Pre- and Postnatal Exposures to Tobacco Smoking and Survival of Childhood Acute Lymphoblastic and Myeloid Leukemias in California, United States

Leukemia is the most common childhood cancer comprised mainly of acute lymphoblastic leukemia (ALL) followed by acute myeloid leukemia (AML). Its incidence has increased in the past decades in the United States, especially in the Latinx population (1, 2). Despite improvements in cancer survival, significant disparities persist by socioeconomic status (SES) and race/ethnicity (3–5). The reasons for differential clinical outcomes in patients with childhood leukemia are not fully explained by disparity in care (6, 7). Pre- and postnatal exposures to tobacco smoking, especially from fathers before conception and passive smoking, have been associated with increased risks of developing childhood ALL and AML (8–10) and ALL with increased number of chromosomal deletions (11, 12). Yet minimal attention has been given to the possible impact of tobacco smoke on clinical outcomes following diagnosis. One study conducted among children with ALL in Spain reported that exposure to tobacco smoke—mainly maternal smoking pre- and postnatally—negatively impacted 5-year survival, relapse, and treatment-related mortality (13). Although based on small numbers of children enrolled (n = 146) and reported deaths (n = 8), these findings were consistent with studies in adults showing reduced survival among smokers treated for leukemia and other cancers, possibly via carcinogenic and immunologic pathways (14–25).

While the prevalence of tobacco smoking has diminished in the United States (26), tobacco use remains highly prevalent in economically deprived households and in populations of color (27–29). Notably children from these various backgrounds also experience worse event-free and overall survival following a leukemia diagnosis (3–5). We tested the hypothesis that pre- and post-natal exposures to tobacco smoke reduce ALL and AML survival in 1,449 children enrolled in a California case–control study.

The California Childhood Leukemia Study (CCLS) is a population-based case-control study originally designed to identify environmental and genetic risk factors of childhood leukemia (1995–2015; ref. 10); only cases were evaluated for this analysis. Patients with childhood leukemia (based on the International Classification of Childhood Cancer) were rapidly identified after diagnosis at 17 hospitals and were eligible if younger than 15 years of age at diagnosis, had an English or Spanish speaking parent, lived in one of the study counties at time of diagnosis, and had no previous cancer. Among 1,709 consented cases, 1,449 were interviewed (85%). About 19% (n = 50) of those without completed interviews were children who died (i.e., 26 ALL and 23 AML).

Interviews were conducted within 6 months of the diagnosis (on average), mainly with biological mothers, using a structured questionnaire on sociodemographic, occupational, residential, medical, and lifestyle factors. Data on child's exposure to tobacco use (defined as a minimum of 100 cigarettes/cigars/pipes in a lifetime) was collected separately from the mother and father when available; if the father was not available, the mother was interviewed regarding father's smoking. We collected information (yes/no) on paternal and maternal smoking (lifetime and before conception), maternal smoking during pregnancy and breastfeeding, and child's passive smoke exposure after birth from the mother (excluding breastfeeding) and/or anyone else (father or others) up to the time of the interview or third birthday, whichever came first. We collected start and end dates of smoking and number of cigarettes per day (cpd) for the preconception, pregnancy, and breastfeeding periods. Histologic leukemia types and white blood count (WBC) at diagnosis were abstracted from medical records and independently validated by a clinician. Determination of cytogenetic subtypes was derived from both medical records and additional testing, including FISH to identify leukemias with high hyperdiploidy and the TEL-AML (ETV6-RUNX1) translocation and multiplex ligation-dependent probe amplification (MLPA) to identify CDKN2A and IKZF1 deletions (10, 12). To assess vital status, electronic death certificate data obtained from the California Department of Public Health Center for Health Statistics and Informatics (1995–2020) were linked to the CCLS database by child's first, middle, and last name, sex, date of birth, mother's maiden name, father's last name, and race/ethnicity, using a probabilistic linkage (Match*Pro Version 2.0.7, SEER). Ambiguous matches, usually due to misspelling of first names or surnames (n = 65), were independently reviewed (by C. Metayer and L.M. Morimoto); matches were considered definitive if consensus was achieved. Out of 1,449 participants, 191 cases were linked to death files (5 were due to external causes). This study was approved by Institutional Review Boards at the University of California, Berkeley, and the California Department of Public Health.

The outcome evaluated was 5-year survival from all causes except external causes. Cases were considered an event if death occurred any time before end of follow-up (12/31/2020); cases were censored if alive or at date of death from external cause. The non-parametric Kaplan–Meier estimator was used to estimate the survival function, and survival curves were plotted to visualize the probability of survival comparing smoking groups. Log-rank tests were performed to test the significance of differences in survival between groups. Cox proportional hazards regression was used to calculate hazard ratios (HR) and 95% confidence intervals (CI) associated with smoking, adjusting for child's sex, race/ethnicity (Latinx, non-Latinx White, non-Latinx Black, non-Latinx Asians, and non-Latinx others), birth year (continuous), parental highest education (dichotomized: high school or lower vs. some college or more), annual household income (6 categories), and risk group (ALL only; categorical: “standard” defined as age >1 year and age <10 years and WBC <50,000/μL; “high” defined as age ≥10 years OR age >1 year and age <10 years and WBC ≥50,000/μL; and “infant” defined as age <1 year). We conducted additional analyses to assess whether other covariates acted as potential confounders including the number of persons supported by the annual household income (as a proxy for SES), year of diagnosis [as proxy for changes in treatment protocols (30) and smoking behavior (31) over time], and hospital study site (as a proxy for treatment modalities and supportive care); none of these affected the risk estimates by more than 10%, and therefore, were not included in the final models (see Directed Acyclic Graph in Supplementary Fig. S1). Stratified analyses were conducted by sex, race/ethnicity (non-Latinx White vs. Latinx), and cytogenetic characteristics [i.e., deletions in CDKN2A and IKZF1 genes, and high hyperdiploidy; data were too sparse to analyze TEL-AML (ETV6-RUNX1) translocation separately with only 3 deaths recorded]. Testing for heterogeneity between groups was conducted using likelihood ratio tests.

The study was approved by Institutional Review Boards for the California Health and Human Services and the University of California, Berkeley and San Francisco, and was conducted according to the U.S Common Rule.

The epidemiologic and clinical data generated in this study are not publicly available due to terms of the informed consent signed when subjects were enrolled to the CCLS study but are available upon reasonable request from the corresponding author. The death data analyzed in this study were obtained from the California Department of Public Health (CDPH) Center for Health Statistics and Informatics (CHSI) and are not publicly available due to terms of CDPH-CHSI.

Among 1,449 interviewed childhood leukemia cases (1,235 ALL, 188 AML, 26 other types), 186 children died of nonexternal causes within 5 years of the diagnosis (i.e., 130 ALL and 52 AML). Causes of death from death certificates were neoplasm/leukemia (n = 170), infection (n = 4), blood/circulatory/immune system disorders (n = 4), respiratory system disorders (n = 3), and others (n = 5). Table 1 shows the distribution of demographic and birth characteristics by vital status. As expected, children who died were more likely to have AML or high-risk ALL, be diagnosed under one year of age, be either Latinx, non-Latinx Asian/Pacific Islanders or non-Latinx Black; and have low parental education and annual household income. Lifetime tobacco smoking was reported in 39% of fathers (65% of which were either Black, Latinx, or Asian/PI, compared with 62% among non-smokers) and 23% of mothers (46% of which were non-Latinx White); paternal smoking was more frequent in households of low education and low income compared with nonsmokers; similarly maternal smoking was more frequent in low-income households (Supplementary Table S1). There was little correlation between maternal and paternal smoking but some level of correlation between pre- and postnatal maternal smoking (Supplementary Fig. S2). Overall, children who died were more likely to have a history of paternal smoking, especially during preconception (29% vs. 22% in deceased vs. alive, P = 0.06) and passive smoking after birth (24.7% vs. 17% in deceased vs. alive, P = 0.15; Table 2). Alive and deceased children were otherwise similar across other categories of parental smoking in univariate analyses.

Paternal preconception smoking was associated with decreased 5-year survival, especially for fathers reporting smoking more than 20 cigarettes per day (P for log-rank test = 0.067; Fig. 1). Table 3 shows the HRs for 5-year survival for childhood AML with adjustment for sex and birthyear (Model A) and with additional adjustment for parental education and annual household income (Model B). In the fully adjusted Model B, the risk of dying for children with AML increased with the amount of preconception paternal smoking (HR = 1.41; 95% CI, 0.95–3.44 and 3.47; 95% CI, 1.37–8.81, respectively for <20 vs. ≥20 cigarettes per day; P_trend = 0.01). Overall, the risk estimates were comparable with and without adjustment for SES. Preconception paternal smoking remains an independent prognostic factor of AML after adjusting for child's passive smoking, although the magnitude of the association decreased (e.g., HR for paternal preconception smoking/20+ cigarettes per day = 2.91, 95% CI, 1.18–7.2). Further adjustment for hospital study site did not substantially modify the risk estimates (e.g., HR for paternal preconception smoking/20+ cigarettes per day = 3.60 (95% CI, 1.38–9.40). There were no notable associations between pre- and post-natal maternal smoking and AML survival.

There were no indications of associations between ALL survival and exposure to tobacco smoke at any time pre- or postnatally in models with and without SES adjustments (Table 4). However, our data suggested differences in smoking-related risk of death by ALL molecular subgroup (favorable vs. poor prognosis; Table 5). Specifically, for paternal preconception smoking, P values for heterogeneity in HRs were 0.07 comparing both high hyperdiploidy status (yes/favorable vs. no/unfavorable) and IKZF1 deletion status (no/favorable vs. yes/unfavorable). When estimating the joint effect of paternal preconception smoking and these molecular types, children treated for high hyperdiploid ALL (favorable) and exposed to paternal preconception smoking were 2.4-fold more likely to die compared with those not exposed (P = 0.17), approaching risk levels close to ALL without high hyperdiploid (unfavorable). A similar trend was seen for the joint effect of IKZF1 deletion negative and paternal preconception smoking, although the association did not reach statistical significance (HR = 1.50, P = 0.27). In contrast, no joint effects were seen for CDKN2A subtype and paternal preconception smoking (Table 5). Child's passive smoking did not confound associations observed for ALL subtypes. Tests for interaction with race and ethnicity reached statistical significance (P = 0.03) for paternal preconception smoking and maternal smoking after birth; for Latinx children the risk was increased (HRs = 1.27, 95% CI, 0.70–2.31 and 1.80, 95% CI, 0.86–3.74, respectively), whereas for non-Latinx White children the risk was decreased (HRs = 0.22, 95% CI, 0.05–1.04 and 0.19, 95% CI, 0.02–1.54, respectively) compared with those not exposed to parental smoking. In general, risk estimates associated with exposure to tobacco smoke at other times were above one for Latinx children and close to or below one for non-Latinx White children; tests for interaction, however, were not statistically significant. Results were similar for boys and girls.

This study is the first one to report that preconception paternal smoking reduces 5-year survival of childhood AML, especially among children whose fathers were heavy smokers. Adjustment for SES and child's passive smoking had little impact on the magnitude of the association indicating that paternal preconception tobacco smoking acted independently. There was a suggestion that children treated for certain favorable molecular subtypes of ALL lost their survival advantage when fathers reported smoking before conception, although these results did not reach statistical significance and need to be replicated. Results for other windows of exposure to maternal tobacco smoking pre- and postnatally were either null or inconclusive.

Studies worldwide (8, 9), including ours in California (10), have documented associations between paternal preconception smoking and increased risks of developing childhood ALL and AML. This is the first study to also report an association with reduced survival in pediatric AML, especially for heavy smokers. Our data also suggested that paternal preconception smoking reduces survival of certain ALL molecular subtypes otherwise known to have favorable prognosis, including those with high-hyperdiploidy and without IKZF1 deletion. These observations were, however, based on small numbers. The biological underpinning that may explain differences in risk by leukemia type remains unclear. The underlying mechanisms by which paternal smoking during the preconception period affect prognosis of certain childhood leukemias may include damage of paternal germ cells that could alter immune and oxidative stress pathways (32). Another potential mechanism is through paternal epigenetic programming of cardiovascular and metabolic pathways in offspring (33) that impact cancer progression, response to treatment, and treatment-related toxicity (34, 35). For example, paternal preconception smoking has been associated with high body weight in offspring (36, 37), suggesting that growth factors or obesity may play a mediating role.

Our data suggested that passive smoking reduced 5-year survival of childhood AML (any smokers) and ALL (Latina mothers), respectively by 1.7- to 1.8- fold, although these associations did not reach statistical significance. This is consistent with studies conducted in adults treated for AML and solid tumors showing worse clinical outcomes among smokers (14–25). Similarly, a study conducted in Spain among 146 children with ALL first reported statistically significant associations between maternal smoking during both pregnancy and after birth and poor clinical outcomes, with a 4-fold increased risk of dying, 8-fold risk of relapse, and a 14-fold increased risk of cumulative treatment-related mortality, after accounting for known prognostic factors (13). In contrast, paternal smoking was not independently associated with worse outcomes. The separate contribution of pregnancy versus postnatal tobacco smoking could not be assessed since most mothers smoked during both periods. In our study, maternal smoking during pregnancy did not affect childhood leukemia survival. Differences in results between the study in Spain and our study in California may be due to prevalence of smoking, especially for mothers who were twice as likely to smoke in Spain compared with California (44% vs. 23%). Carcinogenic compounds found in tobacco smoke such as benzene may not only damage blood cells, but also increase genetic instability of leukemia cells and modulate immune response. Smoke exposure has been shown to lower immunoglobulin levels and T-cell counts and delay count recovery possibly resulting in prolonged susceptibility to infection and increased risk of bleeding (14, 16, 21, 25).

Our results should be interpreted considering limitations and strengths. Although our data on pre- and post-natal tobacco smoking are detailed and novel for the preconception period, they rely on self-report. We previously assessed the quality of recall in a subset of CCLS participants (38, 39). First, we measured two biomarkers of tobacco smoking during pregnancy [i.e., 1- Cys34 protein adducts (13, 39) and CpG sites in the AHRR, GFI1, and MYO1G genes (38)] in archived newborn blood samples and compared those with maternal self-reported smoking. The strong correlation between the biomarkers and self-report suggested that mother's recall was largely accurate. We also evaluated concordance between the mother's and father's reporting (among 107 leukemia cases and 108 controls) to evaluate for recall bias on father's smoking. The overall agreement for smoking (current, lifetime smoking, and three months before the mother's pregnancy) was high, with kappa statistics ranging from 0.70 to 0.76 (P < 0.05; ref. 40). Agreement was higher among parents of children with leukemia, those with higher education, for non-Latinx White parents, and for those with short intervals between the child's birth and diagnosis (<6 years). Since 69% of our ALL cases were diagnosed under the age of 6, mothers’ recall of father's smoking should not have substantially biased our findings regarding ALL and paternal smoking. Although death at the time of interview was not an exclusion criterion, interviews were not completed for 50 children who died shortly after enrollment in the study. In our cohort, however, the 5-year survival for ALL combined was 91.8% and 69.9% for AML, which is aligned with national data for our study period (1995–2015; refs. 41, 42). Deceased children without interviews were similar to deceased children with interviews with respect to various birth registry data including birthweight, gestational age, and individual sociodemographic characteristics. There was a suggestion; however, that neighborhood income varied between households not interviewed and those interviewed (mean = $38,000 vs. $45,300, respectively; pooled t test P = 0.06), indicating the potential for selection bias among those interviewed.

Although models were adjusted for several sociodemographic characteristics known to influence both exposure to tobacco smoking and survival, there may be some residual confounding by SES. Medical insurance and access to supportive care (especially following hematopoietic stem cell transplantation and related complications such as graft-versus-host disease) may be subject to geographic and sociodemographic variations and differential adverse outcomes (43). However, our data showed that hospital study site, as a proxy for treatment modalities and supportive case, was not confounding the observed associations between tobacco smoking and survival. Also, based on a causal diagram built for our analyses, additional adjustment for medical insurance was not necessary, beyond household income, parental education, and child's race and ethnicity. Altogether, we believe that access to care should not substantially impact our results. On the basis of prior knowledge (44), we examined ALL molecular subtypes that are known to affect prognostic and that were available in our study (i.e., high hyperdiploidy, CDKN2A and IKZF1 deletions). We did not have information or sufficient sample size for other prognostic subtypes such as Ph+ and Ph-like ALL, or those with TEL-AML (ETV6-RNX1) fusion gene. Also, despite a relatively large number of childhood leukemia cases, analyses for rare subtypes (i.e., AML and ALL subgroups) and minority racial and ethnic groups were based on small numbers. Finally, multiple comparisons may have led to false positive results.

In conclusion, our data suggest that paternal preconception smoking has a negative impact on childhood AML survival, adding to the body of evidence from previous studies mostly conducted in adults. Results for ALL subtypes need to be replicated in larger studies. Knowledge of exposure to tobacco smoke should be integrated in treatment plans to ensure that at a minimum, children undergoing cancer treatment are not exposed to the harmful effects of tobacco smoke.

C. Metayer reports grants from TRDRP, grants from NIH, and grants from UK Children with Cancer during the conduct of the study. L.E. Winestone reports grants from TRDRP during the conduct of the study. No disclosures were reported by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIEHS and TRDRP.

C. Metayer: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. L.M. Morimoto: Data curation, software, formal analysis, funding acquisition, writing–review and editing. A.Y. Kang: Resources, data curation, software, funding acquisition, writing–review and editing. J. Sanchez Alvarez: Writing–review and editing. L.E. Winestone: Conceptualization, writing–review and editing.

This research and all authors were supported by the California Tobacco-Related Disease Research Program (TRDRP) grant #T31IP1502. Collection and/or maintenance of the original CCLS data were partly supported by the National Institute of Environmental Health Sciences (NIEHS) grants #P42ES004705 (to C. Metayer), R01ES009137 (to C. Metayer, A.Y. Kang, L.M. Morimoto), and R24ES028524 (to C. Metayer, A.Y. Kang, L.M. Morimoto), and the UK Children with Cancer grant# 2006/052 (to C. Metayer). We thank the families for their participation. We also thank the clinical investigators at the following collaborating hospitals for help in recruiting patients: University of California Davis Medical Center (Dr. Jonathan Ducore), University of California San Francisco (Drs. Mignon Loh and Katherine Matthay), Children's Hospital of Central California (Dr. Vonda Crouse), Lucile Packard Children's Hospital (Dr. Gary Dahl), Children's Hospital Oakland (Dr. James Feusner), Kaiser Permanente Roseville (former Sacramento; Drs. Kent Jolly and Vincent Kiley), Kaiser Permanente Santa Clara (Drs. Carolyn Russo, Alan Wong and Denah Taggar), Kaiser Permanente San Francisco (Dr. Kenneth Leung) and Kaiser Permanente Oakland (Drs. Daniel Kronish and Stacy Month). Finally, we acknowledge the entire California Childhood Leukemia Study staff and the former UCB Survey Research Center for their effort and dedication.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.