Is smoking an independent risk factor for developing cervical intra-epithelial neoplasia and cervical cancer? A systematic review and meta-analysis

ABSTRACT

Introduction: Cervical cancer is the fourth most common form of cancer among women. Smoking tobacco seems to be a risk factor for the development of cervical intra-epithelial neoplasia (CIN) and cervical cancer, but the exact role of smoking in the process of cervical carcinogenesis is not known. The aim of this study is to investigate the relationship between smoking and the development of CIN and cervical cancer. Areas covered: We searched Embase, Medline, Cochrane Central, Web of Science, and Google Scholar for studies on smoking and CIN and cervical cancer, published between 2009 and 2018. The following were the outcomes: CIN3 alone, CIN2 and CIN3 combined, CIN2+, CIN3+, and cervical cancer alone. We included 49 studies in our review and 45 in our meta-analyses. Expert opinion: Based on the available evidence it can be – cautiously – concluded that smoking increases the risk of cervical abnormalities. However, the high risk of bias indicates that for future studies, it will be important to adjust for relevant predictors, to separate CIN from cervical cancer as outcome measures, and to report research methods in detail.

KEYWORDS:

• Cervical intraepithelial neoplasia
• cervical cancer
• meta-analysis
• risk factor
• smoking
• systematic review

1. Introduction

Cervical cancer is the fourth most common form of cancer among women worldwide. Every year, half a million women are diagnosed with this disease [1]. Cervical cancer is preceded by a premalignant stage of cervical intra-epithelial neoplasia (CIN). This premalignant stage makes it possible to detect a cervical abnormality by screening before it develops into cervical cancer [2]. A high-risk human papillomavirus (hrHPV) cervical infection underlies the development of cervical abnormalities [3]. Only if an infection persists, can it lead to the development of a premalignant CIN lesion [4,5], and the progression to cancer can take up to 10 to 15 years.

It is unclear why some hrHPV infections lead to the development of CIN and ultimately cancer, while other hrHPV infections are cleared by the immune system and CIN lesions show regression. Smoking tobacco is assumed to be a significant behavioral risk factor for the persistence of hrHPV infections and the development of CIN and cervical cancer [6]. Other risk factors are, for example, immunocompromised status and hormonal contraception [6]. There are several explanatory hypotheses about the association between smoking and cervical cancer, including a direct oncogenic effect on chemical carcinogenesis, or a carcinogenic effect due to the suppression of cell-mediated immunity [7]. However, the exact role of smoking in cervical carcinogenesis is not known [8].

The aim of this study is to investigate the association between smoking and the development of CIN and cervical cancer, by means of a systematic review and meta-analysis. Previous reviews have looked at the relationship between smoking, CIN, and cervical cancer, but have not examined what the exact risk is and have not looked at different stages of CIN [9–11]. The research questions are: (1) Is smoking an independent risk factor for developing (different stages of) CIN, and if so, what is the risk? (2) Is smoking an independent risk factor for developing cervical cancer, and if so, what is the risk?

2. Method

2.1. Registration

This review was registered with NARCIS under number OND1365476.

2.2. Search strategy

We performed a systematic search using a keyword-based search strategy in five databases on 7 January 2019. We searched in Embase, Medline, Cochrane Central, Web of Science, and Google Scholar. For the search strategy, we used search terms related to smoking, HPV, CIN, and cervical cancer. See supplement 1 for the full search strategy.

2.3. Selection of studies

We used the following inclusion criteria: (1) articles published from 2009 onwards; (2) articles written in English; (3) participants in the studies are between 18 and 65 years old; (4) the independent variable smoking refers to active smoking of tobacco; (5) the studies concern the development or exacerbation of CIN and/or the development of cervical cancer in combination with tobacco smoking; and (6) the studies use original data. We used the following exclusion criteria: (1) 20% or more of the study population have HIV; and (2) 20% or more of the study population are pregnant or in the postnatal period.

The articles found were initially independently assessed by two coders for suitability on the basis of the title and abstract. In the case of non-matching codes, a third coder was decisive. The full text of the selected articles was subsequently assessed for suitability by two independent coders. In the case of non-matching codes, a third coder was decisive. To ensure rapid completion of the selection process, five coders worked on the selection process simultaneously (BvS, OvdH, TM, GN, and GJM).

Figure 1 shows a flow chart of the selection process. The search yielded 4,011 articles after duplicates were removed. We screened 220 full-text articles, of which 171 were excluded. Finally, 49 articles met all inclusion criteria. Twenty-one studies were about smoking as a risk factor for CIN, 19 studies examined the question whether smoking is an independent risk factor for developing cervical cancer, and 14 studies combined CIN and cervical cancer as outcome.

Figure 1. Flowchart of the selection process

2.4. Data extraction and assessment

A standardized data extraction form was used for data extraction. The data extraction from the selected articles was carried out by one researcher per article (BvS, OvdH, TM, GN, and GJM) and was checked by another researcher when performing the risk of bias assessment. The form was used to collect data on the study method, the characteristics of the participants in the study, the outcomes, the results and the mentioned conflicts of interest and the funder of the study, if applicable.

For the assessment of the risk of bias, a standardized form was used based on the risk of bias form by the Cochrane Collaboration. Two independent coders assessed the risk of bias and in the case of non-matching assessments, a third coder was decisive (OvdH, TM, and GN). In each case, the coders indicated whether the risk of bias was high, low or unclear, the justification for this assessment, and the source of this justification in the text of the article. To assess the risk of bias, the following aspects were examined: (1) bias due to a non-representative or ill-defined sample of patients; (2) bias due to insufficiently long, or incomplete follow-up, or differences in follow-up between treatment groups; (3) bias due to ill-defined or inadequately measured outcome; and (4) bias due to inadequate adjustment for important prognostic factors.

For each question, two independent coders (OvdH and GN) carried out an assessment using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. In the case of non-matching codes, a third coder was decisive (TM). Evidence based on RCTs receive a quality rating ‘High’, observational studies a quality rating ‘Low’, and other studies (e.g. case reports) a quality rating ‘Very low’. There are five factors that can downgrade the quality of evidence, namely risk of bias, inconsistency, indirectness, imprecision and publication bias. There are three factors that can upgrade the quality: high effect, dose–response relationship, and plausible confounding. For each factor, the quality can be reduced by one or two levels. If the factor is a serious limitation, the quality goes down by one level; if the factor is a very serious limitation, the quality goes down by two levels.

Studies were divided according to the studied stages of CIN or cervical cancer. Before data collection, we planned to divide the studies into four groups according to the outcomes studied: CIN1, CIN2, CIN3, and cervical cancer. After data collection, we found that few studies examined CIN1 and many studies combined CIN and cervical cancer as outcomes. Therefore, the subgroups included in the study are: CIN3 alone, CIN2 and CIN3 combined, CIN2+ (which includes CIN2, CIN3 and cervical cancer), CIN3+ (which includes CIN3 and cervical cancer), and cervical cancer alone.

2.5. Meta-analysis

After data extraction, the results of each included article concerning the relationship between smoking and cervical abnormalities were converted to odds ratio (OR) if the original article was reported in other effect measures (as far as the available data allowed it). This allowed us to better compare the effects of the individual studies. If available, data from adjusted analyses were used. Otherwise data from crude analyses were used. Where possible, results from ‘never’ compared to ‘current’ smokers were included in the analyses. If data from current smokers were unavailable, results from ‘never’ compared to ‘ever’ smokers were included. Some studies were unclear about whether they used current or ever smoking (they just called it ‘smoking’), these studies were also included in the meta-analysis. The calculations were performed using the online tool MedCalc.

Meta-analyses were performed in Review Manager 5.3 (RevMan). ORs were calculated with the inverse variance method. When more than five eligible studies were included in the analyses, a random-effects model was used [12,13]. A random-effects model was also used when a meta-analysis consisting of five or less studies showed moderate or substantial heterogeneity according to the I² test (<40% = no heterogeneity; 40% to 70% = moderate heterogeneity and ≥70% = substantial heterogeneity) [13–15]. A fixed-effects model was used when a meta-analysis consisting of less than six studies showed no substantial heterogeneity.

Publication bias was checked with funnel plots using the Egger test [16] when 10 or more studies were eligible for a meta-analysis. We performed sensitivity analyses in which we only pooled studies with a low risk of bias, i.e. studies which scored a low risk of bias on at least three of the four domains or a high risk of bias on a maximum of one of the four domains.

3. Results

3.1. Smoking as a risk factor for CIN

3.1.1. Study characteristics

The study characteristics are shown in Table 1. Of the 21 studies about smoking as a risk factor for CIN, nine were cross-sectional studies that included between 266 and 12,048 respondents [17–25]. Three cross-sectional studies had a different study objective than examining the relationship between smoking and CIN [23–25]. Nine case–control studies included between 186 and 4,522 respondents [26–34]. Two longitudinal studies with 150 and 1,485 respondents [35,36] and a large prospective cohort study with 308,036 respondents were included [37].

Table 1. Characteristics and results of the included studies

First author, publication year	Study design, setting, country	Number of participants, age range	Determinant	Outcome	Smoking results	Results
Smoking as a risk factor for CIN
Bassal, 2016	Case–control study, a tertiary hospital and a health maintenance organization, Israel	n = 178, age range not reported	Smoking (never, ever)	CIN3	OR = 2.1, 95% CI = 1.1–4.1	Ever smoking is associated with CIN3 in univariate analyses (OR = 2.4, 95% CI = 1.3–4.4, p < 0.01) and in multivariate analyses (OR = 2.1, 95% CI = 1.1–4.1, p = 0.02).
Collins, 2010	Longitudinal cohort study, Birmingham Brook Advisory Centre, UK	n = 1,485, 15–19 years	Smoking (never, ever)	High-grade CIN	HR = 1.33 (95% CI = 0.77–2.30)^a	N = 28 high-grade CIN in incident cohort. N = 56 high-grade CIN in all women.
Daily, 2018	Cross-sectional, Birmingham colposcopy clinic, Alabama, USA	n = 1,058, 21–24 years	Smoking (never, former, current, nonsmoker, unknown)	CIN2/3	OR = 1.64 (95% CI = 1.20–2.25)	Patients with and without CIN 2/3 differed in smoking history (p < .001) In univariate analysis, smoking (OR = 1.64, 95% CI = 1.20–2.25) was associated with CIN 2 or 3 on colposcopic biopsy.
Dimitrov, 2012	Case–control study, study, setting not reported, Macedonia	n = 186, age range not reported	Smoking	CIN	OR = 1.72 (95% CI = 1.08–2.73)	Smoking is associated with a higher risk of CIN. RR = 1.718, 95% CI = 1.080–2.734, p < 0.025.
Grimm, 2011	Case-control study (prospective), hospital, Austria	n = 412, age range not reported	Smoking (yes, no)	CIN	OR = 3.3 (95% CI = 1.6–6.6)	Smoking was independently associated (p = 0.001) with a higher risk of CIN: OR = 3.3; 95% CI = 1.6–6.6.
Guarisi, 2009	Prospective cohort study, hospital, Argentina/Brazil	n = 150, age range not reported	Smoking (never, current/past)	CIN1+ and/or CIN2+	OR = 1.43 (95% CI = 0.57–3.61)^b	A total of 21 cases of incident CIN were detected during follow-up; 11 among 67 smokers and 10 in 83 nonsmokers. Smoking history was not an independent predictor of incident CIN (adjusted HR = 1.16; 95% CI = 0.46–2.95).
Kim, 2012	Case–control study, The Center for Uterine Cancer and the Center for Cancer Prevention and Early Detection, Korea	n = 392, 25 years or older	Smoking (never, ever)	CIN	OR = 2.49 (95% CI = 1.21–5.15)	Smoking was associated with CIN (OR = 2.49, 95% CI = 1.21–5.15, p = 0.014) compared to controls.
Longatto-Filho, 2011	Observational study (combination population-based, cross-sectional and prospective cohort study), clinics, Brazil and Argentina	n = 12,048, 14–67 years	Smoking (never, ever)	CIN2/3	OR = 1.01 (95% CI = 0.67–1.51)	Ever been a smoker was not significantly associated with CIN2/3 (AOR = 1.01, 95% CI = 0.67–1.51, p = 0.963).
Luhn, 2013	Cross-sectional study, University Colposcopy Clinic, USA	n = 2,783, 18 years or older	Smoking (never, ever)	CIN 3 vs CIN2	OR = 1.95 (95% CI = 1.48–2.58)	Compared to CIN2, CIN3 was positively associated with smoking (ever vs. never): OR = 1.95 (95% CI = 1.48–2.58).
Oh, 2016	Case–control study, university hospitals, Korea	n = 678, <35 – ≥55 years	Smoking (never, current, past)	CIN2/3	OR = 1.56 (95% CI = 0.72–3.41)	The CIN groups included a higher proportion of tobacco smokers (p = 0.004). In both multinomial and logistic regression analyses, smoking status was not associated with CIN grade.
Palma, 2010	Case–control study, hospital, Italy	n = 192, 20–77 years	Smoking (no, yes)	HSIL	OR = 1.16 (95% CI = 0.43–3.10)	Smokers did not have an increased risk for LSIL (OR = 1.41; 95% CI = 0.48–4.14; OR adjusted for age = 2.43; 95% CI = 0.52–11.26), HSIL (OR = 1.11; 95% CI = 0.45–2.75; OR adjusted for age = 1.16; 95% CI = 0.43–3.10).
Poomtavorn, 2011	Cross-sectional study, hospital, Taiwan	n = 266, 20–60 years	Smoking (never, current)	High-grade dysplasia	OR = 0.48 (95% CI = 0.05–4.72)	There was no correlation between smoking history and high-grade dysplasia. The OR of high-grade dysplasia was 0.478 (95% CI = 0.048–4.721) in never-smoking patients in comparison with current smokers.
Roura, 2014	Prospective cohort study, centers, Denmark, France, Germany, Greece, Italy, the Netherlands, Norway, Spain, Sweden, UK	n = 308,036, 35–70 years	Smoking (never, ever, past, current)	CIN3/CIS	OR ever vs never = 1.91 (95% CI = 1.65–2.20)^b OR current vs. never = 2.35 (95% CI = 2.00–2.76)^b	Multivariate HR for the association between ever smoking and risk of CIN3/CIS: HR = 1.8 (95% CI = 1.5–2.0); current smoking and risk of CIN3/CIS: HR = 2.1 (95% CI = 1.8–2.5).
Russo, 2011	Cross-sectional study, hospital, Brazil	n = 531, 18–48 years	Smoking (never, ever)	High-grade CIN	OR = 1.92 (95% CI = 1.13–3.27)	A statistically significant association between high-grade CIN and smoking was found. On average, the increase in the odds for high-grade CIN was 80% for smoking: crude OR = 1.80; 95% CI = 1.23–2.62; p < .01. A multivariate analysis confirmed that smoking was a risk factor (adjusted OR = 1.92; 95% CI = 1.13–3.27; p = .02).
Spiryda, 2009	Cross-sectional study, health women’s center, USA	n = 89, 15–59 years	Smoking (no, yes)	CIN2/3 (compared to CIN1)	OR = 0.34 (95% CI = 0.19–0.64)^b	CIN1: Smokers: 45%, non smokers: 55%; CIN2/3: Smokers: 22%, non smokers: 78%; difference significant (p < 0.05).
Tomita, 2011	Case–control study, hospital, Brazil	n = 684, 21–65 years	Tobacco use (never, current/former)	CIN3	OR = 3.40 (95% CI = 2.42–4.78)^b	165 of 231 cases (71.4%) are former or current smokers. 192 of 453 controls (42.4%) are former or current smokers. No statistical test is performed for this difference.
Vidal, 2014	Cross-sectional study, university clinics, USA	n = 572, 18 years or older	Smoking (no, yes)	CIN	OR = 1.41 (95% CI = 0.85–2.33)^b	Seventy-three percent of women with CIN 1,2,3 reported being current cigarette smokers, compared with 15% of women without visible CIN lesions (p = 0.25).
Wang, 2009	Cross-sectional study, university center, USA	n = 1,378, 18–81 years	Smoking (never, current)	CIN3 vs <CIN2	OR = 2.8 (95% CI 2.0–3.9)	Women who were current smokers were more likely than women who were never smokers to have CIN3 compared to less than CIN2 (OR = 2.80, 95% CI = 2.04–3.85).
Xi, 2011	Cross-sectional study, planned parenthood clinics, USA	n = 211, 18–50 years	Smoking (no, yes)	CIN2/3	OR = 1.73 (95% CI = 1.00–3.00)^b	Women with, compared to without, CIN2/3 were more likely to self-report as current smokers (p = 0.05).
Xu, 2018	Case–control study, setting not applicable, Australia	n = 4,522, 30–44 years	Smoking (never, current)	CIN2/3	OR = 1.43 (95% CI = 1.14–1.80)	Current-smokers had higher risk of CIN 2/3 than never-smokers (OR = 1.43; 95% CI = 1.14–1.80).
Zhang, 2012	Case–control study, hospital, China	n = 329, 24–79 years	Smoking (no, yes)	CIN	OR = 0.11 (95% CI = 0.01–1.90)^b	0% of CIN patients smoked and 5.6% of controls. The difference in smoking between CIN patients and controls was not significant (p = 0.067).
Smoking as a risk factor for cervical cancer
Abbas, 2014	Case–control study, Medical University, India	n = 408, 30–70 years	Smoking (never, ever)	Cervical cancer risk	OR = 2.32 (95% CI = 0.95–5.70)^b	13 of 87 cases smoked versus 9 of 128 controls. Difference was not tested.
Bassal, 2016	Case–control study, a tertiary hospital and a health maintenance organization, Israel	n = 80, age range not reported	Smoking (never, ever)	Cervical cancer	OR = 2.4 (95% CI = 0.3–17.2)	Ever smoking is associated with cervical cancer in univariate analysis (OR = 4.2, 95% CI = 1.3–14.0, p = 0.02) and in multivariate analyses controlled for age and depression/anxiety (OR = 5.5, 95% CI = 1.2–25.8, p = 0.03), but not in multivariate analyses controlled for age, age at first intercourse, and depression/anxiety (OR = 2.4, 95% CI = 0.3–17.2, p = 0.38).
Currin, 2009	Cross-sectional study, cancer registries, UK	n = 2,231, 25–64 years	Smoking prevalence, no individual data	Cervical cancer at PCT level	IRR = 1.02 (95% CI = 1.00–1.02)^c	There was a significant positive correlation between the age-standardized incidence of cervical cancer and smoking prevalence at PCT level (Spearman r = 0.62; p < 0.001). Univariate: Smoking prevalence IRR = 1.04 (95% CI = 1.02–1.06). Multivariate: Smoking prevalence IRR = 1.02 (95% CI = 1.00–1.02).
Fernandes, 2011	Case–control study, hospital, Brazil	n = 158, 15–65 years	Smoking (no, yes)	Cervical cancer	OR = 1.25, 95% CI = 0.35–4.53)	Women who were smokers, presented a higher risk of developing cervical cancer (OR = 11.24, 95% CI = 5.09–24.83) that was not significant in a multivariate analysis (adjusted OR = 1.25, 95% CI = 0.35–4.53).
Gutman, 2009	Case–control study, Medical Center, Israel	n = 166, age range not reported	No details on smoking data	Cervical cancer	No effect size or absolute numbers reported	Smoking was the only independent risk factor for cervical cancer (χ^2 = 13.1083, p = 0.0003).
Kim, 2012	Case–control study, The Center for Uterine Cancer and the Center for Cancer Prevention and Early Detection, Korea	n = 392, 25 or older	Smoking (never, ever)	Cervical cancer	OR = 3.42 (95% CI = 1.59–7.38)	Smoking was associated with ICC (OR = 3.42; 95% CI = 1.59–7.38, p = 0.002) compared to controls.
Meng, 2011	Case–control study, medical university, China	n = 733, age range not reported	Smoking (no, yes)	Cervical cancer	OR = 3.94 (95% CI 2.76–5.62)^b	39.4% of cases were smoker, 14.2% of controls were smoker. This difference was significant (p < 0.001).
Palma, 2010	Case–control study, hospital, Italy	n = 192, 20–77 years	Smoking (no, yes)	cervical cancer	OR = 1.91 (95% CI = 0.54–6.82)	Smokers did not have an increased risk for cervical cancer (OR = 1.06; 95% CI = 0.40–2.80; OR adjusted for age = 1.91; 95% CI = 0.54–6.82).
Reis, 2011	Case–control study, hospital, Turkey	n = 1,259, age range not reported	Smoking (never, ever)^d	Cervical cancer	OR = 1.62 (95% CI = 1.16–2.27)^b	Univariate analysis showed the following: cases smoking vs. controls smoking: 28.7% vs. 19.9%; χ^2 = 8.040, p = 0.004. In multivariate analysis, smoking was not found to increase the risk of cervical cancer.
Roszak, 2014	Case–control study, cancer center or university hospital, Poland	n = 741, age range not reported	Smoking (never, ever)	Cervical cancer	OR = 1.10 (95% CI = 0.84–1.44)^b	An increase in cervical cancer risk was seen among patients with a positive history of tobacco smoking. 35% of cases were smoker and 33% of controls.
Roszak, 2017	Case–control study, cancer center, Poland	n = 969, age range not reported	Smoking (never, ever)	Cervical cancer risk	OR = 1.09 (95% CI = 0.84–1.43)^b	35.3% of cases with cervical cancer ever smoked; 33.3% of controls ever smoked. Difference was not tested.
Roura, 2014	Prospective cohort study, centers, Denmark, France, Germany, Greece, Italy, the Netherlands, Norway, Spain, Sweden, UK	n = 308,036, 35–70 years	Smoking (never, ever, past, current)	Cervical cancer	OR ever vs never = 1.85 (95% CI = 1.45–2.37)^b OR current vs never = 2.23 (95% CI = 1.68–2.96)^b	In the cohort analyses smoking status showed a twofold increased risk of ICC. Multivariate HR for the association between ever smoking and risk of ICC: HR = 1.7 (95% CI = 1.3–2.2); current smoking and risk of ICC: HR = 1.9 (95% CI = 1.4–2.5).
Sharma, 2011	Case–control study, hospital, India	n = 617, age range not reported	Smoking (no, yes)	Cervical cancer	OR = 2.09 (95% CI = 1.19–3.67)	The risk of cervical cancer was found to be twice in smokers (OR = 2.09, 95% CI = 1.19–3.67; p = 0.01).
Singh, 2009	Case–control study, medical university, India	n = 310/312, age range not reported	Tobacco use in any form (no, yes)	Cervical cancer	OR = 4.50 (95% CI = 2.44–8.31)^b	Among cervical cancer cases 37.2% were tobacco users while only 11.6% females in control group used tobacco.
Singhal, 2016	Case–control study, two hospitals, India	n = 300, age range not reported	Smoking (no, yes)	Precancer and cancer together	OR for cancer only = 8.37 (95% CI = 3.12–22.43)^b	Smoking cases (pre-cancer and cancer together) 42% versus controls 3%: OR = 11.3 (95% CI = 4.3–29.4). Smoking cases (cancer only) 33% versus controls 3% (no OR calculated by authors).
Zhang, 2012	Case–control study, hospital, China	n = 329, 24–79 years	Smoking (no, yes)	Cervical cancer	OR = 0.88 (95% CI = 0.27–2.89)^b	5.0% of cervical cancer patients and 5.6% of controls smoked (p = 1.000).
Zhang, 2013	Matched case–control comparison study, maternal and child service center, China	n = 165, 35–65 years	Smoking (almost, sometimes, hardly)	Cervical cancer	OR = 12.71 (95% CI = 0.001–145.27)	Living habit (smoking, drinking history, physical training) and social psychological factors (adverse event of life, moodiness, emotion regulation capacity) were not directly associated with cervical cancer.
Zidi, 2017	Case–control study, oncology institute, Tunisia	n = 276, age range not reported	Smoking (no, yes)	Cervical cancer	OR = 7.12 (95% CI = 2.97–17.04)	There was a significant association between the incidence of CC and smoking (p < 0.001; OR = 7.12; 95% CI = 2.97–17.04).
Zidi, 2018	Case–control study, clinic/hospital, Tunisia	n = 600, age range not reported	Smoking (no, yes)	Cervical cancer (early FIGO stages, advanced FIGO stages)	Early FIGO stages: OR = 14.00 (95% CI = 7.82–25.32) Advanced FIGO stages: OR = 23.85 (95% CI = 13.52–42.06)	Smoking habit (early FIGO stages: p = 0.000; OR = 14; 95% CI = 7.82–25.32 or advanced FIGO stages: p = 0.000; OR = 23.85; 95% CI = 13.52–42.06) is a significant factor for cervical cancer clinical course modulation.
Smoking as a risk for CIN and cervical cancer (combined)
Boldrini, 2014	Cross-sectional study, University Hospital, Brazil	n = 291, 30–48 years (interquartile range)	Smoking (yes, no)	High-grade lesions/cervical cancer	OR = 2.4 (95% CI = 1.14–5.18)	In the multivariate analyses, being a tobacco user (OR = 2.4, 95 CI% = 1.14–5.18) was a factor associated with high-grade lesions/cervical cancer.
Castro, 2009	Nested case–control study, the VIP project, Sweden	n = 2,736, 20.1–69.7 years	Smoking (never, former, current)	CIN3, cervical cancer	OR current smokers versus never smokers = 3.01 (95% CI = 2.38–3.80)	Being a former smoker (vs. never being a smoker) was a risk factor for CIN3/cervical cancer (OR = 1.8, 95% CI = 1.35–2.41). Begin a current smoker (vs. never being a smoker) was a risk factor for CIN3/cervical cancer (OR = 3.01, 95% CI = 2.38–3.80).
Fang, 2018	Prospective cohort study, hospital, China	n = 1,531, 21–23 years	Smoking (never, ever)	CIN3+ (=CIN3 or worse)	HR = 1.41 (95% CI = 1.07–1.85)^a	An increased risk was observed for CIN3+ (HR = 1.41; 95% CI = 1.07–1.85) in women who had ever smoked in comparison to those who had never smoked. Current smokers observed a subsequent CIN3+ risk (HR = 1.44; 95% CI = 1.04–1.89; in the fully adjusted model) in comparison to former smokers.
Gargano, 2012	Observational longitudinal study, hospital, USA	n = 1,658, 18–69 years	Smoking (never, former, current)	CIN3+ versus ≤CIN2	OR current versus never smokers = 2.17 (95% CI = 1.46–3.24)^b OR ever versus never smokers = 2.36 (95% CI = 1.68–3.33)^b	35% of CIN3+ cases currently smoked versus 20% of ≤CIN2 cases (RR = 1.5, 95% CI = 1.1–2.1). 47% of CIN3+ cases ever smoked versus 28% of ≤CIN2 cases.
Girianelli, 2009	Cohort study, Family Health Program of Duque de Caxias and Nova Iguaçu cities, Brazil	n = 1,236, 25–59 years	Smoking (never, former, present)	CIN2+	OR = 4.29 (95% CI = 1.28–14.37)^b	After adjustment for covariates, the presence of CIN2 or more advanced lesions was associated with smoking (HR = 7.7; 95% CI = 2.0–29.6).
Jensen, 2012	Cohort study (prospective), population-based, Denmark	n = 1,353, 22–32 years	Smoking (never, ever, former, current)	CIN3+	OR current versus never smoking = 1.52 (95% CI = 1.10–2.09)^b OR ever versus never smoking = 1.40 (95% CI = 1.03–1.89)^b OR current versus former smoking = 1.32 (95% CI = 0.91–1.92)^b	Women who had ever smoked had an increased risk for CIN3+ (HR unadjusted = 1.36; 95% CI = 1.03–1.80). After adjustment for length of schooling, ever given birth, and HPV type, the risk estimate was increased at borderline statistical significance (HR = 1.32; 95% CI = 1.00–1.76). When women who had ever smoked (former smokers and current smokers), only current smokers were significantly more likely to have a subsequent CIN3+ in the fully adjusted model (HR = 1.39; 95% CI = 1.03–1.87).
Jia, 2012	Matched case–control study, three towns in Wufeng, China	n = 1,040, age range not reported	Smoking (never, ever)	CIN2, CIN3, or cervical cancer	OR = 0.92 (95% CI = 0.41–2.06)	Smoking and drinking history were not directly associated with the incidence of cervical cancer or CIN2/3 (p ≥ 0.05).
Jurczyk, 2018	Case–control study, Gynecological and Obstetrical Clinical Hospital and at the Greater Poland Center of Oncology, Poland	n = 200, age range not reported	Smoking (no, yes)	CIN3,cervical cancer	OR = 1.78 (95% CI = 1.01–3.15)	Women with cervical cancer were more frequently smokers (OR = 1.78, 95% CI = 1.01–3.15, p = 0.046).
Lee, 2015	Observational (combination of case–control and cohort study), community/hospital, Taiwan	n = 1,523, 20–75 years	Pack-years of smoking (0, 1–10, ≥11)	CIN2+	OR = 5.1 (95% CI = 1.7–15.7)	A smoking history of ≥11 pack-years was a significant predictor of CIN2+ in the HPV-testing non included model (OR = 5.1; 95% CI = 1.7–15.7). In the HPV-testing-contained model, women who had a smoking history of ≥11 pack-years were more likely to develop CIN2+ (OR = 5.5; 95% CI = 1.4–21.7).
Mazarico, 2015	Cross-sectional study, hospital, Spain	n = 1,007, age range not reported	Smoking (no, yes)	CIN2+	OR = 1.64 (95% CI = 1.33–1.88)	The risk of CIN2-3 or cervical carcinoma increased 1.64 times among smoking women compared to nonsmoking (OR = 1.64, 95% CI = 1.33–1.88, p = 0.05).
Morgan, 2015	Retrospective case-control study, University, USA	n = 1,160, age range not reported	Smoking (never vs former/current)	CIN2+	OR = 1.52 (95% CI = 0.90–2.58)	Smoking history provided an adjusted OR of 1.52 (95% CI = 0.90–2.58) with a p value of 0.12 for CIN2+ .
Sarian, 2009	Multicenter population-based cohort study, centers, Argentina, Brazil	n = 12,076, 15–60 years	Smoking (never, current)	CIN2+	OR = 1.8 (95% CI = 1.3–2.5)	Being a current smoker increased the risk of CIN2+ at baseline (OR = 1.8, 95% CI = 1.3–2.5, p < 0.01).
Tavares, 2016	Case–control study, healthcare center, Brazil	n = 234, age range not reported	Smoking (no, yes)	HSIL	OR = 3.57 (95% CI = 1.10–11.59)	We found an association of smoking with the susceptibility to HSIL in the study population (OR = 3.569, 95% CI = 1.0994–11.5864, p = 0.03).
Zhang, 2016	Cross-sectional study, setting not reported, China	n = 1,315, 21–68 years	Smoking (no, yes)	Cervical lesions	OR = 6.78 (95% CI = 1.20–38.23)	Smoking was significantly associated with cervical lesions (OR = 6.778, 95% CI = 1.202–38.226, p = 0.030).

^aNo absolute number, OR calculation not possible; ^bself-calculated; ^cno individual data; ^dEver = smoked at least one cigarette a day for at least 1 year.

3.1.2. Risk of bias

Of the 21 studies on smoking as a risk factor for CIN, 11 studies showed a high risk of bias and 11 an unclear risk of bias on at least one of the examined areas (supplement 2). Only five studies showed a low risk of bias on all relevant examined areas. The funnel plot is largely symmetrical (supplement 8A).

3.1.3. Results on smoking as a risk factor for CIN

The results of 20 studies could be pooled to examine the association between smoking and CIN2/3 or CIN3 (Figure 2). We found a significant positive association with an OR of 1.67 (95% CI = 1.37–2.04) between smoking and CIN2/3 or CIN3. The heterogeneity index I² was substantial with 75%. The study by Spiryda et al. [25] was included in the pooled analysis but used respondents with CIN1 as comparison group. The study by Collins et al. [36] could not be included (because no OR was reported) and found no significant increased risk of CIN2/3 among women who smoked (HR = 1.33 (95% CI 0.77–2.30). When examining the 15 studies about smoking and CIN2/3 separately from the 5 studies about smoking and CIN3, we found an OR of 1.40 for CIN2/3 (supplement 3) and an OR of 2.46 for CIN3 (supplement 4). When only including studies with a low risk of bias (14 studies), we still found a significant association between smoking and CIN2/3 or CIN3 with an OR of 1.88 (95% CI = 1.56–2.27).

Figure 2. Association between smoking and CIN2/3 or CIN3 (20 studies)

3.2. Smoking as a risk factor for cervical cancer

3.2.1. Study characteristics

A total of 19 studies examined the question whether smoking is an independent risk factor for developing cervical cancer. Among these 19 studies, 12 studies had a different research objective but also showed results about smoking in relation to cervical cancer. These studies generally involved a selected patient population. For the most part, these were observational case–control studies in which a certain risk factor was studied and shown in relation to smoking. In addition, there were seven studies that looked at smoking as a possible risk factor for developing cervical cancer as a primary study objective. Seventeen studies were case–control studies with between 80 and 1,259 respondents [27–30,38–50]. One study was a cross-sectional study with 2,231 respondents [51]. And one study was a prospective cohort study with 308,036 respondents [37].

3.2.2. Risk of bias

Of the 19 studies on smoking as a risk factor for cervical cancer, 16 showed a high risk of bias and 7 an unclear risk of bias on at least one of the examined areas (supplement 2). Only two studies showed a low risk of bias on all relevant examined areas. The funnel plot is largely symmetrical (supplement 8B).

3.2.3. Results on smoking as a risk factor for cervical cancer

Sixteen studies could be pooled to examine the association between smoking and cervical cancer (Figure 3). We found a significant positive association with an OR of 2.65 (95% CI = 1.81–3.88). Heterogeneity was substantial (I² = 88%). No pooling was possible for the studies by Gutman et al. [40] and Currin et al. [51] (because no OR was reported) and Zhang et al. [49] (because it was unsure whether ‘hardly smoking’ was the same as no smoking). Gutman et al. and Currin et al. showed positive associations between smoking and cervical cancer. Zhang et al. found no association. When only including studies with a low risk of bias (nine studies), we still found a significant association between smoking and cervical cancer with an OR of 3.05 (95% CI = 1.73–5.38).

Figure 3. Association between smoking and cervical cancer (16 studies)

3.3. Smoking as a risk factor for CIN and cervical cancer (combined)

3.3.1. Study characteristics

Fourteen studies combined CIN and cervical cancer as outcome measure. Nine of these studies examined the relationship between smoking and CIN2+ (i.e. CIN2, CIN3, and cervical cancer) and five studies examined the relationship between smoking and the development of CIN3+ (i.e. CIN3 and cervical cancer). Three studies were cross-sectional studies with between 291 and 1,315 respondents [52–54]. Six studies were case–control studies with between 200 and 2,736 respondents [55–60]. Four studies were cohort studies with between 1,236 and 12,076 respondents [61–64] and one study was a longitudinal study with 1,628 respondents [65].

3.3.2. Risk of bias

For the 14 CIN2+ and CIN3+ studies, 10 showed a high risk of bias and 6 an unclear risk of bias on at least one of the examined areas (supplement 2). Only two studies showed a low risk of bias on all relevant examined areas. The funnel plot is largely symmetrical (supplement 8 C).

3.3.3. Results on smoking as a risk factor for CIN and cervical cancer (combined)

Of the 14 studies that combined CIN and cervical cancer as outcome measure, 1 study could not be included. Fang et al. [63] found positive associations between smoking and CIN3+ (but did not report an OR or the numbers per group). When combining all 45 studies about CIN and cervical cancer that could be pooled (Figure 4), we found a significant positive association between smoking and CIN or cervical cancer with an OR of 2.03 (95% CI = 1.72–2.39). Heterogeneity was substantial (I² = 81%). When examining the studies with different outcomes separately, we found an OR of 1.93 for CIN2+ (supplement 5), an OR of 2.10 for CIN3+ (supplement 6), and an OR of 2.48 for CIN3, cervical cancer, and CIN3+ combined (supplement 7). When only including studies with a low risk of bias (26 studies), we still found a significant association between smoking and CIN and cervical cancer (combined) with an OR of 2.26 (95% CI = 1.83–2.79).

Figure 4. Association between smoking and CIN2/3, CIN3, cervical cancer, CIN2+ or CIN3+ (45 studies)

3.4. Quality of evidence

The quality of evidence was assessed by using the GRADE approach. For both research questions, the certainty of evidence was judged to be ‘very low’. Reasons for this were the fact that all studies were observational studies, the high risk of bias due to inadequate adjustment for important prognostic factors, insufficient description of the CIN/carcinoma determination in some studies, insufficient description of the study population in some studies (especially studies about cervical cancer), the fact that CIN and cervical cancer were often combined as outcome measures, and that different types of control groups were used in the different studies.

4. Conclusion

The majority of studies show an increased risk of cervical abnormalities among people who smoke. Our meta-analysis confirms this positive correlation between smoking and development of CIN and cervical cancer. However, given the low certainty in the effect measures in the literature found, future studies should adjust for relevant predictors, separate CIN from cervical cancer as outcome measures, and report research methods in detail.

5. Expert opinion

In this systematic review and meta-analysis, we examined the literature published between 2009 and 2018 on smoking as a possible risk factor for the development of cervical abnormalities. The majority of studies show a significant positive association between smoking and CIN2/3 and CIN2+. For CIN3 and CIN3+, the literature shows a stronger increased risk in almost all studies. The relationship between smoking and cervical cancer is also confirmed by most of the included studies. This is consistent with previous literature reviews and meta-analyses [9–11]. It was not feasible for us to systematically review studies published before 2009, but an example of a case–control study published in 1998 found an increase in risk of cervical cancer with increased years of smoking and numbers of cigarettes smoked per day [66]. Another case–control study published in 2003 found that long duration smoking (20 years or more) was associated with a two-fold increase in the risk of squamous cell carcinoma but not with the risk of adenocarcinoma [67].

Our review focused on the relationship between smoking and the development of CIN and cervical cancer, but several studies also described dose–response relationships. Most of these studies found an increased risk of CIN(+) when the length of the period that women smoked, the number of pack-years, or the number of cigarettes per day increased [20,26,34,36,37,54,56,63,64]. Only one study reported a non-significant increased risk of CIN when the number of cigarettes per day increased [32]. Only one large cohort study examined a dose–response relationship for cervical cancer [37]. This study used data from 10 countries and found an increased risk of cervical cancer for women who smoke 20 years or longer and who smoked 10 cigarettes per day or more. It therefore seems that the more and longer that women smoke, the higher their risk of CIN and cervical cancer.

Based on the results of our review, one could hypothesize that quitting smoking may reduce the increased risk of CIN and cervical cancer for women who smoke. Some studies have examined this and indeed found this reduction for CIN(+) [20,36,61–63], while others found a non-significant trend toward a reduction [34,37,64], and some other studies did not find a reduction [21,65]. For cervical cancer, one study examined and found a reduction in the risk of cervical cancer for women who had quit smoking [37]. These mixed results preclude us from drawing definitive conclusions about the benefits of quitting smoking for the reduction in risk of CIN and cervical cancer for women who smoke. However, given the many adverse health effects of smoking, our opinion is that health professionals should always advise women who smoke to stop smoking, even without strong evidence of the impact on cervical abnormalities. Giving advice to stop smoking is not routine practice for all health professionals yet, mainly because it is felt to be too time-consuming and may jeopardize the patient-professional relationship [68,69]. A ‘Very Brief Advice’ may be a solution as this method is quick and non-confrontational [70].

It is important to note that some of the conducted studies included in our review had important limitations, resulting in a high risk of bias and a low confidence in the reported evidence. Many studies did not (adequately) adjust the analyses for important confounders like an HPV infection, hormonal contraception, parity, and sexual (risk) behaviors. Without adjusting the analyses for these confounders, we cannot be certain that smoking is an independent risk factor for the development of cervical abnormalities. Another important limitation of many studies was an insufficient description of both the determination of smoking status and CIN/carcinoma, which makes it unfeasible to assess the risk of bias due to measurement issues. Few studies focused on the relationship between smoking and CIN1 and, therefore, we cannot draw a conclusion about this relationship. A relatively large number of studies combined CIN and cervical cancer as outcome measures, making it impossible to describe results separately. Although the funnel plots appear largely symmetrical, our review may be affected by publication bias. Non-significant results of studies investigating the association between smoking and cervical cancer probably have a lower chance of being published than studies with significant results. Finally, many studies that examined cervical cancer were additionally limited because of ill-defined samples. The meta-analyses showed substantial heterogeneity. We think it is important that future studies adjust for relevant predictors, separate CIN from cervical cancer as outcome measures, and report research methods in detail. Besides the authors themselves, editors and reviewers can play an important role by asking authors to make these adjustments in their analyses and reports, before publication in peer-reviewed journals.

Article highlights

• Smoking tobacco seems to be a risk factor for development of cervical intra-epithelial neoplasia (CIN) and cervical cancer, but the exact role of smoking in the process of cervical carcinogenesis is not known.

• Previous reviews have looked at the relationship between smoking, CIN, and cervical cancer, but have not examined what the exact risk is, and have not looked at different stages of CIN.

• In this systematic review and meta-analysis, most studies showed an increased risk of CIN and cervical cancer among smokers.

• Many studies had a high risk of bias, e.g. not adjusting for relevant predictors or not separating CIN from cervical cancer.

• It can be – cautiously – concluded that smoking increases the risk of cervical abnormalities.

Declaration of interest

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Contributor statement

Gera E. Nagelhout: Conceptualization, Methodology, Investigation, Writing original draft, Funding acquisition. Renée M. F. Ebisch: Conceptualization, Investigation, Writing original draft. Olga L. van der Hel: Investigation, Writing review and editing. Gert-Jan Meerkerk: Investigation, Writing review and editing. Tessa Magnée: Investigation, Writing review and editing. Thomas M. de Bruijn: Formal analysis, Writing review and editing. Barbara van Straaten: Investigation, Writing review and editing, Project administration.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

We thank Gerdien B. de Jonge, Biomedical Information Specialist of Medical Library Erasmus MC, for her help with the literature search. We also thank Ammerins Moss-de Boer from Babylonia translations for translating our article from Dutch to English and Denise van den Broek for help with the tables.

Supplementary material

Supplemental data for this article can be accessed here.

Additional information

Funding

This work was supported by a grant from the Dutch National Health Care Institute (grant number 1044/001/2018).