AMERICAN JOURNAL OF EPIDEMIOLOGY
Vol. 105, No. 6
Copyright © 1977 by The Johns Hopkins University School of Hygiene and Public Health
Printed in USA.
MATERNAL SMOKING AND BIRTH WEIGHT DEBRA T. SILVERMAN1 Silverman, D. T. (Biometry Branch, National Cancer Institute, Bethesda, MD 20014). Maternal smoking and birth weight. Am J Epidemiol 105:513-521, 1977. To determine whether it is the smoker or smoking per se that explains the observed reduction in birth weight of infants of mothers who smoked during pregnancy, data collected on 1016 pairs of successive births which occurred between 1946 and 1963 in Washington County, Maryland, were studied. The mean birth weight difference between first and second members of pairs of consecutive live births of the same mother was examined in relation to her smoking habits during both pregnancies. A multiple regression technique was used to simultaneously adjust for the effects of maternal smoking habits during pregnancy, infant sex, maternal age, and birth order on birth weight difference. There were no significant differences in mean birth weight difference among pairs in which the mother smoked throughout both pregnancies, pairs in which the mother was a nonsmoker during both pregnancies, and pairs in which the mother smoked during the second pregnancy of the pair, but not the first. When birth weights of only first members of the pair were studied, the mean birth weight of infants of women who were nonsmokers throughout both pregnancies was higher than that of infants of women who smoked during both pregnancies. Infants of first members of pairs in which the mothers smoked only during the second pregnancy tended to have birth weights which were lower than that of infants of nonsmokers and higher than that of infants of smokers. These findings neither confirm nor deny the hypothesis that the smoker rather than the smoking per se causes a reduction in birth weight. birth weight; maternal smoking
This study examines the relationship between maternal smoking and birth
weight. The association of these two factors has been extensively investigated (1-
Received for publication July 26, 1976, and in final form December 16,1976. 1 Biometry Branch, National Cancer Institute, NIH, Bethesda, MD 20014. Much of this work was completed in partial fulfillment of requirements for the degree of Master of Science from the Johns Hopkins University School of Hygiene and Public Health. The author acknowledges the invaluable contributions of Dr. James Tonascia and Dr. George W. Comstock, Johns Hopkins School of Hygiene and Public Health, to this study and in the preparation of the manuscript. Gratitude is expressed also for the assistance of Dr. Charles C. Brown and Dr. Robert N. Hoover, National Cancer Institute, Ms. Mary B. Meyer, Johns Hopkins School of Hygiene and Public Health, and Dr. J. David Godwin, U. of Calif, at San Diego; and the staffs of the Washington County Hospital record room, the Division of Vital Statistics of the Maryland State Health Department, and the Washington County Health Department for providing access to their records.
32
> ^ & » n o w generally accepted that maternal smoking is related to a reduction birth weight. The critical issue is m , ., , . , .. whether smoking causes a reduction in birth weight (the causal hypothesis) or whether smokers are a Self-selected group 6 ,, ,.~ _ , . * t h a t &&** fr°m nonsmokers in ways unrelated to smoking, including the product i o n o f i o w e r birth weight babies (the self, ,. , ,, . . selection hypothesis), Current investigations have attempted to r e S olve this issue by examining the as... _ , . , . , ?., ... sociation of change in smoking habits with birth weight. Among the most recent of these studies are the 1972 report by Butler , ._. ,, , n c . o -n •/• i_ -n • A i e t al - (5> o n ™e 1 9 5 8 British Perinatal Survey and Yerushalmy's research based 513
514
DEBRA T. SILVERMAN
on the experience of 5466 white and 3205 black mothers over a six-year period at the Kaiser Hospital in Oakland, California (29). Butler et al. found that smoking before pregnancy had no significant effect on birth weight, but the average number of cigarettes smoked regularly after the fourth month of pregnancy did have an effect after controlling for a number of intervening variables. In contrast, Yerushalmy found that future smokers, during the period before they started to smoke, had the same frequency of low birth weight infants as mothers who smoked during all their pregnancies, and had a greater frequency than mothers who never smoked. He interpreted this result as evidence that the smoker rather than the smoking per se causes a decrease in birth weight (29). However, the validity of his conclusion has been questioned because adjustments for extraneous variables (e.g., birth order, maternal age, infant sex) that may affect birth weight were not made (33-37). In defense of the causal hypothesis, Butler et al. argued that if smoking rather than the smoker causes a decrease in birth weight one would "expect the mothers who gave up smoking during pregnancy to show differences in birth weight . . . of their offspring compared with those who continued to smoke" (5). They considered their observation that a change in smoking habits during pregnancy placed the infant in the birth weight category associated with the new smoking habits incompatible with Yerushalmy's hypothesis. The present study also attempts to deal with the question of causality vs. selfselection by studying birth weight differences in pairs of infants born to the same mother. The mean birth weight difference of pairs of successive pregnancies in which the mother smoked during the second pregnancy but not the first was compared to the mean birth weight difference when the mother smoked throughout both pregnancies, and also to the mean
difference when the mother did not smoke during either pregnancy. In addition, the comparisons were adjusted for several extraneous variables which may influence birth weight. If smoking per se causes the birth weight reduction, then the mean birth weight difference for infants of mothers who changed their smoking habits should be significantly larger than that if the mother's smoking habits remained the same. Further, the birth weight should decrease from births in which the mother did not smoke to births in which the mother smoked. MATERIALS AND METHODS
This study was conducted in Washington County, western Maryland. There were three sources of data: questionnaires completed in a special census of households, certificates of births to Washington County residents, and records of Washington County Hospital where nearly all Washington County births occur. The special census was conducted in 1963 by the Johns Hopkins School of Hygiene and Public Health, the Washington County Health Department, and the National Cancer Institute (38). Families for the present study were selected from the special census. To be eligible, a family had to include more than one living child born during a 17-year period prior to the census date of July 15, 1963, and the family record had to include a completed smoking history for the mother of the children. Nonwhite families were excluded because there were too few for a meaningful analysis. A total of 5490 families were eligible for this study. Maternal smoking habits were ascertained for all pregnancies culminating in live births which occurred between July 16, 1946 and July 15, 1963 and for which the children were still living at the time of the special census. Each pregnancy was classified as "nonsmoking" or "smoking" according to the special census information, which included whether or not a person had ever smoked, and if so, the age at
MATERNAL SMOKING AND BIRTH WEIGHT
which smoking started. If a mother was a smoker in 1963, the pregnancies which occurred before she started smoking were classified as "nonsmoking." All pregnancies of smokers occurring after the date of starting smoking had to be classified as "smoking" because information on periods of temporary abstinence was not requested. Because of this deficiency, systematic misclassification occurred: pregnancies of smokers who abstained during pregnancy were classified as smoking pregnancies. Such a classification error would result in an overestimate of the birth weight of infants of smokers, if abstinence from smoking during pregnancy is in fact associated with heavier infants. Persons who said they were ex-smokers in 1963 were not asked when they stopped smoking. Consequently, their study experience had to be terminated at the time they started to smoke, and only their children born before that time were included in the study. There were only 13 mothers in this category. Their pregnancies were added to the group classified as "nonsmoking." Of the 5490 eligible families, 3415 mothers were nonsmokers and 1721 were smokers throughout all their pregnancies in the study period (table 1). Of the remaining 354, the mother started to smoke sometime after her first pregnancy in 215, and durTABLE 1
Derivation of study sample from eligible families in 1963 census population, by smoking history of mothers during pregnancies Smoking history of mother during pregnancies
Eligible families
Sample families
Nonsmoker throughout Started smoking sometime after 1st pregnancy or during birth year Smoker throughout
3415 354*
201 354*
1721
215
Total -
5490
770
* 215 families in which the mother became a smoker after her first pregnancy, and 139 families in which the mother started to smoke the same year a pregnancy occurred.
515
ing the year of one of her pregnancies in 139. Because the study was based on the results of consecutive pairs of pregnancies terminating in children living in 1963, it was also necessary to classify each pair according to whether the mother smoked during neither pregnancy, only the second pregnancy, or both pregnancies. Pairs of pregnancies occurring to women who did not change their smoking habits were easily classified as nonsmoking or smoking. Pairs of pregnancies occurring to the 215 women who started to smoke some time after their first pregnancy could be nonsmoking, smoking only during the second pregnancy of the pair, or smoking, depending on the time the mother started to smoke in relation to her pregnancies. The 139 pregnancies occurring during the same year the mother started to smoke could not be classified, nor could the pairs to which these pregnancies belonged. However, these women were able to contribute nonsmoking or smoking pairs because each woman in this group had at least two pregnancies either before or after the year in which she started to smoke. Because of financial and time limitations, only a sample of the eligible families could be studied. A total of 770 families were included in the sample, distributed as shown in table 1. All of the 215 families in which the mother became a smoker sometime after her first pregnancy were included because this was the only group that could produce pairs of primary interest to the investigation, i.e., pairs of infants born to the same mother, one born during a nonsmoking pregnancy and the other during a smoking pregnancy. All of the 139 families in which the mother started to smoke the same year a pregnancy occurred were also included, even though this group could yield only homogeneous nonsmoking or smoking pairs. Every 17th nonsmoking family was selected, as was every eighth family in which the mother smoked throughout her
516
DEBRA T. SILVERMAN
pregnancy experience, yielding subsamples of 201 and 215, respectively. A total of 2434 births resulting in children living in 1963 occurred to these 770 sample families. A search was made for the birth certificates of each of these 2434 children to obtain their birth weights and the ages of their mothers at the time of birth. Nearly 20 per cent of the births had to be excluded for reasons shown in table 2. Forty-six twins were not studied because twins have a different birth weight distribution than single births. For 36 births, the stepmother was misnamed as the natural mother on the special census. Twenty-five had to be excluded because the birth weight could not be found on the birth certificate or in the local hospital records. Adopted children were also excluded because the natural mother's smoking history could not be ascertained. The largest number of exclusions, 356, occurred because the birth certificate could not be found. Presumably almost all of this group were births to persons who moved to Washington County after the birth of the child. The 1958 births in the study group consisted of 1016 pairs of births. (Seventy-four children were second members of one pair and first members of another pair.) The derivation of these study pairs according to the mother's smoking history and to the smoking classification of the pair is shown in table 3. There were only 143 pairs for TABLE 2
Births to sample families included in study and excluded births, by reason for exclusion Births
No. of births
In study Excluded Twin births Incorrect mother Birth weight not stated Adopted child Birth certificate not found
1958 476
Total births
2434
% 80.5 19.5
46 36 25 13 356
1.9 1.5 1.0 0.5 14.6 100.0
which there was evidence that the mother did not smoke during the first pregnancy but did smoke during the second pregnancy. RESULTS
Birth weight differences between first and second members of birth pairs. The mean birth weights for both the first and second members of the birth pairs for each smoking group are presented in table 4. The mean birth weight of the first pair member was not significantly different from that of the second member regardless of the mother's smoking habits. However, in birth pairs in which the mother smoked only during the second pregnancy, there was a greater weight gain from the first pair member to the second than in the other smoking groups. When only pairs of the first and second born infants were considered, the same trends persisted. To allow for the effects of other intervening variables that might influence the relationship between smoking and birth weight difference, a multiple regression analysis was performed. Unfortunately, as often happens in retrospective studies, some important independent variables were not available, e.g., maternal weight gain during pregnancy. However, data for the following independent variables were at hand: maternal smoking habits during pregnancy; infant sex; maternal age at time of birth; and birth order. Birth order was defined as the number of previous live births as of July 1963 plus the birth in question. Thus, a first born infant had birth order one. If a child had died prior to July 1963, the birth orders of offspring born after this child would be inaccurate because birth order had to be defined in terms of live births. The dependent variable was the mean birth weight difference, d, where d equals the birth weight of the second member of a pair minus the birth weight of the first member of the same pair. A binary variable multiple regression method of analyzing classificatory data in-
517
MATERNAL SMOKING AND BIRTH WEIGHT TABLE 3
Allocation of birth pairs in study by mother's history of smoking during all of her pregnancies and during each pair of pregnancies Smoking history of mother during pregnancies
Nonsmoker throughout all Started smoking sometime after 1st pregnancy Smoker throughout all Started smoking during year of one pregnancy Total
Smoking history of mother during pair of pregnancies %
2nd only
%
25.8 30.4
262 68
68.6 17.8
0 143
0.0 100.0
0 98
0.0 20.0
282 163
27.8 16.0
0 52
0.0 13.6
0 0
0.0 0.0
282 111
57.4 22.6
1016
100.0
382
100.0
143
100.0
491
100.0
%
Both
%
Neither
262 309
Total
TABLE 4
Mean birth weight for first and second members of birth pair by imother's smoking history 1st pair member Smoking history of mother during pair of pregnancies
Nonsmoker during both pregnancies Smoker during 2nd but not first Smoker during both preg-
No. of pairs
2nd pair member
Difference: 2nd pair member minus 1st pair member Mean birth SD weight difference (gm)
Mean birth weight (gm)
8D*
Mean birth weight (gm)
SD
382
3451.0
531.8
3473.0
547.8
22.0
561.1
143
3341.5
538.2
3410.2
566.0
68.7
571.6
491
3265.5
526.4
3257.7
530.2
-7.8
531.3
nancies
* SD, standard deviation.
troduced by Suits (39) and developed by Feldstein (40) was used. A modification of Feldstein's procedure was necessary because the dependent variable in this study, birth weight difference, was a continuous rather than a classificatory variable. Four pairs with incomplete observations were excluded from the regression analysis. The crude and adjusted mean birth weight differences for each subclass of independent variable are presented in table 5. The F-ratios, computed using the extra sum of squares principle (41), were used to assess the independent effect of each variable on birth weight difference after simultaneously adjusting for the effects of the other intervening variables. Adjustment did not result in any dramatic changes from the crude mean birth weight differences. The well-known excess birth weight among males was confirmed. The tendency for older women to have smaller babies was also noted in this material.
After adjustment for other variables, there was a slight increase in weight with birth order to birth order three, and essentially no change thereafter. Adjustment made little change in the mean birth weight differences for the smoking history subclasses. There was essentially no difference in the mean birth weights between first and second members of pairs of infants born to smoking mothers, and very little mean birth weight difference for pairs born to nonsmoking mothers. When the mother had smoked only during the second pregnancy, the second infant was 64 gm heavier on the average than the first member of the pair. The smoking group to which a mother belonged had no statistically significant effect on birth weight difference between two consecutive pregnancies (F (2, 995) = 0.73, p > 0.25). When only pairs of first and second born infants were considered, the same trends persisted.
518
DEBRA T. SILVERMAN TABLE 5
Crude and adjusted mean birth weight differences, by independent variable Mean birth weight difference Independent variable
Smoking history during pair of pregnancies Neither Second only Both Sex of pair MF FF MM FM
Maternal age during pair of pregnancies (years)
Vol. 105, No. 6
Copyright © 1977 by The Johns Hopkins University School of Hygiene and Public Health
Printed in USA.
MATERNAL SMOKING AND BIRTH WEIGHT DEBRA T. SILVERMAN1 Silverman, D. T. (Biometry Branch, National Cancer Institute, Bethesda, MD 20014). Maternal smoking and birth weight. Am J Epidemiol 105:513-521, 1977. To determine whether it is the smoker or smoking per se that explains the observed reduction in birth weight of infants of mothers who smoked during pregnancy, data collected on 1016 pairs of successive births which occurred between 1946 and 1963 in Washington County, Maryland, were studied. The mean birth weight difference between first and second members of pairs of consecutive live births of the same mother was examined in relation to her smoking habits during both pregnancies. A multiple regression technique was used to simultaneously adjust for the effects of maternal smoking habits during pregnancy, infant sex, maternal age, and birth order on birth weight difference. There were no significant differences in mean birth weight difference among pairs in which the mother smoked throughout both pregnancies, pairs in which the mother was a nonsmoker during both pregnancies, and pairs in which the mother smoked during the second pregnancy of the pair, but not the first. When birth weights of only first members of the pair were studied, the mean birth weight of infants of women who were nonsmokers throughout both pregnancies was higher than that of infants of women who smoked during both pregnancies. Infants of first members of pairs in which the mothers smoked only during the second pregnancy tended to have birth weights which were lower than that of infants of nonsmokers and higher than that of infants of smokers. These findings neither confirm nor deny the hypothesis that the smoker rather than the smoking per se causes a reduction in birth weight. birth weight; maternal smoking
This study examines the relationship between maternal smoking and birth
weight. The association of these two factors has been extensively investigated (1-
Received for publication July 26, 1976, and in final form December 16,1976. 1 Biometry Branch, National Cancer Institute, NIH, Bethesda, MD 20014. Much of this work was completed in partial fulfillment of requirements for the degree of Master of Science from the Johns Hopkins University School of Hygiene and Public Health. The author acknowledges the invaluable contributions of Dr. James Tonascia and Dr. George W. Comstock, Johns Hopkins School of Hygiene and Public Health, to this study and in the preparation of the manuscript. Gratitude is expressed also for the assistance of Dr. Charles C. Brown and Dr. Robert N. Hoover, National Cancer Institute, Ms. Mary B. Meyer, Johns Hopkins School of Hygiene and Public Health, and Dr. J. David Godwin, U. of Calif, at San Diego; and the staffs of the Washington County Hospital record room, the Division of Vital Statistics of the Maryland State Health Department, and the Washington County Health Department for providing access to their records.
32
> ^ & » n o w generally accepted that maternal smoking is related to a reduction birth weight. The critical issue is m , ., , . , .. whether smoking causes a reduction in birth weight (the causal hypothesis) or whether smokers are a Self-selected group 6 ,, ,.~ _ , . * t h a t &&** fr°m nonsmokers in ways unrelated to smoking, including the product i o n o f i o w e r birth weight babies (the self, ,. , ,, . . selection hypothesis), Current investigations have attempted to r e S olve this issue by examining the as... _ , . , . , ?., ... sociation of change in smoking habits with birth weight. Among the most recent of these studies are the 1972 report by Butler , ._. ,, , n c . o -n •/• i_ -n • A i e t al - (5> o n ™e 1 9 5 8 British Perinatal Survey and Yerushalmy's research based 513
514
DEBRA T. SILVERMAN
on the experience of 5466 white and 3205 black mothers over a six-year period at the Kaiser Hospital in Oakland, California (29). Butler et al. found that smoking before pregnancy had no significant effect on birth weight, but the average number of cigarettes smoked regularly after the fourth month of pregnancy did have an effect after controlling for a number of intervening variables. In contrast, Yerushalmy found that future smokers, during the period before they started to smoke, had the same frequency of low birth weight infants as mothers who smoked during all their pregnancies, and had a greater frequency than mothers who never smoked. He interpreted this result as evidence that the smoker rather than the smoking per se causes a decrease in birth weight (29). However, the validity of his conclusion has been questioned because adjustments for extraneous variables (e.g., birth order, maternal age, infant sex) that may affect birth weight were not made (33-37). In defense of the causal hypothesis, Butler et al. argued that if smoking rather than the smoker causes a decrease in birth weight one would "expect the mothers who gave up smoking during pregnancy to show differences in birth weight . . . of their offspring compared with those who continued to smoke" (5). They considered their observation that a change in smoking habits during pregnancy placed the infant in the birth weight category associated with the new smoking habits incompatible with Yerushalmy's hypothesis. The present study also attempts to deal with the question of causality vs. selfselection by studying birth weight differences in pairs of infants born to the same mother. The mean birth weight difference of pairs of successive pregnancies in which the mother smoked during the second pregnancy but not the first was compared to the mean birth weight difference when the mother smoked throughout both pregnancies, and also to the mean
difference when the mother did not smoke during either pregnancy. In addition, the comparisons were adjusted for several extraneous variables which may influence birth weight. If smoking per se causes the birth weight reduction, then the mean birth weight difference for infants of mothers who changed their smoking habits should be significantly larger than that if the mother's smoking habits remained the same. Further, the birth weight should decrease from births in which the mother did not smoke to births in which the mother smoked. MATERIALS AND METHODS
This study was conducted in Washington County, western Maryland. There were three sources of data: questionnaires completed in a special census of households, certificates of births to Washington County residents, and records of Washington County Hospital where nearly all Washington County births occur. The special census was conducted in 1963 by the Johns Hopkins School of Hygiene and Public Health, the Washington County Health Department, and the National Cancer Institute (38). Families for the present study were selected from the special census. To be eligible, a family had to include more than one living child born during a 17-year period prior to the census date of July 15, 1963, and the family record had to include a completed smoking history for the mother of the children. Nonwhite families were excluded because there were too few for a meaningful analysis. A total of 5490 families were eligible for this study. Maternal smoking habits were ascertained for all pregnancies culminating in live births which occurred between July 16, 1946 and July 15, 1963 and for which the children were still living at the time of the special census. Each pregnancy was classified as "nonsmoking" or "smoking" according to the special census information, which included whether or not a person had ever smoked, and if so, the age at
MATERNAL SMOKING AND BIRTH WEIGHT
which smoking started. If a mother was a smoker in 1963, the pregnancies which occurred before she started smoking were classified as "nonsmoking." All pregnancies of smokers occurring after the date of starting smoking had to be classified as "smoking" because information on periods of temporary abstinence was not requested. Because of this deficiency, systematic misclassification occurred: pregnancies of smokers who abstained during pregnancy were classified as smoking pregnancies. Such a classification error would result in an overestimate of the birth weight of infants of smokers, if abstinence from smoking during pregnancy is in fact associated with heavier infants. Persons who said they were ex-smokers in 1963 were not asked when they stopped smoking. Consequently, their study experience had to be terminated at the time they started to smoke, and only their children born before that time were included in the study. There were only 13 mothers in this category. Their pregnancies were added to the group classified as "nonsmoking." Of the 5490 eligible families, 3415 mothers were nonsmokers and 1721 were smokers throughout all their pregnancies in the study period (table 1). Of the remaining 354, the mother started to smoke sometime after her first pregnancy in 215, and durTABLE 1
Derivation of study sample from eligible families in 1963 census population, by smoking history of mothers during pregnancies Smoking history of mother during pregnancies
Eligible families
Sample families
Nonsmoker throughout Started smoking sometime after 1st pregnancy or during birth year Smoker throughout
3415 354*
201 354*
1721
215
Total -
5490
770
* 215 families in which the mother became a smoker after her first pregnancy, and 139 families in which the mother started to smoke the same year a pregnancy occurred.
515
ing the year of one of her pregnancies in 139. Because the study was based on the results of consecutive pairs of pregnancies terminating in children living in 1963, it was also necessary to classify each pair according to whether the mother smoked during neither pregnancy, only the second pregnancy, or both pregnancies. Pairs of pregnancies occurring to women who did not change their smoking habits were easily classified as nonsmoking or smoking. Pairs of pregnancies occurring to the 215 women who started to smoke some time after their first pregnancy could be nonsmoking, smoking only during the second pregnancy of the pair, or smoking, depending on the time the mother started to smoke in relation to her pregnancies. The 139 pregnancies occurring during the same year the mother started to smoke could not be classified, nor could the pairs to which these pregnancies belonged. However, these women were able to contribute nonsmoking or smoking pairs because each woman in this group had at least two pregnancies either before or after the year in which she started to smoke. Because of financial and time limitations, only a sample of the eligible families could be studied. A total of 770 families were included in the sample, distributed as shown in table 1. All of the 215 families in which the mother became a smoker sometime after her first pregnancy were included because this was the only group that could produce pairs of primary interest to the investigation, i.e., pairs of infants born to the same mother, one born during a nonsmoking pregnancy and the other during a smoking pregnancy. All of the 139 families in which the mother started to smoke the same year a pregnancy occurred were also included, even though this group could yield only homogeneous nonsmoking or smoking pairs. Every 17th nonsmoking family was selected, as was every eighth family in which the mother smoked throughout her
516
DEBRA T. SILVERMAN
pregnancy experience, yielding subsamples of 201 and 215, respectively. A total of 2434 births resulting in children living in 1963 occurred to these 770 sample families. A search was made for the birth certificates of each of these 2434 children to obtain their birth weights and the ages of their mothers at the time of birth. Nearly 20 per cent of the births had to be excluded for reasons shown in table 2. Forty-six twins were not studied because twins have a different birth weight distribution than single births. For 36 births, the stepmother was misnamed as the natural mother on the special census. Twenty-five had to be excluded because the birth weight could not be found on the birth certificate or in the local hospital records. Adopted children were also excluded because the natural mother's smoking history could not be ascertained. The largest number of exclusions, 356, occurred because the birth certificate could not be found. Presumably almost all of this group were births to persons who moved to Washington County after the birth of the child. The 1958 births in the study group consisted of 1016 pairs of births. (Seventy-four children were second members of one pair and first members of another pair.) The derivation of these study pairs according to the mother's smoking history and to the smoking classification of the pair is shown in table 3. There were only 143 pairs for TABLE 2
Births to sample families included in study and excluded births, by reason for exclusion Births
No. of births
In study Excluded Twin births Incorrect mother Birth weight not stated Adopted child Birth certificate not found
1958 476
Total births
2434
% 80.5 19.5
46 36 25 13 356
1.9 1.5 1.0 0.5 14.6 100.0
which there was evidence that the mother did not smoke during the first pregnancy but did smoke during the second pregnancy. RESULTS
Birth weight differences between first and second members of birth pairs. The mean birth weights for both the first and second members of the birth pairs for each smoking group are presented in table 4. The mean birth weight of the first pair member was not significantly different from that of the second member regardless of the mother's smoking habits. However, in birth pairs in which the mother smoked only during the second pregnancy, there was a greater weight gain from the first pair member to the second than in the other smoking groups. When only pairs of the first and second born infants were considered, the same trends persisted. To allow for the effects of other intervening variables that might influence the relationship between smoking and birth weight difference, a multiple regression analysis was performed. Unfortunately, as often happens in retrospective studies, some important independent variables were not available, e.g., maternal weight gain during pregnancy. However, data for the following independent variables were at hand: maternal smoking habits during pregnancy; infant sex; maternal age at time of birth; and birth order. Birth order was defined as the number of previous live births as of July 1963 plus the birth in question. Thus, a first born infant had birth order one. If a child had died prior to July 1963, the birth orders of offspring born after this child would be inaccurate because birth order had to be defined in terms of live births. The dependent variable was the mean birth weight difference, d, where d equals the birth weight of the second member of a pair minus the birth weight of the first member of the same pair. A binary variable multiple regression method of analyzing classificatory data in-
517
MATERNAL SMOKING AND BIRTH WEIGHT TABLE 3
Allocation of birth pairs in study by mother's history of smoking during all of her pregnancies and during each pair of pregnancies Smoking history of mother during pregnancies
Nonsmoker throughout all Started smoking sometime after 1st pregnancy Smoker throughout all Started smoking during year of one pregnancy Total
Smoking history of mother during pair of pregnancies %
2nd only
%
25.8 30.4
262 68
68.6 17.8
0 143
0.0 100.0
0 98
0.0 20.0
282 163
27.8 16.0
0 52
0.0 13.6
0 0
0.0 0.0
282 111
57.4 22.6
1016
100.0
382
100.0
143
100.0
491
100.0
%
Both
%
Neither
262 309
Total
TABLE 4
Mean birth weight for first and second members of birth pair by imother's smoking history 1st pair member Smoking history of mother during pair of pregnancies
Nonsmoker during both pregnancies Smoker during 2nd but not first Smoker during both preg-
No. of pairs
2nd pair member
Difference: 2nd pair member minus 1st pair member Mean birth SD weight difference (gm)
Mean birth weight (gm)
8D*
Mean birth weight (gm)
SD
382
3451.0
531.8
3473.0
547.8
22.0
561.1
143
3341.5
538.2
3410.2
566.0
68.7
571.6
491
3265.5
526.4
3257.7
530.2
-7.8
531.3
nancies
* SD, standard deviation.
troduced by Suits (39) and developed by Feldstein (40) was used. A modification of Feldstein's procedure was necessary because the dependent variable in this study, birth weight difference, was a continuous rather than a classificatory variable. Four pairs with incomplete observations were excluded from the regression analysis. The crude and adjusted mean birth weight differences for each subclass of independent variable are presented in table 5. The F-ratios, computed using the extra sum of squares principle (41), were used to assess the independent effect of each variable on birth weight difference after simultaneously adjusting for the effects of the other intervening variables. Adjustment did not result in any dramatic changes from the crude mean birth weight differences. The well-known excess birth weight among males was confirmed. The tendency for older women to have smaller babies was also noted in this material.
After adjustment for other variables, there was a slight increase in weight with birth order to birth order three, and essentially no change thereafter. Adjustment made little change in the mean birth weight differences for the smoking history subclasses. There was essentially no difference in the mean birth weights between first and second members of pairs of infants born to smoking mothers, and very little mean birth weight difference for pairs born to nonsmoking mothers. When the mother had smoked only during the second pregnancy, the second infant was 64 gm heavier on the average than the first member of the pair. The smoking group to which a mother belonged had no statistically significant effect on birth weight difference between two consecutive pregnancies (F (2, 995) = 0.73, p > 0.25). When only pairs of first and second born infants were considered, the same trends persisted.
518
DEBRA T. SILVERMAN TABLE 5
Crude and adjusted mean birth weight differences, by independent variable Mean birth weight difference Independent variable
Smoking history during pair of pregnancies Neither Second only Both Sex of pair MF FF MM FM
Maternal age during pair of pregnancies (years)
