeterminate). Finally, we estimated the impact of length of labor on risk of infant leukemia overall, ALL, and AML.

For the subset of participants with both interview and medical records data, we estimated sensitivity and specificity of our classification of PLCD and ECD from interview data using medical record data as the gold standard.

Infants diagnosed with leukemia at very young ages may have compromised health at birth, which may confer an increased risk of fetal distress and other indications for birth by CD. To assess the possibility of reverse causation in our data, we conducted sensitivity analyses excluding infants diagnosed at <3 months of age.

Interviews were completed for 443 (64%) eligible cases (264 ALL, 172 AML, 7 other) and 324 (47%) eligible controls. Participation rates and reasons for nonparticipation have been published elsewhere (16–18). Sex, birth weight, gestational age, and plurality were all similar between cases and controls. Mothers of cases were more likely to have completed less than a high school education (33.7% of case mothers compared to 28.2% of control mothers). Mothers of cases were also more likely to be Hispanic and less likely to be Caucasian than mothers of controls (Table 1).

Cesarean delivery due to any indication was associated with an elevated estimated effect for risk of infant ALL (OR, 1.52 and 95% CI, 1.02–2.25). We did not note an association between CD and infant AML (OR, 1.02 and 95% CI, 0.64–1.62). When we classified PLCD and ECD using interview data, birth by ECD was associated with a 2-fold increased risk of infant ALL. The point estimate for the association between PLCD and infant ALL was elevated, although not statistically significant. We did not observe an association between either PLCD or ECD and infant AML (Table 2).

Using data from medical records to classify type of CD, both PLCD and ECD were associated with an increased risk of infant ALL (OR, 2.04 and 95% CI for PLCD 1.00–4.15; OR, 1.80 and 95% CI for ECD, 0.93–3.51). Neither PLCD nor ECD was associated with infant AML (Table 2).

When we estimated the association between length of labor and infant leukemia, we did not find statistically significant associations between precipitous labor (<3 hours) or long labor (>10 hours) and infant leukemia overall, ALL, or AML (Table 3).

Analyses stratified by MLL rearrangement status revealed no association between CD and MLL⁻ leukemia. ORs for the association between CD and MLL⁺ leukemia were slightly elevated but not statistically significant (Table 4).

When we assessed the accuracy of our categorization of PLCD and ECD from interview data, using medical record data as the gold standard, we found that the sensitivity and specificity for PLCD were 81% (95% CI, 64%–98%) and 89% (95% CI, 77%–100%), respectively. The sensitivity and specificity for ECD were 86% (95% CI, 73%–99%) and 85% (95% CI, 69%–100%), respectively.

We conducted sensitivity analyses excluding infants diagnosed at <3 months of age because these infants may have compromised health at birth, which may confer an increased risk of fetal distress and other indications for birth by CD. These analyses did not alter our results for the associations between infant leukemia and CD, PLCD, or ECD, and the associations between CD and infant ALL remained statistically significant (Supplementary Table S1).

We observed an elevated risk of infant leukemia, specifically infant ALL, following CD. We did not observe a difference in risk according to whether the delivery was PLCD or ECD.

Although there is a growing body of literature on health outcomes following birth by CD, including childhood leukemia, few previous studies have examined the impact of CD on infant leukemia specifically. A large pooled analysis reported an elevated risk estimate for infant leukemia following PLCD (OR, 2.62 and 95% CI, 0.96–7.19); however, this result was based on few exposed cases and controls (10). A registry-based study of CD and childhood leukemia did not observe an association between leukemia diagnosed at age <2 years and CD overall; however, among the subset of cases and controls with data on whether a birth was PLCD or ECD, the authors reported a modestly increased effect estimate for risk of leukemia diagnosed at age <2 years after both PLCD and ECD (11). Our findings suggest that birth by CD overall is associated with elevated risk of infant leukemia, although we did not observe a difference in our results according to whether the delivery was PLCD or ECD. This is in contrast with previous studies of leukemia in older children that suggest birth by PLCD, but not ECD, is associated with increased risk of leukemia.

Two distinct mechanisms have been proposed to explain the association between birth by CD and adverse health outcomes. The first involves the stress response which labor and delivery elicit in the fetus. Concentrations of both catecholamine and cortisol are increased by a factor of 1.5–3x in neonates born by vaginal delivery compared with those born by CD before the onset of labor (19, 20). In contrast, neonates born by emergency CD exhibit postpartum cortisol levels similar to those observed in neonates born by vaginal delivery (21). Increased cortisol levels in the neonate activate the hypothalamic–pituitary–adrenal (HPA) axis, which has a negative feedback relationship with immune and inflammatory reactions (22). The role of the HPA axis and increased cortisol levels in reducing risk of ALL was previously hypothesized by Schmiegelow and colleagues (23) as part of the adrenal hypothesis of childhood leukemia etiology. This hypothesis suggests that increased cortisol levels in early life mimic the anti-leukemic effects of glucocorticosteroids, which are used in treatment of childhood leukemia, including in infants. Furthermore, increased cortisol may suppress the Th1-mediated pro-inflammatory response to infections by promoting production of anti-inflammatory Th2 cytokines, including IL-4 and IL-10 (23). This impact on the Th1/Th2 balance may decrease the proliferative stress on extant preleukemic cells. Therefore, it is possible that exposure to elevated cortisol levels during labor and delivery plays a role in mitigating ALL risk for infants harboring preleukemic cells that have arisen in utero. Children born by vaginal delivery and ECD are generally exposed to comparable cortisol levels during labor and delivery, whereas children born by PLCD exhibit significantly reduced cortisol levels at birth (20, 21). We did not observe differences in our results based on whether births were PLCD or ECD. The chemosensitivity of infant leukemia cells to corticosteroids, and therefore cortisol, may depend on subtype. In vitro studies have demonstrated that MLL⁺ ALL cells have enhanced chemoresistance to corticosteroids, whereas MLL⁺ AML cells do not demonstrate a chemoresistant phenotype (24–26). We did not have sufficient power to examine MLL⁺ ALL and AML separately, and our results stratified by MLL rearrangement status suggested that infants born by CD may have elevated risk of MLL⁺ leukemia, but this analysis was underpowered to achieve statistical significance. We examined duration of labor as another potential intrapartum exposure relevant to infant leukemia risk. We did not observe statistically significant associations between length of labor and infant ALL or AML. Furthermore, previous studies have indicated that duration of labor may not correlate with umbilical cord blood cortisol concentrations (20, 21).

A second potential mechanism is differential microbiota colonization following birth by CD as compared to vaginal delivery. Growing evidence suggests a significant role of the gut microbiome broadly in human health, and immune system development, in particular (27). Studies of germ-free mice show compromised development of the mucosal immune system and reduced numbers of both IgA-producing plasma cells and IgG in germ-free mice compared with animals of the same strain that are free of only specific pathogens (28, 29). These mice also exhibit abnormalities of the spleen and lymph nodes, including modified structure and poorly formed B- and T-cell zones (30). Intestinal microbiota influence early postnatal immune development through interactions with intestinal Toll-like receptors and production of suppressive cytokines, transforming growth factor-β (TGF-β), and IL-10, which play a critical role in producing a balanced Th1 and Th2 immune response (31, 32). Microbiota colonization occurs during the first moments of life, and mode of birth has been shown to alter both composition (33) and diversity (34) of intestinal microbiota in humans. These differences persist through the first 6 to 12 months of life (35), a critical period for immune development. Furthermore, studies have suggested that differential microbiota colonization may impact risk of autoimmune disorders (36), chronic diseases (37), infection (38), and many types of adult cancer (39).

In addition, CD may alter constitution of the microbiome through altered breastfeeding practices. Infants born by vaginal delivery are breastfed earlier and are more likely to be breastfed than those born by CD (40). Breastmilk contains diverse microbes from the mother's gut and plays a crucial role in early microbiota colonization (41). We did not adjust for breastfeeding in these analyses since establishment of breastfeeding may be dependent on the overall health of the infant (42), which may be compromised in infants diagnosed with leukemia shortly after birth. Thus, in this population, an inverse association between breastfeeding and infant leukemia may be present due to reverse causation.

Incidence rates of CD have increased sharply over the last several decades, both in the United States and worldwide (43). The World Health Organization recommends that no more than 15% of births should occur by CD (44); however, most developed regions have CD rates well above that number, with some as high as 40% (45). The risks of CD without medical indication have been well documented and include both short- and long-term effects on the offspring. These include impaired lung function, altered metabolism and blood pressure during infancy, as well as risk of obesity and both hepatic- and immune-related conditions during childhood and adulthood (6). Rates of infant leukemia have also risen during this time, and the causes of this increase are unknown.

There are some limitations to this study. Because this study is case–control in design, it is possible that the control group is not representative of the source population of cases with respect to exposure distribution. Given that this study relied on active participation, we had a 69% participation rate among cases and 67% among controls (18). It is possible that selection bias is present in our dataset. However, the cesarean delivery rates noted in controls show the expected frequency and trend based on national US rates for the birth years represented. We also cannot preclude the possibility that the observed associations are due to confounding by indication or other unmeasured confounding factors. It is possible that some maternal or fetal pathology that increases risk of CD also predisposes the infant to leukemia. We did not have sufficient statistical power to assess the possibility of confounding by indication. However, we observed an association among ALL cases only and it is unlikely that any confounding factors would differ by type of leukemia. Within analyses based on questionnaire data, we categorized PLCD and ECD based on the indication for CD, as recorded in the interview. It is possible that some of these births were misclassified. However, when we assessed the accuracy of our categorization of PLCD and ECD from interview data, using medical record data as the gold standard, we found very good sensitivity and specificity for both PLCD and ECD classifications. Furthermore, we expect that any misclassification would be nondifferential and therefore would bias our results toward the null. We have offered plausible biological mechanisms that could explain the association if it is indeed causal. This study is the largest population-based case–control study conducted on infant leukemia to date. Additional strengths of this analysis include the availability of detailed medical record data for a portion of our participants to confirm PLCD and ECD classification and assess duration of labor. Furthermore, we expect that our main exposure variable, mode of birth, is recalled with nearly 100% accuracy, as supported by a validation of self-report versus medical record in our study population (46).

We observed an association between CD and risk of infant ALL. Future larger studies with detailed information on molecular subtype, including MLL rearrangement status, and indication for CD will be important to further examine this association with enhanced statistical power. Because our results were consistent across classifications of PLCD or ECD, this suggests that altered microbiota colonization may play a role in development of leukemia in infants, but a clear biological mechanism has yet to be determined.

No potential conflicts of interest were disclosed.

Conception and design: E.L. Marcotte, L.G. Spector

Development of methodology: E.L. Marcotte, L.G. Spector

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.A. Roesler

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.L. Marcotte, M.R. Richardson, L.G. Spector

Writing, review, and/or revision of the manuscript: E.L. Marcotte, M.R. Richardson, M.A. Roesler, L.G. Spector

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.G. Spector

Study supervision: E.L. Marcotte, L.G. Spector

This work was supported by the Children's Oncology Group and the National Institutes of Health Grants R01 CA79940 (J.A. Ross), U10 CA13539 (W.A. Bleyer), U10 CA98543 (P.C. Adamson), U10 CA180886 (P.C. Adamson) and the Children's Cancer Research Fund (J.A. Ross), Minneapolis, Minnesota.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.