Association between Cigarette Smoking and Histotype-Specific Epithelial Ovarian Cancer: A Review of Epidemiologic Studies

In the United States, ovarian cancer is the fifth leading cause of cancer-related deaths among females (1). As the deadliest gynecologic cancer, ovarian cancer has an overall 5-year survival rate of 47% (2). At the time of diagnosis, approximately 60% of women present with distant stage of the disease that is characterized by 5-year survival rate of only 29% (2).

The majority of ovarian tumors tend to originate from the ovarian epithelium (3). Epithelial ovarian tumors include four main histotypes: serous (70%), endometrioid (10%), clear cell (13%), and mucinous (3%; ref. 4). These subtypes could be further subdivided into benign, borderline, and malignant (3, 5). Benign tumors are characterized by a lack of intense proliferation and invasiveness while borderline tend to exhibit atypical proliferation with the absence of invasiveness, and malignant being characterized by invasiveness (3). Because epithelial ovarian cancer accounts for nearly 90% of all malignant ovarian tumors (3), epidemiologic research focuses mainly on this subtype of ovarian cancer.

According to a recent theory, histologic subtypes of ovarian cancer differ by the origin of disease and may originate outside the ovary (6). It has been proposed that high-grade serous cancer can originate in the fallopian tube epithelium; and endometrioid and clear cell cancers may develop from endometrial epithelium implanted on ovarian surface through a retrograde menstrual flow (7, 8). At the same time, the origin of mucinous carcinoma is not clear and some argue that this subtype may arise at the tuboperitoneal junction (8). Such etiologic heterogeneity could explain differences in the strength of associations with various risk and survival factors across different histologic subtypes. For instance, BRCA mutation is associated with an increased risk of high-grade serous ovarian cancer, while endometriosis is associated with risk of ovarian cancer of endometrioid or clear cell histotypes (9, 10).

Cigarette smoking represents an exposure that has been demonstrated to be linked to ovarian cancer with associations varying across the histotypes. In fact, while epidemiologic studies reported either inverse, positive, or no association between smoking and overall ovarian cancer risk, the same and some additional studies have demonstrated that smoking is associated with increased risk of mucinous ovarian cancer (11–37). At the same time, epidemiologic evidence on the association between smoking and risk of ovarian cancer across other histotypes has been inconclusive. Although some studies suggested an inverse association between smoking and risk of endometrioid, serous, and clear cell ovarian cancer, the results were, in general, not statistically significant (11–36, 38). Meanwhile, the role of smoking on ovarian cancer survival and, specifically, in relation to histotype-specific associations has been unclear because very few studies have been conducted on the topic (39–41).

Currently, there is a need for a better understanding of the relationship between smoking and histotype-specific ovarian cancer in the context of a high mortality of ovarian cancer and smoking being one of the leading causes of death in the United States (42, 43) and its ability to impact the survival of patients with cancer (44, 45). A thorough understanding of the link between smoking and ovarian cancer could allow for the development of more targeted and, therefore, more effective preventive and therapeutic measures.

The results of the most recent pooled analysis of epidemiologic data on smoking and histotype-specific ovarian cancer risk were published in 2016 (35). However, there still is a need to summarize evidence on this topic accumulated by today and to compare and contrast the results of various studies. Therefore, to understand the impact of smoking on ovarian cancer risk and survival after diagnosis and to emphasize epidemiologic research that recognized heterogeneous nature of ovarian cancer by examining histotype-specific associations, we conducted a systematic review that summarized the results of the studies published in the past 20 years.

We conducted a literature search through PubMed on smoking and ovarian cancer, using a combination of key words such as “ovary or ovarian,” “cancer or neoplasm or carcinoma,” and “tobacco or smoke or smoking.” Studies included were published from 1997 to 2018 and written in English language. Eligible studies included in this review met all of the following criteria: (i) studies with human subjects; (ii) observational studies; (iii) studies investigating the association between cigarette smoking and epithelial ovarian cancer risk and survival; and (iv) studies presenting data on ORs, RRs, or HRs. Furthermore, we decided to focus on the studies published from 1997 since 1996 was the year when it was first hypothesized that histologic subtypes could be etiologically different, specifically, that mucinous ovarian cancer was distinct from the other histotypes (46). Therefore, in this review, we included only those studies that, in addition to reporting overall estimates, presented histotype-specific associations or only focused on histotype-specific associations.

We identified 778 records through our initial search of the PubMed database, and among them 30 studies met the criteria mentioned above (Fig. 1). One additional study was identified through a reference list and added to our review. Overall, of 31 studies that were eligible to be included in the review, 28 were risk and three were survival studies.

Characteristics of 16 case–control, seven cohort, five pooled/meta-analyses, and three survival studies included in this review are listed in Tables 1, 2, 3, and 4, respectively. The tables included information on country where the studies were conducted, sample size, and the main findings. We also presented results of the studies in Figs. 2–5 according to the following sequence: Fig. 2, if these studies reported an overall association for ovarian cancer risk associated with cigarette smoking; Figs. 3, 4, 5, if these studies presented estimates for serous, mucinous, and/or endometrioid histotypes, respectively. We presented results for invasive tumors when studies distinguished between borderline and invasive tumor types.

To increase comparability among the study results and to avoid a problem of a limited power, we presented the results for the smoking status variable that had only three categories: never smokers/nonsmokers; current smokers, and former smokers. Two studies, one case–control by Kuper and colleagues (37) and one pooled analysis by Wentzensen and colleagues (35) did not parameterize smoking by employing this variable. Therefore, the results of these analyses were discussed in a separate section and included in the tables, but were not presented in the figures.

As shown in Fig. 2, 10 case–control studies reported on the association between current smoking and epithelial ovarian cancer (11–20). One study by Green and colleagues observed a significantly increased risk of overall ovarian cancer (OR = 1.8; 95% CI, 1.3–2.5) when borderline and invasive cases were combined in the analyses (19). When these subtypes were examined separately, current smoking was associated with elevated risk of ovarian cancer in both cases (OR = 1.7; 95% CI, 1.2–2.4 and OR = 2.4; 95% CI, 1.4–4.1 for invasive and borderline tumors, respectively). In the study by Riman and colleagues, smoking more than 10 cigarettes per day among current smoking was associated with increased risk of ovarian cancer (OR = 1.66; 95% CI, 1.04–2.63), while smoking less than that was not associated with ovarian cancer (OR = 1.20; 95% CI, 0.74–1.95; ref. 20). Two case–control studies by Baker and colleagues (14) and Riman and colleagues (17) found a decreased risk of ovarian cancer associated with current smoking. The other six case–control studies found no significant associations for current smoking (11–13, 15, 16, 18). Four case–control studies had no data on overall ovarian cancer risk (29–31, 36), therefore, these results were not shown in Fig. 2.

Seven cohort studies presented results on overall ovarian cancer risk (21–26, 32). Among them only one reported an increased risk associated with current smoking (HR = 1.4; 95% CI, 1.0–1.8), when borderline and invasive cases were combined (25). The association was more pronounced for borderline (HR = 2.7; 95% CI, 1.2–5.7) and more attenuated for invasive tumors (HR = 1.1; 95% CI, 0.8–1.7) when these two types were analyzed separately.

One pooled study that used data from the Ovarian Cancer Association Consortium (OCAC) by Faber and colleagues (27) and one meta-analysis by Beral and colleagues (28) presented results on the overall ovarian cancer risk associated with current smoking status. While Faber and colleagues did not observe significant association between current smoking and overall invasive ovarian cancer risk, they found an increased risk of borderline ovarian tumors among current smokers (OR = 1.36; 95% CI, 1.13–1.64; ref. 27). Beral and colleagues found increased risk of overall ovarian cancer among current smokers (RR = 1.07; 95% CI, 1.00–1.14), with borderline and invasive cases combined together (28). Three other studies either did not present the estimates for the overall ovarian cancer (33, 34) or did not use the current/former/never parameterization of smoking variable (35).

As shown in Fig. 2, two case–control studies by Green and colleagues (19) and Pan and colleagues(16) found approximately 30% increase in overall ovarian cancer risk among former smokers. Rossing and colleagues reported increased risk of ovarian cancer among those who discontinued smoking 6–15 years prior to reference date (OR = 1.3; 95% CI, 1.0–1.8 and OR = 1.8; 95% CI, 1.1–2.9 for all and borderline tumors, respectively; ref. 12). No association was observed if smoking was discontinued within 2–5 years or more than 15 years prior to reference date. In the study by Kelemen and colleagues (11), a significant increase of ovarian cancer risk was observed among former smokers (OR = 1.71; 95% CI, 1.30–2.25) and, in particular, among ex-smokers who quit within two years prior to reference date (OR = 4.24; 95% CI, 2.44–7.36), as no significant results were observed among former smokers who quit more than two years prior to reference date (11). Furthermore, among former smokers, greater number of pack-years (not more than 20 pack-years) and smoking more than 20 years was also associated with increased risk of ovarian cancer (OR = 1.93; 95% CI, 1.29–2.89 and OR = 1.87; 95% CI, 1.28–2.75, respectively). No significant association for former smoking was observed in the other case–control analyses (refs. 13–15, 18, 20; Fig. 2).

Only the cohort study by Gram and colleagues (25) and meta-analysis by Beral and colleagues(28) found an increased risk of overall ovarian cancer among former smokers (HR = 1.3; 95% CI, 1.0–1.7; and RR = 1.06; 95% CI, 1.00–1.13, respectively), while other cohort or pooled/meta-analyses studies observed no association (21–24, 26, 27, 32).

As shown in Fig. 3, 12 case–control studies estimated the association between smoking and serous ovarian cancer risk and the majority of studies found a decreased but nonsignificant risk of serous ovarian cancer associated with current smoking (11, 13, 15, 18, 20, 30) and one study reported null association (36). Two case–control studies observed a significantly decreased risk of serous ovarian cancer associated with current smoking [OR = 0.52; 95% CI, 0.29–0.95 for Baker and colleagues (14) and OR = 0.53; 95% CI, 0.33–0.88 for Riman and colleagues (17)] if women were currently smoking 11 or more cigarettes per day.

Three studies examined the association within nonmucinous histotype (16, 19, 31). However, because a large proportion of nonmucinous ovarian cancer is being presented by serous cancer (33), these results were included among the results for the studies that investigated the associations for serous histotype (Fig. 3). In the study by Green and colleagues (19), an increased risk of nonmucinous cancer was observed among current smokers (OR = 1.6; 95% CI, 1.1–2.3). On the contrary, Modugno and colleagues (31) and Pan and colleagues (16) observed no association between current smoking and nonmucinous ovarian cancer.

Six cohort studies reported results on association between serous ovarian cancer risk and smoking (21–26), but only one cohort study by Gram and colleagues found an increased risk when borderline and invasive cases were combined together (HR = 1.6; 95% CI, 1.0–2.7; ref. 25). However, the association became nonsignificant when referent group included never and passive smokers instead of just never smokers (HR = 1.3; 95% CI, 0.9–1.9). No pooled or meta-analysis studies found any significant association between serous ovarian cancer risk and current smoking (27, 28, 33, 34).

As shown in Fig. 3, among 12 case–control studies only in a study by Kelemen and colleagues it was found that former smoking is significantly associated with an increased risk of serous ovarian cancer (OR = 1.97; 95% CI, 1.46–2.66), driven largely by those who quit within two years of diagnosis (OR = 5.48; 95% CI, 3.04–9.86; ref. 11). In addition, in the study by Riman and colleagues former smoking was inversely associated with risk of serous cancer (OR = 0.72; 95% CI, 0.52–0.98; ref. 17). One cohort study found an increased risk among former smokers (HR = 1.6; 95% CI, 1.0–2.6; ref. 25). Similarly to current smokers, in this study, confidence interval included a null value when the referent category was changed from including just never users to including both never and passive smokers. No other case–control, cohort or pooled studies observed significant results among former smokers.

As shown in Fig. 4, 13 case–control studies presented estimates of the association between current smoking and mucinous ovarian cancer risk. Three of the eight studies combined borderline and invasive mucinous cases together (13, 31, 36) and two presented estimates for borderline tumors only (15, 20). Seven studies reported a significantly increased risk of mucinous ovarian cancer associated with current smoking with estimates ranging from 2.1 to 2.9 (15, 16, 19, 20, 29, 31, 36). In the study by Soegaard and colleagues, increased risk for current smoking was observed when compared to never and former smokers combined (OR = 1.78; 95% CI, 1.01–3.15); however, the association became slightly attenuated and nonsignificant when current smoking was compared with never smokers (13). Five studies observed no association for mucinous ovarian cancer (11, 14, 17, 18, 30).

Seven cohort studies presented results on mucinous ovarian cancer risk associated with current smoking, and four of them found significant positive association between mucinous ovarian cancer risk and current smoking (21, 23, 26, 32) (Fig. 4). Two cohort studies by Gram and colleagues and Gates and colleagues found positive but nonsignificant associations (24, 25). Among the aforementioned six studies three combined borderline and invasive mucinous cases in their analyses (23, 24, 32). The cohort study by Licaj and colleagues reported an increased risk of borderline mucinous tumors among current smokers (HR = 2.17; 95% CI, 1.06–4.45) but did not find a significant association for invasive mucinous ovarian cancer (22). Four pooled/meta-analysis studies reported results on mucinous ovarian cancer risk associated with current smoking and all four found a significantly increased risk, and the estimates ranged between 1.31 to 2.4 (refs. 27, 28, 33, 34; Fig. 4).

As shown in Fig. 4, only one case–control study by Zhang and colleagues found a significantly increased risk of mucinous ovarian cancer among former smokers (OR = 2.5; 95% CI, 1.1–5.4; ref. 30) and no association was observed in other studies (11, 13–18, 20, 29, 31, 36). In the case–control study by Green and colleagues (19), the authors did not find any significant association between former smoking and invasive mucinous cancer risk but they observed a significantly increased risk of borderline mucinous ovarian cancer (OR = 2.6; 95% CI, 1.2–5.2). Only one cohort study by Tworoger and colleagues reported an increased risk of mucinous ovarian cancer among former smokers (RR = 2.02; 95% CI, 1.15–3.55; ref. 32). No other cohort study, pooled or meta-analysis observed the association between former smoking and risk of mucinous ovarian cancer (Fig. 4).

Eight case–control studies presented results on endometrioid ovarian cancer risk and none of them observed any statistically significant association for current smoking, although, almost all the studies reported inverse associations (Fig. 5; refs. 11, 13, 14, 17, 18, 30, 36, 38). Similarly, among the five cohort studies that conducted the analyses on investigating the relationship between current smoking and risk of endometrioid ovarian cancer, all the studies observed nonsignificant inverse association (Fig. 5; refs. 21–24, 26). All of the four pooled/meta-analysis studies observed an inverse association for current smoking, and estimates ranged between 0.73 and 0.84 (Fig. 5; refs. 27, 28, 33, 34). However, only results from the meta-analysis by Beral and colleagues reached statistical significance (RR = 0.81; 95% CI, 0.72–0.92; ref. 28).

As shown in Fig. 5, for the association between former smoking and endometrioid cancer risk, results from the case-control studies were mixed in terms of direction and none of the estimates were statistically significant (11, 13, 14, 17, 18, 30, 36, 38). Only four cohort studies presented results associated with former smoking and all of them observed an inverse association (21, 23, 24, 26), but only the results from study by Gates and colleagues (24) were significant (RR = 0.59; 95% CI, 0.39–0.90). Finally, among the three pooled/meta-analysis studies that presented results associated with former smoking, the associations reported were inverse but all the confidence intervals included the null value (Fig. 5; refs. 27, 28, 34).

Four case-control studies presented results on clear cell ovarian cancer risk and all reported nonsignificant inverse associations for current smoking (11, 14, 17, 18). Similarly, among these studies, for former smoking, the associations were inverse except for the study by Kelemen and colleagues (11) where the authors reported the OR = 1.95 and 95% CI = 0.52–7.29. None of the cohort studies investigated the relationship between smoking and ovarian cancer risk according to this histologic subtype.

Four pooled/meta-analysis studies assessed the association between smoking and clear cell ovarian cancer risk; and all of them observed a significantly decreased risk among current smokers [pOR = 0.74; 95% CI, 0.56–0.98 for Faber and colleagues (27); RR = 0.80; 95% CI, 0.65–0.97 for Beral and colleagues (28); and RR = 0.6; 95% CI, 0.3–0.9 for Jordan and colleagues (33)]. In the study by Kurian and colleagues (34), the estimate was, although inverse, nonsignificant. For former smoking, Faber and colleagues (27) observed an inverse association between former smoking and risk of clear cell ovarian cancer. In the studies by Beral and colleagues (28) and Kurian and colleagues(34), the associations were inverse and nonsignificant.

There were three studies, two case–control (12, 37) and one pooled analysis (35) that did not present the results by the common variable for smoking status with current/former/never categories. The study by Kuper and colleagues(37) reported, however, an increased risk of ovarian cancer associated with a prolonged exposure to smoking. In fact, for smoking for 20–40 years, OR was 1.45; 95% CI, 1.04–2.01 in the overall group. They also reported an increased risk of invasive serous cancer associated with years of tobacco use and elevated risk of mucinous ovarian cancer for women who smoked more than 40 cigarettes per day. In another case–control study, although the results for current/former tobacco use for the overall study population were included, the results for mucinous and serous tumors were not presented by this smoking variable (12). Instead, a variable “years since last smoked” with the categories ≤15 and >15 years was used, with the estimated OR = 2.6; 95% CI, 1.6–4.2 for borderline mucinous, OR = 2.7; 95% CI = 1.1–6.5 for invasive mucinous, and OR = 1.4; 95% CI = 1.1–1.9 for invasive serous for smoking within 15 years before the diagnosis or reference date (12).

Finally, in one the largest prospective studies of smoking and ovarian cancer to date, based on data from the Ovarian Cancer Cohort Consortium, smoking was associated with an increased risk of mucinous ovarian cancer, RR = 1.27; 95% CI = 1.10–1.59 for ever smoking and RR = 1.26; 95% CI 1.08–1.46 per 20 pack-years (35). At the same time, the authors reported a decreased risk of clear cell cancer, RR = 0.72; 95% CI 0.55–0.94. No association was observed in the overall sample and among other histotypes.

Only three studies investigated the impact of cigarette smoking on ovarian cancer survival overall and according to histological subtypes (39–41) with one study being based on consortium data (39). Kelemen and colleagues found significantly increased risk of mortality among both current and former smokers receiving adjuvant chemotherapy; however, this association was limited to cases diagnosed with mucinous ovarian cancer (40). Similarly, in the study by Kim and colleagues (41), it was observed that increased risk of death associated with smoking was limited to mucinous cancer (HR = 2.52; 95% CI, 1.01–6.33). In the pooled analysis, Praestegaard and colleagues (39) also reported significantly increased risk of mortality among current smokers diagnosed with mucinous cancer (pHR = 1.91; 95% CI, 1.01–3.65). Also, for serous ovarian cancer, both current and former smoking was associated with increased mortality (pHR = 1.11; 95% CI, 1.00–1.23 and pHR = 1.12; 95% CI, 1.04–1.20, respectively). In the overall study sample, the associations were of a similar magnitude among both current and former smokers (pHR = 1.17; 95% CI, 1.08–1.28 and pHR = 1.10; 95% CI, 1.02–1.18, respectively).

Epidemiologic evidence summarized in current review suggests a positive association between current cigarette smoking and mucinous ovarian cancer risk. Certain biological mechanisms could explain the observed findings for mucinous histotype. One of them refers to carcinogenic nature of smoking. In fact, according to Surgeon General's Report of 2014, there are more than 7,000 chemicals and 69 known carcinogens in tobacco and tobacco smoke (47). The metabolizing intermediates of these foreign chemicals and carcinogens can bind to DNA and form DNA adducts which can lead to miscoding of DNA replication and permanent mutations (47). Permanent mutations in oncogenes like KRAS or tumor suppressor genes like TP53 can trigger further mutations and lead to carcinogenic processes (47). Significantly elevated levels of DNA adducts have been detected in various tissues among smokers (48).

Moreover, the mechanism underlying the association between smoking and mucinous ovarian cancer relates to the fact that most primary mucinous cancer cells resemble intestinal epithelial cells (33), and that, in many cases, ovarian mucinous carcinomas has a histochemical profile similar to gastrointestinal cancers (4). As smoking has been consistently shown to be associated with mucinous gastrointestinal cancers and adenomatous colorectal polyps, a precursor of colorectal cancer (33), it could also be implicated in the development of mucinous ovarian cancer the same way it may cause mucinous cancers of gastrointestinal system.

It is also important to note that many studies included in current review observed associations of a similar magnitude between smoking and borderline mucinous tumors (12, 15, 19–22, 28). Compared with invasive mucinous tumors, the association between current smoking and borderline mucinous tumors was, generally, more pronounced. The explanation to this observation could be that latent period from precursor lesions to cancer development might take decades and that cigarette smoking might cause carcinogenesis in the early phase (49).

The results of the studies included in this review also suggest that smoking could be associated with a reduced risk of clear cell and endometrioid ovarian cancer. Clear cell cancer shares similar genetic profile with endometrial epithelium, and case studies have found that a high percentage of women with ovarian cancer of this histotype also tend to have endometriosis (3, 8). Therefore, it was postulated that, similar to endometriosis, most clear cell cancer might develop from endometrium through retrograde menstrual flow (8). Smoking is also related to the decreased risk of endometriosis and decreased risk of endometrial cancer especially among postmenopausal women (50, 51). Because endometriosis and endometrial cancer are both estrogen-dependent, antiestrogenic effect of smoking has been suspected to be the underlying biological mechanism for decreased risk of both conditions among smokers (50, 52).

As it was mentioned before, endometrioid and clear cell ovarian cancer might share similar etiology, and, similar to clear cell cancer, endometrioid cancer has been found to be associated with endometriosis (53–55). Therefore, antiestrogenic effect of smoking could potentially explain a decreased risk of endometrioid and clear cell ovarian cancer among smokers. It is important to emphasize, however, that, even though the inverse associations were observed, they were mostly nonsignificant even among pooled or meta-analysis studies. One explanation of that could be a lack of statistical power to be able to detect association. It could also be that misclassification of endometrioid ovarian cancer cases could attenuate the estimated measure of association. Many cases of endometrioid subtype, especially of high grade, diagnosed in the past are now considered serous ovarian cancer (4). As a result, the association between smoking and endometrioid ovarian cancer could have been attenuated in the earlier studies due to the lack of association between smoking and serous ovarian cancer.

The results of epidemiologic studies included in this review suggest that cigarette smoking is not associated with risk of serous ovarian cancer. Two counterbalanced mechanisms were proposed to explain the null association between serous cancer and smoking. On one hand, carcinogenic effect of smoking might increase serous cancer risk. On the other hand, its antiestrogenic effect might decrease serous cancer risk, because nonmucinous cancer has been reported to be more responsive to hormonal factors than mucinous cancer (33, 56). As a result, the ultimate effect might become neutral (33).

What requires further attention is that, although the associations were mostly nonsignificant, there was some discrepancy of the direction of the estimated associations depending on the study design. In fact, among current smokers, the estimates were mostly inverse for case–control studies and mostly positive among cohort studies. Such difference could have several explanations. One of them is the control selection. For instance, Baker and colleagues used hospital controls who usually smoke more than general population and this could attenuate or even reverse the actual association between smoking and cancer risk (14). Another reason is a possible survival bias affecting the findings of the case–control studies. In fact, in case–control studies, there could have been a higher proportion of less severe and less aggressive cases, and, quite likely, a higher proportion of low-grade serous ovarian cancer tumors. There is evidence of low-grade serous ovarian cancer being linked to endometriosis (55), which, similarly to proposed mechanisms for the association between smoking and clear cell and endometrioid cancers, could explain inverse associations among serous ovarian tumors observed in case–control studies.

In our review, only three studies reported on the histotype-specific associations between smoking and ovarian cancer survival. All three studies observed increased risk of mortality within mucinous subtype (39–41), and one reported on adverse survival among women diagnosed with serous histotype (39). Such findings correspond to the results of the studies reporting adverse prognostic impact of smoking on a variety of cancers including lung, head and neck, and breast cancer (57, 58). However, more studies should be conducted to make a more definite conclusion about smoking and survival during the postdiagnostic period among patients diagnosed with certain histotypes of ovarian cancer.

The unfavorable influence of smoking on survival among patients with ovarian cancer could be due to both direct and indirect mechanisms. In in vitro model, cigarette smoke extract promoted cancer cell proliferation and increased VEGF expression, which led to increased angiogenesis (59). Tobacco carcinogens could induce genetic mutations that could lead to more aggressive cancer (60). In addition, smoking might increase cancer mortality through promoting inflammation and suppressing immune function (47, 60). Long-term smokers usually have other smoking-associated conditions such as chronic obstructive pulmonary disease and cardiovascular disease. Furthermore, smoking is associated with presence of other adverse lifestyle habits and comorbidities that may contribute to worse survival.

As mentioned before, studies on epithelial ovarian cancer face a common limitations of a small number of cases and insufficient statistical power that would prevent from conducting analysis stratified by histologic subtypes, and, specifically, for the least frequently occurring histotypes. For example, in the EPIC cohort study, which is, to date, one of the largest prospective cohort studies on smoking and epithelial ovarian cancer, there were only 83 mucinous cases (23). Moreover, further stratification of the sample by categories of smoking exposure lowers the number of cases in each category of these smoking variables. Analyzing such associations by additionally stratifying by histologic subtypes becomes nearly impossible due to restricted power. Because the main aim of this review was to investigate epidemiologic evidence on the associations between smoking and histotype-specific ovarian cancer, we were able to present only the findings for the current/former/never smoking variables. Moreover, a limited number of cases for each histotype may have resulted in the inability to detect the presence of association if one existed or in widening of confidence intervals resulting in nonsignificant findings.

Future studies may need to address this important point by emphasizing other variables that characterize exposure to smoking such as duration, intensity, and frequency of smoking and evaluate presence of a dose–response relationship.

One way to alleviate the problem of insufficient sample size is to combine several histotypes together. For instance, merging together borderline and invasive mucinous cases was a common practice in earlier studies when assessing risk of mucinous ovarian cancer associated with smoking (23, 32, 36). If, as reported by some studies in this review, the association between smoking and mucinous ovarian cancer is stronger among borderline tumors compared with invasive tumors, then combining these two types could lead to overestimation of the true association between smoking and invasive mucinous cancer. Similarly, combining all the nonmucinous histotypes into one “nonmucinous ovarian cancer” category could have obscured a possible association between current smoking and endometrioid histotype.

Another common limitation of ovarian cancer studies, as it was pointed out earlier, is misclassification of cases according to histologic subtype. Very few individual studies included in this review reported having systematic or central pathology review. But even if such pathology review were to be implemented, misclassification of cases could have only been mitigated not eliminated. As it was mentioned earlier, there was misclassification of serous subtype as endometrioid ovarian cancer in earlier studies. There were also misclassification of primary and secondary ovarian cancer, and misclassification of borderline and invasive tumors, especially among mucinous tumor (61). Compared with the other histologic subtypes of epithelial ovarian cancer, mucinous cancer is relatively uncommon, accounting for about 2%–10% of all primary ovarian cancers, but in earlier studies mucinous ovarian cancer was diagnosed much more frequently (29). For example, in two most recent case–control studies, invasive mucinous cases accounted for no more than 5% of total invasive cases (11, 12), but represented more than 12% in three earlier case–control studies (16, 30, 36).

Advances in pathologic techniques have led to a subsequent drop in the number of mucinous cancers cases, particularly because it was determined that some primary mucinous cancers, especially of advanced stage, were actually cases of secondary cancer metastasized from gastrointestinal or cervical cancer (4). As a result of this misclassification, the association between smoking and mucinous ovarian cancer risk could have been overestimated because smoking is an established risk factor for both colorectal and cervical cancers (50). According to Seidman and colleagues, another reason for more diagnoses of invasive mucinous tumor in the past was that some borderline mucinous tumors were misclassified as invasive (62). Such misclassification could have also explained some variation in the measures of association observed within the studies of a similar design.

One of the possible solutions to attenuate a problem of a restricted statistical power and, partially, misclassification could be combining the histotypes according to the origin of disease, into type I and type II tumors (6). To our knowledge, none of the studies examined attempted to combine the histotypes according to this proposed model. To do so, future studies may need to focus on a more thorough pathologic distinction between the ovarian cancer cases, paying a particular attention to grade of the tumors.

Third, misclassification of smoking exposure may occur due to recall bias that could be unintentional or intentional. According to Kelemen and colleagues, underreporting of smoking behavior among current smokers was common and it could have attenuated the positive association between current smoking and cancer risk and mortality (40). Missing information on smoking could also cause misclassification of smoking exposure. For instance, cohort studies often collected information on smoking at enrollment but no updated information tends to be collected during follow-up period when some patients can change their smoking behavior. Studies on survival also reported possibility of misclassification of smoking exposure due to missing information on smoking after diagnosis (60, 63).

Fourth, selection bias with case–control studies could occur due to underrepresentation of smokers among controls because controls who are willing to participate tend to be healthier than nonresponders. This selection bias could lead to overestimation of the positive association between smoking and risk of mucinous ovarian cancer. Selection bias could also stem from underrepresentation of advanced cases. In the case–control study by Kelemen and colleagues, 14% of newly diagnosed cases were nonresponders due to rapid death after diagnosis (11). Another case–control study had 11.4% of non-responders due to death (16), while many other studies did not report such statistics. Serous ovarian cancer is usually diagnosed at advanced stage, while endometrioid, clear cell, and mucinous ovarian cancers are usually detected at a lower stage (4), so underrepresentation of advanced cases might disproportionally affect the histotype of serous cancer. If, as reported, smoking is associated with worse survival primarily among cases at advanced stage, underrepresentation of advanced cases could lead to attenuated association between smoking and risk of serous ovarian cancer. Survival bias could explain differences in the measures of associations observed across case–control studies.

Finally, sources of our review were all from published literature, so the interpretation of the results could have been affected by publication bias. There was also overlapping of results as some individual studies were a part of one or more pooled or meta-analysis included in the current review. Such duplication of the results should be remembered when interpreting the results of these studies.

In conclusion, no significant association between cigarette smoking and overall ovarian cancer risk was consistently observed among studies included in this review. The only significant association observed across case–control, cohort, and pooled studies was an increased risk of mucinous ovarian cancer associated with current smoking. However, we cannot rule out the fact that this association could be overestimated due to misclassification of mucinous cases and inclusion of borderline tumors. There was also some evidence on a decreased, although nonsignificant, risk of clear cell and endometrioid ovarian cancer associated with smoking, Overall, smoking appeared to have a negative impact on ovarian cancer survival, but its subtype-specific impact was not clear due to the insufficient amount of epidemiologic evidence.

International epidemiologic collaborations formed in the past decade have dramatically increased sample size and statistical power in ovarian cancer research, therefore, allowing further investigation of the links between such important lifestyle factors as smoking with uncommon histologic subtypes including endometrioid, clear cell, and mucinous (64). However, to have a better understanding of the relationship between smoking habits and ovarian cancer, future studies should be conducted with a particular attention being paid to the collection of a detailed information on smoking habits both prior and after the diagnosis and a more thorough classification of ovarian tumors.

No potential conflicts of interest were disclosed.

A.N. Minlikeeva was supported by NIH/NCI 4R25CA113951 and NIH/NCI P50CA159981. K.B. Moysich was supported by NIH/NCI (2R25CA113951, R01CA095023, R01CA126841, P50CA159981) and the Roswell Park Alliance Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.