Maternal Diet and Acute Lymphoblastic Leukemia in Young Children

With the exception of ionizing radiation (1) and some rare genetic abnormalities (2, 3), causes of childhood acute lymphoblastic leukemia (ALL) have not been identified (4, 5). Two lines of evidence point to the intrauterine environments as playing a major etiologic role: (a) leukemia clone-specific chromosomal translocations are present at birth in children who have later developed leukemia (6, 7) and (b) birth weight has been frequently found to be associated with risk of ALL (8-14), particularly among children ages <5 years (15-17), although null results have also been reported (18-20). The first line of evidence tends to incriminate prenatal exposures. The second line of evidence points to quantitative or qualitative aspects of maternal diet during pregnancy 21, 22), although hormonal factors may also be important (23). Accordingly, children ages <5 years with ALL represent an appropriate group for the documentation of the maternal diet-childhood ALL association, if such an association actually exists.

We have undertaken a nationwide study of ALL among children ages <5 years in Greece with focus on maternal diet during the index pregnancy. We have not included cases of infant leukemia, because the majority of them has a specific genetic abnormality in the 11q23 chromosome band involving the MLL gene (24, 25).

A nationwide network comprising all six Childhood Hematology-Oncology Departments has been established in the mid-1980s and has conducted several epidemiologic investigations concerning childhood hematologic malignancies (26-28). For the present study, all 171 cases of ALL ages 1 to 4 years (12-59 completed months of age) first diagnosed anywhere in Greece from January 1999 to June 2003 were eligible. Five ALL cases diagnosed among infants during the study period were not included based on the protocol. For 21 cases diagnosed in one of the two departments located in Thessaloniki, data were not available, but their exclusion is unlikely to have introduced selection bias because it was based on administrative reasons. For another 9 cases from the remaining five departments, consent for inclusion in the study was not obtained. Thus, 141 cases were eligible from the two Departments of “Aghia Sophia” General Children's hospital and the single Department of Children's Hospital of Athens “Kyriakou,” from the Department of the American Hellenic Educational Progressive Association Hospital in Thessaloniki, and from the Department of the University Hospital in Heraklion, Crete. The initially enrolled hospitals are all five children's hospitals in Greece, and although children are also admitted in pediatric wards of general hospitals, the bulk of childhood morbidity is dealt in the children's hospitals.

An attempt was made to match each ALL case with one control of the same gender and similar age (±6 months), concurrently hospitalized in the same institution for minor conditions and without a history of cancer or overt nutritional or metabolic disorder. For 10 ALL cases, the mother was not available or no suitable control child was found. In 7 instances, control children could not be enrolled because of inaccessibility of their mothers, but they were appropriately substituted. Thus, the study was eventually based on 131 individually matched pairs of ALL cases and control children. The admission diagnoses of the control children were mild respiratory conditions (29 controls), viral infections (33 controls), allergy (16 controls), gastrointestinal or genitourinary conditions (18 controls), nervous system conditions (12 controls), and injuries (23 controls). The Ethics Committee of the University of Athens Medical School approved the study protocol and all procedures were in accordance with the Helsinki declaration for human rights.

The mothers of the 131 case-control pairs who have consented to participate were interviewed in person by trained interviewers, the same for each case-control pair. Interviews took place in the respective health care settings. The questionnaire used covered sociodemographic variables and an extensive section assessing maternal dietary intakes, including typical portion sizes, during the index pregnancy. Specifically, the mothers were asked to indicate the average frequency of consumption during pregnancy, per month, per week, or per day, of the indicated portion sizes of 157 food or beverage items. The dietary questionnaire has been previously validated among adult men and nonpregnant women (29).

For the analysis, the frequency of intake of each food item was translated into average daily quantity of intake (in g/d) and the food items were combined into nine groups in a variation of the scheme recommended by Davidson and Passmore (30) and regularly used in nutritional epidemiologic studies in Greece (31, 32). The food groups were cereals and starchy roots, sugars and syrups, pulses and nuts, vegetables, fruits, meat and meat products, fish and seafood, milk and dairy products, and butter and margarine. Some cooked meals were allocated into more than one food groups (e.g., “pastitsio” was allocated 50% into cereals and 50% into meats). Total energy intake was also calculated by multiplying the energy content of the typical portion of each food item by the frequency the food item was consumed and adding the products over all food items (33). Intakes of energy (in kcal/d) and intakes of each of the nine food groups (in g/d) were then distinguished into quintiles based on the respective distributions of the cases and controls combined, apart from intake of butter and margarine that could only be distributed in tertiles.

For the statistical analysis, cases and controls were distributed by marginal quintiles. A χ statistic (the square root of χ² with 1 df) was used to assess the direction and the statistical significance of the association between maternal consumption of foods of a particular group and ALL risk in the offspring, without adjustment for covariates. Adjustment for the matching variables, birth weight (continuously in 500 g increments), as well as maternal age (continuously in 3-year increments), years of schooling (in three categories, ordered), occupation (yes/no), energy intake (continuously, in increments of 1 SD among controls), and tobacco smoking during pregnancy (yes/no), was accomplished by modeling the data through conditional logistic regression. Covariates were chosen as possible predictors of either maternal dietary intakes or disease risk and thus as conceivable confounders of the association maternal dietary intakes and childhood ALL in the present data set. To this core model, intake of each of the nine food groups was alternatively added (in quintiles, ordinally, except for butter and margarine). The SAS statistical package was used in all instances (34).

Of the 131 cases of ALL, 75 were boys and 56 were girls. Within the age group under investigation, there were more cases ages 2 to 3 years than younger and older ones. There is evidence that maternal smoking during pregnancy increases the risk for ALL among children ages 1 to 4 years and suggestive evidence for a positive association between maternal age at birth and risk of this disease. In contrast, in this data set, birth weight is not related to ALL risk (Table 1). These data, however, are not mutually adjusted and the indicated associations are explored further on.

In Table 2, cases and controls are compared with respect to consumption of each of the nine food groups as well as total energy intake. The food group associations in this table are not adjusted for energy intake, maternal age at birth, or birth weight, nor do they accommodate the matched design of the study. Nevertheless, there is evidence in the data that increased maternal consumption of sugars and syrups as well as of meat and meat products increases the risk of ALL in the offspring, whereas increased maternal consumption of fruits and perhaps vegetables reduces the risk.

The data in Table 3 indicate that after mutual adjustment there are significant positive associations of ALL at ages 1 to 4 years with maternal age at birth as well as with tobacco smoking and energy intake during pregnancy. In this data set, maternal years of schooling, as an indicator of socioeconomic status, maternal occupation, and birth weight, are not associated with ALL risk. Introduction, one at a time, of the nine food groups under study in ordered quintiles (tertiles for butter and margarine) in the model presented in Table 3 reveals several significant associations with ALL risk: inverse for fruits, vegetables, and fish and seafood and positive for sugars and syrups and meat and meat products (Table 4). Controlling for maternal occupation in specific job categories (professionals, white color nonprofessional workers, manual workers, and no occupation besides homework) and for tobacco smoking according to whether mothers were actually smoking during the index pregnancy had no effect on the odds ratio (OR) estimates given in Table 4.

We have also run several models with mutual adjustment of two or more of the food groups indicated in Table 4. In general, the direction of the associations did not change, but their strength was reduced (ORs tended toward the null) and the corresponding 95% confidence intervals (95% CI) increased because of the underlying intercorrelations and the overdetermination of the models (data not shown).

In a nationwide case-control study in Greece on ALL among children ages 12 to 59 months, we have found evidence that maternal consumption during pregnancy of increased quantities of vegetables, fruits, and fish and seafood is associated with reduced risk of the disease in the offspring, whereas increased maternal consumption of meat and meat products and sugars and syrups is associated with increased risk of ALL among their young children. A marginal inverse association was also noted with respect to maternal consumption of milk and dairy products. In essence, our results indicate that a diet generally considered as “healthy” (35) for adults may, if consumed during pregnancy, also reduce the risk of ALL among offspring.

Strengths of the present study are its nationwide coverage; its satisfactory size, considering that it refers to a relatively rare disease in a relatively small country; the smooth cooperation on the part of the children's mothers in the hospital environment; the use of a dietary questionnaire that has been validated, although not among pregnant women; the high comparability between cases and controls in the interviewing conditions; and control, in the analysis, for all available variables that could have confounding potential. The study has also several weaknesses, including those inherent in case-control investigations. An additional weakness was enrollment of hospital, rather than general population, controls. In Greece, few women in the general population are willing to discuss issues concerning the health of their children with essentially unknown persons, not withstanding their credentials. Hospital controls, however, were enrolled among those attending the large pediatric hospitals, which are treating the bulk of childhood morbidity in Greece and in which the participating pediatric hematology/oncology units were situated (36). Care was also taken that hospital controls had diagnoses that have not been linked to maternal, as contrasted to own diet. We have no information on actual income (the question is considered too sensitive), and we could not ascertain how well the diet of control women approximates the diet of pregnant women in the general Greek population (there are no relevant studies that have used the food frequency questionnaire employed in the present investigation). It has not been possible to inquire about intake of illicit drugs and we have had considerable difficulties in ascertaining olive oil intake, because this food is consumed almost universally. Lastly, misclassification of dietary exposures is certainly present, but it is likely to be nondifferential and thus unlikely to generate false associations or exaggerate genuine ones.

The Greek diet most closely approximates the traditional Mediterranean diet. This diet is characterized by high intake of vegetables, legumes, fruit, and cereals; a high intake of olive oil; a low intake of saturated lipids; a moderately high intake of fish; a low to moderate intake of dairy products (mostly in the form of cheese or yogurt); and a low, but rapidly increasing during the last few decades, intake of meat and meat products (37). In this investigation, we have focused on food groups rather than nutrients, in line with the strategy adopted in the early studies investigating the relation of diet to adult onset chronic diseases, including cancer (38).

In our study, maternal age at birth was positively associated with ALL risk in the offspring, in line with reports from several recent larger investigations (39-41). We have not been able to document in this data set the association of birth weight and childhood ALL (14). It is not unusual, however, to fail to document a relatively weak association in a study with moderate statistical power, as it has also happened with the birth weight and ALL association in other investigations (18-20). The positive association in our data between maternal smoking and ALL in the offspring has occasionally been reported in other investigations (42) but has not been documented in several others (43). The higher total energy intake during pregnancy of the mothers of ALL cases in comparison with those of control children may reflect relative overreporting, which was controlled for in the analysis, or may reflect a genuine phenomenon that needs to be evaluated in future investigations.

Few investigations have examined maternal diet during or immediately before pregnancy in relation to ALL in the offspring. The results of these studies as well as those of our investigation are remarkably consistent in spite of differences in methodologic and sample characteristics. Blot et al. (44) reviewed in 1999, among other issues, the limited at the time evidence concerning consumption during pregnancy of cured meat, a source of potentially carcinogenic N-nitroso compound, and childhood malignancies, including ALL. They have noted that some of the studies, despite using limited dietary questionnaires, were indicative of a positive association. Thompson et al. (45) reported that offspring of women who during their pregnancies received supplements with folate (naturally found in several leafy vegetables) had lower risk of ALL. Additionally, Jensen et al. (46) have found that increased maternal intake immediately before the index pregnancy (and inferentially, during that pregnancy) of vegetables and fruits was associated with decreased risk of ALL.

In conclusion, we have found evidence that young children of women who during their index pregnancy tend to consume what is currently considered to be a “healthy” diet, which is a diet high in vegetables, fruits, fish, and seafood and low in meat and meat products, sugars, and syrups, have a lower risk of ALL. These results are consistent, but more striking, with those previously reported from other investigators. If confirmed, our findings would indicate that the incidence of ALL among young children could be reduced by maternal adherence during pregnancy to the generally accepted principles concerning a healthy diet throughout life.

We thank A. Trichopoulou for making the European Prospective Investigation of Cancer questionnaire available for the purpose of this study.