Epidemiology of childhood acute myeloid leukemia


  • Susan E. Puumala PhD,

    Corresponding author
    1. Center for Health Outcomes and Prevention Research, Sanford Research, Sioux Falls, South Dakota
    2. Department of Pediatrics, Sanford School of Medicine, University of South Dakota, Sioux Falls, South Dakota
    • Sanford Research Center, 2301 E 60th Street North, Sioux Falls, SD 57104.
    Search for more papers by this author
  • Julie A. Ross PhD,

    1. Division of Pediatric Epidemiology and Clinical Research, Department of Pediatrics, University of Minnesota, Minneapolis, Minnesota
    2. University of Minnesota Masonic Cancer Center, Minneapolis, Minnesota
    Search for more papers by this author
  • Richard Aplenc MD, PhD,

    1. Children's Hospital of Philadelphia, Philadelphia, Pennsylvania
    Search for more papers by this author
  • Logan G. Spector PhD

    1. Division of Pediatric Epidemiology and Clinical Research, Department of Pediatrics, University of Minnesota, Minneapolis, Minnesota
    2. University of Minnesota Masonic Cancer Center, Minneapolis, Minnesota
    Search for more papers by this author

  • Conflict of interest: Nothing to declare.


Although leukemia is the most common childhood cancer diagnosis, the subtype, acute myeloid leukemia (AML), is less common and fewer etiologic studies exist. This review summarizes the major risk factors for AML. We searched the literature using PubMed for articles on childhood AML and reviewed 180 articles. While few risk factors are definitive, we identified several with consistent evidence of a possible effect. Thorough analysis of genetic and epigenetic factors is missing from this literature and methodological issues are unresolved. Future studies should more closely examine causal mechanisms, improve exposure measurement, and include analysis using genetic and epigenetic factors. Pediatr Blood Cancer 2013; 60: 728–733. © 2013 Wiley Periodicals, Inc.


Leukemia is the most common of pediatric cancers accounting for about 30% of diagnoses 1. There are two main subtypes; acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML). AML is less common, accounting for approximately 18% of childhood leukemia diagnoses 1. The etiology of the two subtypes is likely quite different based on both cell lineage and epidemiological studies of incidence and risk factors. Few risk factors have been conclusively determined for childhood AML.

AML develops through a transformation of hematopoietic progenitor cells that leads to a block in differentiation, increasing the number of progenitor cells and decreasing the number of mature blood cells 2. Evidence for multiple mutations required for AML development has been demonstrated from backtracking studies that have found mutations associated with childhood AML present at birth, yet development of leukemia takes place sometimes years later 3.

Many studies have included an analysis of risk factors for childhood AML, but few have definitively emerged. Only in utero exposure to ionizing radiation is considered an established cause of pediatric de novo AML 4. Evidence about risk factors for AML is limited due to small sample sizes and improper grouping of AML with ALL. This review synthesizes what is known about childhood AML, identifies gaps in current knowledge and suggests directions for future studies.

Descriptive Epidemiology

The incidence of childhood AML in the United States was estimated at 7.7 cases per million children aged 0–14 in 2005–2009 with some indication of an increase in incidence over time 5. Incidence peaks in infants less than 1 year of age with a rate of 18.4 per million, declining to 4.3 per million for ages 5–9 years and increasing up to 7.7 per million for children ages 10–14 years (Fig. 1) 5. Little variation is seen by racial/ethnic groups in the US, with the exception of a possible increased rate in Hawaiians 6 and potentially higher incidence of a specific subtype of AML, acute promyelocytic leukemia (APL), in Hispanic/Latino children 7. In a report using data from the Surveillance, Epidemiology, and End Results (SEER) Program, Asian and Pacific Islanders had the highest rate of childhood AML (8.4 per million), followed by Hispanics (8.1 per million), Caucasians (7.5), and African Americans (6.6) 1.

Figure 1.

Incidence rates for AML by age at diagnosis, U.S. SEER 2005–2009: adapted from data presented in Howlader et al. 5.

Worldwide incidence of childhood AML varies with annual standardized incidence rates ranging from 2 per million in Kuwait to 14.4 per million for the Maori in New Zealand 6. Differences in rates could either be due to differences in incidence or differences in ascertainment methods between registries. Countries with available data observed rates between 5 and 8 per million. Increased incidence in Maori, Hawaiians, and Pacific Islanders in general could suggest shared genetic predisposition.


The 5 year survival rate for children <15 years of age at diagnosis was estimated at 64.3% in the US (2002–2008) 5. AML subtypes have very different prognoses with 5 year survival expectations ranging from 22% to 90% 2. Higher survival rates are seen in children with APL (due to the sensitivity to all-trans-retinoic acid and arsenic trioxide) and in children with other specific mutations 2. Lower survival is seen in children with mutations such as FLT3-ITD, monosomy 7, del-5q, and poor disease response.


We searched Medline to identify articles that studied childhood AML etiology. The search string was (“childhood leukemia” or “pediatric leukemia” or “childhood leukaemia” or “paediatric leukaemia”) AND (etiology OR aetiology OR risk factors OR epidemiology OR case–control), yielding 1,983 articles (through 6/30/2012). Articles were considered if they reported etiologic analysis results for childhood AML separately from childhood ALL or all leukemias. References of selected articles were also explored for additional studies. Overall, 180 articles were identified that performed etiologic analysis of childhood AML or acute non-lymphoblastic leukemia either in an original study or a meta-analysis. Only selected risk factors are presented here. If meta-analysis was available for particular exposure, only those results were reported.


Genetic Factors

The most common genetic factor for development of AML is trisomy 21. Children with Down syndrome (DS) have an increased risk of childhood AML and a 500-fold increased risk of developing a specific subtype of AML, acute megakaryocytic leukemia 8. A small proportion of childhood AML cases are associated with other genetic syndromes such as Fanconi anemia, Bloom syndrome, ataxia telangiectasia, Shwachman–Diamond syndrome, and familial monosomy 7 9.

Two studies found a significant association between family history of hematologic cancers in first or second degree relatives and childhood AML 10, 11. An association between any cancer in first or second degree relatives and childhood AML was found in one study, but not replicated in others 11–13.

The majority of genetic studies on childhood AML involve examining variation in single nucleotide polymorphisms (SNPs). No genome-wide association studies have been performed for predisposition to childhood AML. Previous studies have been based on candidate genes, typically examining genes involved with xenobiotic metabolism or folate metabolism. While some positive results have been found, additional confirmation is needed.

The glutathione S-transferase (GST) genes are involved in the metabolism of various carcinogens in tobacco smoke including benzene. Two genes have been examined in childhood AML (GSTT1 and GSTM1). Null genotypes (no enzyme activity in homozygotes) in these two genes are quite prevalent with 50% of non-Hispanic whites null for GSTM1 and 20% null for GSTT1 14. While studies involving the GSTT1 null genotype found no significant association, the GSTM1 null genotype has been associated with a small increased risk 15–18. The NAD(P)H:quinone oxidoreductase 1 (NQO1) gene plays a role in preventing toxicity due to benzene exposure and a SNP (rs1800566) results in complete reduction in enzymatic activity for homozygotes 19. In a meta-analysis of three original studies of childhood AML, no statistically significant association was observed 20. Many genes in the cytochrome P450 (CYP) family also contribute to detoxification, and CYP1A1, CYP2D6, and CYP2E1 have been investigated in relation to childhood AML. In three studies of a SNP (rs4646903) in CYP1A1, no association was found 15–17. One study found a possible association between a SNP (rs2031920) in CYP2E1 (OR = 4.9; 95% CI 1.6, 15.2, for heterozygotes vs. homozygous wild type with no cases or controls homozygous for the variant allele), while the association with SNPs in CYP2D6 (rs35742686 and rs3892097) was not significant 15.

Four studies have explored the relationship between SNPs (rs1801133 and rs1801131) in the methylenetetrahydrofolate reductase (MTHFR) gene and childhood leukemia; three finding no association 16, 21, 22 and one finding possible association with rs1901133 (OR = 2.2; 95% CI 1.0, 4.8) 23. One examined other genes in this pathway along with maternal genotype. In affected children, mutations in the 5-methyltetrahydrofolate-homocysteine methyltransferase (MTR) gene (rs1805087; OR = 2.74; 95% CI 1.07, 7.01) and the thymidylate synthetase (TS) gene (1494 del 6; OR = 2.04; 95% CI 1.03, 4.03) were both significantly associated with increased risk of childhood AML 21, while no associations were observed with maternal polymorphisms in these genes.

Parental Factors

Parental age

Differing classification, use of continuous variables, and correlation between maternal and paternal age make studies of parental age challenging to synthesize, but overall the data supports an increased risk with older maternal age. Three studies found a statistically significant association with advanced maternal age, one being a very large pooled study with 804 cases that included data from the other two (OR = 1.08; 95% CI 1.01, 1.16, for a 5 year increase in age) 24–26. An additional study found a significant effect of maternal age in children with DS and AML for women aged ≥35 years compared with women <30 years old (OR = 2.63; 95% CI 1.26, 5.49) 27. Others found positive but non-significant associations 28–36, a negative but non-significant associations 37–39, and null associations 40. When examining large studies with over 200 cases of childhood AML, almost all found a positive association 24, 28, 30, 33 (one did not report the direction of the effect 41), but only in the large pooled study was this association statistically significant 24.

Only one study found a significant relationship between paternal age and childhood AML, but this finding was only significant when not adjusted for maternal age 24. Other studies found non-significant positive associations with older paternal age 28, 30, 32, 33, 36, a non-significant inverse associations with older paternal age 37–39, or an unspecified non-significant difference 41. For studies with over 200 cases, results were mixed as well. Overall, the evidence is inconclusive for an association between older paternal age and childhood AML.

Prior fetal loss

Overall, there appears to be an increased risk of AML with prior fetal loss. However, most studies examining prior fetal loss were somewhat small, including less than 100 cases. Four studies indicated significant associations between prior fetal loss and childhood AML, including the largest study with over 200 cases of childhood AML 32, 37, 42, 43. Within those studies, one used number of miscarriages and found an increase risk for two or more compared to none (OR = 11.3; 95% CI 2.3, 56) 42, two looked at any fetal loss compared to none (OR = 2.7; 95% CI 1.0, 6.8; OR = 2.0; 95% CI 1.3, 2.9) 37, 43 and one split early and late fetal loss (<20 weeks, >20 weeks) and found an association only with early fetal loss (OR = 2.3; 95% CI 1.0, 5.2; 2 or more vs. none) 32. In contrast, one study found a significant protective effect of prior fetal loss (OR = 0.6; 95% CI 0.4, 1.0; any vs. none) 35. Among other studies with non-significant results, two were positive 29, 44, three were negative 33, 38, 40, and the rest were null or had mixed findings 25, 27, 31, 45.

Birth order

Some evidence of an increased risk of childhood AML with increasing birth order exists, although this could be due in part to a maternal age effect. Three studies found significant associations and had sample sizes of over 100 cases in each study 33, 34, 44. One found a significant linear trend for increasing birth order in infants (P = 0.04) 44, another reported a significant association for three or more previous live births compared to one or two (OR = 1.9; 95% CI 1.0, 3.3) 33, and the third found an association for three or more previous live births compared to none (OR = 1.6; 95% CI 1.0, 2.8) 34. Studies with non-significant associations have suggested an increased risk with increasing birth order or prior live births 25, 28, 38–40, 46, but others found inconsistent results or possible negative associations 27, 29, 30, 36, 37, 41, 47, 48.

Periconceptional and Prenatal Exposures


Meta-analysis results support a positive association for any alcohol during pregnancy with a combined OR of 1.56 (95% CI 1.13, 2.15) overall and 2.68 (95% CI 1.85, 3.89) for those diagnosed with AML under 5 years of age 49. Analysis of specific beverages singled out wine (OR = 1.67; 95% CI 1.21, 2.32) 49. Several studies have attempted to quantify the number of drinks, but dose response patterns have been inconsistent between studies.


A meta-analysis found no association between maternal smoking and childhood AML (OR = 0.99; 95% CI 0.90, 1.09) 50. The only prospective study of maternal smoking and childhood leukemia did not find a significant effect overall, but did find an increased risk for women smoking 10 or more cigarettes per days compared to non-smokers (OR = 2.28; 95% CI 1.05, 4.94) 51. Most studies examining the risk between paternal smoking and childhood AML have been null 35, 52–63, although a few studies have indicated a possible increase in risk 59, 62, 63. Differing time periods and smoking patterns are likely important, but have not been well studied.

Dietary DNA topoisomerase II inhibitors

Two studies have shown a possible association between maternal dietary intake of DNA topoisomerase II inhibitors and infant AML, although both were relatively small studies with less than 100 cases. A 10-fold increased risk of AML in infants whose mothers consumed medium or high levels of DNA topoisomerase II inhibitors (beans, fresh vegetables, canned vegetables, fruit, soy, regular coffee, black tea, green tea, cocoa, and wine) 64 was reported in an initial test of this hypothesis. The larger second study examined the relationship within the subgroup of infants with an MLL translocation, and again found some indication of a higher risk for increasing levels in the AML MLL+ subgroup (OR = 3.2; 95% CI 0.9, 11.9, highest vs. lowest quartile) 65.


Antibiotic use during pregnancy has been examined either as a group or by specific drugs. One study in infants found that maternal use of metronidazole, used to treat vaginal infections, was associated with a possible increased risk (OR = 4.66; 95% CI 0.89, 24.24) 66, while another study in infants found no association with maternal use of any specific antibiotic during pregnancy, although a positive, non-significant association was observed for metronidazole 67. Other studies have been mixed with one finding a significant association for any antibiotic use (OR = 3.20; 95% CI 1.72, 5.96) 68 and two others finding no relationship 69, 70. All studies were similar in size with 70–100 cases of AML and most used maternal interviews or questionnaires.

While it is possible that the underlying condition may be the “true” etiologic risk factor rather than the use of antibiotics, four studies found no significant association between maternal infections and childhood AML for urinary tract or lower genital tract infections 47, 71–73, although both studies based on medical records found a non-significant elevated OR of 4.0 (95% CI 0.85, 18.8) 73 and 1.34 (95% CI 0.32, 5.56) 47 for risk of childhood AML 73. An additional study found no association between viral infections during pregnancy and childhood AML 70. Current data are limited, but could suggest some relationship with either metronidazole use or urinary tract or lower genital tract infections.


Evidence of an association with benzene has been mixed likely due to the heterogeneity of sources and issues of measurement. Sources of benzene include cigarette smoke, car exhaust, industrial emissions as well as from building materials, paints and adhesives. Sources of parental benzene exposure can be divided into direct exposure (e.g., occupational or household product use) and inferred exposure (e.g., occupation title or traffic density). Only one study indicated an increased risk for direct paternal occupational exposure to petroleum products (OR = 2.4; 95% CI 1.3, 4.1 for over 1,000 days of exposure vs. none) and solvents (OR = 2.0; 95% CI 1.2, 3.8, over 1,000 days of exposure vs. none), the relationship was similar for time periods before and during the index pregnancy, but was reduced for exposures after the index pregnancy 74. Others have found no association with paternal occupational or household exposure to benzene 38, 75. One study found an association for direct maternal occupational exposure to benzene or gasoline (OR = 4.0; 95% CI 1.8, 8.3 and OR = 2.1; 95% CI 1.1, 4.3, respectively) 38. Others indicated non-significant but positive relationships for occupational or household exposure benzene, hydrocarbons, petroleum products, paints, or solvents 70, 74–76.

Apart from parental smoking, few studies have examined inferred parental exposures. One study found a significant association with paternal employment as a mechanic 74. Another study found a significant effect of proximity to repair garages and gas stations during pregnancy 77. A third study found a significant association with either maternal or paternal employment in tire production (OR = 7.6; 95% CI 1.7, 34.2 and OR = 10.1; 95% CI 2.2, 46.0, respectively) 78. A recent review of the association between benzene and childhood AML found no consistent pattern of increased risk of childhood AML for any individual exposure which might suggest higher levels of benzene 79.


In two meta-analyses, which included different studies, both found an increased risk for maternal occupational exposure to pesticides (OR = 2.7; 95% CI 1.1, 6.8 and OR = 2.6; 95% CI 1.4, 5.0) and no association for paternal occupational exposure 80, 81. However, the studies of paternal exposure were more heterogeneous with different time periods and types of exposures.

The relationship between household pesticide use during pregnancy and childhood AML is less clear with two meta-analyses producing different results depending on study inclusion for exposure to any pesticide (OR = 2.3; 95% CI 1.5, 3.5 and OR = 1.4; 95% CI 0.8, 2.6) 82, 83. However, both studies show a significant relationship with insecticides (OR = 3.1; 95% CI 1.5, 6.8 and OR = 1.9; 95% CI 1.3, 2.6) 82, 83. In addition, one study indicated that prenatal exposure to the insecticide propoxur (as measured in meconium) was associated with a specific translocation in cord blood 84 which is seen in about 12% of childhood AML cases 2.


Ionizing radiation in utero is a well established cause of childhood leukemia, including AML 4. This observation led to decreased use of X-rays in pregnant women and, combined with the lower dose now received, is thought to not contribute substantively to the current burden of childhood leukemia. However, in recent case–control studies, while no significant results have been observed, in seven of the nine studies OR estimates were greater than one 34, 38, 48, 70, 73, 78, 85–87.

Child Factors

Birth weight

The evidence suggests that there may be an increased risk of childhood AML in children with both low and high birth weight. Two meta-analyses have explored the relationship between AML and birth weight. The first included four studies and indicated a non-significant increased risk for high birth weight (OR = 1.3; 95% CI 0.7, 2.2, for ≥4,000 vs. <4,000 g) 88. The more recent analysis, which included nine studies combining different definitions of low and high birth weight, found an increased risk for both high and low birth weight (OR = 1.24; 95% CI 1.16, 1.33; OR = 1.50; 95% CI 1.05, 2.13, respectively) 89. A recent study, not included in the meta-analyses, suggested that the relationship between birth weight and childhood AML was nonlinear, and found increased risks both for low and high birth weight 90.


A meta-analysis combining data from eight studies found that long term breast feeding (>6 months) had a protective effect on the development of childhood AML (OR = 0.85; 95% CI 0.73, 0.98) while short-term breast feeding (6 months or less) also had a protective effect, but did not reach statistical significance (OR = 0.90; 95% CI 0.80, 1.02) 91.


Prior studies have uncovered possible associations between various risk factors and childhood AML (Table I). Notable findings have been observed for maternal age, birth weight, prior fetal loss, birth order, maternal exposure to pesticide, and maternal alcohol use in pregnancy. To date, genetic studies have focused on candidate genes in small studies lacking the ability to fully characterize genetic variance and have been unable to examine complex genetic interactions. Future research should develop biomarkers for difficult exposures, explore causal mechanisms rather than risk factors, analyze genetic contributions more fully, and include analysis of specific AML subtypes.

Table I. Summary of Risk Factors for AML
Generally accepted risk factorsSuggestive of increased riskSuggestive of decreased riskLimited evidence
Down syndromeOlder maternal ageLong term breastfeedingPaternal exposure to benzene
Fanconi anemiaIncreasing birth order Parental smoking
Familial monosomy 7Prior fetal loss Maternal exposure to benzene
Ataxia telangiectasiaMaternal alcohol use Maternal use of antibiotics
Shwachman–Diamond syndromeMaternal exposure to pesticides Maternal dietary consumption of DNA topoisomerase II inhibitors (infant)
Bloom syndromeHigh birth weight
Ionizing radiation in uteroLow birth weight

One of the biggest challenges in measuring maternal exposures during pregnancy is accurate measurement. Retrospective case–control studies are prone to both differential recall of exposures and selection bias. Precise measurement of exposures that occurred many years ago is also problematic. Development of biomarkers for these exposures that could be measured in neonatal blood spots or other relevant samples would provide a more accurate account of exposures during pregnancy. Epigenetic biomarkers of environmental exposures, such as DNA methylation, are starting to be developed. Low level benzene exposures have been linked with a decrease in global DNA methylation levels 92 and in utero exposure to cigarette smoke has been linked to a decrease in global DNA methylation levels 93. While these studies are not specific enough for biomarkers, other studies are looking at DNA methylation in specific genes. One study found that DNA methylation in a set of genes accounted for 78% of the variation in birth weight 94, so specific epigenetic biomarkers could be possible in the future. This may be important for childhood AML as some studies have suggested fewer genetic alterations, opening a potential larger role for epigenetic disruptions 95, 96. However, it is not currently known if epigenetic alterations are initiating events in childhood AML or a result of downstream effects of genomic changes.

Factors such as maternal age, birth weight, birth order, and prior fetal loss are not causes themselves but potential markers of some unknown factor. We need to move beyond these exposure variables and examine causative mechanisms to better understand the development of childhood AML. Causal mechanisms could involve alterations in DNA methylation or chromatin structure both of which may be affected by maternal age as well as birth order and prior fetal loss 97. Genetic alterations could also be more likely for children born to older fathers affecting the observed association with maternal age and birth order 98. Fetal loss may also be associated with genetic alterations or exposure to environmental factors. Birth weight represents a combination of genetics and in utero factors 99. Maternal factors including diabetes, pre-pregnancy body mass index, and weight gain during pregnancy are related to birth weight 100. Maternal and fetal genetic factors are also important with maternal genetic effects estimated to account for 22% of the variability and fetal genetic effects estimated to account for 31% of the variability in birth weight 99. Further analysis should include innovative methods to explore the underlying causes of these observed associations.

A thorough examination of genetic variation is necessary for a fuller understanding of childhood AML etiology. Previous research has focused only on candidate genes and no comprehensive genome-wide study has been conducted either in children or their mothers. A next generation sequencing approach could lead to discoveries of genes not anticipated to be involved in childhood AML development a priori and would provide new genetic leads for future studies. It will be important to examine genomes of affected children and their mothers given the large contribution of the maternal genome to the intrauterine environment. Few studies have examined the possibility of maternal genetic effects in childhood AML development and none have yet found significant genetic main effects, although a very small number of genes were examined 21, 22. More complex genetic studies are also necessary to assess the possibility of synergistic interactions between genes and environmental factors both in children and their mothers. Little has been done to examine these types of interactions due to small study sizes. Given the possible relationship between childhood AML and environmental toxicants, these complex effects could contribute significantly to childhood AML etiology. These studies will require collaboration between investigators and cooperative pediatric oncology groups to assemble enough case to produce meaningful results.

In summary, the epidemiological data to date have produced several possible risk factors for childhood AML. Future studies should build on this research by exploring potential causes of observed relationships between maternal age, birth weight, and prior fetal loss; developing innovative measurement tools for complex and difficult to measure exposures such as alcohol, pesticides, and benzene; examining the etiology by AML subgroups; and fully exploring the role of genetic factors including interactions. Additional research in these areas will greatly enhance the knowledge of childhood AML etiology.


Dr. Ross was supported in part by NIH K05 CA157439.