Gynecologic Oncology 103 (2006) 1122–1129www.elsevier.com/locate/ygyno

Review

Does smoking increase risk of ovarian cancer? A systematic review

Susan J. Jordan a,b,⁎, David C. Whiteman a, David M. Purdie a,Adèle C. Green a, Penelope M. Webb a

a Cancer and Population Studies Group, Queensland Institute of Medical Research, Brisbane, Queensland, Australiab School of Population Health, University of Queensland, Brisbane, Australia

Received 24 April 2006Available online 26 September 2006

Abstract

Objectives. Although early reports suggested that smoking was not associated with ovarian cancer risk, recent studies have reported positiveassociations for cancers of the mucinous subtype. We sought to clarify the relationship between smoking and ovarian cancer by histologicalsubtype.

Methods. We conducted a systematic literature review and meta-analysis of studies investigating the association between smoking and risk ofthe different histological subtypes of epithelial ovarian cancer. Eight population-based case–control studies, one pooled analysis of case–controlstudies, and one cohort study met the inclusion criteria. Summary relative risks (RR), 95% confidence intervals (CI), and tests for heterogeneitywere generated from random effects models.

Results. Combined, these studies included a total of 910 women with mucinous and 5564 with non-mucinous ovarian cancers. There was asignificant doubling of risk of mucinous ovarian cancer in current smokers compared to never smokers (summary RR 2.1, 95%CI 1.7–2.7), butno increased risk of serous (1.0, 95%CI 0.8–1.2) or endometrioid (0.8, 95%CI 0.6–1.1) cancers and a significant risk reduction for clear cellcancers (0.6, 95%CI 0.3–0.9). The risk of mucinous cancer increased with increasing amount smoked but returned to that of never smokerswithin 20–30 years of stopping smoking.

Conclusions. Meta-analysis suggests that current smoking doubles a woman's risk of developing mucinous ovarian cancer. Stopping smokingreturns the risk to normal in the long term. Smoking may thus be one of the few modifiable factors offering potential for primary prevention ofmucinous ovarian cancer.© 2006 Elsevier Inc. All rights reserved.

Keywords: Ovarian cancer; Histological subtype; Smoking; Meta-analysis; Risk factor

Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1123Materials and methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1123

Statistical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1123Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1123

Serous cancers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1125Mucinous cancers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1125Endometrioid and clear cell cancers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1126

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1126Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1128References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1128

⁎ Corresponding author. Queensland Institute of Medical Research, PO Royal Brisbane Hospital, QLD 4029, Australia. Fax: +61 7 3845 3502.E-mail address: susan.jordan@qimr.edu.au (S.J. Jordan).

0090-8258/$ - see front matter © 2006 Elsevier Inc. All rights reserved.doi:10.1016/j.ygyno.2006.08.012

1123S.J. Jordan et al. / Gynecologic Oncology 103 (2006) 1122–1129

Introduction

Epithelial ovarian cancer is a major cause of cancer deathamong women [1] and although important risk factors for thedisease have been identified, few are readily modifiable [2].Tobacco smoking is a modifiable cause of many types of cancer,but whether it influences a woman's risk of ovarian cancer is notclear. Earlier reports mostly suggested that smoking was notassociated with an increased risk of ovarian cancer [3–5]. Thereare, however, several different histological subtypes of ovariancancer and evidence is accumulating to suggest that thesesubtypes, particularly the mucinous tumours, may arise viadifferent causal pathways [2,6,7]. Recent studies investigatingthe relation between smoking and ovarian cancer haveevaluated the different histological subtypes separately, report-ing positive associations for mucinous tumours but not the othersubtypes [8,9]. Results from a recently published pooledanalysis of 10 case–control studies also suggest a specificrelationship between smoking and invasive mucinous ovariancancers [10]. This latter analysis was limited however, beingrestricted to US-based studies conducted mostly in the 1970sand 1980s and evaluating neither the effects of the amount ofsmoking nor time since stopping smoking. In addition, therelationship between smoking and borderline (low malignantpotential) mucinous cancers was not assessed, and the results ofa number of more recent population-based studies conductedoutside the US were not included.

Smoking rates among women, particularly girls and youngwomen, are increasing in most developed and developingregions [11,12] thus clarification of the relationship betweensmoking and mucinous ovarian cancers and the effects ofsmoking cessation is timely. Hence, we have conducted asystematic literature review and meta-analysis of the associationbetween tobacco smoking and risk of epithelial ovarian cancer.Our aims were firstly to quantify the association separately forthe different histological subtypes of ovarian cancer andinvasive versus borderline tumours and, secondly, to evaluatethe effects of amount and duration of smoking and, for pastsmokers, time since quitting smoking.

Materials and methods

We undertook a Medline® (1966–March 2005) search using the searchterms “ovary or ovarian”, “cancer or neoplasm or carcinoma,” and “smokingor tobacco” and the search strategy developed by Haynes and colleagues [13]to identify etiologic studies. We identified additional studies by searching thereference lists of identified papers. Studies were considered eligible if theypresented data pertaining to the relationship between smoking and epithelialovarian cancer separately for the different histological subtypes of ovariancancer and included adjustment for major ovarian cancer risk factorsincluding age and parity and/or oral contraceptive use. The followinginformation was extracted for all eligible studies: year of publication; countryin which the study was conducted; type of study (cohort or case–control); thesampling frame for controls (population-based or hospital-based/other);numbers of cases and controls (or cohort size); and whether the studyincluded invasive and/or borderline tumours. For the present review, weexcluded hospital-based case–control studies a priori, because it is widelyconsidered that smoking habits among hospitalised women do not reflectthose of women in general and could thus lead to biased estimates of risk[14].

From the eligible studies that met the inclusion criteria, we extracted relativerisks and confidence intervals for the association between smoking and ovariancancer, and recorded the factors adjusted for in the analyses. Where available inthe published report, data were extracted relating to smoking status (never, ever,current or past), number of cigarettes smoked per day, duration of smoking inyears, pack-years of smoking (where one pack-year is equivalent to smoking 20cigarettes per day for one year) and time since quitting smoking. For accuracy,the extracted data were checked independently by two authors (SJ and PW).

Statistical analysis

We grouped the results from each study according to smoking status (never,past, current smoker) and number of pack-years smoked as these measures weremost frequently reported. Some overlapping of the categories of number ofpack-years smoked across studies was required to accommodate the differentcut-points set by each study. The lowest category included women who hadfewer than ten pack-years of smoking. Women in the medium category hadbetween five and 24 pack-years of smoking, while those in the high category had20–30 pack-years of smoking. The very high category included women whoreported 30 or more pack-years of smoking.

For our primary analyses, we included all ovarian tumours, both borderlineand invasive, because several of the studies did not distinguish the two tumourtypes [8,9,15]. We then conducted secondary analyses examining borderline andinvasive tumours separately, restricted to those studies which presented suitabledata [16–21].

We estimated summary relative risks (RR) and 95% confidence intervals(CI) for the association between smoking and ovarian cancer from randomeffects models which were also used to test for heterogeneity of effect acrossstudies. When the outcome is rare, as for ovarian cancer, the odds ratio in a case–control study is a good approximation of the rate ratio from a cohort study. Wethus combined the odds ratios from the case–control studies with the rate ratiofrom the single cohort study. The results are also presented graphically usingforest plots where each square and line represents the relative risk andconfidence interval from a single study. The size of the square is inverselyproportional to the variance of the logarithm of the RR, thus studies contributingmore information are represented by larger squares. The diamond at the bottomof each plot represents the summary RR and CI for the studies combined. Allanalyses were conducted using the STATA statistical package [22].

Results

Overall we identified 33 reports from 30 individual studiesthat evaluated the association between smoking and ovariancancer [3–5,8,9,15–21,23–42], 10 of which had considered thedifferent histological subtypes separately [8,9,15–21,41]. Oneof these studies had used hospitalised patients as controls [41]and was excluded from the present analyses leaving eightreports from seven population-based case–control studies[8,9,15–20] and one prospective cohort study [21]. Resultsfor invasive and borderline tumours were published separatelyfor one of the case–control studies [17,20] thus for the purposesof clarity the two reports will be considered as separate studies.In addition, we identified a pooled analysis that combined theraw data from 10 US-based case–control studies [10] includingone of the population-based studies above [8] and another ninestudies that had not previously reported sub-type specificresults. As 80% of the control subjects included in the pooledanalysis were recruited from the general population and only20%were hospital-based, we have, where possible, included thepooled data in our analysis in place of the single report byMarchbanks et al. [8]. To ensure that inclusion of this minorityof hospital controls did not unduly affect our results, we alsorepeated the analyses excluding the pooled data and have

1124 S.J. Jordan et al. / Gynecologic Oncology 103 (2006) 1122–1129

presented those results in the text. The included studies werethus undertaken in the USA (3940 cases), Canada (896 cases),Australia (794 cases) and Sweden (844 cases).

Collectively, the eligible case–control studies and the pooledanalysis included data from 6020 women with incidentepithelial ovarian cancer (including 878 with mucinoustumours) and 16,863 control women (Table 1). Five includedwomen with either borderline or invasive tumours[8,9,15,18,19], two additional studies and the pooled analysisincluded only women with invasive cancer [16,20], while onewas restricted to women with borderline tumours [17]. Thecohort study included 454 women diagnosed with invasiveepithelial ovarian cancer (including 32 mucinous cases) in acohort of 89 835 women [21].

Seven percent of the case women in the cohort study hadmucinous tumours while the corresponding percentages in thecase–control studies were somewhat higher (due to theinclusion of borderline tumours), ranging between 9% and

Table 1Characteristics of studies that have published results for the relationship between sm

Study, country Design Cases/Controls

Years ofdiagnosis

Invas

Kuper 2000, USA [15] Case–control 549/516 1992–1997 Invasborde

Marchbanks 2000, USA [8] Case–control 447/3868 1980–1982 Invasborde

Green 2001, Australia [18] Case–control 794/855 1990–1993 Invasborde

Riman 2001, Sweden [17] Case–control 193/3899 1993–1995 Borde

Modugno 2002, USA [9] Case–control 767/1367 1994–1998 Invasborde

Goodman 2003, USA(Hawaii and Los Angeles)[19]

Case–control 558/607 1993–1999 Invasborde

Terry 2003, Canada [21] Cohort 454/89,835 1980–2000 Invas

Pan 2004, Canada [16] Case–control 442/2135 1994–1997 Invas

Riman 2004, Sweden [20] Case–control 651/3873 1993–1995 Invas

Kurian 2005, USA [10] b Pooled analysisof case–controlstudies

2066/7484 1973–2001 Invas

Abbreviations: E, endometrioid; M, mucinous; NM, non-mucinous; O, other; S, sera Adjusting factors: all studies adjusted for age with additional adjustment for (

education; as noted. When adjustment was not made for listed confounders this wapoint estimates.b Includes the data for invasive cancer from Marchbanks et al., 2000. Analyses i

20% (excluding the study of borderline tumours in which42% were mucinous). Smokers were defined variously by thedifferent studies as those who had smoked daily for a periodof six or more months [9,19], those who had smoked dailyfor 12 or more months [18], or those who had smoked morethan 100 cigarettes in their lives [8,15,16]. Current smokerswere generally defined as women who had been smokingwithin 6 to 12 months of diagnosis (cases) or interview(controls). Two reports divided current smokers into thosewho smoked 10 or fewer cigarettes per day and those whosmoked more than 10 per day [17,20], so we included theresults for those who smoked more than 10 per day in ouranalyses of current smokers. In the one cohort study, thesmoking data relate to status at the time of recruitment intothe cohort [21]. Among the control women, the proportion ofever-smokers ranged from 38% to 55%, while 13% to 40% werecurrent smokers. Five of the case–control studies reported datafor amount of smoking measured variously as pack-years (4

oking and the different histological subtypes of epithelial ovarian cancer

iveness Histologicalsubtypes

Proportionof smokersamong controls(person-years)

Adjusting factors a

Never Current 1 2 3 4 5

ive andrline

M 14% 48% Not given Yes Yes Yes Yes YesS 56%O 30%

ive andrline

M 18% 44% 36% Yes Yes No No NoS 52%E 19%O 11%

ive andrline

M 14% 62% 16% No Yes No Yes YesNM 86%

rline only M 42% 51% 28% Yes Yes No No NoS 57%O 1%

ive andrline

M 15% 46% 23% Yes Yes Yes Yes NoNM 85%

ive andrline

M 20% 60% 13% Yes Yes No Yes YesNM 80%

ive M 7% 51% 22% Yes Yes No No YesS 40%E 15%O 38%

ive M 16% 51% 20% Yes No No No YesNM 84%

ive M 9% 57% 22% Yes Yes No No NoS 51%E 27%O 12%

ive M12% 46% 32% Yes Yes No No NoS 52%E 18%O 18%

ous.1) parity; (2) oral contraceptive use; (3) family history; (4) tubal ligation; (5)s generally because inclusion of these factors did not substantially change the

nclude either the pooled analysis (where possible) or Marchbanks et al.

1125S.J. Jordan et al. / Gynecologic Oncology 103 (2006) 1122–1129

studies), cigarettes per day (3 studies), and/or duration ofsmoking (2 studies). The variable pack-years was thus used inthis review to evaluate the association with amount of smoking.

Serous cancers

Three studies and the pooled analysis presented results forserous cancers separately. For a further four studies, weincluded the published risk estimates for non-mucinous tumoursin this analysis because serous cancers account for a highproportion of such cancers. The results for current and pastsmokers compared to never smokers are shown in Figs. 1(a) and(b), respectively. Five of the studies, including the pooledanalysis, found no association between smoking and risk ofserous ovarian cancer while one found a significant 60%increased risk of serous cancer for current smokers [18] andanother reported a marginally significant 28% increase in riskassociated with past but not current smoking [16]. One studyreported significant reductions in risk of serous invasive cancersamong both former and current smokers [20].

There was no significant heterogeneity between the reportedrelative risks for serous tumours in past smokers (heterogeneityχ2 =10.7 on 7 degrees of freedom [43], p=0.2), but there wassome heterogeneity among the results for current smokers(χ2 =14.7, df=7, p=0.04). Overall, compared to never smokers,

Fig. 1. Individual and summary relative risks and 95%CI of non-mucinous epitheliacancer for (c) current and (d) past smokers compared to never smokers.

the summary relative risk of serous tumours was 1.0 (95%CI0.8–1.2) for current smokers and 1.0 (95%CI 0.9–1.2) for pastsmokers. We repeated the meta-analysis excluding the data fromthe pooled analysis [10] and obtained identical results with asummary OR of 1.0 (95%CI 0.8–1.2) for current smokers and1.0 (95%CI 0.9–1.2) for past smokers.

Of the four case–control studies that evaluated the relationwith pack-years, three reported no association [8,9,16] whileone found a significant 60% increased risk of serous tumoursamong women with more than 30 pack-years of smoking (datanot shown) [18]. One of the remaining case–control studiesreported a significant increase in risk of invasive serous cancerswith years of tobacco use (p for trend=0.05), but not for numberof cigarettes smoked per day [15], while the single cohort studyalso reported an increased risk of serous cancers but only amongwomen who had smoked for more than 40 years [21]. Incontrast, one study reported a significant reduction in risk ofinvasive serous ovarian cancer with increasing number ofcigarettes smoked per day [20].

Mucinous cancers

The results of the studies that presented relative risks ofmucinous cancers for current and past smokers compared tonever smokers are shown in Figs. 1(c) and (d), respectively. One

l ovarian cancer for (a) current and (b) past smokers, and of mucinous ovarian

Fig. 2. Relative risks and 95%CI of mucinous cancer in relation to pack-years ofsmoking compared to never smokers.

Fig. 3. Relative risks and 95%CI of mucinous cancer in relation to time sincestopping smoking in past smokers compared to never-smokers.

1126 S.J. Jordan et al. / Gynecologic Oncology 103 (2006) 1122–1129

of the studies [19] found no significant association betweensmoking and risk of mucinous ovarian cancer and a secondstudy found no significant association for invasive mucinouscancers [20] although in both cases the point estimate wasgreater than one. The other studies all reported statisticallysignificant two to three-fold increases in risk of mucinousovarian cancer with current smoking. There was no significantheterogeneity between the results of the different studies(current smokers χ2 =6.5, df=7, p=0.5; past smokers: hetero-geneity χ2 =7.5 on df=7, p=0.4). Women who were currentsmokers had a significant 2.1-fold increased risk of mucinouscancers compared to never smokers (95%CI 1.7–2.7), while thesummary relative risk for past smoking versus never smokingwas 1.1 (95%CI 0.9–1.4). When the data from the previouspooled analysis were excluded and the analysis repeated, almostidentical results were seen with a summary OR of 2.1 (95%CI1.6–2.8) for current smokers and 1.2 (95%CI 0.9–1.5) for pastsmokers. Although they did not present relative risks, Kuper etal. also reported a ‘doubling in risk’ of mucinous cancer forwomen who smoked more than an average of 40 cigarettes perday [15].

There was also a suggestion that the relative risk associatedwith smoking was somewhat higher for borderline than invasivemucinous tumours. The summary of relative risk for borderlinemucinous tumours derived from the three studies that evaluatedthis was 2.4 (95%CI 1.4–4.1) [17–19] while a fourth studyreported a 2.5-fold increase in risk for ever-smoking forborderline tumours [8]. In comparison, the estimate for invasivemucinous cancer derived from the same three studies was 1.4(95%CI 0.8–2.4) while combining the data from all six studiesthat had presented relevant data [10,16,18–21] gave an OR of2.0 (95%CI 1.5–2.7).

All four studies that presented results by pack-years smokedreported a significant increase in risk of mucinous ovariancancer with increasing pack-years (Fig. 2) [8,9,16,18]. Com-pared to women who had never smoked, the summary relativerisk for women in the group reporting the lowest number ofpack-years smoked was 1.5 (95%CI 1.0–2.4); in the medium

group it was 2.4 (95%CI 1.8–3.2); in the high group it was 2.9(95%CI 2.0–4.1); and in the very high group it was 2.5 (95%CI1.5–4.2). The same four studies also examined risk of mucinousovarian cancer with time since quitting smoking. All four foundthe risk decreased with increasing time since stopping smokingand approached that in never smokers by the longest categorystudied (Fig. 3).

Endometrioid and clear cell cancers

Only four studies presented separate results for endometrioidcancers (a total of 692 cases) and three for clear cell cancers(total 231cases). Two of the studies that evaluated endometrioidcancers and the pooled analysis reported a non-significantinverse association with current smoking [10,19,21] while onestudy reported a 10% increase in risk [20]. There was nosignificant heterogeneity between the study results (χ2 =2.4,df=3, p=0.5) and the summary relative risk versus neversmoking was 0.8 (95%CI 0.6–1.1). The two studies thatreported results for clear cell cancers and the pooled analysis allreported non-significant inverse associations with currentsmoking [10,19,20]. There was no significant heterogeneitybetween the study results (χ2 =0.3, df=2, p=0.9) and combin-ing the data gave a statistically significant inverse associationwith risk of clear cell ovarian cancer with a summary relativerisk of 0.6 (95%CI 0.3–0.9) for current smokers compared tonever smokers.

Discussion

Nine population-based studies and one pooled analysishave considered the association between cigarette smokingand the risk of the different histological subtypes of ovariancancer and all except one study reported sub-type specificrelative risks. Six individual studies and the pooled analysisreported a significant two to three-fold increase in risk of

1127S.J. Jordan et al. / Gynecologic Oncology 103 (2006) 1122–1129

mucinous cancer among current smokers with two othersreporting a non-significant 20% increase in risk [19,20] that,in one study, was restricted to women smoking more than 10cigarettes per day [20]. The final study reported a ‘doublingin risk' of mucinous cancer among heavy smokers [15].Combining the results of the studies that presented subtype-specific relative risks suggests that current smokers havearound a two-fold increased risk of developing mucinousovarian cancer. This risk increases with increasing amountsof smoking but returns to baseline risk within 20–30 years ofstopping smoking. Only one study reported a significantincrease in risk of non-mucinous cancer among smokers [18]with two others reporting increases that were restricted toborderline serous tumours [19] or women who had smokedfor more than 40 years [21]. One study found a significantreduction in risk of invasive serous cancer among smokers[20]. Although few studies reported results separately forendometrioid and clear cell cancers [10,19,21], currentsmokers may be at somewhat reduced risk of developingthese types of tumours. Overall, therefore, smoking does notappear to increase risk of the non-mucinous types ofepithelial ovarian cancer although an increase in risk ofserous cancer associated with very long-term smoking cannotbe ruled out.

While systematic reviews and meta-analyses can provide abetter assessment of the relationship between specific factorsand outcomes by increasing the power to investigate raredisease and allowing examination of the relationship acrossvarying study settings, they cannot eliminate potential sourcesof error inherent in the included studies. All except one of thestudies included in the current review used case–control datathus the potential for both selection and recall bias must beconsidered. It is possible that controls who agreed to take part inthe studies may have had a lower prevalence of smoking thanthe general population. Controls may also have under-estimatedtheir level of smoking. Such bias could lead to falsely elevatedodds ratios in these studies, but would be expected to affect theresults for all tumour types equally and so cannot explain thestrikingly different patterns seen for mucinous and non-mucinous cancers, or the apparent inverse association withclear cell and possibly endometrioid tumours. Of note, the onehospital-based case–control study that considered this relationby histological subtype also reported increased risks formucinous but not serous or endometrioid cancers amongsmokers [41]. Because our meta-analysis was based onpublished data, we were unable to independently investigatethe effects of confounding or interaction. All of the includedstudies had, however, either adjusted for the major confoundersor noted that adjustment did not affect the results thus it isunlikely that the results are due to confounding. Theconsistency between the findings for mucinous cancer fromthe pooled analysis, based on US women mostly diagnosed inthe 1970s and 80s (OR 2.4, 95%CI 1.5–3.8) [10], and theresults of the current meta-analysis excluding the pooledanalysis (OR 2.1, 95%1.6–2.8), based on women from theUS and elsewhere who were diagnosed mostly in the 1990s,further supports the validity of the results.

In addition, use of the published data has required that wemake decisions such as that to over-lap categories of pack-yearsof smoking in order to accommodate the different cut-points setby each study. Decisions of this type have to be made in allmeta-analyses of observational studies when different investi-gators have used different cut-points. In practice, anymisclassification as a result of the overlapped categories isonly likely to have attenuated the true association.

Could the results be due to publication bias with authorsonly submitting results for publication or editors onlyaccepting papers if they found some association withsmoking? The pooled analysis [10] included previouslyunpublished subtype-specific results from nine studies ofovarian cancer. We repeated our meta-analysis including andexcluding these data and our findings did not substantiallychange suggesting that publication bias does not explain theresults. Although there are almost certainly further unpub-lished results, it is unlikely that these would have such stronginverse effects for smoking and mucinous cancer that theywould negate the strong positive association observed amongthe published data.

The diagnosis of mucinous ovarian cancer is, however,complicated by the fact that mucinous tumours from the colon,appendix and pancreas, and occasionally the stomach andcervix, may present clinically as primary ovarian cancer [44].This problem has only been appreciated relatively recently thusit is likely that, in older studies in particular, a proportion oftumours classified as invasive mucinous ovarian primarieswere really gastrointestinal or possibly cervical cancers. Couldthe observed association between smoking and mucinousovarian cancer simply represent an association betweensmoking and metastatic intestinal cancer? In two of the studiesincluded in this review [8,18], a group of three pathologistsreviewed all of the cases centrally to classify the histologicalsubtype of each cancer. It is likely that the histologicalclassification is most reliable in these studies and both reportedstrong associations between smoking and mucinous ovariancancer, although the possibility of misclassification remains.In addition, the effects appear to be somewhat stronger forborderline tumours, which are much less likely to be mis-classified [44].

There was no evidence of statistical heterogeneity betweenthe results for mucinous cancer in the different studies. Onlytwo studies reported no association between smoking and riskof mucinous ovarian cancer; however, the confidence limitsaround their estimates (95%CI 0.66–2.26 [19] and 0.50–2.79among women who smoked more than 10 cigarettes per day[20]) are compatible with a two-fold increase in risk ofmucinous ovarian cancer among current smokers. The maindifference between the first of these studies and the others is thehigh proportion (41%) of Asian women, although the authorsreported that their results did not differ appreciably betweenAsian and White women. That study also had the lowestproportion of current smokers at 13% and the highest proportionof mucinous tumours (20%). The second study included fewwomen with invasive mucinous cancer thus the confidencelimits around the estimates were very wide [20].

1128 S.J. Jordan et al. / Gynecologic Oncology 103 (2006) 1122–1129

Among premenopausal women, smoking is associated withlower levels of circulating oestrogen, delayed conception, andearly menopause while postmenopausal smokers have elevatedlevels of androgens [45,46]. Cigarette smoke also containsnumerous carcinogenic chemicals, including benzo [a]pyrenewhich has been shown to initiate ovarian tumours in mice [47]and smoking-related DNA adducts have been detected ingranulosa-lutein cells from women exposed to cigarette smoke[48]. An association between smoking and risk of ovariancancer is therefore biologically plausible [8,18].

Histologically, mucinous ovarian tumours resemble cervicaladenocarcinomas or intestinal epithelium. Although there is norelation between smoking and adenocarcinoma of the cervix[49], cigarette smoking has consistently been associated withmucinous cancers of the gastrointestinal tract. Gastric andpancreatic cancer are both classed as smoking-related diseasesand, although the relation with colon cancer is less clear,smoking is strongly associated with development of adenoma-tous colorectal polyps, a pre-cursor of colon cancer [50]. It isthus plausible that a positive association between smoking andovarian cancer could be specific for the mucinous subtype. Incontrast, endometrioid and clear cell ovarian cancers arehistologically similar to their endometrial counterparts andsmoking has been associated with a reduced risk of endometrialcancer, presumably because of its anti-oestrogenic effects [50].Non-mucinous ovarian cancers are more strongly related tohormonal and reproductive factors than mucinous cancers [6].Thus the anti-oestrogenic effects of smoking might counteractthe carcinogenic effects for these subtypes, as for endometrialcancer.

The finding is also in keeping with results of other studiessuggesting that mucinous tumours are etiologically distinctfrom the other subtypes. In particular, women with mucinoustumours are less likely to have a family history of breast orovarian cancer [7,10,51] and are very unlikely to have BRCAmutations [52]. In addition, some have found that the effects ofreproductive risk factors are less pronounced for mucinouscancers [6,7,53] although other studies have not supported this[10,54].

In conclusion, our systematic review incorporating data from910 women with mucinous ovarian cancer suggests thatsmoking approximately doubles a woman's risk of developingmucinous ovarian cancer and that this risk increases withincreasing amounts of smoking. It also suggests that this effectis reversible and that within 20–30 years of stopping smoking, awoman's risk of mucinous ovarian cancer will return to that of anever smoker. Tobacco is thus one of the few known modifiablefactors that offer some potential for primary prevention ofovarian cancer.

Acknowledgments

We would like to thank Nirmala Pandeya for conducting thestatistical analysis and generating the forest plots. The authorswere supported by an Australian Postgraduate Award from theUniversity of Queensland (SJ) and by Senior Research Fellow-ships from the National Health and Medical Research Council

of Australia (DP, DW) and Queensland Cancer Fund (PW). Thiswork was partly funded by the National Breast Cancer Centre,Australia.

The authors have no conflicts of interest to declare. SusanJordan (corresponding author) had full access to all of the dataand had the final responsibility for submitting the manuscriptfor publication.

The studywas conceived by PWebb,DWhiteman andDPurdieand designed by S Jordan, P Webb, D Whiteman, and A Green. SJordan conducted the literature searches and S Jordan and P Webbreviewed the studies and extracted the data. All authors contributedto drafting the manuscript.

References

[1] Ferlay J, et al. GLOBOCAN 2002: cancer incidence, mortality andprevalence worldwide. IARC Cancer Base, vol. 5. Lyon: IARCPress;2004. version 2.0.

[2] Bertone-Johnson ER. Epidemiology of ovarian cancer: a status report.Lancet 2005;365(9454):101–2.

[3] Whittemore A, et al. Personal and environmental characteristics related toepithelial ovarian cancer: II. Exposures to talcum powder, tobacco,alcohol, and coffee. Am J Epidemiol 1988;128(6):1228–40.

[4] Hartge P, et al. A case–control study of epithelial ovarian cancer. Am JObstet Gynecol 1989;161(1):10–6.

[5] Franks A, et al. Cigarette smoking and the risk of epithelial ovarian cancer.Am J Epidemiol 1987;126(1):112–7.

[6] Purdie D, et al. Reproduction-related risk factors for mucinous andnonmucinous epithelial ovarian cancer. Am J Epidemiol 2001;153(9):860–4.

[7] Risch H, et al. Differences in risk factors for epithelial ovarian cancer byhistologic type. Results of a case–control study. Am J Epidemiol 1996;144(4):363–72.

[8] Marchbanks P, et al. Cigarette smoking and epithelial ovarian cancer byhistologic type. Obstet Gynecol 2000;95:255–60.

[9] Modugno F, Ness RB, Cottreau C. Cigarette smoking and the risk ofmucinous and nonmucinous epithelial ovarian cancer. Epidemiology 2002;13(4):467–71.

[10] Kurian AW, et al. Histologic types of epithelial ovarian cancer: have theydifferent risk factors? Gynecol Oncol 2005;96(2):520–30.

[11] Mackay J, Eriksen M. The Tobacco Atlas. Geneva: World HealthOrganisation; 2002.

[12] US Department of Health and Human Services. Women and Smoking. Areport of the Surgeon General. In: U.D.o.H.a.H. Services, editor.Rockville, MD: Office of the Surgeon General; 2001.

[13] Haynes RB, et al. Developing optimal search strategies for detectingclinically sound studies in MEDLINE. J Am Med Inform Assoc 1994;1(6):447–58.

[14] Rothman KJ. Epidemiology—An Introduction. Oxford Univ. Press; 2002.[15] Kuper H, et al. Population based study of coffee, alcohol and tobacco use

and risk of ovarian cancer. Int J Cancer 2000;88(2):313–8.[16] Pan SY, et al. Association of cigarette smoking with the risk of ovarian

cancer. Int J Cancer 2004;111(1):124–30.[17] Riman T, et al. Risk factors for epithelial borderline ovarian tumors:

results of a Swedish case–control study. Gynecol Oncol 2001;83(3):575–85.

[18] Green A, et al. Cigarette smoking and risk of epithelial ovarian cancer.Cancer Causes Control 2001;12:713–9.

[19] Goodman MT, Tung KH. Active and passive tobacco smoking and the riskof borderline and invasive ovarian cancer (United States). Cancer CausesControl 2003;14(6):569–77.

[20] Riman T, et al. Some life-style factors and the risk of invasive epithelialovarian cancer in Swedish women. Eur J Epidemiol 2004;19(11): 1011–9.

[21] Terry P, et al. Cigarette smoking and the risk of invasive epithelial ovariancancer in a prospective cohort study. Eur J Cancer 2003;39:1157–64.

1129S.J. Jordan et al. / Gynecologic Oncology 103 (2006) 1122–1129

[22] Stata. Corp. Statistical Software: Release 7.0, in Corp Stata. CollegeStation, TX: Stata Corporation; 2001.

[23] West RO. Epidemiologic study of malignancies of the ovaries. Cancer1966;19(7):1001–7.

[24] Trichopoulos D, Papapostolou M, Polychronopoulou A. Coffee andovarian cancer. Int J Cancer 1981;28(6):691–3.

[25] La Vecchia C, et al. Coffee drinking and the risk of epithelial ovariancancer. Int J Cancer 1984;33(5):559–62.

[26] Booth M, Beral V, Smith P. Risk factors for ovarian cancer: a case–controlstudy. Br J Cancer 1989;60(4):592–8.

[27] Newhouse ML, et al. A case control study of carcinoma of the ovary. Br JPrev Soc Med 1977;31(3):148–53.

[28] Cramer DW, et al. Dietary animal fat in relation to ovarian cancer risk.Obstet Gynecol 1984;63:833–8.

[29] Slattery ML, et al. Nutrient intake and ovarian cancer. Am J Epidemiol1989;130(3):497–502.

[30] Williams RR, Horm JW. Association of cancer sites with tobacco and alcoholconsumption and socioeconomic status of patients: interview study from theThird National Cancer Survey. J Natl Cancer Inst 1977;58(3): 525–47.

[31] Tzonou A, et al. The epidemiology of ovarian cancer in Greece: a case–control study. Eur J Cancer Clin Oncol 1984;20(8):1045–52.

[32] Stockwell H, Lyman G. Cigarette smoking and the risk of femalereproductive cancer. Am J Obstet Gynecol 1987;157:35–40.

[33] Yen ML, et al. Risk factors for ovarian cancer in Taiwan: a case–controlstudy in a low-incidence population. Gynecol Oncol 2003;89(2): 318–24.

[34] Tverdal A, et al. Mortality in relation to smoking history: 13 years' follow-up of 68,000 Norwegian men and women 35–49 years. J Clin Epidemiol1993;46(5):475–87.

[35] Mori M, et al. Anthropometric and other risk factors for ovarian cancer in acase–control study. Jpn J Cancer Res 1998;89(3):246–53.

[36] Engeland A, et al. Smoking habits and risk of cancer other than lungcancer: 28 years' follow-up of 26 000 Norwegian men and women. CancerCause Control 1996;7:497–506.

[37] Doll R, et al. Mortality in relation to smoking: 22 years' observations onfemale British doctors. Br J Med 1980 (5th April):967–9.

[38] Baron J, et al. Cigarette smoking in women with cancers of the breast andreproductive organs. JNCI 1986;77:677–80.

[39] PolychronopoulouA, et al. Reproductive variables, tobacco, ethanol, coffee andsomatometry as risk factors for ovarian cancer. Int J Cancer 1993;55(3):402–7.

[40] Purdie D, et al. Reproductive and other factors and risk of epithelialovarian cancer: an Australian case–control study. Survey of Women'sHealth Study Group. Int J Cancer 1995;62(6):678–84.

[41] Zhang Y, et al. Cigarette smoking and increased risk of mucinous epithelialovarian cancer. Am J Epidemiol 2004;159(2):133–9.

[42] Niwa Y, et al. Cigarette smoking and the risk of ovarian cancer in theJapanese population: findings from the Japanese Collaborate Cohort study.J Obstet Gynaecol Res 2005;31(2):144–51.

[43] Struewing JP, et al. Prophylactic oophorectomy in inherited breast/ovariancancer families. Monogr Natl Cancer Inst 1995;17:33–5.

[44] Tavassoli FA, Devilee P. World Health Organization Classification ofTumours. Pathology and Genetics of Tumours of the Breast and FemaleGenital Organs. Lyon: IARC Press; 2003.

[45] The Practice Committee of the American Society for ReproductiveMedicine Smoking and infertility. Fertil Steril 2004;81(4):1181–6.

[46] Tanko LB, Christiansen C. An update on the antiestrogenic effect ofsmoking: a literature review with implications for researchers andpractitioners. Menopause 2004;11(1):104–9.

[47] Mattison DR, Thorgeirsson SS. Smoking and industrial pollution, and theireffects on menopause and ovarian cancer. Lancet 1978;1(8057): 187–8.

[48] Zenzes MT, Puy LA, Bielecki R. Immunodetection of benzo[a]pyreneadducts in ovarian cells of women exposed to cigarette smoke. Mol HumReprod 1998;4(2):159–65.

[49] Berrington de Gonzalez A, Sweetland S, Green J. Comparison of riskfactors for squamous cell and adenocarcinomas of the cervix: a meta-analysis. Br J Cancer 2004;90(9):1787–91.

[50] U.S. Department of Health and Human Services. The health consequencesof smoking: a report of the Surgeon General. 2004, Centers for DiseaseControl and Prevention, National Center for Chronic Disease Control andPrevention, Office on Smoking and Health: Atlanta, GA.

[51] Purdie D, et al. The different etiologies of mucinous and non-mucinousepithelial ovarian cancers. Gynecol Oncol 2003;88(1 Pt 2):s145–8.

[52] Risch HA, et al. Prevalence and penetrance of germline BRCA1 andBRCA2 mutations in a population series of 649 women with ovariancancer. Am J Hum Genet 2001;68(3):700–10.

[53] Kvale GH, Nilssen IS, Beral V. Reproductive factors and risk of ovariancancer: a prospective study. Int J Cancer 1988;42:246–51.

[54] Wittenberg J, et al. Reproductive risk factors for mucinous and non-mucinous epithelial ovarian cancer. Epidemiol 1999;10(6):761–3.