We conducted a neighborhood-based, matched case-control study based on patients from 6 Colombian hospitals. Eligible cases consisted of all children younger than 15 years who were newly diagnosed with ALL between January 1, 2000, and March 30, 2005, and were identified through the review of institutional registries. All cases had histological and clinical verification. These registries had high levels of completeness and, if necessary, were periodically reviewed. The hospitals were 6 tertiary care centers in Bogotá and Bucaramanga, 2 major Colombian cities, and patients were residents of Bucaramanga, Bogotá, Tunja, and their closest municipalities (those located up to 2.5 hours away by car). Neighborhood-based controls were individually matched to cases (1 control per case) by sex and age at diagnosis (within or equal to 2 y of the case's age at diagnosis for cases younger than 3 y and within or equal to 3 y for cases aged 3-14 y). Controls were chosen from a house-to-house search; children recruited for the control group had to be healthy and living in the same neighborhood as the case. This area was expanded if the case child lived in a rural area where the chance to find his or her control was considered low (2 cases) or in a potentially dangerous neighborhood (1 case). Children who were adopted, who had been diagnosed with Down syndrome or any previous neoplastic disease, or who were living with people other than their biological parents were excluded.

We calculated sample size using a procedure for matched case-control studies described by Schlesselman (12). Exclusion criteria were observed in 13 (8.1%) initially eligible cases, and another 12 (8.2%) were not analyzed because the parents refused to participate. Overall, at least 1 parent of 135 cases and 113 matched controls participated. However, we based our analysis on information from 170 subjects (85 case-control pairs) with participation of both parents. Preliminary analysis suggested that unadjusted results with only these children were similar to results obtained by using the complete group of participants. If only 1 parent had been interviewed, reasons for nonparticipation of the other parent were inability to find him or her, unwillingness of the other parent to participate, and separation or death of the other parent.

We collected data about 3 periods of the child's development: the last 24 months before conception (both parents), pregnancy (mother only), and childhood (from birth to ALL diagnosis for cases or equivalent age for controls) (mother only). These periods were calculated from the date of birth. A questionnaire-based face-to-face interview was used. This instrument (Appendices A and B) was developed by using structured questions based on a review of scientific literature. For some specific topics, a last open-ended question was added (eg, "Did you take any other medication during these last 24 months before that pregnancy?"). We asked parents about socioeconomic and demographic characteristics, work and medical history, smoking, consumption of alcoholic beverages and psychoactive drugs, and environmental exposures. To improve the likelihood that these questions would be answered, sections had a short heading with an introduction to the specific topic showing that such exposure could happen to anyone. In Colombia, socioeconomic status (SES) is classified from 1 (low) to 6 (high) in urban zones. For this study, the SES of parents who lived more than half of each period in low-income neighborhoods (including rural areas, SES1, and SES2) was classified as low and the SES of the other parents as medium/high. Nonmedical interviewers and field coordinators were trained in how to use the questionnaire through didactic sessions and a pilot study. They were not informed about the objectives or methods of this study. After cases were identified, parents were contacted by telephone or in person and the research was explained. Parents of controls were initially contacted at the time of the house-to-house search. All parents were interviewed in their houses, the hospital, or any other place they chose (eg, offices). The mean duration of the interviews was 80 minutes for mothers and 45 minutes for fathers.

We used a job-exposure matrix. An occupational medicine physician who was also trained in occupational hygiene classified the parental exposure to hydrocarbons on the basis of the main industrial activity of each workplace where parents worked before (both parents) or during the index pregnancy (mother only). The physician was masked to the case or control status. We defined hydrocarbons classified as carcinogenic (group 1) and probably carcinogenic (group 2A) to humans by the International Agency for Research on Cancer as the occupational exposures of interest (13) (Appendix C). The main industrial activity of each workplace that parents reported was defined by the International Standard Industrial Classification (14). Parental occupational exposure to hydrocarbons (agents) and subgroups of hydrocarbons was defined by analyzing the primary economic activity of each company where the parents worked by using data from the National Occupational Exposure Survey (15).

The difference between 2 proportions from paired data was tested by using the McNemar test or Stuart-Maxwell statistic. The difference between 2 means was assessed by using the paired t test; if distributions were not normally distributed, the Wilcoxon signed rank test was used. Odds ratios and their 95% confidence intervals were determined by using conditional logistic regression. The first step for variable selection was to discard the exposures with a P value of more than .25 in crude analysis. Possible risk factors reported in other studies (eg, mother’s age) were retained for an additional step independently of their P values. The assumption of linearity was assessed for each quantitative variable (eg, duration of breastfeeding) by plotting the log odds ratio against previously ordered categories of each, and nonlinear variables were categorized according to previous literature or the optimization method. The change-in-estimate method (16) was used for modeling, but the variable indicating parental occupational exposure to hydrocarbons during preconception (24 mo) was retained in all processes of model fitting. Preconception exposure to hydrocarbons was defined as 1) neither parent exposed, 2) father only exposed, 3) mother only exposed, and 4) both parents exposed. The likelihood ratio was used to test the significance of the difference between the fitted model and a reduced model. Pairs with subjects determined to be poorly fit were identified by using graphical methods explained by Hosmer and Lemeshow (17). Significance was set at P < .05. Stata SE version 9.0 (StataCorp LP, College Station, Texas) was used for analysis. Written informed consent was obtained from the parents of all participating children. This study was approved by the Industrial University of Santander ethical committee, the Javeriana University ethical committee, and the institutional review boards of the participating hospitals.

The median difference in age between cases and controls was 1.12 years (interquartile range, 0.52-1.91 y). For 7 cases, we were unable to find a control of the same sex, but comparative analyses with and without these pairs suggested that their exclusion did not affect the risks found in this study. These 7 pairs were included in the analysis. The other variables matched.

The distribution of parental socioeconomic and demographic characteristics between cases and controls was similar for each period (Table 1). Parental occupational exposures to hydrocarbons of interest at home and in other workplaces were combined. The prevalence of parental occupational exposure to most of these hydrocarbons was higher among cases than among controls. Unadjusted significant associations were found between childhood ALL and preconception maternal exposures to mineral oil, aliphatic, and aromatic subgroups, and to the following specific agents: trichloroethylene, benzene, epichlorohydrin, ethylene oxide, and diesel engine exhaust (Table 2a). Preconception paternal exposure to the mineral oil subgroup and trichloroethylene agents also showed significant effects. Maternal exposures to any specific hydrocarbon during the index pregnancy did not significantly affect the crude risk of ALL (Table 2b).

The storage of chemical products in 1 or more of the mother's houses and the exposure to 1 or more types of pesticides during pregnancy, but not other environmental exposures, were significantly associated with an increased risk of childhood ALL (Table 3).

In most of the conditions investigated, the prevalence of maternal and paternal diseases before the child's conception was lower than 10% and did not reach significance. These results were similar to the prevalence of maternal diseases during the index pregnancy. The crude association of some parental diseases with childhood ALL is shown in Table 3. Paternal smoking before conception and passive maternal smoking during pregnancy were significantly associated with childhood ALL. Parental consumption of alcohol did not show any significant association with childhood ALL. No significant associations were found between childhood ALL and parental exposure to psychoactive drugs (Table 3).

The mean breastfeeding duration was higher in the case group (19.7 mo) than in the control group (15.2 mo). Controls were significantly more likely than cases to have been breastfed for 6 to 11 months (Table 4).

After the model-fitting process, parental occupational exposure to 1 or more hydrocarbons classified as carcinogenic or probably carcinogenic before the child's conception was associated with an increased risk of childhood ALL. The risk was also higher with preconception parental exposure to smoking, elevated maternal age at index birth, and low maternal SES during index pregnancy (Table 5). The likelihood ratio test showed that the fitted model was better than 2 reduced models (the first one containing only intercept and the second one containing intercept plus explanatory variable). The multivariable analyses also showed that the parental occupational exposure to some of the more prevalent hydrocarbons classified as carcinogenic or probably carcinogenic to humans is related to a higher risk of childhood ALL after adjusting for nonoccupational confounders described in Table 5 (Table 6).