Active and Passive Cigarette Smoking and the ...

2 downloads 0 Views 681KB Size Report
Active and Passive Cigarette Smoking and the Occurrence of Breast Cancer. Timothy L Lash12 and Ann Aschengrau1. Using a case-control design, the authors ...
American Journal of Epidemiology Copyright © 1999 by The Johns Hopkins University School of Hygisne and Public Health All rights reserved

Vol. 149, No. 1 Printed in U.S.A.

ORIGINAL CONTRIBUTIONS

Active and Passive Cigarette Smoking and the Occurrence of Breast Cancer

Timothy L Lash12 and Ann Aschengrau1

breast neoplasms, etiology; smoking, adverse effects; tobacco smoke pollution, adverse effects

The well-established risk factors for breast cancer offer few opportunities for intervention and account for less than half of all breast cancer cases (1). Most involve aspects of a woman's reproductive course that are intimately related to her lifestyle and culture (difficult to predict and thus to change) or are currently immutable (e.g., genotype). Were tobacco smoke a cause of breast cancer, changing a woman's exposure to it would present an opportunity for meaningful preventive intervention. However, until recently, exposure to tobacco smoke has not been thought to cause breast cancer (2). Novel perspectives on measuring the association of tobacco smoke with the occurrence of breast cancer (3) and studies of genetically susceptible populations exposed to tobacco smoke (4-6) argue for further investigation. Received for publication January 26,1998, and accepted for publication August 12, 1998. Abbreviation: Cl, confidence interval. 1 Boston University School of Public Health, Boston University Medical Center, Boston, MA. 2 Department of Medicine, Boston University Medical Center, Boston, MA. Reprint requests to Timothy L. Lash, Department of Medicine, Boston University Medical Center, 88 East Newton Street, F433, Boston, MA 02118.

Few breast cell mutagens have been identified, but estrogen-induced mitogenesis has been well characterized (7). To cause cancer, mutagenesis must occur while breast tissue is susceptible and prior to mitogenesis. As depicted in figure 1, breast tissue development and differentiation determine its susceptibility to mutagenesis; breast cells derived from type 1 or type 2 lobules are susceptible to chemical mutagens before menopause occurs, but breast cells derived from type 3 lobules are immune to the mutagens that have been tested thus far (8). This model suggests that the time period of exposure to breast carcinogens ought to determine susceptibility to carcinogenesis. Before menopause occurs, when the proportion of type 1 or type 2 lobules is high, breast tissue exposed to mutagenic events should be susceptible. Promotion of the exposed and susceptible tissue by mitogenic estrogen compounds should further increase the risk of tumor development (9), whereas inhibition of mitogenesis should reduce the risk (7). In epidemiologic studies of breast cancer etiology, the definitions of index and reference conditions should reflect this model of breast carcinogenesis. For example, the timing of exposure to tobacco smoke relative to milestones of breast tissue development may

Downloaded from aje.oxfordjournals.org by guest on July 15, 2011

Using a case-control design, the authors studied female residents of five Massachusetts towns between 1983 and 1986. The objective was to measure the associations between breast cancer occurrence and exposure to active and passive cigarette smoke. Until recently, exposure to tobacco smoke has not been thought to cause breast cancer. Novel perspectives on measuring the association of tobacco smoke with the occurrence of breast cancer and studies of genetically susceptible populations argue for further investigation. In this study, the authors found that ever-active smokers had an odds ratio of 2.0 (95 percent confidence interval (Cl) 1.1-3.6) when compared with never-active, never-passive smokers. Women who smoked only before their first pregnancy (odds ratio = 5.6, 95 percent Cl 1.5-21) and women who quit smoking 5-15 years before their index year (odds ratio = 3.9, 95 percent Cl 1.4-10) were at the highest risk. Passive-only smokers had an odds ratio of 2.0 (95 percent Cl 1.1-3.7) when compared with never-active, never-passive smokers. Among those women who were exposed to passive smoke before age 12 years, the odds ratios were 4.5 (95 percent Cl 1.2-16) for passive-only smokers and 7.5 (95 percent Cl 1.6-36) for ever-active smokers. Women who were first exposed to passive smoke after age 12 years had lower, although still elevated, odds ratios. The pattern of associations between exposure to cigarette smoke and the occurrence of breast cancer comports with a model of breast carcinogenesis. Am J Epidemiol 1999;149:5-12.

6

Lash and Aschengrau

later stages of breast tissue development. These careful definitions of exposure to tobacco smoke revealed associations with the risk of breast cancer that would have been obscured by using simpler definitions.

be important in defining exposure (figure 1). Early exposure, especially before a woman's first term pregnancy, may cause breast cancer as a result of genotoxic mechanisms (10), whereas later exposure may prevent breast cancer because of an antiestrogenic effect of tobacco smoke (11, 12). On balance, lifelong exposure may enable these effects to cancel. Also, definitions of index and reference conditions should take into account exposure to both active and passive smoking (13). Before conducting the present analyses, we hypothesized index conditions for active and passive exposure to cigarette smoke that would cause breast cancer while the breast tissue contains primarily type 1 and type 2 lobules or those that would prevent breast cancer as a result of antiestrogenic mechanisms during

MATERIALS AND METHODS

Our data collection methods have been described in detail elsewhere (14). Briefly, from 1983 to 1986, we identified 334 incident cases of breast cancer that occurred among permanent female residents of five Massachusetts towns and were reported to the state cancer registry. We used three methods to select a single set of female controls from the base population of permanent residents of the towns during 1983-1986. First, we used random-digit dialing to select an age-

BREAST TISSUE DEVELOPMENT AND SUSCEPTIBILITY TO CHEMICAL CARCINOGENESIS

Sexual maturity

During pregnancy or gradually with premenopausal aging

Lobule type 3: low susceptibility to carcinogens, low doubling rate Postlactational involution

Causation: initiation of carcinogenesis by active or passive smoking Prevention: antiestrogenic effect after initiation

Causation: reduced susceptibility to initiation Prevention: primarily antiestrogenic effect

Lactation

Lobule type 4: not characterized

FIGURE 1. Model of the susceptibility of breast tissue to tobacco smoke. The size of each left-pointing arrow represents the hypothesized magnitude of the causal effect of tobacco smoke exposure during the corresponding stage of breast tissue development.

Am J Epidemiol Vol. 149, No. 1, 1999

Downloaded from aje.oxfordjournals.org by guest on July 15, 2011

Causation: initiation of carcinogenesis by primarily passive smoking Prevention: antiestrogenic effect after initiation

Lobule type 1: most susceptible to carcinogens, highest doubling rate

Lobule type 2: intermediate susceptibility to carcinogens, intermediate doubling rate

HYPOTHESIZED EFFECT OF TOBACCO SMOKE ON BREAST CARCINOGENESIS

Cigarette Smoking and Breast Cancer

Am J Epidemiol

Vol. 149, No. 1, 1999

passive smoke and breast cancer occurrence, we divided passive-only smokers into categories demarcated by their age at first exposure in one analysis and by their first term pregnancy in a second analysis. We created categories of age at first exposure to correspond to milestones of breast tissue development. The initial category included women whose first exposure occurred before age 12 years or before the onset of puberty. In the present study, 90 percent of the 446 controls whose age at menarche was known were aged 12 years or older at menarche. The second category included women whose first exposure occurred between age 12 and 20 years, the typical period of breast tissue development. The third category included women whose first exposure occurred at age 21 years or older, after breast tissue development but not necessarily before their first pregnancy. In the present study, 81 percent of the 551 controls with at least one term pregnancy had their first term pregnancy at age 21 years or older. We also grouped passive-only smokers by their total duration of exposure to passive smoking. Finally, we measured the association between passive smoking and breast cancer occurrence among ever-active smokers. To do so, we divided ever-active smokers into the same categories of age at first exposure to passive smoke as those described above for passive-only smokers. Unless otherwise specified in the tables in this paper, we adjusted the odds ratios for confounding by age (80 years at index year), parity (0, 1, 2, or >3 term pregnancies), family history of breast cancer (whether or not the mother or a sister had breast cancer), body mass index (25 kg/m2 based on self-reported usual adult body weight), history of benign breast disease, history of breast cancer other than the index diagnosis, and history of radiation therapy. Controlling for these confounders influenced the odds ratios by 10 percent or more in at least one analysis. We considered but did not control for age at first birth, hormone replacement therapy, or menopausal status at index year, because controlling for these potential confounders did not influence the odds ratios by 10 percent or more after we controlled for the confounders listed above. We used the exposure odds ratios to estimate the relative risks. We estimated the adjusted odds ratios and their corresponding 95 percent confidence intervals by using multivariate logistic regression. RESULTS

The distribution of selected demographic characteristics and confounding variables among the cases and controls is shown in table 1. Approximately 90 percent of the cases and controls were postmenopausal, so the

Downloaded from aje.oxfordjournals.org by guest on July 15, 2011

and sex-stratified sample of living subjects younger than age 65 years who resided in the towns during those years. Second, we selected an age- and sexstratified random sample of living subjects aged 65 years or older by using lists provided by the Health Care Financing Administration. Third, we selected a random sample of deceased subjects, stratified by sex, age, and year of death, by using a list furnished by the Massachusetts Department of Vital Statistics and Research of all resident deaths in the towns from 1983 through 1989. We selected deceased controls to approximately balance the proportion of proxy interviews completed by next-of-kin respondents in the case and control groups. Trained interviewers conducted structured interviews to obtain information on demographic characteristics, smoking, medical conditions, and reproductive events. Of the original 334 breast cancer cases identified, 33 were never found, 6 were ineligible to participate, and 30 could not be interviewed because physicians refused to allow us to contact them. Thus, we interviewed 265 cases. We interviewed 75 percent of eligible random-digit-dialing controls, 76 percent of eligible Health Care Financing Administration controls, and 79 percent of eligible next-of-kin controls, yielding a total of 765 interviewed controls. We conducted 86 percent of all interviews by telephone and the remainder in person. Each case's index year was set as the year of breast cancer diagnosis. We randomly assigned an index year to controls so the distribution of index years for controls matched that for cases. We categorized cases and controls by three types of exposure to cigarette smoke: any history of active cigarette smoking (ever-active smokers), any history of passive exposure to cigarette smoke in the residence but no history of active cigarette smoking (passiveonly smokers), and no history of active cigarette smoking or passive exposure to cigarette smoke in the residence (nonsmokers, the reference condition throughout). We did not measure exposure to passive cigarette smoke outside of the residence. To measure the association between first exposure to active cigarette smoking and the occurrence of breast cancer, we divided ever-active smokers into categories by their age at first exposure in one analysis and by their first term pregnancy in a second analysis. We also categorized ever-active smokers by their duration of smoking in years and by their usual intensity of smoking in number of cigarettes smoked per day. To measure the association between smoking cessation and the occurrence of breast cancer, we grouped everactive smokers according to the number of years between cessation and the index year. To measure the association between first exposure to

7

8

Lash and Aschengrau

TABLE 1. Characteristics of cases and controls studied to determine an association between cigarette smoking and breast cancer risk, Massachusetts, 1983-1986 Cases

Age (years) at index year 80 Type of interview Self (living subjects) Proxy (deceased subjects) Age at menarche (years) 15 Data missing

No.

%*

No.

%

31

38 82 71 43

12 14 31 27 16

53 82 252 213 163

7 11 33 28 21

177 88

67 33

417 346

55

66 76 33 90

25 29 12 34

133 213 100 317

31 234

12 88

63 700

8 92

45

17

28 13

42

Body mass index (kg/m2) 25 Data missing

18

7

185 52

70

49 492

20

187

10

4

35

6 64 25 5

76 31 57 95 6

29 12

175 106 165 305 12

23 14 22 40 1

Parity (term pregnancies) 0 1

2 >3 Data missing Age at first birth (years) No birth 30 Data missing

21 36

2

76 139

29 52

34 16

13

History of benign breast disease Yes No Data missing

26 213 26

10 80 10

Mother or sister with breast cancer Yes No Data missing

46 197 22

17 74

6

8

175 452 99 37

23

107 594

14 78 8

62

59 13

5

60

8

619 84

81 11

* Some percentages have been rounded.

following results may not apply to premenopausal women. Active cigarette smoking

Ever-active smokers had an adjusted odds ratio of 2.0 (95 percent confidence interval (CI) 1.1-3.6) when compared with nonsmokers (table 2). The odds ratios of breast cancer declined with increasing intensity and

Passive-only cigarette smoking

Passive-only smokers had an adjusted odds ratio of 2.0 (95 percent CI 1.1-3.7) (table 3). This odds ratio for passive-only smokers approximately equals that for ever-active smokers, which emphasizes the importance of using never-active, never-passive smokers as the reference population. Had the never-active, neverpassive smokers been combined with the passive-only smokers to form the reference group, the odds ratio associated with ever-active smoking would have been 1.2 (95 percent CI 0.8-1.7). The odds ratios of breast cancer varied inversely with the duration of exposure to passive smoke. Passive smokers who lived with an active smoker for 20 or fewer years had an odds ratio of 3.2 (95 percent CI 1.5-7.1). Those who lived with an active smoker for more than 20 years had an odds ratio of 2.1 (95 percent CI 1.0-4.1). The odds ratios for passive-only smokers did not depend on whether their exposure preceded or followed their first pregnancy, as did the odds ratios for ever-active smokers. However, an earlier age at first exposure to passive smoke did confer an elevated odds ratio of breast cancer. Women who were first exposed to passive smoke before age 12 years, between age 12 Am J Epidemiol

Vol. 149, No. 1, 1999

Downloaded from aje.oxfordjournals.org by guest on July 15, 2011

Menopausal status Premenopausal Postmenopausal

Controls

duration of ever-active smoking consistent with an antiestrogenic mechanism, although the overall trend was weak. The association between smoking and breast cancer risk depended on whether women smoked before or after their first pregnancy. Women who smoked only before their first pregnancy had an odds ratio of 5.6 (95 percent CI 1.5-21). Women who smoked only after their first pregnancy had an odds ratio of 2.1 (95 percent CI 1.1—4.0). Women who began to smoke before their first pregnancy and continued to smoke after their first pregnancy did not share the elevated odds ratio (odds ratio = 1.1, 95 percent CI 0.6-2.0). The association between smoking and breast cancer risk did not depend on the age at which active smoking began, so this variable could not be considered a surrogate measure for initiation and termination of smoking prior to the first pregnancy. Women who quit smoking less than 5 years before their index year or were smoking during their index year had an odds ratio of 2.3 (95 percent CI 0.8-6.8). Women who quit smoking 5-15 years before their index year, the period during which estrogen promotion of smoking-initiated breast cancer may be most potent, had an odds ratio of 3.9 (95 percent CI 1.4-10). Women who quit smoking more than 15 years before their index year had an odds ratio of 2.2 (95 percent CI 1.0-4.9).

Cigarette Smoking and Breast Cancer

9

TABLE 2. Measured odds ratios of breast cancer associated with exposure to active cigarette smoking, Massachusetts, 1983-1986 Exposure condition

Never active, never passive Ever active^

Case:control ratio

40:139

OR*

95% Clf

1

2.0

1.1-3.6

84:160 16:42

2.1 1.6

1.0-4.6 0.6-4.3

Duration of smoking (years)H 0-19 20-39 >40

34:54 46:117 54:147

2.6 1.5 2.4

1.2-5.5 0.7-3.2 1.1-5.5

Smoking relative to first term pregnancy Only before first pregnancy Only after first pregnancy Both before and after first pregnancy

7:6 63:110 57:175

5.6 2.1 1.1

1.5-21 1.1-4.0 0.6-2.0

Cessation before index year (years)§,H 15

22:75 33:54 82:209

2.3 3.9 2.2

0.8-6.8 1.4-10 1.0-4.9

Age smoking started (years)§,H 21

28:75 60:138 47:106

2.4 2.3 2.4

0.8-7.2 1.0-5.5 1.0-5.7

* OR, odds ratio; adjusted for age, history of radiation therapy, body mass index, history of mother or sister with breast cancer, history of breast cancer, parity, and history of benign breast disease. t Cl, confidence interval. $ Also adjusted for usual number of alcoholic drinks per day. § Also adjusted for duration of active smoking (OR of 10 additional years of active smoking = 0.93, 95% Cl 0.76-1.14). H Also adjusted for number of cigarettes per day (OR of 10 additional cigarettes per day = 0.85, 95% Cl 0.66-1.09).

and 20 years, or at age 21 years or older had odds ratios of 4.5 (95 percent Cl 1.2-16), 3.8 (95 percent Cl 1.1-13), and 2.4 (95'percent Cl 0.9-6.1), respectively. We measured the same relation among women who were ever-active smokers and who lived with another active smoker. In this group, women who were first exposed to passive smoke before age 12 years, between age 12 and 20 years, or at age 21 years or older had odds ratios of 7.5 (95 percent Cl 1.6-36), 3.9 (95 percent Cl 0.8-20), and 4.7 (95 percent Cl 1.6-14), respectively. DISCUSSION

The effect of tobacco smoking on the occurrence of breast cancer has been the topic of some debate. Its effect on the occurrence of other cancers is well known, yet most studies (2, 15-19) of the association between smoking and breast cancer have shown either a weakly positive or a null effect. Two studies (20, 21) found a protective effect. A direct association between passive tobacco smoking and the occurrence of breast cancer has been observed more consistently (3, 10, Am J Epidemiol Vol. 149, No. 1, 1999

13). A recent series of letters in the Journal (22-24) suggested that the disparate association between active smoking and breast cancer occurrence may arise from the failure to exclude passive smokers from the reference group in most studies, from differential distribution of genetically susceptible subpopulations across the studies, or both. A previous study (3) that also separated active and passive smokers measured an excess risk of breast cancer among both ever-active smokers and passive-only smokers. The definitions of index conditions used in the earlier study preclude direct comparison with our results. Our study found that passive exposure to cigarette smoke appears to affect the first stage of breast carcinogenesis. First exposure at an age prior to breast tissue development confers the highest risk. First exposure during adolescence or as a young adult confers an intermediate risk, and first exposure as an adult confers the lowest risk. This pattern of declining risk with age at first exposure to passive smoke likely reflects the declining susceptibility of breast tissue to chemical mutagens, as depicted in figure 1. Age at first exposure

Downloaded from aje.oxfordjournals.org by guest on July 15, 2011

137:338

Cigarettes per day (no.)§ 20

10

Lash and Aschengrau TABLE 3. Measured odds ratios of breast cancer associated with exposure to passive cigarette smoke in the residence, Massachusetts, 1983-1986 Exposure condition

Case:control ratio

OR*

95% Clf

Never active, never passive

40:139

1

Passive onlyt

80:267

2.0

1.1-3.7

Duration of passive smoke exposure (years) 20

28:56 43:148

3.2 2.1

1.5-7.1 1.0-4.1

Smoking relative to first term pregnancy All before first pregnancy All after first pregnancy Both before and after first pregnancy

6:15 35:102 21:63

2.8 2.4 2.2

0.8-9.9 1.2-5.1 1.1-4.7

Age (years) at first exposure, passive-only smokers§ 21

14:25 11:30 34:118

4.5 3.8 2.4

1.2-16 1.1-13 0.9-6.1

Age (years) at first exposure to passive smoke, ever-active smokers§,H 21

26:33 10:31 46:105

7.5 3.9 4.7

1.6-36 0.8-20 1.6-14

to active cigarette smoking did not reflect the same pattern, possibly because few women were active smokers when their breast tissue was most susceptible. The risks from first being exposed to passive smoke at an early age then declined with the duration of exposure to passive cigarette smoke and the intensity of exposure to active cigarette smoking. Furthermore, women whose entire exposure to active cigarette smoking preceded their first pregnancy were at the highest risk. Their breast tissue would have been susceptible to mutagenesis while they smoked and would not have benefitted from an antiestrogenic effect with later exposure to cigarette smoke. Women who began to smoke before their first pregnancy and continued to smoke after their first pregnancy did not share the elevated risk. Finally, women who quit smoking during a plausible induction period of 5-15 years before diagnosis, when estrogen might play an important role in breast cancer promotion, also had a higher risk. These measurements might be attributable to alternative explanations. We ascertained exposure by retrospective survey, so the study design was susceptible to recall bias. However, the substantial associations that were found were within the strata defined by time periods calculated from a series of responses. We do not expect these derived exposures to be susceptible to recall bias. Furthermore, neither active nor passive

exposure to cigarette smoke has been closely related to breast cancer risk, so recall of exposure should not depend on disease status. However, the widely held perception that smoking causes cancer may contribute to some disease-dependent recall of exposure to tobacco smoke. We considered whether misclassification of passive smoking might account for the measurements that we observed. We did not determine whether respondents were exposed to passive smoke in the workplace. The reference population of never-active, never-passive smokers may have included respondents who were exposed to passive smoke in the workplace, primarily as adults. Workplace exposure of the reference group to passive smoke could have been antiestrogenic, reducing the risk of breast cancer and biasing the odds ratios away from the null value. To reach this conclusion, it would be necessary to assume that exposure to passive smoke in the residence is inversely related to the frequency of exposure to passive smoke in the workplace. If workplace exposure to passive smoke were similarly divided between those with a history of passive smoking in the residence and those without such a history, then no differential bias would exist. Model misspecification provides another opportunity for misclassification. In many analyses, we included a term for duration of exposure to passive smoke, duraAm J Epidemiol

Vol. 149, No. 1, 1999

Downloaded from aje.oxfordjournals.org by guest on July 15, 2011

* OR, odds ratio; adjusted for age, history of radiation therapy, body mass index, history of mother or sister with breast cancer, history of breast cancer, parity, and history of benign breast disease. t Cl, confidence interval. t Also adjusted for usual number of alcoholic drinks per day. § Also adjusted for duration of passive smoking (OR of 10 additional years of passive smoking = 0.96, 95% Cl 0.76-1.22). I Also adjusted for duration of active smoking (OR of 10 additional years of active smoking = 0.93, 95% Cl 0.76-1.14) and number of cigarettes per day (OR of 10 additional cigarettes per day = 0.85, 95% Cl 0.66-1.09).

Cigarette Smoking and Breast Cancer

Am J Epidemiol

Vol. 149, No. 1, 1999

explained that the downward trend shows a change in the prevalence of a causal or protective risk factor other than those already identified. Our data suggest one possible explanation. The unidentified factor may be the antiestrogenic potency of tobacco smoke among women with a lifelong history of either active smoking or exposure to passive smoke. However, as ex-smokers become more prevalent and active smokers less prevalent (26), women will be exposed to only the initiating stages of active and passive cigarette smoke, and the birth cohort trend may reverse direction. The odds ratios measured in this study, although imprecise, comport with an underlying biologic model of breast carcinogenesis. Cigarette smoking causes cancer in organs not in direct contact with smoke (27), but it may also be antiestrogenic (11, 12). Taken together, these observations suggest both a real association and a need for further examination of the relation between exposure to cigarette smoke and the occurrence of breast cancer. Future studies might focus on segregation of the effects of passive smoking and active smoking, the minimum duration and intensity of active and passive smoking necessary to initiate breast carcinogenesis, the interaction between time period of exposure and milestones of breast tissue development, or the precise time period before diagnosis when women are the most susceptible to quitting smoking.

ACKNOWLEDGMENTS

The authors thank Drs. Rebecca Silliman, Laura Green, and Edmund Crouch for their comments on drafts of the manuscript.

REFERENCES 1. Madigan MP, Ziegler RG, Benichou J, et al. Proportion of breast cancer cases in the United States explained by wellestablished risk factors. J Natl Cancer Inst 1995;87:1681-5. 2. Palmer JR, Rosenberg L. Cigarette smoking and the risk of breast cancer. Epidemiol Rev 1993;15:145-56. 3. Morabia A, Bernstein M, H6ritier S, et al. Relation of breast cancer with passive and active exposure to tobacco smoke. Am J Epidemiol 1996;143:918-28. 4. Ambrosone CB, Freudenheim JL, Graham S, et al. Cigarette smoking, N-acetyltransferase 2 genetic polymorphisms, and breast cancer risk. JAMA 1996;276:1494-501. 5. Shields PG, Ambrosone CB, Graham S, et al. A cytochrome P4502E1 genetic polymorphism and tobacco smoking in breast cancer. Mol Carcinog 1996; 17:144-50. 6. Ambrosone CB, Shields PG. Molecular epidemiology of breast cancer. Prog Clin Biol Res 1997;396:93-9. 7. Spicer DV, Pike MC. Sex steroids and breast cancer prevention. J Natl Cancer Inst Monogr 1994;16:139^}7. 8. Russo J, Russo M. Toward a physiological approach to breast cancer prevention. Cancer Epidemiol Biomarkers Prev 1994;3:353-64.

Downloaded from aje.oxfordjournals.org by guest on July 15, 2011

tion of active smoking, and/or intensity of active smoking. These terms all indicated a downward trend in the odds ratios with increasing exposure. We interpret this downward trend as a measure of the antiestrogenic potency of cigarette smoke. Under these models, those women whose duration of smoking was 0 years and whose intensity of smoking was 0 cigarettes per day would have the highest risk, which is not logically possible. We expect that some minimum exposure must be necessary to confer the initial risk, which is then mitigated by an antiestrogenic effect. These data are not sufficient to measure the minimum exposure necessary. The associations between breast cancer occurrence and age at first exposure to passive smoke persisted, although they were reduced, in models with no measure of duration or intensity of exposure. Many participants, particularly those who had proxy respondents, could not be characterized according to some covariates. For example, we did not know the age at menarche for 407 of 1,028 women or the age at menopause for 254 of 934 postmenopausal women. Age at menopause depends on smoking status (12), so we would not have adjusted our odds ratios for confounding by that variable in any event. Adjusting for premenopausal or postmenopausal status at index year had no effect on the odds ratios. We adjusted for age at menarche in the subset of 621 women for whom this information was reported; the odds ratios migrated away from the null value. However, the odds ratios increased for this subset of women relative to the odds ratios measured in the complete data set, whether or not the measure of age at menarche was included. This finding suggests that the changes in estimated associations are more properly attributable to selection of the subset of subjects than to control of confounding by age at menarche. The level of precision was low for some measures of association. The widest 95 percent confidence interval that did not include the null value was 1.6-36. It applied to the odds ratio of 7.5 estimated for everactive smokers who were first exposed to passive smoke before age 12 years. If the odds ratios were null, the interval would have been 0.2^.8, which is still wide. This imprecision, in combination with a coherent picture of effects based on underlying biology, the known carcinogenic effect of cigarette smoking, and its observed antiestrogenic potency, emphasizes the importance of examining these exposures in larger studies. Should the measurements described in this paper prove to be valid and accurate, then certain implications would merit consideration. Tarone et al. (25) measured a downward trend in US breast cancer mortality rates for all birth cohorts beginning in about 1940. They

11

12

Lash and Aschengrau

9. Pike MC, Spicer DV, Dahmoush L, et al. Estrogens, progestogens, normal breast cell proliferation, and breast cancer risk. Epidemiol Rev 1993;15:17-35. 10. Smith SJ, Deacon JM, Chilvers CED, et al. Alcohol, smoking, passive smoking, and caffeine in relation to breast cancer risk in young women. Br J Cancer 1994;70:112-19. 11. MacMahon B, Trichopoulos D, Cole P, et al. Cigarette smoking and urinary estrogens. N Engl J Med 1982;307:1062-5. 12. Baron JA, La Vecchia C, Levi F. The antiestrogenic effect of cigarette smoking in women. Am J Obstet Gynecol 1990; 162:502-14. 13. Wells AJ. Breast cancer, cigarette smoking, and passive smoking. (Letter). Am J Epidemiol 1991;133:208-10. 14. Aschengrau A, Ozonoff D, Coogan P, et al. Cancer risk and residential proximity to cranberry cultivation in Massachusetts. Am J Public Health 1996;86:1289-96. 15. Braga C, Negri E, La Vecchia C, et al. Cigarette smoking and the risk of breast cancer. Eur J Cancer Prev 1996;5:159-64. 16. Baron JA, Newcomb PA, Longnecker MP, et al. Cigarette smoking and breast cancer. Cancer Epidemiol Biomarkers Prev 1996;5:399-403. 17. Ranstam J, Olsson H. Alcohol, cigarette smoking, and the risk of breast cancer. Cancer Detect Prev 1995; 19:487-93. 18. Calle EE, Miracle-McMahill HL, Thun MJ, et al. Cigarette smoking and risk of fatal breast cancer. Am J Epidemiol 1994;139:1001-7.

19. Bennicke K, Conrad C, Sabroe S, et al. Cigarette smoking and breast cancer. BMJ 1995;310:1431-3. 20. Vessey M, Baron J, Doll R, et al. Oral contraceptives and breast cancer: final report of an epidemiological study. Br J Cancer 1983;47:455-62. 21. O'Connell DL, Hulka BS, Chambless LE, et al. Cigarette smoking, alcohol consumption, and breast cancer risk. J Natl Cancer Inst 1987,78:229-34. 22. Wells AJ. Re: "Breast cancer, cigarette smoking, and passive smoking." (Letter). Am J Epidemiol 1998,147:991-2. 23. Morabia A, Bernstein M, Heritier S. Re: "Smoking and breast cancer: reconciling the epidemiologic evidence by accounting for passive smoking and/or genetic susceptibility." (Letter). Am J Epidemiol 1998; 147:992-3. 24. Whidden P. Re: "Relation of breast cancer with passive and active exposure to tobacco smoke." (Letter). Am J Epidemiol 1998; 147:994. 25. Tarone RE, Chu KC, Gaudette LA. Birth cohort and calendar period trends in breast cancer mortality in the United States and Canada. J Natl Cancer Inst 1997,89:251-6. 26. Surveillance for selected tobacco-use behaviors—United States, 1900-1994. CDC surveillance summaries, November 18, 1994. MMWR Morb Mortal Wkly Rep 1994;43. (Publication no. SS-3). 27. DeVita VT, Hellman S, Rosenberg SA. Cancer: principles and practice of oncology. 3rd ed. PhOadelphia, PA: JB Lippincott Co, 1989. Downloaded from aje.oxfordjournals.org by guest on July 15, 2011

Am J Epidemiol

Vol. 149, No. 1, 1999