Accumulating evidence on passive and active smoking and breast cancer risk

Abstract

The aim of the study was to examine the risk of breast cancer associated with passive and active smoking and to explore risk heterogeneity among studies. Nineteen of 20 located published studies of passive smoking and breast cancer risk among women met basic quality criteria. Pooled relative risk estimates for breast cancer were calculated for 1) life-long non-smokers with regular passive exposure to tobacco smoke and 2) women who smoked. They were compared to women categorized as never regularly exposed to tobacco smoke. The pooled risk estimate for breast cancer associated with passive smoking among life-long non-smokers was 1.27 (95% confidence interval (CI), 1.11–1.45). In the subset of 5 studies (all case-control studies) with more complete exposure assessment (quantitative long-term information on the 3 major sources of passive smoke exposure: childhood, adult residential and occupational), the pooled risk estimate for exposed non-smokers was 1.90 (95%CI, 1.53–2.37). For the 14 studies with less complete passive exposure measures the risk was 1.08 (95%CI, 0.99–1.19) overall, 1.16 for 7 case-control and 1.06 for 7 cohort studies, although dose-response results in 3 of 4 Asian cohort studies suggested increased risk. The overall premenopausal breast cancer risk associated with passive smoking among life-long non-smokers was 1.68 (95%CI 1.33–2.12), and 2.19 (95% CI 1.68–2.84) for the 5 of 14 studies with more complete exposure assessment. For women who had smoked the breast cancer risk estimate was 1.46 (95%CI 1.15–1.85) when compared to women with neither active nor regular passive smoke exposure; 2.08 (95% CI 1.44–3.01) for more complete and 1.15 (95% CI 0.92–1.43) for less complete passive exposure assessment. Studies with thorough passive smoking exposure assessment implicate passive and active smoking as risk factors for premenopausal breast cancer. Cohort studies with thorough passive smoking assessment would be helpful and studies exploring biological mechanisms are needed to explain the unexpected similarity of the passive and active risks. © 2005 Wiley-Liss, Inc.

Breast cancer is one of the most commonly diagnosed cancers among women in Western countries.1 Despite the well-characterized role of genetics,2 and reproductive risk factors,2 and physical activity3 and the suggested role of alcohol,4 more than half of breast cancer risk remains unexplained.5

Although recent analyses have suggested increased active smoking risk among women with particularly long and/or prepregnancy exposure,6, 7 most studies that have examined active smoking and breast cancer have concluded that there is little indication of increased risk.4, 8 Paradoxically, most studies of breast cancer and passive smoking (second-hand smoke, environmental tobacco smoke or involuntary smoking) have suggested an increase in breast cancer risk for exposure to tobacco smoke.9, 10 The issue is complicated by the fact that although many of the studies observe increased breast cancer risk with passive smoking, 3 large American cohort studies found little increase in risk.

The analysis presented here focuses on passive smoking risk, unlike a thorough recent review which targeted active smoking and breast cancer,11 and goes beyond earlier passive smoking-breast cancer reviews9, 10 in 3 important ways: inclusion of 7 more recently published studies, separate risk estimation for premenopausal women and quantitative exploration of the substantial heterogeneity in the observed relative risks in the individual studies.

Material and methods

Studies of passive smoking and breast cancer were identified though MEDLINE from December 1966 through December 2004, inclusive using the terms “tobacco smoke pollution”, “passive smoking”, and “environmental tobacco smoke”, each combined with “breast neoplasms” and with “breast cancer”. Reviews of the literature8, 9, 10, 11, 12 were also examined for reference to further studies. Copies of all located studies were obtained and a translation was obtained of a study in Chinese.13

For inclusion here, studies had to be published and meet the following minimum study quality criteria: a) data collection utilizing established epidemiologic designs (case-control or cohort study), b) some quantitative measure of exposure to passive smoking; and c) allowance for comparisons of passive smoking risk among never-smoking women.

To explore heterogeneity of the study results, the impact on risk estimates were examined with regard to 1) study design (case-control vs. cohort), 2) endpoints (breast cancer incidence vs. mortality), 3) the population studied (Asian, European or North American); 4) menopausal status at diagnosis (premenopausal or postmenopausal); 5) confounder control (control for other breast cancer risk factors) and 6) exposure assessment (the completeness of the passive smoking measure). Risk by population was studied because breast cancer general risk factor profiles for Asian women have varied somewhat from those of Western women. There was an attempt made to locate in each paper risk ratios for passive smoking and for active smoking, where exposed women were compared to women who had never either smoked nor reported any period of regular exposure to second-hand smoke. For several of the earlier studies, the desired comparisons were published in letters by Wells12, 14 based on personal communications between Wells and the individual study authors.

The thoroughness of the passive smoking exposure assessment was examined because many studies did not account for lifetime regular exposure at home and at work and the resultant exposure misclassification has the potential to seriously dilute risks, primarily through substantial contamination of the non-exposed referent genes.15 Studies that collected quantitative long-term information on the 3 major potential lifetime sources of passive smoking exposure (childhood exposure from parents, adult residential exposure and adult occupational exposure)16 were considered as “unlikely” to have missed important sources of passive exposure. Studies not collecting data on all 3 passive exposure sources or only collecting a cross-sectional measure of any of them (e.g., current occupational exposure) were considered to be “likely” missing important passive exposure.

Pooled relative risk estimates were calculated using the method of DerSimonian and Laird,17 which weighs each study using the within and between study variance. A random effects model was chosen because it is considered to be more conservative and generalizable than a fixed effects model. For case-control studies, the odds ratio was used to approximate the relative risk. The evaluation of active smoking risk was made using never-smokers with no regular exposure to passive smoking as the referent group.

Results

Table I summarizes basic study characteristics. Nineteen13, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35 of the 20 published studies of breast cancer and passive smoking met the minimum study quality criteria for inclusion in the analyses. One study was eliminated36 because there were internal contradictions in the presented data that could not be resolved through contact with the author. Table II summarizes the passive smoking exposure assessment in each study.

Table I. Published Epidemiologic Studies of Passive Smoking and Breast Cancer. Study Characteristics

Study	Place	Years	Study type	Outcome	Age range	Cases (never-smokers)	Non-diseased for referent	Other risk factors controlled for referent1
1 BBD, benign breast disease; BMI, body mass index; FHBC, family history of breast cancer. 2 Delfino et al. cases and controls selected from 391 women who had suspicious breast masses detected clinically or mammographically. Cases were the 113 where histopathology indicated breast cancer. Controls were the 278 with benign breast disease. Controls for this analysis were from the 110 labeled “low risk controls,” i.e., those with “normal breast or benign breast disease histopathologies with nonproliferative changes”.
Hirayama 199218	Japan	65–81	Cohort	Death	>40	115	91,540	Age
Sandler et al., 198519	USA - North Carolina	79–81	Case-control	Diagnosis	15–59	32	177	Age, Race, Gender, FHBC
Smith et al., 199420	United Kingdom	85–88	Case-control	Diagnosis	<36	94	100	Age, region, age at menarche, parity, age at 1^st full term pregnancy, breastfeeding, oral contraceptives, BBD, FHBC, alcohol (age 18)
Morabia et al., 199621	Switzerland	92–93	Case-control	Diagnosis	<75	126	620	Age, education, BMI, age at menarche, age at 1^st live birth, oral contraceptives, FHBC, BBD
Millikan et al., 199822	USA - North Carolina	93–96	Case-control	Diagnosis	>20	247	253	Age, race, age at menarche, parity, FHBC, BBD, alcohol
Lash et al., 199923	USA - Massachusetts	83–86	Case-control	Diagnosis	All	120	406	Age, BMI, parity, BBD, FHBC, history of breast cancer, radiation therapy
Jee et al., 199924	South Korea	94–97	Cohort	Diagnosis	18–65	138	157,298	Age, husband's age, socioeconomic status, residency, husband's vegetable consumption, husband's occupation.
Zhao et al., 199913	China - Chengdu	94–97	Case-control	Diagnosis	26–82	252	259	Age, menarche to menopause >35 years, oral contraceptives, BBD, breast feeding, soybean intake
Delfino et al., 200025	USA - California	NR	Modified2 Case-control	Diagnosis	>39	64	602	Age, menopausal status, FHBC
Johnson et al., 200026	Canada - 8 provinces	94–97	Case-control	Diagnosis	20–74	608	727	Age, province, education, age at menarche, age at 1^st pregnancy (> 5 months), live births, breastfeeding, BMI, height, alcohol, physical activity (2 years before interview)
Wartenberg et al., 200027	USA, Cancer Prevention Trial II	82–94	Cohort	Death	30–70+	669	146,488	Age, race, education, BMI, age at 1^st live birth, age at menarche, age at menopause, spontaneous abortions, oral contraceptives, estrogen replacement therapy, history of breast cysts, alcohol, dietary fat, vegetable intake, woman's occupation, spouse's occupation
Liu et al., 200036	China	94–96	Case-control	Diagnosis	24–55	186	186	Age, age at first birth, age at menarche, age at menopause, BBD, education, family history of cancer, history of suffering hospitalized disease, height, marital status, occupation, radiation exposure, socioeconomic status, stress.
Nishino et al., 200128	Japan	84–92	Cohort	Diagnosis	>40	67	9671	Age, age at first birth, age at menarche, number of live births, alcohol, BMI, fruit intake, vegetable consumption
Egan et al., 200229	USA, Nurses Health Study	82–96	Cohort	Diagnosis	36–61	1,138	78,206	Age, region, age at menarche, age at first pregnancy, age at menopause, height, BBD, BMI, FHBC, alcohol, vegetable consumption, estrogen replacement therapy
Lash et al., 200230	USA-Massachusetts	87–93	Case-control	Diagnosis	All	305	249	Age, age at first pregnancy, parity, BBD, BMI, FHBC, history of breast cancer, radiation therapy, history of breast cancer, BBD, alcohol
Kropp et al., 200231	Germany	NR	Case-control	Diagnosis	<51	197	454	Age, education, region, age at menopause, FHCB, lactation, BMI, alcohol
Reynolds et al., 200432	USA - California	95–00	Cohort	Diagnosis	All	1,174	76,534	Age, race, age at menarche, age at first full-term pregnancy, parity, menopausal status, BMI, FHBC, alcohol consumption, physical activity, hormone replacement therapy
Shrubsole et al., 200433	China, Shanghai	96–98	Case-control	Diagnosis	25–64	1,013	1,117	Age, education, household income, age at menarche, age at first live birth, age at menopause, menopausal status, BMI, FHBC, physical activity, history of fibroadenoma
Gammon et al., 200434	U.S.A., Long Island, New York State	96–97	Case-control	Diagnosis	24–98	598	627	Age, BBD, BMI at age 20, FHBC, history of fertility problems, number of pregnancies, menopausal status, weight in year prior to reference date
Hanaoka et al., 200435	Japan, 14 administrative districts across Japan	90–99	Cohort	Diagnosis	40–59	162	20,011	Age, education, public health center, employment status, BMI, FHBC, BBD, age at menarche, number of births, menopausal status, hormone use, alcohol

Table II. Published Epidemiologic Studies of Passive Smoking and Breast Cancer. Passive Smoking Exposure Assessment

Study	Summary of exposure measures	Childhood exposure	Adult residential exposure	Occupational exposure	Other exposure	Important exposure missed?
1 Current occupational exposure in 1982 collected, but only husband's smoking history used for main analysis and husbands history not used in analysis of 1982 cross-sectional exposure.
Hirayama, 1992	Husband's smoking history	No	Husband's smoking history	No		Likely
Sandler et al., 1985	Years smoked by spouse	No	Years smoked by spouse	No		Likely
Smith et al., 1994	Lifetime residential and occupational	Detailed history	Detailed history	Detailed history		Unlikely
Morabia et al., 1996	Lifetime residential and occupational and social	Detailed history	Detailed history	Detailed history	Social	Unlikely
Millikan et al., 1998	Residential	Years with smoker at home	Any housemate who smoked.	No	No	Likely
Lash et al., 1999	Lifetime residential	Yes	Yes	No		Likely
Zhao et al., 1999	Lifetime passive smoking history	Yes	Yes	Yes	Yes	Unlikely
Jee et al., 1999	Husband's smoking history	No	Husband's smoking history	No		Likely
Delfino et al., 2000	Adult residential	No	Adult residential	No	No	Likely
Johnson et al., 2000	Lifetime residential and occupational	Number of smokers in each residence	Number of smokers in each residence	For each job: Number of smokers who smoked regularly in immediate work area	No	Unlikely
Wartenberg et al., 2000	Husband's smoking history	No	Husband's smoking history	Current in 1982 only1	No	Likely
Liu et al., 2000.	Childhood, youth, adult, home, work, Number of smokers, cigarettes/day	Yes	Yes	Yes	No	Unlikely
Nishino et al., 2001	Currently living with smoker(s) in 1984	No	Husband, wife, father, mother, children or others living in the household who smoke (currently in 1984)	No	No	Likely
Egan et al., 2002	Parental, years lived as adult with a smoker, current (1982) home and work	Mother, father or both parents smoked	Years lived with smoker, current in 1982	Current, in 1982 only	No	Likely
Lash et al., 2002	Lifetime residential	Yes	Yes	No	No	Likely
Kropp et al., 2002	Years exposed to age 50	Years exposed	Years exposed	Years exposed	No	Unlikely
Reynolds et al., 2004	Lifetime residential	Yes	Yes	No	No	Likely
Shrubsole et al., 2004	Husband and workplace	No	Husband's smoking history	During prior 5 years	No	Likely
Gammon et al., 2004	Lifetime residential	Yes	Yes	No	No	Likely
Hanaoka et al., 2004	Lived with any regular smoker, and exposure outside of the home in 1990	Lived with any smokers before age 20	Lived with any smokers after age 20	Current, in 1990 only	Current, in 1990 only	Likely

Passive smoking and breast cancer in never smokers

The pooled estimate for breast cancer risk associated with passive smoking in women who had never smoked was 1.27 (95% CI 1.11–1.45), based on the 19 included studies (Table III). Individual study risk estimates were heterogeneous (test for heterogeneity, p < 0.001). Stratification of the studies by continent, time period of publication or outcome measure (breast cancer mortality or incidence) did not consistently separate studies into those with higher observed risk from those with little increase in risk. Study method was more predictive, with 7 of the 12 case-control studies suggesting increased risk. While, the 3 large American cohorts27, 29, 32 suggested little increase in risk, 3 cohort studies from Asia18, 24, 35 suggested small elevations in the risk estimates and a 4th a decrease in risk.28

Table III. Summary Breast Cancer Risk Estimates for Exposure to Second Hand Smoke1 among Women Who Never Smoked

Study	Important source(s) of passive exposure not collected	Relative risk (RR)2	95% Confidence interval		Test for heterogeneity	Number of studies
			lower	upper
1 For several studies, summary overall risk estimates had to be calculated using component risks and confidence intervals reported in the paper and combined using Equation 16–8 from Rothman (1986)^62. For several of the earlier studies, risk estimates for the desired comparisons were published in letters by Wells (1991, 1992, 1998) after Wells had personal communication with the authors. Combined estimates: Hirayama 1992, (Wells, 1998), combined counts for age 50–59 and 60–69; For Smith et al. (1994) estimated overall passive smoking risk calculated by summarizing unadjusted lifetime exposure categories (1-200, >200 cigarette-years); Wartenberg et al., 2000, combined estimates for ex-smoking and current smoking husbands; Johnson et al., 2000, combined estimates for pre and postmenopausal risks. 2 Odds ratios assumed to be a reasonable approximation for the relative risk in case-control studies. 3 Summary relative risk estimates were calculated using the method of DerSimonian and Laird.
Cohort
Hirayama, 1992	Likely	1.32	0.83	2.09
Jee et al., 1999	Likely	1.3	0.9	1.8
Wartenberg et al., 2000	Likely	1.0	0.8	1.2
Nishino et al., 2001	Likely	0.58	0.32	1.1
Egan et al., 2002	Likely	1.07	0.88	1.30
Reynolds et al., 2004	Likely	0.94	0.82	1.07
Hanaoka et al., 2004	Likely	1.1	0.8	1.6
Case-control
Sandler et al., 1985	Likely	1.62	0.76	3.44
Millikan et al., 1998	Likely	1.3	0.9	1.9
Lash et al., 1999	Likely	2.0	1.1	3.7
Delfino et al., 2000	Likely	1.86	0.81	4.27
Lash et al., 2002	Likely	0.85	0.62	1.20
Shrubsole et al., 2004	Likely	1.0	0.8	1.3
Gammon et al., 2004	Likely	1.04	0.81	1.35
Case-control
Smith et al., 1994	Unlikely	2.53	1.19	5.36
Morabia et al., 1996	Unlikely	2.3	1.5	3.7
Zhao et al., 1999	Unlikely	2.36	1.66	3.66
Johnson et al., 2000	Unlikely	1.48	1.06	2.07
Kropp et al., 2002	Unlikely	1.61	1.08	2.39
Summary RR3 all studies		1.27	1.11	1.45	p < 0.001	19
Summary RR - important passive exposure sources collected		1.90	1.53	2.37	p = 0.247	5
Summary RR - important passive exposure sources missed		1.08	0.99	1.19	p = 0.101	14
Cohort studies with important passive exposure sources missed		1.06	0.97	1.16	p = 0.398	7
Case-control studies with important passive exposure soucrces missed		1.16	0.95	1.42	p = 0.104	7

The studies reporting increased risk, however, were best differentiated from studies not suggesting increased risk by the quality of the passive smoking exposure measure. When the summary was limited to the 5 studies (all case-control studies),13, 20, 21, 26, 31 which enumerated lifetime measures of exposure to the major sources of lifetime passive smoke exposure (childhood residential, adult residential and occupational), the summary risk estimate was 1.90 (95% CI 1.53–2.37). Among the studies that did not enumerate all 3 major passive exposure sources the risk estimate was 1.08 (95% CI 0.99–1.19), with cohort and the case-control studies yielding estimates of 1.06 and 1.16, respectively.

The pooled RR estimate for premenopausal breast cancer risk was 1.68 (95% CI 1.33–2.12), based on the 14 studies where premenopausal risk estimates or estimates for younger women (age ≤50) were available (Table IV). Ten of the 14 individual risk estimates suggested increased risk. All 5 of the studies that had better passive exposure assessment measures included a premenopausal risk estimate and each was statistically significant at the 95% confidence level. These 5 studies yielded a premenopausal summary risk estimate of 2.19 (95% CI 1.68–2.84).

Table IV. Premenopausal Breast Cancer Summary Risk Estimates for Exposure to Second Hand Smoke Among Women Who Never Smoked

Study	Important source(s) of passive exposure not collected	Relative risk (RR)1	95% Confidenceinterval		Test for heterogeneity	Number of studies
			Lower	Upper
1 Odds ratios assumed to be a reasonable approximation for the relative risk in case-control studies. 2 Summary relative risk estimates were calculated using the method of DerSimonian and Laird. For several of the earlier studies, the desired comparisons were published in letters by Wells (1991, 1992, 1998) after personal communication with the authors. For Smith et al. (1994) estimated overall passive smoking risk calculated by summarizing unadjusted lifetime exposure categories (1–200, >200 cigarette-years). Lash reports there are too few premenopausal never smoking cases to make meaningful estimates (personal correspondence with author KCJ, 2001). For Zhao, separate pre- and post-menopausal risk estimates provided by Zhao (personal correspondence with author KCJ, 2001). For Wartenberg premenopausal estimate is actually women who were under age 50 in 1982. Only some of the women would be premenopausal at diagnosis (unspecified number). Calculated combined risk for current smoking husbands plus ex-smoking husbands. For Reynolds, estimate is for women pre- or perimenopausal.
Cohort
Hirayama, 1992	Likely	1.5	0.5	4.2
Jee et al., 1999	Likely
Wartenberg et al., 2000	Likely	1.14	0.82	1.59
Nishino et al., 2001	Likely		Older cohort
Egan et al., 2002	Likely		Not reported
Reynolds et al., 2004	Likely	0.93	0.71	1.22
Hanaoka et al., 2004	Likely	2.6	1.3	5.2
Case-control
Sandler et al., 1985	Likely	7.1	1.6	31.3
Millikan et al., 1998	Likely	1.5	0.8	2.8
Lash et al., 1999	Likely		Older cohort
Delfino et al., 2000	Likely	2.69	0.91	8
Lash et al., 2002	Likely		Older cohort
Shrubsole et al., 2004	Likely	1.1	0.78	1.56
Gammon et al., 2004	Likely	1.21	0.78	1.90
Case-control
Smith et al., 1994	Unlikely	2.53	1.19	5.36
Morabia et al., 1996	Unlikely	3.6	1.6	8.2
Zhao et al., 1999	Unlikely	2.56	1.63	4.01
Johnson et al., 2000	Unlikely	2.3	1.2	4.6
Kropp et al., 2002	Unlikely	1.61	1.08	2.39
Summary RR2 all studies		1.68	1.33	2.12	p = 0.002	14
Summary RR - important passive exposure sources collected		2.19	1.68	2.84	p = 0.363	5
Summary RR - important passive exposure sources missed		1.35	1.07	1.71	p = 0.081	9
Cohort studies with important passive exposure sources missed		1.29	0.93	1.81	p = 0.051	4
Case-control studies with important passive exposure sources missed		1.48	1.00	2.21	p = 0.095	5

Active smoking and breast cancer risk after control for passive smoking

Active smoking risk ratios were reported in 13 of the 19 included studies (Table V). The pooled estimate for women who had smoked was 1.46 (95% CI 1.15–1.85) compared to women who never were regularly exposed to tobacco smoke. Six of the individual estimates that suggested increased risk were statistically significant at the 95% confidence level. Individual study risk estimates were again heterogeneous and best differentiated by the quality of the passive smoking exposure measure: with incomplete coverage of major sources of lifetime passive exposure (8 studies), active smoking risk was estimated at 1.15 (95% CI 0.92–1.43), and with more complete passive exposure assessment (5 case-control studies), active smoking risk was estimated at 2.08 (95% CI 1.44–3.01), 2.11 (95% CI 1.31–3.40) for premenopausal women (data not shown).

Table V. Summary Breast Cancer Risk Estimates for Active Smoking Compared to Women Never Smokers, Never Regularly Exposed to Second Hand Smoke1

Study	Important source(s) of passive exposurenot collected	Relative risk(RR)2	95% Confidence interval		Test for heterogeneity	Number of studies
			Lower	Upper
1 For several of the earlier studies, the desired comparisons were published in letters by Wells (1991, 1992, 1998) after personal communication with the authors. For Johnson et al., pre and post-menopausal results were combined. For Reynolds, combined estimate for ex and current smokers provided by Reynolds (personal communication with M. Miller., 2005). 2 Odds ratios assumed to be a reasonable approximation for the relative risk in case-control studies. 3 Summary relative risk estimates were calculated using the method of DerSimonian and Laird.
Cohort
Hirayama, 1992	Likely	1.59	1.01	2.52
Jee et al., 1999	Likely	—
Wartenberg et al., 2000	Likely	—	—	—
Nishino et al., 2001	Likely	—	—	—
Egan et al., 2002	Likely	—		—
Reynolds et al., 2004	Likely	1.06	0.92	1.21
Hanaoka et al., 2004	Likely	1.7	1.0	3.0
Case-control
Sandler et al., 1985	Likely	1.21	0.58	2.52
Millikan et al., 1998	Likely	—
Lash et al., 1999	Likely	2.0	1.1	3.6
Delfino et al., 2000	Likely	0.97	0.50	1.87
Lash et al., 2002	Likely	0.72	0.55	0.95
Shrubsole et al., 2004	Likely	—	—	—
Gammon et al., 2004	Likely	1.13	0.91	1.42
Case-control
Smith et al., 1994	Unlikely	2.69	0.98	4.12
Morabia et al., 1996	Unlikely	3.0	1.9	4.8
Zhao et al., 1999	Unlikely	3.54	1.36	9.18
Johnson et al., 2000	Unlikely	1.7	1.2	2.4
Kropp et al., 2002	Unlikely	1.31	0.91	1.89
Summary RR3 all studies		1.46	1.15	1.85	p < 0.001	13
Summary RR - important passive exposure sources collected		2.08	1.44	3.01	p = 0.031	5
Summary RR - important passive exposure sources missed		1.15	0.92	1.43	p = 0.010	8
Cohort studies with important passive exposure sources missed		1.35	0.91	1.99	p = 0.059	3
Case-control studies with important passive exposure sources missed		1.05	0.76	1.45	p = 0.026	5

Discussion

This analysis suggests that women regularly exposed to passive smoking and women who smoke may have increased breast cancer risk when compared to women never regularly exposed to tobacco smoke, particularly premenopausal risk. Furthermore, the completeness of the passive smoking measure impacts substantively on the level of risk observed. Concerns raised about the plausibility of the relationship center on 1) the biological plausibility of observing breast cancer risks of similar magnitude for passive and active smoking, 2) the fact that studies of active smoking (where no passive smoking data were collected) observe small, if any, increases in overall risk,4, 8 3) the fact that the 3 large American cohort studies do not observe risk increases and 4) the potential for bias, particularly recall bias among the case-control studies.

Biological plausibility

Cigarette smoke mutagens have been reported in the breast fluid of nonlactating women,37, 38 and nicotine has been found in greater concentrations in the breast fluid of smokers than in the plasma.39 Animal studies demonstrate that mammary tumours in rats, mice and/or hamsters can be induced by a number of tobacco smoke carcinogens including the 6 benzene-based aromatic hydrocarbons: the nitrosamine N-Nitrosamine-n-butyl-amine; the aliphatic compounds acrylonitrile, 1,3-Butadine, urethane and vinyl chloride; and the arylamines 4-aminobiphyenyl and ortho-Toluidine.40, 41, 42 Thus it is biologically plausible that exposure to tobacco smoke is related to breast cancer.

Similar active and passive risk?

If a monotonic dose-response relationship is assumed, it is paradoxical that observed breast cancer risk is similar for passive smoking and active smoking, given the apparently large difference in the degree of tobacco smoke exposure between smokers and those exposed only to second-hand smoke. However, studies of passive smoking demonstrate that a high percentage of never smokers often have regular long-term exposure to second-hand smoke. Pirkle et al.43 showed explicitly in the NHANES III study that while 88% of the U.S. nonsmoking population had detectable levels of cotinine in their blood, a dependable biologic indicator of exposure to tobacco smoke, only 40% reported passive smoking exposure. If risks are similar for passive and active smoking, and a large percentage of a nonsmoker control group has passive exposure, then in any study ignoring passive smoking, the active smoking risk could be largely canceled out by the passive risk expressed (but unaccounted for) in the control group.

Confounding, selection bias or information bias are potential explanations that might account for the relatively small association and relatively homogenous effect between both active and passive smoking and breast cancer occurrence, without yielding any biologic discrepancy to explain the similarity. If there were bias in the passive smoking measure, it could create an artificially elevated passive risk that would also increase active risk to about the same degree. The cohort design minimizes the potential for recall, selection and information bias. It is unfortunate that none of the cohort studies except Reynolds et al.32 collected detailed lifetime passive smoking exposure information and Reynolds et al. have only reported on residential exposure. However, 3 of the 4 Asian cohort studies suggest an increased passive smoking risk.18, 24, 35 Two of the case-control studies with better passive smoking measures assessed the potential for such biases through an independent measure and did not find such bias.21, 26 (See more detailed discussion below under confounding.) Assessments of bias done in relation to passive smoking and lung cancer and heart disease have also failed to turn up strong biases as an explanation for those relationships. As well, if the observed effect were only the result of bias, one would expect to observe similar premenopausal and postmenopausal risk, rather than the higher premenopausal risk generally observed. The similarity of the summary pooled risks for cohort studies (all likely missing important sources of passive exposure) and the subset of case-control studies likely missing important exposure, argues against recall bias as the explanation for the elevated risks associated with those studies with more complete exposure assessment.

Morabia et al.44 discuss in detail possible mechanisms by which passive and active smoking might come to have similar impact on breast cancer risk despite the large difference in apparent level of exposure. In particular, they point to research by Vineis et al.45 who found that women exposed to passive smoking had more DNA adducts than active smokers. This suggests that passive smoking may be a different type of exposure (see below), not just a weak equivalent of active smoking.44 Passive risk could also be magnified by a “low dose effect” similar to that proposed for colon cancer,46 where the modifying effect of a genotype might be more apparent at low doses.

Furthermore, women who smoke are at higher risk than non-smokers for conditions related to estrogen deficiency such as osteoporotic fracture, earlier menopause and at lower risk of endometrial cancer, fibrocystic disease and vomiting during pregnancy, which are conditions relating to excess estrogen.47 With total estrogen a surrogate for breast cancer risk, the antiestrogenic effects associated with active smoking might depress the level of breast cancer risk related to tobacco smoke in active smokers but not be strong enough in women passively exposed to depress their tobacco-related risk.

Levels and sources of carcinogenic exposure: passive vs. active smoking

Because the idling cigarette burns at a much lower temperature than when a cigarette is being drawn on, combustion is less complete and the side stream smoke at the tip is a much richer source of at least 20 known carcinogens and dozens of toxic chemicals that are possible or probable carcinogens than the same volume of mainstream smoke inhaled by a smoker through the cigarette and filter. For example, side stream smoke contains 8–10 times as much benzene, 2.5 to 20 times as much Benzo[a]pyrene, 7.2 times as much cadmium, 13–30 times as much nickel, 1.0 to 22 times as much NNK [4-(methylnitrosamino)-(3-pyridyl)-1-butanone] and 1.1–15.7 times as much tar.48

The concentration of cotinine in urine is often used to estimate exposure of non-smokers to tobacco smoke and usually measures at about 1% of the level observed for active smokers.49 However, a recent study actually measured the metabolites of the tobacco-specific carcinogen 4(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK) and reported that the levels in the urine of the women with male partners who smoked were, on average, about 5.6% of the levels in their husband's urine.50 This high level was found even though the women's cotinine levels were only 0.6% of their smoking husband's cotinine levels.50

The 3 American cohort studies: the large potential impact of passive smoking exposure misclassification

Although cohort studies are generally considered a superior method to case-control studies because they avoid recall bias, they too can have serious limitations.15 The 3 large American cohort studies27, 29, 32 may have suffered from substantial exposure misclassification because many women who had had regular second-hand smoke exposure may have been categorized as unexposed to second-hand smoke, thus seriously contaminating the referent non-exposed group. Specifically, in the main analysis by Wartenberg et al. of the American Cancer Prevention Study-II (CPS-II) cohort,27 passive smoking exposure was based exclusively on the husband's smoking history. In the secondary analysis, the exposure measure was based on women's self-reported passive smoking exposure in 1982 and thus ignored historical exposure (including exposure from the 32.6% of husbands who were former smokers, historic workplace exposure and all childhood exposure.) A review of passive smoking exposure studies by Hammond51 found that approximately 30% of U.S. workers not exposed at home were exposed in the workplace. In a passive smoking-lung cancer study with detailed second-hand smoke exposure assessment in 5 U.S. metropolitan areas,16 among the never-smoking female controls, 64% reported passive exposure in childhood, 14% nonspousal adult household exposure, 24% social exposure and 60% exposure from coworkers, and these exposures were often for many years. Substantial nondifferential exposure misclassification can heavily dilute risk estimates.15 It has been demonstrated how exposure misclassification in the Wartenberg cohort analysis could have diluted an underlying RR of 2.0 for breast cancer among passive smoking-exposed women to 1.14.52

Three aspects of the second large American cohort study, the Harvard Nurses Study,29 may have contributed to substantial misclassification of the lifetime passive smoking exposure.53 First, the only measure of occupational passive smoke exposure collected was current exposure in 1982, causing the analysis to be dependent on a single proxy measure to represent lifetime occupational exposure for the likely 15 to 40 years of employment before 1982 and employment until 1996. The very limited study of passive smoke exposure in hospitals54, 55 concurs with what nurses have suggested anecdotally, that many hospital cafeteria, nursing stations and/or medical staff lounges historically had high levels of second-hand smoke. Second, the analysis of childhood exposure ignores adult exposure and the adult exposure analysis ignores childhood exposure. Third, as a cohort defined by being registered nurses in 1976, almost the entire cohort of never-active-smokers may have had regular (unrecorded) exposure to passive smoking for several years during their in-school and/or in-hospital nursing training. Over half the nurses in the cohort reported having been active smokers and most of those would have been smoking during nursing training, as more than 85% of the nurses who had smoked reported smoking by age 22, and 2/3s smoked for at least 20 years.29 Most nurses probably took their training as young adults before having children, a potentially critical period for exposure.29

The other American cohort study, of teachers in California,32 collected detailed lifetime passive smoking exposure information but the published analysis reported only on residential exposure—analysis of occupational and other exposures to ETS have not been published yet. As well, the risk estimate for younger women was based on women pre- or perimenopausal at study initiation, which could have obscured a premenopausal-at-diagnosis risk.

Dose-response relationship with passive smoking exposure

Among the 5 case-control studies with the higher quality exposure measures, 2 reported a dose-response relationship and the other 3, the British,20 Swiss21 and Chinese study by Zhao13 reported overall risk estimates of 2.53 (95% CI 1.19–5.36), 2.3 (95% CI 1.7–3.5) and 2.36 (95%CI 1.66–3.66), respectively. The Canadian study26 observed a dose-response gradient for premenopausal breast cancer with ORs of 1.5 (95% CI 0.5–4.4), 2.0 (0.9–4.5), 2.9 (1.3–6.6) and 3.0 (1.3–6.6) for increasing levels of total passive smoking exposure (p for trend 0.03). The postmenopausal dose-response was more modest with ORs with increasing exposure of 1.1, 1.3 and 1.4. Kropp et al.31 found risks of 1.42 and 1.83 (95% CI 1.16–2.87) for 2 levels of total lifetime exposure: 1–50 hours/day years and >50 hours/day years (p for trend 0.009).

Among the Asian cohort studies 3 of 4 suggested a dose-response relationship. The Hirayama cohort study found an overall risk of 1.32 for breast cancer mortality14 but observed a relative risk of 1.73 (90% CI 1.12–2.66) for Japanese never-smoking women whose husbands smoked more than 20 cigarettes per day.18 The South Korean cohort study found an overall RR of 1.2 for wives of ex-smoking husbands, 1.3 for wives of current smokers and a risk of 1.7 (1.0–2.8) for wives of current smokers who had lived with their husband's smoking for at least 30 years.24 In Hanaoka's cohort study35 in Japan premenopausal breast cancer relative risks were 1.6 (95% CI 0.9–2.7) for any history of residential exposure, 2.3 (95% CI 1.4–3.8) for current occupational and/or public exposure everyday and 2.6 (95% CI 1.3–5.2) for both a residential history and public/occupational exposure everyday.

Among other case-control studies, Scrubsole33 found a dose-response relationship with occupational passive smoking exposure (p for trend 0.03), with the highest occupational exposure associated with an odds ratio of 1.6 (95% CI 1.0–2.5). Although Gammon et al.34 in the U.S. did not observe a dose-response relationship, they did observe a risk estimate of 2.1 (95% CI 1.47–3.02) for women who had never smoked but had lived for more than 27 years with a spouse who smoked.

Other Potential Sources of Bias in the Published Studies

Recall bias can occur in a case-control study if cases recall past exposure differently than healthy controls. The case-control studies of passive smoking and breast cancer have tended to observe higher risks than the cohort studies, so this is a possibility. However, reviews of recall bias in studies of passive smoking and lung cancer56, 57, 58, 59, 60 and passive smoking and heart disease (reviews discussed and summarized by Thun et al.61) have concluded that recall bias is unlikely to have had an important effect on those observed relationships. The Swiss21 and Canadian26 case-control studies that reported increased risk, were 2 of the 5 studies that had better quality exposure measures and each assessed recall bias. The Swiss study asked all women about their perception of passive smoking risk but found little difference in perception between cases and controls.21 The Canadian study was part of a multicancer site study. When lung cancer risk was assessed using the same target control group, observed lung cancer risks associated with passive smoking were consistent with those in the lung cancer-passive smoking literature.26 Both assessments suggested that recall bias was unlikely the explanation for the observed excess risk.21, 26 Furthermore, with the passive smoking-breast cancer relationship, the observed premenopausal risks are larger than for either passive smoking and heart disease or lung cancer, making it less likely that recall bias would be strong enough to be responsible for the observed relationships.

Confounding is unlikely to explain the association. The major known breast cancer risk factors were controlled for in most of the studies. Alcohol might confound the results to a small degree but the impact on risk observed with alcohol is small (7% increase per drink per day),4 the percentage of woman reporting high alcohol consumption is also small, and alcohol was controlled for in many of the studies suggesting increased risk with passive smoking.

Publication bias occurs when studies with positive results are more likely to be published than those with negative results. With 13 of 19 studies suggesting increased risks, and more importantly, all 5 with relatively complete exposure measures suggesting increased risk, there would have to be a number of unpublished studies with good exposure measures that were all negative for publication bias to be a reasonable explanation.

In any questionnaire-based estimate of passive smoking exposure, there will be misclassification, as years of exposure, number of smokers and hours of exposure per day to second-hand smoke are only proxies for actual amount of passive exposure, a measure that is dependent on many other factors as well (proximity to smokers, room size, ventilation etc.). There are no biological markers available to measure long-term or lifetime exposure. The direction of expected bias often depends on the prevalence of the exposure and the specificity of exposure classification. In the current circumstances, nondifferential misclassification of passive smoking will tend to attenuate risk estimates.15

Conclusion

In conclusion, this review of breast cancer risk evaluating studies that include measures of passive smoking found that studies with more comprehensive passive exposure assessment suggest passive and active smoking are risk factors for premenopausal breast cancer. More studies of passive and active smoking and breast cancer, in particular cohort studies, are needed that quantify the major lifetime potential sources of passive smoking; further studies with incomplete ascertainment of major sources of passive smoking may not contribute to better understanding of the risks. More work is needed to try to explain the biological mechanisms that would result in the observation of a similar magnitude of risk associated with both active and passive smoking. In the mean time, it would be prudent to warn women that current epidemiologic research that includes comprehensive exposure measures suggests an association between passive and active smoking and premenopausal breast cancer risk.

Acknowledgements

The author is indebted to A. Judson Wells, Ph.D., for his pioneering and ongoing work on passive smoking's relationship to lung cancer, heart disease and breast cancer, for his input on this paper and many discussions of passive smoking and breast cancer with the author over the last 7 years, and for the inspiration Dr. Wells provides by example, by still having such an energetic, engaged and inquiring mind at age 88. Thanks are also due to K. Cantor, PhD, U.S. National Cancer Institute, for feedback on the manuscript, M. Miller, MD, MPH, Office of Environmental Health Hazard Assessment (OEHHA), California Environmental Protection Agency for his input on the work and R. Broadwin, MPH (also OEHHA) for statistical support.