American Journal of Epidemiology
Copyright O 1996 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 144, No. 11
Printed In U.SA.
Maternal Smoking During Pregnancy and Childhood Cancer
Mark A. Klebanoff, John D. Clemens, and Jennifer S. Read
The association between maternal smoking during pregnancy and childhood cancer was investigated using
prospectively collected data from 54,795 livebom children in the Collaborative Perinatal Project (1959-1966).
Cases of cancer had a histologic diagnosis and/or a compatible clinical course. There were 51 children with
cancer, for a cumulative incidence of cancer of 1.1 per 1,000 by 96 months of age. Maternal smoking was
determined at each prenatal visit; 52% of mothers reported smoking at one or more visits. By age 8 years,
cancer had occurred in 1.4 per 1,000 children whose mothers did not smoke during pregnancy, compared with
0.9 per 1,000 children whose mothers smoked (p = 0.15 by log rank test); the hazard ratio was 0.67 (95%
confidence interval (Cl) 0.38-1.17). There was no dose-response effect of smoking compared with nonsmokers (hazard ratio for one to 10 cigarettes/day = 0.45, more than 10 cigarettes/day = 0.83). The hazard ratio for
leukemia among children whose mothers smoked was 0.82 (95% Cl 0.31-2.11); the hazard ratio for cancers
other than leukemia was 0.60 (95% Cl 0.30-1.20). Adjustment did not change the hazard ratio substantially.
Although the relatively small number of cases precluded extensive study of individual types of cancer, the
authors conclude that maternal smoking during pregnancy is not associated with an increased risk of
childhood cancer in this cohort. Am J Epidemiol 1996; 144:1028-33.
carcinogens; cohort studies; neoplasms; pregnancy; smoking
Tobacco smoking accounts for approximately 30
percent of the cancer deaths in the United States (1).
However, the role of prenatal smoke exposure in the
pathogenesis of cancer during childhood is less certain
(2). Cancer during childhood is a rare event, and most
studies of childhood cancer have used the case-control
design. Retrospective studies of a risk factor that is
socially disapproved, such as maternal smoking during
pregnancy, may be particularly susceptible to recall or
reporting bias. Mothers of children with cancer may be
more likely than mothers of normal children to recall
smoking during pregnancy, falsely elevating the risk
of smoking. Conversely, even in the presence of accurate recall, guilt may make case mothers more likely
than control mothers to deny the presence of a risk
factor widely known to cause cancer, falsely reducing
the risk of smoking.
There have been only three prospective cohort studies of maternal prenatal smoking and childhood canReceived for publication April 16, 1996, and accepted for publication July 25, 1996.
Abbreviations: Cl, confidence interval; CPP, Collaborative Perinatal Project.
From the National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, MD.
Reprint requests to Dr. Klebanoff, Division of Epidemiology, Statistics, and Prevention Research, National Institute of Child Hearth
and Human Development, National Institutes of Health, 6100 Building, Room 7B03, Bethesda,. MD 20892-7510.
cer. One (3) found a significantly increased relative
risk of 2.47 for cancer among children of smokers,
another found a mildly increased risk of 1.3 (4), and a
third found a relative risk of 0.99 (5). The present
study is the fourth prospective cohort study of maternal prenatal smoking and childhood cancer.
MATERIALS AND METHODS
The subjects of this study were members of the
Collaborative Perinatal Project (CPP) cohort, participating in a prospective, multicenter study designed to
evaluate risk factors for various neurodevelopmental
disorders of childhood (6-8). Pregnant women were
enrolled from 1959 to 1966 at 12 university-affiliated
medical centers in the United States. Their children
were followed to age 7 years (approximately 80 percent (7)) or 8 years (approximately 36 percent (8)) and
were examined according to a common protocol. Each
child's guardian was interviewed at regular intervals
during the first 2 years of life and annually thereafter.
Whenever study personnel learned (usually by interview of the guardian) that a child had died or had been
hospitalized, the relevant medical and, when appropriate, autopsy records were obtained.
The methods used to ascertain potential cases of
cancer have been described in detail previously (9,
10). Potential cases were ascertained by several mechanisms. Study identification numbers of children with
1028
Maternal Smoking and Childhood Cancer
cancer were obtained from the authors of a previous
investigation of cancer and exposure from radiographic examination in this cohort (11). In addition,
diagnostic summary forms were completed for all
subjects at birth and at 1 and 7 years of age. All
summary forms reporting a neoplasm were identified.
The study records of these children were reviewed to
obtain a specific diagnosis. Reviews were done by two
pediatricians (M. A. K. and J. S. R.) who were unaware of the mother's smoking habit. Definite cases
were required to have a histologic diagnosis of cancer
noted on a medical record summary or a clinical
course (including treatment administered) consistent
with the diagnosis.
A previous analysis from this cohort (9) excluded
cases of cancer diagnosed in stillboms or on the first
day of life. As maternal smoking might be associated
with cancer in these children, the records of three
stillborn infants and seven liveborn infants with neoplasia noted on the first day of life were reviewed by
investigators who were blinded to maternal smoking
history. Three cases of cancer diagnosed on the first
day of life in liveborn children (leukemia, Wilms'
tumor, and hepatoblastoma) were included after this
review.
Women were usually registered in the study at their
initial prenatal visit and were seen subsequently according to local policy at each site. The median gestation at registration was 17-20 weeks for white and
21-24 weeks for black women; approximately 75 percent of women of both races had registered by 28
weeks (7). At study registration and again at each visit,
the women were asked how many cigarettes per day
they were currently smoking. Women who reported
smoking at any visit were considered to be smokers,
and their maximum daily consumption was used in
analysis of dose response.
To account for the varying time of follow-up, survival analysis methods were used. The children entered the cohort at birth and were followed until the
earliest of the dates cancer was diagnosed, their last
computerized CPP form was completed, or they
reached 96 months of age. Differences in cumulative
incidence were tested using the log rank test, with
statistical significance defined as a two-tailed p value
of 0.05. The hazard ratio was estimated from the
proportional hazards model; 95 percent confidence
intervals were derived from the standard errors of the
regression coefficients. The proportionality assumption was tested by including both fixed and timedependent terms for smoking and was found to be
valid (12). Categorical variables were compared using
the chi-squared test, and continuous variables were
Am J Epidemiol
Vol. 144, No. 11, 1996
1029
compared using the t test for means and the Wilcoxon
rank sum test for medians.
RESULTS
The cohort comprised 44,621 women, who gave
birth to 54,795 liveborn children. Maternal smoking
data were available for 54,306 births. At study registration, 25,353 women (47 percent) were current
smokers, and an additional 3,120 women reported
smoking on at least one subsequent visit, for a total of
28,473 women (52 percent) who smoked during pregnancy. Characteristics of smokers and nonsmokers are
shown in table 1. Women who smoked during pregnancy were less educated, more likely to be white,
more likely to be primigravid, less likely to receive
diagnostic radiation during pregnancy, less likely to
breast feed their infants, younger, of lower socioeconomic status, taller, and lighter than nonsmokers. Children of women who smoked were on average 143 g
lighter at birth and were followed a median of 4
months less than children of nonsmokers. All of these
differences were statistically significant. Gestational
age, birth date, and sex of the children of smokers and
nonsmokers were comparable.
Fifty-one children developed cancer, corresponding
to a life-table probability of 1.1 per 1,000 children
followed to 96 months of age. The leukemias were the
most common malignancy (n = 17), although the
records did not allow the reliable differentiation of
lymphoblastic from other types of leukemia. There
were eight brain tumors, seven Wilms' tumors, six
neuroblastomas, five lymphomas (four non-Hodgkin),
three retinoblastomas, two each of hepatoblastoma and
rhabdomyosarcoma, and one fibrosarcoma.
The cumulative incidence of cancer for children of
mothers who smoked and did not smoke during pregnancy is presented in figure 1 (data on maternal smoking were missing for one case). Children of women
who smoked during pregnancy had a cumulative cancer incidence of 0.9 per 1,000 at 96 months of age,
compared with 1.4 per 1,000 for children of women
who did not smoke (p = 0.15). The corresponding
hazard ratio was 0.67 (95 percent confidence interval
(CI) 0.38-1.17). There was no evidence of a doseresponse association with smoking (hazard ratio for
one to 10 cigarettes = 0.45, for more than 10 cigarettes = 0.83). Children of smokers were not at increased risk of developing leukemia (0.82, 95 percent
CI 0.31-2.11) or other cancers (0.60, 95 percent CI
0.30-1.20). There were too few cases of individual
types of nonleukemic cancer for analysis.
Most of the characteristics in table 1 exhibited neither practically nor statistically significant associations
with cancer. However, mothers of children with cancer
1030
Klebanoff et a).
TABLE 1. Maternal and Infant characteristics associated wHh smoking during pregnancy,
Collaborative Perinatal Project, 1959-19661
Characteristic
Smokers
Nonsmokers
No.
%
No.
%
Race*
White
Black
Other
14,613
12,317
1,543
51
43
5
10,265
12,098
2,470
40
51
10
Education (years completed)*
<12
12
>12
17,055
8,174
3,244
60
29
11
13,726
7,964
4,143
53
31
16
7,345
6,122
4,679
3,394
6,815
26
22
16
12
24
7,700
5,481
3,828
2,742
5,997
30
21
15
11
23
Diagnostic radiation In pregnancy*
None
Any
Pelvimetry
11,892
16,651
1,611
42
58
6
9,353
16,465
1,704
36
64
7
Type of feeding in hospital*
Breast only
Mixed
Formula
No data
1,403
2,003
21,233
3,834
5
7
75
14
1,867
2,367
17,895
3,704
7
9
69
14
Infant sex
Male
Female
14,451
13.980
51
49
13,075
12,716
51
49
Prior pregnancies*
0
1
2
3
:>4
Mean (SO*)
Mean (SD)
Maternal age (years)
24.1 (5.8)*
24.3 (6.2)*
Education (years)
10.5 (2.4)*
10.7 (2.7)*
Sodoeconomlc lndex§
Maternal height (cm)
Prepregnant weight (kg)
Infant blrthwelght (g)
GestatJonal age (weeks)
4.6 (2.1)*
4.7 (2.2)*
161-2(6.8)*
160.7 (7.0)*
58.5(11.2) *
3,069
(581)'
38.6 (3.2)
59.5 (12.0)*
3,212
(576)*
38.6 (3.1)
• p < 0.001.
t Median date of delivery for smokers, February 4, 1963, and for nonsmokers, February 14, 1963;
median age at last contact for smokers, 87 months, and for nonsmokers, 91 months (p < 0.001).
t SD, standard deviation.
§ Myrianthopoulos NC, French KS. An application of the US Bureau of the Census sodoeconomlc index
to a large, diversified patient population. Soc Sd Mod 1968;2:283-99.
had slightly greater prepregnant weights than mothers
of children without cancer (63.0 vs. 59.1 kg; p =
0.052), and children with cancer were born earlier in
the study than children without cancer (April 8, 1962,
vs. February 2, 1963; p = 0.02). Infant birth weight
was not associated with childhood cancer (p = 0.45).
Because of the small number of cases, the hazard ratio
for cancer among children of smokers compared with
children of nonsmokers was adjusted in sequence for
each of the factors in table 1 except infant birth weight
and gestational age, as these outcomes are themselves
strongly and probably causally associated with smoking. Results of these adjustments are shown in table 2.
Adjustment had little effect on the hazard ratio, although increased maternal prepregnant weight and
earlier delivery date were significantly associated with
increased risks of childhood cancer in these analyses
(p = 0.04 for both associations).
Am J Epidemiol
Vol. 144, No. 11, 1996
Maternal Smoking and Childhood Cancer
0
6
12
18
24
30
36
42
48
64
60
66
72
78
84
90
1031
96
Age (months)
Non-smoker
Smoker
FIGURE 1. Cumulative incidence of cancer among children of women who did and did not smoke during pregnancy, Collaborative Perinatal
Project, 1959-1966.
At registration, women who currently were nonsmokers were not asked about smoking since their last
menstrual period. However, they were asked if they
had smoked at least 100 cigarettes, and those who had
were asked their age when they stopped. By comparing this age with their age at study registration, we
TABLE 2. Adjusted hazard ratio* for cancer among children
of women who smoked during pregnancy compared with
children of nonsmoking women, Collaborative Perinatal
Project, 1959-1966
Characteristic
controlled
Hazard
ratio tor
smokers
96% Cl*
V8.
nonsmokere
Maternal racet
Maternal age*
Maternal education*
Maternal soctoeconornic Index*
Maternal height*
Maternal prepregnancy weight*
Prior pregnancies*
Diagnostic radiation during
pregnancy§
PeMmetry§
Infant feeding In hospital!)
Infant sex#
Date of delivery*
0.63
0.67
0.67
0.67
0.61
0.69
0.67
0.36-1.11
0.38-1.17
0.38-1.18
0.38-1.17
0.34-1.08
0.39-1.21
0.38-1.17
0.67
0.66
0.68
0.66
0.67
0.38-1.17
0.38-1.16
0.39-1.19
0.38-1.17
0.38-1.17
* a , confidence Interval.
t Coded as white, black, other.
* Continuous variable.
§ Coded as any, none.
U Coded as bottle only, breast only, mixed, unknown.
* Coded as male, female.
Am J Epidemiol
Vol. 144, No. 11, 1996
estimate that the occurrence of any smoking during
pregnancy might have been as high as 58 percent.
When women whose age at stopping was the same or
1 year less than their current age were reclassified as
smokers, the hazard ratio was even less indicative of a
positive association between maternal smoking and
childhood cancer (hazard ratio 0.57, 95 percent CI
0.33-1.005).
Data were not collected on maternal smoking after
the study child's birth, and some of the children unexposed in utero to maternal smoking might have been
exposed postnatally. However, comparing smoking
status during successive pregnancies with the 8,442
women who had second CPP pregnancies showed that
88 percent of women who did not smoke at any time
during their first pregnancy also did not smoke at any
time during their second, and that 91 percent of those
who smoked during their first pregnancy also smoked
during their second. Although the women's smoking
habits may have changed due to the pregnancy itself,
it therefore seems unlikely that a large fraction of
children unexposed to maternal tobacco smoke prenatally was exposed postnatally.
DISCUSSION
This study found that children of women who
smoked during pregnancy were not at increased risk of
developing cancer in their first 8 years; the upper
confidence limits indicate that even moderate in-
1032
Klebanoff et al.
creases in risk are unlikely. Although there have been
many case-control studies of maternal smoking and
childhood cancer, there have been only three other
cohort studies. These cohort studies reported relative
risks of 2.47 (3), 1.3 (4), and 0.99 (5). When the four
cohort studies were combined in a random effects
model (13), there was significant heterogeneity between studies (Q statistic = 8.8, df = 3, p < 0.05), and
a summary effect measure would not be meaningful.
The CPP was not designed to study childhood cancer. Therefore, several indirect mechanisms were used
to identify cases, and underascertainment is possible.
Based on national age-, race- and sex-specific incidence data for 1980-1984 (14), the expected probability of developing cancer by age 8 years in a cohort
with the racial makeup of the CPP is 1.2 per 1,000,
which is nearly identical to our observed incidence of
1.1 per 1,000, suggesting that there were very few
missed cases.
Nearly 20 percent of study children had younger
siblings who were also enrolled in the CPP. Due to the
rarity of childhood cancer and the lack of any instances of siblings developing cancer, the potential
nonindependence of the data is not likely to influence
the results. To verify this, the crude hazard ratio was
adjusted, utilizing SUDAAN (15) software, for the
possible concordance of risk among siblings. As the
corrected 95 percent confidence limits for the hazard
ratio were 0.37 and 1.19, virtually identical to the
limits of 0.38 and 1.17 observed when nonindependence was ignored, this was not addressed further.
The CPP did not collect data on paternal smoking or
on smoking by other members of the mother's household or at her workplace. Paternal smoking has been
associated with childhood cancer in some case-control
studies (16) and might exert an effect either through
damage to sperm or through passive exposure of the
mother. Although unlikely to have much additional
effect on the exposure of women who already smoke,
maternal passive smoke exposure would result in misclassification of women with some smoke exposure as
being unexposed. Such misclassification might bias an
elevated risk of maternal smoking toward the null.
If the fetus or child were vulnerable to tobaccoinduced cancer only at a specific point in development, the definition of smokers as women who reported smoking at any interview during pregnancy
might be too crude. However, of the women who were
nonsmokers at their initial visit, only 3,120 (11 percent) reported smoking at one or more subsequent
visits. In addition, 81 percent of the women who
smoked at their initial visit reported smoking at all
subsequent visits. It is also probable that most women
who smoked during pregnancy smoked postnatally,
and most women who did not smoke during pregnancy
did not smoke postnatally. Therefore, the vast majority
of the women who smoked are likely to have smoked
during the entire pregnancy and postnatal periods,
minimizing the possibility of misclassification of maternal smoking during a time when the embryo or child
was vulnerable to carcinogenesis.
As in all studies relating prenatal exposures to postnatal outcomes, the role of pregnancy and postnatal
losses should be considered. Spontaneous abortions
comprise 30 percent of all pregnancies (17), and maternal smoking is associated with an increased risk of
spontaneous abortion (18) and infant mortality (19).
An increased risk of mortality from nonmalignant
causes among fetuses and children destined to develop
cancer had they survived could distort the association
between prenatal tobacco exposure and childhood cancer (20). As it cannot be determined which fetuses or
children dying of other causes would have developed
cancer had they survived, this hypothesis cannot be
tested.
Childhood cancer is very different from cancer in
adults. Epithelial tumors comprise the majority of cancers in adults compared with a small minority of
tumors in children of the age range studied (21). As
smoking is associated primarily with epithelial tumors
in adults, it may not be biologically plausible to expect
a positive association between in utero tobacco exposure and childhood cancer. However, prenatal and
early childhood tobacco exposure may be associated
with an increased risk of cancer in adulthood (22); as
this cohort was not followed to adulthood, this hypothesis could not be tested.
In summary, this study found that although the small
number of cases limits the ability to draw conclusions
regarding specific types of cancer, children exposed in
utero to maternal smoking were not at increased risk of
cancer in general. However, maternal smoking during
pregnancy is associated with an increased risk of a
variety of adverse outcomes, and there are many other
valid reasons for pregnant women to refrain from
smoking.
ACKNOWLEDGMENTS
The authors are thankful to Dr. Rebecca DerSimonian for
calculating the pooled relative risk, confidence intervals,
and heterogeneity of the prospective studies.
REFERENCES
1. Fielding JE. Smoking: health effects and control. N Engl
JMed 1985;313:491-8.
Am J Epidemiol Vol. 144, No. 11, 1996
Maternal Smoking and Childhood Cancer
2. Tre'daniel J, Boffetta R, Little J, et al. Exposure to passive
smoking during pregnancy and childhood, and cancer risk: the
epidemiologic evidence. Paediatr Perinat Epidemiol 1994;8:
233-55.
3. Golding J, Paterson M, Kinlen LJ. Factors associated with
childhood cancer in a national cohort study. Br J CanceT
1990;62:304-8.
4. Neutel CI, Buck C. Effect of smoking during pregnancy on the
risk of cancer in children. J Natl Cancer Inst 1971;47:59-63.
5. Pershagen G, Ericson A, Otterblad-Olausson P. Maternal
smoking in pregnancy: Does it increase the risk of childhood
cancer? Int J Epidemiol 1992;21:l-5.
6. Niswander KR, Gordon M. The women and their pregnancies.
Philadelphia, PA: W. B. Saunders, 1972.
7. Myrianthopoulos NC. Malformations in children from one to
seven years: a report from the Collaborative Perinatal Project
New York, NY: Alan R. Liss, 1985.
8. Lassman FM, Fisch RO, Vetter DK, et al. Early correlates of
speech, language, and hearing. Littleton, MA: PSG Publishing
Company, 1980:37-56.
9. Klebanoff MA, Read JS, Mills JL, et al. The risk of childhood
cancer following neonatal vitamin K exposure. N Engl J Med
1993;329:905-8.
10. Read JS, Clemens JD, Klebanoff MA. Moderate low birth
weight and infectious disease mortality during infancy and
childhood. Am J Epidemiol 1994;140:721-33.
11. Shiono PH, Chung CS, Myrianthopoulos NC. Preconception
radiation, intrauterine diagnostic radiation, and childhood neoplasia. J Natl Cancer Inst 1980;65:681-6.
12. Kalbfleisch JD, Prentice RL. The statistical analysis of failure
time data. New York, NY: John Wiley and Sons, 1980:134-5.
Am J Epidemiol
Vol. 144, No. 11, 1996
1033
13. DerSimonian R, Laird N. Meta-analysis in clinical trials.
Control Clin Trials 1986;7:177-88.
14. Robison LL. General principles of the epidemiology of childhood cancer. In: Pizzo PA, Poplack DG, eds. Principles and
practice of pediatric oncology. Philadelphia, PA: J. B. Lippincott, 1993:3-10.
15. Shah BV, Bamwell BG, Bieler GS. SUDAAN user's manual,
release 6.40. Research Triangle Park, NC: Research Triangle
Institute, 1995.
16. John EM, Savitz DA, Sandier DP. Prenatal exposure to parents' smoking and childhood cancer. Am J Epidemiol 1991;
133:123-32.
17. Wilcox AJ, Weinberg CR, O'Connor JF, et al. Evidence of
early loss of pregnancy. N Engl J Med 1988;319:189-94.
18. Harlap S, Shiono PH. Alcohol, smoking, and incidence of
spontaneous abortions in the first and second trimester. Lancet
1980;2:173-6.
19. Malloy MH, Kleinman JC, Land GF, et al. The association of
maternal smoking with age and cause of infant death. Am J
Epidemiol 1988;128:46-55.
20. Hook EB, Regal RR. Conceptus viability, malformation and
suspect mutagens or teratogens in humans: the Yule-Simpson
paradox and implications for inferences of causality in studies
of mutagenicity or teratogenicity limited to human livebirths.
Teratology 1991;43:53-9.
21. Miller RW, Myers MH. Age distribution of epithelial cancers.
(Letter). Lancet 1983;2:1250.
22. Sandier DP, Everson RB, Wilcox AJ, et al. Cancer risk in
adulthood from early life exposure to parents' smoking. Am J
Public Health 1985;75:487-92.