International Journal of Epidemiology, 2014, 1355–1366
doi: 10.1093/ije/dyu160
Reprints and Reflections
Reprints and Reflections
The relationship of parents’ cigarette smoking
to outcome of pregnancy—implications as to
the problem of inferring causation from
observed associations1
J Yerushalmy
Child Health and Development Studies, Division of Biostatistics, School of Public Health, University of
California, Berkeley; the Kaiser Foundation Research Institute; and the Permanente Medical Group,
Oakland, California.
Supported by Grant No. HD-00718 of the National Institutes of Health.
The author is indebted to Dr. Bea J. van den Berg for assistance in analysis, to Mrs. Dorothy. Friedman who performed
the programming and computer tabulations, and to Mr. Paul Shalmy for assistance in organization and in methods of
presentation.
Abstract
The relationship of parents’ cigarette smoking to outcome of pregnancy—implications as
to the problem of inferring causation from observed associations. Amer J Epidem
1971;93:443–456. Nearly 10 000 white and more than 3000 black women were interviewed early in pregnancy on a variety of medical, genetic, environmental, and behavior
variables. The increase in the incidence of low-birth-weight among infants of smoking
mothers was confirmed. However, a number of paradoxical findings were observed
which raise doubts as to causation. Thus, no increase in neonatal mortality was noted.
Rather, the neonatal mortality rate and the risk of congenital anomalies of low-birthweight infants were considerably lower for smoking than for nonsmoking mothers.
These favorable results cannot be explained by differences in gestational age, nor does a
“displacement” hypothesis appear reasonable. Among other findings which could not
easily be explained: The healthiest low-birth-weight infants were found for couples
where the wife smoked and her husband did not smoke; the most vulnerable were produced by couples where the wife did not smoke and the husband smoked. There were
great differences in mode-of-life characteristics between smokers and nonsmokers. The
latter were more likely to use contraceptive methods, to plan the baby, less likely to drink
coffee and hard liquor, and in general appeared to live at a much slower and moderate
pace than the smokers. Most puzzling difference is that of age at menarche, which was
1 Yerushalmy J. The relationship of parents’ cigarette smoking
to outcome of pregnancy—implications as to the problem of
inferring causation from observed associations. American
Journal of Epidemiology 1971;93:443–456. Reprinted with
permission
1355
1356
International Journal of Epidemiology, 2014, Vol. 43, No. 5
lower for smoking mothers. These paradoxical findings raise doubts and argue against
the proposition that cigarette smoking acts as an exogenous factor which interferes with
intrauterine development of the fetus.
Key words: birth weight, low; congenital anomalies; contraception; menstruation, menarche; mortality, neonatal;
pregnancy; prenatal influences; smoking
Introduction
The search for identification of causes of chronic diseases
and conditions often starts with uncontrolled, or poorly
controlled, observational studies. A well-controlled experimental study on human beings usually entails long-term
changes in personal habits and behavior patterns of a large
number of people. Such a task is formidable if not impossible-even the most cooperative human beings are not that
malleable. Instead, research on the risks associated with
environmental factors such as smoking, diet, emotional
stress begins with observations of self-formed subgroups as
found in the population (e.g. smokers, nonsmokers, and
past smokers). These subgroups seldom satisfy the basic
rule for valid scientific inference—that, a priori, groups
being compared be alike in all pertinent characteristics.
Observational studies, therefore, rarely establish with reasonable certainty that the observed differences are due to
the characteristics under investigation and not to extraneous factors which differentiate the groups. They provide
useful leads but not definitive conclusions about causes.
The phenomenon of self-selection is the root of many of
the difficulties. Were all other complications eliminated,
the inequalities between groups which result from selfselection would still leave in doubt inferences on causality.
For example, in the study of the relationship of cigarette
smoking to health, if we assume well-conducted investigations in which, (a) large random samples of the population
have been selected and the individuals correctly identified
as smokers, nonsmokers, or past smokers, (b) that the
problem of non-response did not exist, (c) that the population had been followed long enough to identify all cases of
the disease in question, (d) that no problems of misdiagnosis and misclassification existed, (e) and that no one in
the population had been lost from observation, then even
under these ideal conditions, the inferences that may be
drawn from the study are limited because the individuals
being observed, rather than the investigator, made for
themselves the crucial choice: smoker, nonsmoker, or past
smoker.
These shortcomings notwithstanding, observational
studies remain the first line of attack in probing for causes
of chronic diseases and conditions. What is needed, however, is to develop auxiliary and complementary methods
which would help overcome their built-in limitations.
To date, epidemiologists and biostatisticians have not responded fully to this challenge. We have made only a meager beginning in refining the tools and re-defining the
standards that would aid the process of inferring causation
from observed associations. It may be hoped that the time
is approaching when we can undertake this task more
calmly and constructively than has been the case for the
past few years.
A helpful first step is to see if the observed association is
specific to the disease under study. For it is intuitively recognized that if the association is specific there is a greater
likelihood of an etiologic relationship than if it is not. An
association which is nonspecific, diffuse, manifesting itself
in a large and varied number of diseases, increases the
probability that the observed differences are due to extraneous factors which differentiate the self-formed groups. It
then becomes crucial to study in great detail the differences
and similarities of persons with the environmental factor
under study and those without it.
Obviously, specificity of association does not by itself
prove causation and lack of specificity does not prove that
the observed association is necessarily not one of cause and
effect.
The consideration of specificity is only a first step.
Many other auxiliary methods of study and newer methodologies need to be developed which would help in coming
to grips with these problems and assist in arriving closer
to an understanding of, and solution to, the problem of
inferring causation from observed association.
Smoking and pregnancy
In the last decade a number of investigators reported that
women who smoked gave birth to a larger proportion of
prematures than women who did not smoke.1–10 In most
studies “prematures” were defined by the birth weight criterion, and referred to low-birth-weight infants weighing
2500 grams or less. Most of the investigators stopped with
this observation. The implication was serious, for it is well
known that infants weighing 2500 grams or less suffer a
neonatal mortality rate—the risk of dying in the first
month of life—which is more than 20 times as high as that
International Journal of Epidemiology, 2014, Vol. 43, No. 5
of heavier infants. For example, the only set of available
national data of the relation of birth weight to mortality
reported a neonatal mortality rate of around 7.8 per 1000
for infants weighing more than 2500 grams at birth, and a
rate of 174 per 1000 for infants of low-birth-weight.11 The
finding that nearly twice as many infants of smoking mothers are of low-birth-weight than are those of nonsmoking
mothers implies that infants of smoking mothers suffer
considerably higher neonatal mortality rates.
Several years ago we put this implication to the test
after accumulating a sufficient number of pregnancies in
our Child Health and Development Studies.12 We confirmed the observation of an increased proportion of lowbirth-weight infants for smoking mothers. But we explored
the problem further and found no increase in the neonatal
mortality rate, and encountered a number of other paradoxes. We are now in a position to investigate the phenomenon in greater detail on a larger sample of pregnancies.
The purpose of this paper is to investigate the relationship of parental smoking to outcome of pregnancy and to
other variables. The findings are used also to illustrate the
problems of inferring causation from observed associations
by investigating differences between smokers and nonsmokers along a number of environmental, behavior, and
biologic characteristics.
Material and method
The information is derived from our Child Health and
Development Studies (CHDS)—a comprehensive investigation of all pregnancies that occurred between 1960–1967,
among women in the Kaiser Foundation Health Plan in the
San Francisco-East Bay area. The Kaiser Health Plan is a
prepaid medical care program. The members represent a
broadly based group which is not atypical of an employed
population. It is deficient only in the two extremes:
the very affluent and the very indigent portions of the
population.
The women were interviewed on a variety of medical,
genetic, and environmental subjects, including behavior
variables, such as smoking, drinking, use of contraceptive
methods and the like. The interview took place early in
pregnancy. The information was thus derived prospectively
before the woman knew the outcome of the pregnancy.
The child was followed to evaluate his physical and mental
development including survival and the development of
congenital anomalies.
The interviewed group comprised some 15 000 pregnancies. This study is based on single live born infants
among the whites and the blacks. The numbers of members
in the other ethnic groups in the sample were too small and
1357
Table 1. Per cent of white and black single liveborn infants
weighing 2500 grams or less (low-birth-weight) according to
smoking status of mother
Smoking Status
All mothers
Nonsmoker
Smoker
Nonsmoker:
Never smoked
Past smoker
No of liveborn
infants
% with birth weight
2500 gm
White
Black
White
9793
6067
3726
3290
2219
1071
4.4
3.2*
6.4*
7.9
5.8*
12.3*
4252
1815
1763
456
3.2
3.5
6.0
5.0
Black
*Difference between nonsmoker and smoker statistically significant:
p < .00001.
were therefore left out. The study is based on 9793 pregnancies among the whites and 3290 among the blacks.
Incidence of low-birth-weight
Table 1 and figure 1 show the incidence of lowbirth-weight among infants of smoking and nonsmoking
mothers. Smokers include all women who were smoking
one or more cigarettes per day during the pregnancy.
Nonsmokers include those who never smoked and those
who stopped smoking either before or during the pregnancy. It is seen that women who smoked had approximately twice as many low-birth-weight infants per 1000
single live births as women who did not smoke. The differences are statistically highly significant (p < .00001). (All
statements of significance in this paper are derived by the
x2 test.) There was a regular increase with the number of
cigarettes smoked. The more cigarettes per day the women
smoked, the larger the proportion of low-birth-weight
infants. Women who were past smokers, that is, those who
stopped smoking, were much like those who never
smoked, with respect to the proportion of low-birth-weight
infants. Almost identical relationships were found for
whites and for blacks. These are the striking findings which
corroborate those of a number of other investigators.
Exploring the problem further, the neonatal mortality
rates of infants of smoking mothers were compared with
those of the nonsmoking mothers (table 2). There was no
difference in the risk of early death between the two groups
among whites. The rate was somewhat but not significantly higher for smokers among black gravidas.
The next step was to compare the neonatal mortality
rate of the low-birth-weight infants of smoking and nonsmoking mothers. It is seen in table 3 and figure 2 that the
low-birth-weight infants of smoking mothers survived
1358
International Journal of Epidemiology, 2014, Vol. 43, No. 5
Figure 1. Per cent of single liveborn infants weighing 2500 grams or less (low-birth-weight) according to smoking status of mother. (1–4, 5–14, and
15þ refer to number of cigarettes smoked.)
Table 2. Neonatal mortality rates for single liveborn infants
Table 3. Neonatal mortality rates for single liveborn low-
according to smoking status of mother
birth-weight infants according to smoking status of mother
Smoking Status
Smoking Status Single liveborn infants weighing 2500 gm at birth
Neonatal mortality rate per 1000 live births
White
Black
All mothers
11.1
18.5
Nonsmoker
11.0
17.1
Smoker
11.3
21.5
Never smoked
10.3
16.4
Past smoker
12.7
19.7
Nonsmoker:
No
Neonatal Deaths
No
White
Rate per 1000
Black
White
Black
White
Black
All mothers
434
261
70
41
161.3
157.1
Nonsmoker
197
129
43
26
218.3*
201.6
Smoker
237
132
27
15
113.9*
113.6
134
106
28
22
209.0
207.5
63
23
15
4
238.1
173.9
Nonsmoker:
Never smoked
considerably better than the low-birth-weight infants
of nonsmoking mothers. The difference is significant
for whites (p < .005), and approaches significance
(.05 < p < .06) for blacks.
The lower rate for the low-birth-weight infants of
smokers explains why the total neonatal mortality rate
of infants of smoking mothers is not higher than that of
infants of nonsmoking mothers. However, it presents a
puzzling phenomenon—presumably, mother’s cigarettesmoking increases the proportion of low-birth-weight infants but these low-birth-weight infants are much healthier
than those of the nonsmokers. Further investigations are
therefore indicated.
First, the answer might lie with pregnancies that terminated in fetal deaths, especially early fetal deaths, or abortions. William F. Taylor studied our fetal deaths and among
other variables included also that of smoking.13 Figure 3
presents by lunar month of pregnancy, the probabilities of a
Past smoker
Difference statistically significant (p < .005).
fetal death in that month and in all subsequent months for
infants of white smokers and nonsmokers. It is seen that
there was no difference in these probabilities between the
two groups.
Another important hazard to which low-birth-weight
infants are especially exposed is that of severe congenital
anomalies. A comparison of this risk for low-birth-weight
infants of smoking and nonsmoking mothers is shown in
table 4. It is seen that the risk was greater for infants of
nonsmoking mothers compared to that of infants of smoking mothers. The difference is significant for whites
(.02 > p > .01) but not for blacks.
A natural question arises whether the differences between the smoking and nonsmoking groups are confined
to the low-birth-weight portion of the distribution or
International Journal of Epidemiology, 2014, Vol. 43, No. 5
1359
Table 4. Children with severe congenital anomalies per 1000
single low-birth-weight infants according to smoking status
of mother
Smoking Status
Anomaly rates per 1000
Nonsmokers
Smoker
Nonsmokers:
Never smoked
Past smoker
White
Black
147.2*
71.7*
54.3
45.5
119.4
206.3
37.7
130.4
*Difference statistically significant p < .02.
Figure 2. Neonatal mortality rates for single liveborn low-birth-weight
infants according to smoking status of mother.
Figure 3. Probability of fetal death in specified month of pregnancy and
all subsequent months according to smoking status of white gravidas.
(Source: W.F. Taylor, reference 13.)
Figure 4. Per cent distribution of single liveborn infants by birth weight
according to smoking status of mother.
Table 5. Neonatal mortality rates by birth weight of single
whether it is a more generalized phenomenon relating to
the entire birth weight distribution. Figure 4 shows that
the latter is the case. The entire distribution by birth weight
is different for smokers than for nonsmokers for both
white and black mothers. Infants of smoking mothers
weigh less than infants of nonsmoking mothers.
Table 5 presents the neonatal mortality rate of infants
of smoking and nonsmoking mothers in the different birth
weight groups. It is seen that the favorable rates of infants
of smoking mothers were not confined to the group of
low-birth-weight. They were found in all but one of the
birth weight groups among the whites; among the blacks it
is noted especially in the low-birth-weight groups. In both
liveborn infants according to smoking status of mother
Birth weight (grams) Neonatal mortality rates per 1000
White
Black
Nonsmoker Smoker Nonsmoker Smoker
1500
1501–2000
20011–2500
2501–3000
3001–3500
3501þ
Total
791.7
406.3
78.0
11.6
2.2
3.8
11.0
565.2
346.2
26.6
6.1
4.5
2.6
11.3
740.7
166.7
25.6
4.2
4.3
8.7
17.1
666.7
45.5
21.7
12.6
4.8
9.8
21.5
1360
International Journal of Epidemiology, 2014, Vol. 43, No. 5
Figure 5. Per cent distribution of single liveborn infants by gestational age according to smoking status of mother.
Table 6. Mean birth weight by duration of gestation of single liveborn infants according to smoking status of mother
Gestation (weeks)
Mean birth weight (grams)
White
Black
Nonsmoker a
Smoker b
Difference c ¼ ab
Nonsmoker a’
Smoker b’
Difference c’ ¼ a’b’
2026.0
2848.3
2877.7
2969.6
3166.7
3267.6
3404.4
3529.4
3609.9
3672.5
3650.6
3464.4
1834.1
2573.9
2680.1
2689.0
2932.9
3082.5
3233.4
3347.5
3445.9
3482.9
3449.3
3255.3
191.9
274.4
197.6
280.6
233.8
185.1
171.0
181.9
164.0
189.6
201.3
209.1
2155.5
3093.3
2965.4
3037.3
3079.8
3182.0
3282.9
3399.0
3488.7
3509.2
3405.6
3263.4
2080.3
2567.1
2629.2
2820.1
2970.4
2950.3
3134.8
3252.4
3282.3
3295.5
3311.3
3058.6
75.2
526.2
336.2
217.2
109.4
231.7
148.1
146.6
206.4
213.7
94.3
204.8
33
34
35
36
37
38
39
40
41
42
43þ
Total
racial groups the advantage to low-birth-weight infants of
smoking mothers was not confined to the group as a whole
but was present in each of the subgroups, namely among
the very small infants weighing 1500 grams or less, among
those weighing 1501–2000 grams, and among those
weighing 2001–2500 grams.
Length of gestation
Another variable which must be considered in this connection is the length of pregnancy. It is possible that
low-birth-weight infants of smoking mothers are “more
mature” than are low-birth-weight infants of nonsmoking
mothers and this may explain their lower mortality.
Infants weighing 2500 grams or less of smoking mothers
have been in utero, on the average, three days longer than
the low-birth-weight infants of nonsmoking mothers
(253.3 days compared to 250.4 days). It is therefore
necessary to explore this question further and investigate
whether this is a reasonable explanation for their favorable
mortality.
Figure 5 shows that the distribution by weeks of gestation was approximately the same for infants of smoking
and non-smoking mothers for both whites and blacks.
Apparently, therefore, there has not been a displacement as
far as length of pregnancy is concerned. Moreover, the
mean birth weight of infants of smoking mothers was
lower than that of nonsmoking mothers over the entire
span of the gestational age distribution (table 6). Infants of
smoking mothers weighed approximately 200 grams less
than those of nonsmoking mothers. The difference was of
about the same order of magnitude in nearly each week of
gestation.
When the low-birth-weight infants were separated
into those of less than 37 and those of 37 or more weeks
gestation, the neonatal mortality rate was higher for
International Journal of Epidemiology, 2014, Vol. 43, No. 5
1361
nonsmokers than for smokers in both groups. It was significantly higher for low-birth-weight infants of less than
37 weeks only for whites (table 7). The favourable rates of
low-birth-weight infants of smoking mothers can therefore
not be easily explained on the gestational age hypothesis.
Among the interpretations which imply a causative hypothesis, perhaps the most attractive is that mother’s
smoking causes a displacement from higher to lower birth
weights, and this displacement, or shift, is responsible for
the observed excess of infants of “low-birth-weight” and
for their favorable early mortality. It may, therefore, be of
interest to review in greater detail the consequences of this
hypothesis to determine whether they are reasonable.
It follows from this hypothesis that if the smoking
mothers in the study had not acquired the smoking habit,
their infants would have been distributed by weight and
survival as were the infants of the nonsmoking mothers in
this study. It is therefore possible to determine for each
birth weight group the number of births and deaths which
would have occurred to the smoking mothers if they did
not smoke. We refer to these as the “expected” as distin-
guished from the “displaced,” births. The excess of the
observed numbers of births and deaths among smokers
over these expected numbers provides an estimate of the
magnitude and early mortality of the births which were
presumably “displaced” because of mother’s smoking.
Table 8 provides such an evaluation for white infants
weighing 2500 grams or less at birth. There were a total of
237 infants of smokers in this birth weight group. The “expected” number of low-birth-weight infants among the
3729 smokers, that is, the number which would have
occurred among them if they were nonsmokers, is 121.5.
There were 27 deaths observed among the .237 low-birthweight infants of smoking mothers; the “expected” number of deaths among the “expected” 121.5 infants of
low-birth-weight under the neonatal mortality rates for
nonsmokers is shown in the table to be 26.6.
Consequently, if the causal hypothesis is correct, a total of
115.5 (237–121.5) infants were “displaced,” because of
their mothers’ smoking, into lower birth weight groups,
among whom there was only .4 of one death (27–26.6).
This may be compared to the 26.6 deaths among the 121.5
infants of low-birth-weight who were not displaced but belong in their birth weight group on the basis of the rates
among nonsmokers.
If it is assumed, on the basis of table 6, that smoking reduces the birth weight by some 200 grams, it is possible to
test whether these findings are reasonable. Table 9 shows
the expected number of deaths in the presumed
“displaced” group if they died under the mortality rates of
the group from which they were displaced, namely in each
case the next higher birth weight group, weighing 250
grams more. It is seen that under these conditions 7.5 additional deaths are expected while only .4 occurred. Similar
findings were obtained for black gravidas. The “displacement” hypothesis is therefore unreasonable.
Table 7. Neonatal mortality rates of single low-birth-weight
infants of less than 37 and of 37 or more weeks gestation
according to smoking status of mother
Smoking Status
<37 weeks
37 weeks
336.3*
172.1*
59.5
52.2
279.1
181.8
46.5
18.2
Whites
Nonsmoker
Smoker
Blacks
Nonsmoker
Smoker
*Difference statistically significant at p < .005.
Table 8. Observed, “Expected”* and “Displaced”* single white low-birth-weight infants and neonatal deaths among 3729 smoking mothers
Birth weight
(grams)
Relative
frequency among
nonsmokers (%)
(a)
“Expected”
frequency among
3729 smokers
ðaÞ
ðbÞ ¼ 3729
100
Observed
frequency among
the 3729
smokers (c)
1500
1501–1750
1751–2000
2001–2250
2251–2500
All 2500
.40
.25
.28
.68
1.65
14.9
9.3
10.4
25.4
61.5
121.5
23
11
15
55
133
237
*See text for definition of categories.
“Displaced”*
births (d) ¼
(c) – (b)
8.1
1.7
4.6
29.6
71.5
115.5
Neonatal mortality rates among
nonsmokers (e)
791.67
466.67
352.94
97.56
70.00
Estimate of
neonatal deaths
among
“Expected”
births
Þ ðeÞ
ðfÞ ¼ ðb1000
11.8
4.3
3.7
2.5
4.3
26.6
1362
International Journal of Epidemiology, 2014, Vol. 43, No. 5
Table 9. Estimated number of neonatal deaths of infants
Table 10. Per cent of single liveborn low-birth-weight infants
“Displaced”* to the next lower birth weight groups on the
according to smoking status of gravidas and their husbands
basis of the neonatal mortality of the infants of nonsmoking
mothers in their presumed original birth weight groups
Birth weight
(grams)
1500
1501–1750
1751–2000
2001–2250
2251–2500
“Displaced”* Neonatal
births (a)
mortality rates
of non-smokers
in next higher
birth weight
group (b)
8.1
1.7
4.6
29.6
71.5
115.5
466.67
352.94
97.56
70.00
8.85†
Expected
deaths among
“Displaced”*
ÞðbÞ
ðcÞ ¼ ða1000
3.8
0.6
0.4
2.1
0.6
7.5
*See text for definition.
†
This is the neonatal mortality rate for nonsmokers in the 2501–2750 gram
birth weight.
Figure 6. Per cent of low-birth-weight infants according to smoking status of husbands.
Husband’s smoking and outcome of pregnancy
We turn next to explore the relationship of husband’s
smoking to outcome of pregnancy. The a priori expectation is for independence, because it is difficult to visualize
a biologic mechanism for an association. We first inquire
into the association with incidence of low-birth-weight infants. Figure 6 shows an increase in the proportion of lowbirth-weight infants of smoking compared to nonsmoking
husbands. This is not of the same order of magnitude as is
found for the mothers but nevertheless it is significant
(p < .004 for whites, and p < .02 for blacks). This may, of
course, be a reflection of the association with mother’s
smoking since husband’s and wife’s smoking habits are
Smoking status of husband
White
Nonsmoker
Smoker
Black
Nonsmoker
Smoker
Smoking status of gravida
Nonsmoker
Smoker
3.43
3.10
4.83*
6.74*
5.80
6.11
9.50
13.41
*Difference statistically significant at p < .05.
correlated. When the incidence of low-birth-weight infants
was studied by the smoking habits of both gravida and
husband it was found that while there was no association
with husband’s smoking when the gravida does not smoke,
there was a significant association for whites (p < .05) but
not for blacks with husband’s smoking when the
gravida does smoke. The increase in the proportion of lowbirth-weight infants was found primarily when both husband and wife smoke (table 10).
Of interest is the fact that the relationship of husband’s
smoking to neonatal mortality is different from that of
mother’s smoking. It will be recalled that for the latter
there was no difference in total neonatal mortality, and
that the neonatal mortality rate of the low-birth-weight
was considerably lower for the infants of smoking mothers. Figure 7 shows that the neonatal mortality rate of lowbirth-weight infants of smoking fathers was higher than
that of infants of nonsmoking fathers especially among the
blacks. The differences for the whites are not statistically
significant, but the fact that it operates inversely from that
found for mothers is interesting. Figure 8 shows that the
neonatal mortality rate for infants weighing 2500 grams or
less was highest when the mother did not smoke and the
husband smoked, and was very low when the mother
smoked and the husband did not smoke. The same relationship was found both for whites and for blacks.
Although the figures are too small in certain categories, the
fact that the same relationship was found for each racial
group cannot be ignored.
Comparison of smokers and nonsmokers
We turn next to a comparison of smokers and nonsmokers
along several environmental, behavior, and biologic
variables. Several investigators found marked differences
between smokers and nonsmokers. Thus, the former were
found to be more neurotic, to change jobs and spouses
more often than nonsmokers.14 They were shown to differ
in a number of personality traits.15–17 Perhaps the most
International Journal of Epidemiology, 2014, Vol. 43, No. 5
1363
Figure 9. Per cent not using contraceptive methods by smoking status
of mother. (1–4, 5–14, and 15þ refer to number of cigarettes smoked.)
Figure 7. Neonatal mortality rates of low-birth-weight infants according
to smoking status of husband.
Table 11. Per cent of parents who planned baby according to
smoking status of gravida and husband
Smoking habits
Gravida
Nonsmoker
Smoker
Husband:
Nonsmoker
Smoker
No with information
on planning
% who planned baby
White
Black
White
Black
5698
3450
2078
1028
43.0*
36.1*
22.4
20.4
3757
4349
955
1727
44.6*
37.2*
27.1*
19.8*
*Difference statistically significant at p < .0002.
Figure 8. Per cent of low-birth-weight infants according to smoking status of gravidas and their husbands.
interesting are the findings of Seltzer in a prospective
study of 922 Harvard alumni 13 years out of college whose
physical characteristics were recorded when they were
undergraduates. He found that smokers were significantly
differentiated from nonsmokers in morphological dimensions and proportions.18 He also found significant differences between cigarette smokers, pipe smokers, and cigar
smokers. He noted a statistically significant association between cigarette smoking habits and the strength of what
the anthropologists call the “masculinity component”.19
In our Child Health and Development Studies we
inquired of the gravida in the initial interview about a
number of characteristics in herself and her husband. It is
therefore possible to investigate the relationship of some of
these to smoking habits. Figure 9 shows the proportion not
using contraceptive methods. It is seen that it was much
higher among smoking than among nonsmoking mothers
(p < .001). Among whites, the proportion increases
quantitatively with the number of cigarettes smoked.
A similar relationship was found for husband’s smoking.
Table 11 shows that the proportion stating that the
baby was planned was higher for nonsmoking mothers as
well as for non-smoking husbands.
The profile of drinking habits was different for
smokers and nonsmokers for both gravidas and husbands
(figure 10). Thus, a significantly higher proportion of
smokers drink coffee, beer, and whiskey while the reverse
is true for tea, wine, and milk. The same relationship was
found for whites and blacks.
The association with drinking is found quantitatively
with the number of cigarettes smoked as is illustrated for
coffee in figure 11.
The mean number of drinks per drinker was higher for
smoking than for nonsmoking gravidas and the same for
husbands.
Perhaps the more telling difference is provided by the
proportion of “extremes.” Thus, those who drink relatively large amounts, for example, seven or more cups of
coffee per day or seven or more bottles of beer per week
and so on, were much more frequent among smokers than
1364
International Journal of Epidemiology, 2014, Vol. 43, No. 5
Table 12. Per cent of white gravidas and of husbands drinking
relatively large amounts, according to their smoking status
Smoking status
Gravida:
Nonsmoker
Smoker
Husband:
Nonsmoker
Smoker
7 or more
cups/glasses a day
7 or more
glasses a week
Coffee
Tea
Milk
Beer
7.5*
22.3*
0.4*
0.9*
0.7*
1.4*
5.9*
18.8*
0.3
0.5
2.5
2.9
Wine
Whiskey
2.0*
3.9*
5.9
6.4
2.1
4.9
10.1*
18.2*
9.5*
7.7*
5.1*
8.3*
*Difference statistically significant at p < .005.
Figure 10. Per cent of gravidas and husbands consuming different types
of beverages according to their smoking status.
Figure 12. Twinning rates per 1000 pregnancies according to smoking
status of mother. (1–4, 5–14 and 15þ refer to number of cigarettes
smoked.)
Figure 11. Per cent of gravidas drinking coffee according to their smoking status. (1–4, 5–14, and 15þ refer to number of cigarettes smoked.)
among nonsmokers as may be seen in table 12 for white
gravidas and husbands. The same relationship was found
for blacks.
These findings document the fact that there are major
differences in mode-of-life characteristics between smokers
and nonsmokers. The latter are revealed to be more moderate than the smokers, who are shown to be more extreme
and carefree in their mode-of-life.
Two additional variables which are more in the biologic
domain are available in our studies. These relate to the incidence of twinning and the age at menarche. Figure 12
shows the twinning rates per 1000 deliveries for smoking
and nonsmoking mothers. The rates among whites were
significantly higher for smokers (p < .004), and increased
with number of cigarettes smoked. No such relationship
was found among blacks.
An even more remarkable relationship is shown in
figure 13 which presents for smokers and nonsmokers the
Figure 13. Per cent of gravidas with menarche before age 12 according
to smoking status of mother. (1–4, 5–14, and 15þ refer to number of cigarettes smoked.)
proportion of gravidas who began to menstruate before
age 12. The smokers, in general, began to menstruate at
younger ages than the nonsmokers. The per cent menstruating before age 12 was higher for the smokers and
increased quantitatively with number of cigarettes smoked.
This was true for both whites and blacks. However, the
difference was significant only for whites (p < .0001).
International Journal of Epidemiology, 2014, Vol. 43, No. 5
1365
Figure 14. Reproductive performance of nonsmoking and smoking gravidas contrasted with that of tall and short gravidas.
This phenomenon of early menstruation can obviously not
be construed as having been caused by smoking because at
that age very few of the women smoked. In fact, there were
only 22 white and 4 black gravidas who stated that they
began smoking before age 12. When these are left out the
differences of age at menarche between smokers and nonsmokers persisted.
Comments
The following findings concerning the relationships of parents’ smoking to their reproductive performance cannot
easily be reconciled on a cause-effect basis: Smoking mothers have relatively more low-birth-weight infants but these
infants are much healthier than are low-birth-weight infants of nonsmoking mothers—the low-birth-weight infants of smoking mothers have lower neonatal mortality
rates and lower risks for severe congenital anomalies. On
the other hand, smoking husbands have a somewhat higher
incidence of low-birth-weight infants but the prognosis of
their low-birth-weight infants is poor—they have higher
neonatal mortality rates and higher risks for severe congenital anomalies. The healthiest low-birth-weight infants
are found for couples where the wife smokes and the
husband does not smoke. The most vulnerable lowbirth-weight infants are produced by couples in which the
wife does not smoke and the husband smokes.
Added to these are the great differences in mode-of-life
characteristics between smokers and nonsmokers. The latter
appear to live at a much slower and moderate pace than the
smokers. The two groups were found to be different also in
some biologic characteristics. White smoking gravidas had a
considerably higher twinning rate. Perhaps the most puzzling
difference is that of age of menarche for smoking and nonsmoking gravidas.
These paradoxical findings raise doubts and argue
against the proposition that cigarette smoking acts as an
exogenous factor which interferes with intrauterine development of the foetus.
It was thought of interest to investigate whether it
would be possible to duplicate some of the findings for
smoking and non-smoking gravidas in groups of women
who are differentiated along a specific biologic characteristic. The height variable was selected because it is known
that infants of short women are, on the average, smaller
than those of tall women. Consequently, the 30 per cent
shortest women were compared with the 30 per cent tallest
women in our Child Health and Development Studies.
Figure 14 shows a remarkable parallelism. Smoking and
nonsmoking women are differentiated in their reproductive
performance in an almost identical fashion as short
and tall gravidas. The difference in incidence of lowbirth-weight infants between smoking and nonsmoking
women was identical to that between short and tall
women. The same was true for many of the characteristics
found for low-birth-weight infants—mean length of gestation, neonatal mortality rate and rate for severe congenital
anomalies. Thus the low-birth-weight infants of short
mothers as those of smoking mothers had lower neonatal
mortality and postneonatal mortality and anomaly rates
than low-birth-weight infants of tall mothers or of nonsmoking mothers.
1366
This comparison is presented not as proof that the differences between smokers and nonsmokers are necessarily,
of biologic origin, rather it is to indicate that a biologic
hypothesis is not unreasonable.
One definite conclusion which emerges from these puzzling findings is that the phenomenon of smoking and its
relation to health is complex and calls for continued exploration and vigorous and energetic investigations. It calls
primarily for the development of study designs and methods which would overcome some of the inherent difficulties in investigations on humans.
In the last few years we often heard the disparaging outcries that the evidence on the ill effects of smoking is not
convincing because it is “only statistical.” The evidence may
not be convincing, but not because it is “only statistical,” rather because the evidence is nonstatistical in the sense that
the method of study which produced the evidence violates
the basic principles for valid scientific inference.
References
1. Simpson WJ. A preliminary report on cigarette smoking and the incidence of prematurity. Amer J Obstet Gynec 1957;73:808–815.
2. Lowe CR. Effect of mothers’ smoking habits on birth weight of
their children. Brit Med J 1959;2:673–676.
3. Herriot A, Billewicz WF, Hytten FE. Cigarette smoking in pregnancy. Lancet 1962;1:771–773.
4. Savel LE, Roth E. Effects of smoking on fetal growth. Obstet
Gynec 1962;20:313–316.
5. Zabriskie JR. Effect of cigarette smoking during pregnancy; a
study of 2000 cases. Obstet Gynec 1963;21:405–411.
6. O’Lane JM. Some fetal effects of maternal cigarette smoking.
Obstet Gynec 1963;22:181–184.
Commentary: Smoking in
pregnancy and neonatal
mortality
International Journal of Epidemiology, 2014, Vol. 43, No. 5
7. MacMahon B, Alpert M, Salber EJ. Infant weight and parental
smoking habits. Amer J Epidem 1965;82:247–261.
8. Underwood P, Hester LL, Laffitte T, et al. The relationship of
smoking to the outcome of pregnancy. Amer J Obstet Gynec
1965;91:270–276.
9. Underwood P, Kesler K, O’Lane J, et al. Parental smoking
empirically related to pregnancy outcome. Obstet Gynec
1967;29:1–8.
10. Butler N, Alberman ED. Perinatal Mortality. The Second Report
of the 1958 British Perinatal Mortality Survey. Edinburgh, E. &.
S. Livingston Ltd, 1969.
11. Shapiro S, Schlesinger E, Nesbitt REL. Infant, Perinatal,
Maternal and Childhood Mortality in the United States.
Harvard University Press, 1968, p 51.
12. Yerushalmy J. Mother’s cigarette smoking and survival of infant.
Amer J Obstet Gynec 1964;88:505–518.
13. Taylor WF. The probability of fetal death. Proc. Third
International Conference on Congenital Malformations. Edited
by Clarke Frazer F, Ebling FJG. Excerpta Medica Foundation,
Amsterdam, 1970.
14. Lilienfeld AM. Emotional and other selected characteristics of
cigarette smokers and nonsmokers as related to epidemiological
studies of lung cancer and other diseases. J Nat Cancer Inst
1959;22:259–282.
15. Eysenck HJ, Tarrant M, Woolf M, et al. Smoking and personality. Brit Med J 1960;1:1456–1460.
16. Heath CW. Differences between smokers and nonsmokers. Arch
Intern Med (Chicago) 1958;101:377–388.
17. Thomas CV. Characteristics of smokers compared with nonsmokers in a population of healthy young adults, including observations on family history, blood pressure, heart rate, body
weight, cholesterol and certain psychological traits. Ann Intern
Med 1960;53:697–718.
18. Seltzer CC. Morphologic constitution and smoking. JAMA
1963;183:639–645.
19. Seltzer CC. Masculinity and smoking. Science 1959;130:
1706–1707.
International Journal of Epidemiology, 2014, 1366–1368
doi: 10.1093/ije/dyu161
Harvey Goldstein
Centre for Multilevel Modelling, Graduate School of Education, University of Bristol, 2 Priory Road,
Bristol BS8 1TX, UK. E-mail: [email protected]
Most epidemiologists are familiar with the controversy
that followed the original Doll and Hill research on the
link between cigarette smoking and lung cancer, and the
ill-fated attempts to dismiss the findings as ‘merely correlational’ by many scientists—including those who should
have known better, such as R.A Fisher. Less well known,
however. is the controversy over the effects of smoking in
pregnancy on birthweight and mortality. This short
review, of one of the seminal papers in the area by
Yerushalmy,1 explores a similar controversy that arose
C The Author 2014; all rights reserved. Published by Oxford University Press on behalf of the International Epidemiological Association
V
1366