Smoking Cessation and Risk of Esophageal Cancer by Histological Type: Systematic Review and Meta-analysis

Abstract

Background

Tobacco smoking strongly increases risk of esophageal squamous cell carcinoma and moderately increases risk of esophageal adenocarcinoma. How smoking cessation influences esophageal cancer risk across histological subtypes, time latencies, and geographic regions is not clear.

Methods

Studies were systematically searched on Medline, Embase, Web of Science, Cochrane Library, and ClinicalTrials.gov. Pooled estimates of risk ratios (RRs) were derived using a random effects model. Cochran’s Q test and I² statistic were used to detect heterogeneity.

Results

Among 15 009 studies, 52 fulfilled the inclusion criteria. Using nonsmokers as a reference, risk of esophageal squamous cell carcinoma was lower among former smokers (RR = 2.05, 95% confidence interval [CI] = 1.71 to 2.45) than among current smokers (RR = 4.18, 95% CI = 3.42 to 5.12). Compared with current smokers, a strong risk reduction was evident after five or more years (RR = 0.59, 95% CI = 0.47 to 0.75), and became stronger after 10 or more years (RR = 0.42, 95% CI = 0.34 to 0.51) and 20 or more years (RR = 0.34, 95% CI = 0.25 to 0.47) following smoking cessation. The risk reduction was strong in Western populations, while weak in Asian populations. Using nonsmokers as reference, the risk of esophageal adenocarcinoma was only slightly lower among former smokers (RR = 1.66, 95% CI = 1.48 to 1.85) than among current smokers (RR = 2.34, 95% CI = 2.04 to 2.69). The risk of esophageal adenocarcinoma did not show any clear reduction over time after smoking cessation, with a risk ratio of 0.72 (95% CI = 0.52 to 1.01) 20 or more years after smoking cessation, compared with current smokers.

Conclusions

Smoking cessation time-dependently decreases risk of esophageal squamous cell carcinoma, particularly in Western populations, while it has limited influence on the risk of esophageal adenocarcinoma.

Esophageal cancer is the ninth most common cancer and the sixth leading cause of cancer death globally (1). The overall prognosis is poor (<20% five-year survival) and has not improved much despite intensive research aiming to develop the treatment, which stresses the need for preventive actions. Esophageal cancer has two main histological subtypes, squamous cell carcinoma (ESCC) and adenocarcinoma (EAC). ESCC accounts for 80% of cases globally and is the dominant subtype in Asian countries (2). In many Western countries, however, the incidence of EAC has increased rapidly during the last four decades and now exceeds that of ESCC (3,4).

Tobacco smoking and heavy alcohol consumption are the main risk factors for ESCC, particularly in Western populations (5). Dietary factors, socioeconomic status, exposure to environmental carcinogens, and inherited susceptibility may play a stronger role in the etiology of ESCC in Asian populations compared with Western populations (6–8). The main risk factors for EAC are gastroesophageal reflux disease and obesity, while tobacco smoking is only a moderately strong risk factor (9–11). However, these studies are based on Western populations. In Asia, on the other hand, the incidence rate of EAC remains low and the etiology of EAC has rarely been studied (12).There is a dose-response association between smoking and both subtypes of esophageal cancer, and 49% of ESCC cases are estimated to be attributable to smoking (13–15). Although one literature review and one pooled analysis of 12 studies indicated a decreased risk of esophageal cancer following tobacco smoking cessation (14,16), it is not clear how smoking cessation influences the risk of esophageal cancer across histological subtypes, time latencies between cessation and risk reductions, and geographic areas globally. Yet, such knowledge should be of great importance for public health and health care. Therefore, we conducted a systematic review and meta-analysis that, to the best of our knowledge, is the first study aiming to clarify the role of smoking cessation in relation to the risk of esophageal cancer with separate assessments of the main histological subtypes, time latencies, and geographic regions.

Methods

Search Strategy and Selection Criteria

This systematic literature review and meta-analysis was performed in accordance with the PRISMA statement and MOOSE guidelines (17,18). The search strategy was discussed, and a final search string was agreed upon by all authors. A systematic search was conducted on MEDLINE, Embase, Web of Science, and Cochrane Library databases (up to March 30, 2016) for studies reporting data on the association between tobacco smoking cessation and the risk of esophageal cancer. For the search, we used a combination of three themes of Medical Subject Headings terms and related extended versions: “smoking or tobacco,” “esophageal or oesophageal,” and “cancer, squamous cell carcinoma or adenocarcinoma.” No restrictions in the search strategy were used. Reports on ongoing registered clinical trials from the National Institutes of Health website (http://www.clinicaltrials.gov) were also considered. In addition, we reviewed the reference lists of original studies, review articles, systematic reports, and the two monographs on “Smokeless tobacco and some tobacco-specific N-nitrosamines” and “Tobacco smoking and involuntary smoking” by the International Agency for Research on Cancer (IARC) to identify further studies of potential interest (19,20). The search strategy is presented in more detail in the Supplementary Methods (available online).

Studies fulfilling the following criteria were considered for inclusion in the systematic review: 1) smoking status was ascertained and data were presented as odds ratios (ORs), risk ratios (RRs), hazard ratios (HRs), or in another format from which the relative risk could be estimated; 2) case-control studies, cohort studies, intervention studies, and clinical trials; and 3) original and independent studies with full text. Language restriction was implemented only at the end of the search when only studies published in English were eligible. In the case of multiple reports on the same study population, only the most recent or most informative report with the longest follow-up was considered. We followed a detailed study protocol that was completed before initiation of the search for eligible studies.

One reviewer (Q. L. Wang) conducted the initial search and removed obviously irrelevant articles by screening the titles and abstracts according to the selection criteria. The final decision of articles selected for the review was made by all authors. We contacted the investigators for relevant data if their studies were potentially eligible for this study.

Data Extraction and Quality Assessment

Identified studies were independently assessed by two authors (Q. L. Wang and W. T. Li), and any discrepancies were resolved by joint review of reports to reach consensus, or determined by a third author (S. H. Xie). The following information was collected from the eligible studies into an electronic database: author names, year of publication, geographic origin, number of participants, number and type of case patients (incident or prevalent), participants’ characteristics (ethnic origin, mean age, and sex), histological subtype of esophageal cancer (squamous cell carcinoma or adenocarcinoma), method of ascertainment of case patients, smoking status (non-, former, or current, and years since smoking cessation), control for potential confounding factors (by matching or statistical analysis), and statistical analysis. For case-control studies, we collected information on participation rates and how control subjects were recruited. For cohort studies and clinical trials, information of representativeness of the study participants and the completeness, period, and duration of follow-up were recorded.

The methodological quality of the studies was assessed in terms of selection bias, information bias, and bias from confounding. We quantitatively scored the study quality according to the nine-item Newcastle-Ottawa Scale (21), which includes assessment of the generalizability of the study population, selection of control subjects or nonexposed cohort members, exposure, definition of control subjects or participants at the start of the cohort, adjustment for relevant confounders, outcome, response rate or completeness of follow-up, and rate of loss to follow-up or dropouts. An additional item was added to the scale: if smoking was investigated as the main exposure (1 point) or as a confounding variable only (0 points) (22). The methodological quality assessment could provide a score from 0 to 10 on the final scale, where higher scores represent better quality.

Statistical Analysis

Tobacco smoking status (current, former, or nonsmoker) and time latencies of smoking cessation were analyzed in relation to the risk of ESCC and EAC separately. In most included studies, current smokers were defined as those who smoked at the time of recruitment into the study or those who stopped smoking less than one or two years before recruitment. Former smokers were defined as those who quit smoking one or two years before inclusion, although two studies used five years as the boundary for current and former smokers (23,24). In the analyses of latency time after smoking cessation, some studies used current smokers as the reference group, while other studies used nonsmokers. Therefore, in the meta-analysis, the reference groups were all uniformed to current smokers for an easier understanding of smoking cessation, using the method suggested by Hamling et al (25). We categorized smoking cessation latency into five groups: nonsmokers, less than five years, five to nine years, 10 to 20 years, or more than 20 years. Some studies reported more than one category of duration for the first five years or over 20 years after smoking cessation, for example, multiple categories of less than two years and three to five years for the first five years, or 20 to 29 years and more than 30 years for over 20 years (26–29). In such cases, we combined these categories into a single one, that is, less than five years or more than 20 years, according to the meta-analysis approach, by pooling the estimates for more than two categorizes into one estimate of the risk ratio (30).

Risk ratio was used as the measure of association in the meta-analysis. For some studies, hazard ratio and odds ratio were used as proxies of risk ratio, which was justified by the low incidence of esophageal cancer (30). To take heterogeneity into account, we used a random effects model (Der-Simonian and Laird’s method) to compute the pooled risk ratios, and we also calculated their 95% confidence intervals (CIs) (31).

Statistical heterogeneity across studies was assessed by Cochran’s Q test (a P value < .10 being considered statistically significant for conservativeness of the test) (32) and the I² statistic, which describes the proportion of the total variation in study estimates that is due to heterogeneity rather than chance (33). An I² value of less than 25% indicated low heterogeneity, 25% to 50% indicated moderate heterogeneity, and more than 50% is suggestive of high heterogeneity (34). We conducted stratified analyses by study design (case-control or cohort study), publication year (≤1999, 2000–2009, or ≥ 2010), geographic origin of the study (North America, Europe, Oceania, Asia, or South America), gender (men, women, or unspecified), response rate (≥80%, <80%, or unknown), smoking exposure (main exposure or confounder), tobacco type (cigarettes or unspecified), study quality (low with score <7 or high with score ≥7), and potential confounding factors adjusted for alcohol use, dietary factors, socio-economy, place of residence, body mass index, and gastroesophageal reflux. For case-control studies, we stratified analyses by study design (population-based or hospital-based), case recruitment (incident, prevalent, or unknown), and source of control subjects (neighborhood or unrelated). For cohort studies, we stratified for study design (population-based or hospital-based), follow-up time (<10 years or ≥ 10 years), and assessment of outcomes (record linkage or self-reported).

Publication bias was evaluated using Begg’s and Egger’s tests, as well as visual inspection of the funnel plots (35,36). In addition, exploratory meta-regression was performed to examine potential sources of heterogeneity using the same covariates as in the stratified analysis, where P values of less than .10 were regarded statistically significant. We used sensitivity analyses by removing one study at a time to examine the robustness of the pooled risk ratios. The statistical analyses were conducted using the Comprehensive Meta-Analysis program, version 3.3 (Biostat, Englewood, NJ, USA). All statistical tests were two-sided.

Results

Literature Search and Study Characteristics

The search identified 15 009 studies. Among these, 52 studies fulfilled the inclusion criteria and were enrolled to this meta-analysis (Figure 1) (9,23,24,26–29,37–81). Of the 52 studies, 41 and 23 studies contained smoking data in relation to risk of ESCC and EAC, respectively. Most studies were case-control studies (n = 44), including a total of 11 965 esophageal cancer cases and 47 817 control subjects, and the remaining (n = 8) were cohort studies, with 1185 new esophageal cancer cases among 1 045 947 cohort members. No randomized clinical trials met the inclusion criteria. Most studies were conducted in Europe (n = 22), the United States (n = 10), and Mainland China or Taiwan (n = 7), while the remaining studies were conducted in Japan (n = 3), Brazil (n = 3), Uruguay (n = 3), Argentina (n = 1), Australia (n = 1), Czech Republic (n = 1), and Serbia (n = 1). Two studies from the United Kingdom were performed on the same study population, but they analyzed the two eligible histological types of esophageal cancer in separate studies (45,49). Two studies from Uruguay had partly overlapping study periods (52,53). Some overlap of research centers was possible in three studies from Italy or Switzerland (44,47,48), and two studies from Taiwan were partly overlapping (57,75). Supplementary Tables 1 and 2 (available online) provide an overview of the characteristics of the included case-control studies and cohort studies, respectively.

Figure 1.

Flow chart of study selection. EAC = esophageal adenocarcinoma; ESCC = esophageal squamous cell carcinoma.

Quality Assessment

A detailed study quality assessment is shown in Supplementary Tables 3 to 5 (available online). In brief, of all 41 studies examining ESCC, 22 (53.7%) had a high quality score (≥7) and 19 (46.3%) had a lower quality score (<7). Of all 23 studies analyzing EAC, 17 (73.9%) had a high quality score (≥7) and six (26.1%) had a lower quality score (<7). Thirty (57.7%) of all 52 studies reported different categories of years after smoking cessation, including 18 analyzing ESCC and 12 analyzing EAC. Among these, one study analyzing EAC did not report more categories than less than 26 years of smoking cessation, and was therefore not included in the further analysis (23). Detailed results by duration since smoking cessation in the original studies are presented in Supplementary Table 6 (available online). Eighteen (34.6%) studies reported sex-specific associations. All but 10 (19.2%) studies examined tobacco smoking as the main exposure. Among the 44 case-control studies, 16 were population-based and 31 analyzed incident cancer cases. Among the cohort studies, all but one identified case patients via record linkages, and the longest follow-up was 22.2 years. Adjustment for age and sex was made in all studies except for one (76). Adjustments for other potential confounding factors in the included studies are shown in Supplementary Table 3 (available online).

Smoking Cessation and Esophageal Squamous Cell Carcinoma

In an analysis of the 41 studies assessing ESCC, former smokers had a risk ratio of 2.05 (95% CI = 1.71 to 2.45) (Figure 2), and current smokers had a risk ratio of 4.18 (95% CI = 3.42 to 5.12) (Supplementary Figure 1, available online), compared with nonsmokers. There was a dose-response association between smoking cessation latency time and risk of ESCC (Figure 3A). Compared with current smokers, those who had quit smoking less than five years ago had a risk ratio of 0.96 (95% CI = 0.73 to 1.25), and those who had quit smoking five to nine years ago, 10 to 20 years ago, and more than 20 years ago had risk ratios of 0.59 (95% CI = 0.47 to 0.75), 0.42 (95% CI = 0.34 to 0.51), and 0.34 (95% CI = 0.25 to 0.47), respectively (Figure 3A;Supplementary Figure 2, available online). The risk ratio for those who quit smoking more than 20 years ago was similar to that of nonsmokers, at 0.22 (95% CI = 0.18 to 0.28). In a sensitivity analysis restricted to studies that reported risk ratios of all smoking cessation latency categories, the results were similar to those of the overall analyses (Supplementary Table 7, available online). The meta-analysis revealed substantial heterogeneity across studies for risk ratios in former smokers (I² = 69.6%, P < .001) and current smokers (I² = 85.0%, P < .001).

Figure 2.

Forest plot of risk ratios of esophageal squamous cell carcinoma among former smokers with nonsmokers as reference, stratified by study design. The diamonds represent the effect sizes for the studies combined, the squares represent the effect sizes of individual studies and the weights given to the studies, and the error bars represent the corresponding 95% confidence intervals. CI = confidence interval.

Figure 3.

Risk ratio of esophageal squamous cell carcinoma and adenocarcinoma by duration since smoking cessation, using current smokers as reference. A) Effect sizes for esophageal squamous cell carcinoma. B) Effect sizes for esophageal adenocarcinoma. Error bars = 95% confidence interval.

The results from the stratified analyses are shown in Table 1. All risk ratios in former smokers were lower than risk ratios in current smokers in each stratum for ESCC, although heterogeneity (I² > 50%) was found in most strata. The difference in risk ratio between former smokers and current smokers was most pronounced in studies from North America and Europe, while it was not evident in Asian studies. Among former smokers, women had lower risk ratios of ESCC than men. High-quality studies (score ≥ 7) generated higher risk ratios of ESCC among current smokers and lower risk ratios among former smokers, compared with low-quality studies (score < 7). The results remained stable after adjustment for alcohol use and dietary factors. In hospital-based case-control studies, the risk ratios among both former and current smokers were slightly higher compared with population-based studies (Table 1).

Table 1.

Tobacco smoking status and risk of esophageal squamous cell carcinoma, using nonsmokers as the reference

Study characteristics	Former smokers				Current smokers
	RR (95% CI)	Studies, No.	Pheterogeneity*	I2(%)†	RR (95% CI)	Studies, No.	Pheterogeneity*	I2, %†
Overall	2.05 (1.71 to 2.45)	41	<.001	69.6	4.18 (3.42 to 5.12)	41	<.001	85.0
Study design
Case-control	2.01 (1.67 to 2.43)	37	<.001	70.4	3.81 (3.06 to 4.74)	37	<.001	85.6
Cohort	2.50 (1.29 to 4.85)	4	.03	66.2	6.95 (4.17 to 11.57)	4	.10	52.9
Publication year
≤1999	1.98 (1.54 to 2.54)	8	.26	21.7	5.07 (3.35 to 7.68)	8	<.001	75.3
2000–2009	1.85 (1.47 to 2.34)	21	<.001	67.8	3.62 (2.83 to 4.64)	21	<.001	79.5
≥2010	2.57 (1.69 to 3.91)	12	<.001	78.7	4.29 (2.54 to 7.24)	12	<.001	91.8
Geographic origin
North America	2.45 (1.83 to 3.27)	7	.31	16.3	5.75 (3.56 to 9.26)	7	.002	70.7
Europe	1.75 (1.15 to 2.65)	14	<.001	79.3	4.57 (3.19 to 6.54)	14	<.001	81.5
Oceania	2.18 (1.51 to 3.17)	1	1.00	0.0	4.58 (2.99 to 7.02)	1	1.00	0.0
Asia	2.47 (1.78 to 3.44)	12	<.001	72.6	2.82 (1.81 to 4.39)	12	<.001	91.5
South America	1.67 (1.37 to 2.04)	7	.67	0.0	2.91 (2.41 to 3.50)	7	.47	0.0
Sex‡
Men	2.00 (1.43 to 2.80)	10	.01	57.2	3.77 (2.29 to 6.20)	10	<.001	85.7
Women	1.34 (0.71 to 2.53)	6	.006	69.1	3.85 (2.20 to 6.74)	6	.003	72.2
Unspecified	2.26 (1.86 to 2.76)	29	<.001	66.9	3.94 (3.12 to 4.99)	29	<.001	84.0
Response rate, %
≥80	1.92 (1.56 to 2.36)	15	.05	40.6	4.21 (2.74 to 6.47)	15	<.001	90.4
<80	2.42 (1.70 to 3.45)	10	<.001	70.1	4.80 (3.09 to 7.43)	10	<.001	83.2
Unknown	1.95 (1.36 to 2.78)	16	<.001	79.3	3.51 (2.69 to 4.60)	16	<.001	76.9
Smoking exposure
Main exposure	1.88 (1.57 to 2.24)	33	<.001	60.6	3.97 (3.17 to 4.96)	33	<.001	81.3
Confounder	3.08 (1.90 to 4.98)	8	<.001	81.2	4.35 (2.44 to 7.76)	8	<.001	92.7
Tobacco types
Cigarettes	2.38 (1.58 to 3.60)	15	<.001	82.8	4.02 (3.07 to 5.28)	15	<.001	72.7
Unspecified	1.94 (1.66 to 2.26)	26	.01	41.9	3.96 (2.94 to 5.33)	26	<.001	88.4
Study quality
Low (score < 7)	2.16 (1.63 to 2.86)	19	<.001	74.8	3.60 (2.65 to 4.90)	19	<.001	87.0
High (score ≥ 7)	1.97 (1.57 to 2.48)	22	<.001	64.7	4.45 (3.31 to 5.99)	22	<.001	83.5
Adjusted variables
Alcohol use
Yes	2.09 (1.72 to 2.54)	22	.02	42.3	4.59 (3.67 to 5.74)	19	<.001	60.1
No	2.05 (1.52 to 2.76)	19	<.001	81.1	3.54 (2.56 to 4.88)	21	<.001	90.6
Dietary factors
Yes	1.76 (1.20 to 2.58)	12	<.001	71.0	4.42 (3.11 to 6.28)	11	.001	67.6
No	2.17 (1.78 to 2.66)	29	<.001	69.4	3.91 (3.04 to 5.03)	30	<.001	87.6
Socio-economy
Yes	2.07 (1.73 to 2.49)	18	.11	30.0	4.19 (3.20 to 5.48)	16	<.001	70.9
No	2.04 (1.54 to 2.70)	23	<.001	79.4	3.93 (2.93 to 5.27)	25	<.001	88.8
Place of residence
Yes	2.02 (1.67 to 2.45)	20	.01	46.5	3.78 (2.85 to 5.01)	19	<.001	79.3
No	2.09 (1.55 to 2.82)	21	<.001	78.9	4.26 (3.14 to 5.79)	22	<.001	88.1
Case-control study
Study design
Population-based	1.66 (1.19 to 2.30)	11	<.001	71.3	3.23 (1.98 to 5.29)	11	<.001	90.4
Hospital-based	2.21 (1.77 to 2.75)	26	<.001	68.0	4.10 (3.26 to 5.15)	26	<.001	80.6
Case recruitment
Incident	2.20 (1.78 to 2.73)	28	<.001	73.4	4.21 (3.28 to 5.40)	28	<.001	86.1
Prevalent	1.49 (1.01 to 2.19)	7	.08	47.3	2.90 (1.77 to 4.75)	7	<.001	83.1
Unknown	1.32 (0.73 to 2.38)	2	.60	0.0	2.11 (0.58 to 7.61)	2	.04	77.4
Source of controls
Unrelated	2.07 (1.71 to 2.51)	35	<.001	70.1	4.05 (3.28 to 5.01)	35	<.001	82.9
Neighborhood	1.36 (0.98 to 1.89)	2	.95	0.0	1.30 (0.96 to 1.77)	2	.30	5.3
Cohort study
Study design
Population-based	3.46 (2.26 to 5.31)	3	.79	0.0	6.98 (3.19 to 15.28)	3	.06	65.6
Occupation-based	0.90 (0.40 to 2.00)	1	1.00	0.0	7.60 (4.50 to 12.70)	1	1.00	0.0
Follow-up time, y
<10	4.35 (1.95 to 9.72)	1	1.00	0.0	9.27 (4.04 to 21.29)	1	1.00	0.0
≥10	2.08 (0.92 to 4.70)	3	.03	70.4	6.51 (3.42 to 12.42)	3	.06	64.0
Outcome assessment
Record linkage	2.37 (0.98 to 5.77)	3	.01	77.2	6.16 (3.56 to 10.66)	3	.10	56.2
Self-reported	2.93 (1.18 to 7.32)	1	1.00	0.0	12.67 (4.55 to 35.28)	1	1.00	0.0

*

P values from two-sided Cochran’s Q test. CI = confidence interval; RR = risk ratio.

†

I² statistics indicating the percentage of variation across studies that is due to heterogeneity.

‡

Two studies (37,38) reported risk ratio for men and women separately and combined; one study (39) reported risk ratio for men and women separately.

The meta-regression showed that sex and study quality could explain 11.0% (P = .08) and 13.6% (P = .06) of the heterogeneity in former smokers, respectively (data not shown). The continent where the study was conducted and source of controls (neighborhood-based or unrelated) could explain 9.1% (P = .02) and 19.1% (P = .01) of the heterogeneity in current smokers, respectively. The sensitivity analyses excluding one study at a time showed no substantial changes (Supplementary Figure 3 and 4, available online).

Smoking Cessation and Esophageal Adenocarcinoma

In an analysis of all 23 studies, former smokers had a risk ratio of 1.66 (95% CI = 1.48 to 1.85) (Figure 4) and current smokers had a risk ratio of 2.34 (95% CI = 2.04 to 2.69) (Supplementary Figure 5, available online), compared with nonsmokers. No substantial heterogeneity of the risk ratios was revealed for former smokers (I² = 11.6%, P = .30) or current smokers (I² = 26.0%, P = .13). Studies from North America (n = 10) showed the highest risk ratio among former smokers (RR = 1.87, 95% CI = 1.62 to 2.16) and current smokers (RR = 2.52, 95% CI = 2.14 to 2.96), compared with studies from other continents (Supplementary Table 8, available online).

Figure 4.

Forest plot of risk ratio of esophageal adenocarcinoma among former smokers with nonsmokers as reference, stratified by study design. The diamonds represent the effect sizes for the studies combined, the squares represent the effect sizes of individual studies and the weights given to the studies, and the error bars represent the corresponding 95% confidence intervals. CI = confidence interval.

Compared with current smokers, smoking cessation less than five years ago was associated with a risk ratio of 0.81 (95% CI = 0.52 to 1.26), five to nine years with a risk ratio of 0.87 (95% CI = 0.58 to 1.30), 10 to 20 years with a risk ratio of 0.95 (95% CI = 0.78 to 1.15), and more than 20 years ago with a risk ratio of 0.72 (95% CI = 0.52 to 1.01) (Figure 3B;Supplementary Figure 6, available online).

Publication bias

No publication bias was detected by visual inspection of the funnel plot or by the Begg’s and Egger’s test (Supplementary Figure 7, available online). For ESCC, P values for former smokers using Begg’s and Egger’s test were .17 and .19, respectively. The corresponding P values for current smokers were .09 and .20, respectively. Similarly, no publication bias was found for EAC. The P values for former smokers using Begg’s and Egger’s test were .71 and .63, respectively. The corresponding P values for current smokers were .81 and .33, respectively.

Discussion

This study indicates a strongly decreased risk of ESCC in former smokers compared with current smokers and a clear decline in risk of ESCC already within five years of smoking cessation, which further decreased with each longer latency period of smoking cessation until after 20 years, when the risk was similar to that of nonsmokers. North American populations seem to benefit most from smoking cessation, while Asian populations benefitted the least. There was only a small difference in the risk of EAC comparing former and current smokers, and a slightly decreased risk of EAC was suggested only among those who had stopped smoking more than 20 years ago. The observed associations of the risk of ESCC or EAC in current or former smokers persisted across subgroups stratified by participants’ characteristics and study design.

This meta-analysis has several main strengths, including the extensive search strategy, which should have identified all relevant publications globally. It includes a large number of studies and participants, which provides good statistical power for robust subgroup analyses, including separate analyses of the main histologic types of esophageal cancer and various lengths of smoking cessation periods. There are also limitations; heterogeneity was found across studies investigating the risk of ESCC for former and current smokers. This might have resulted from the large number of included studies, differences in design, population, and quality of the studies, and differences in participants’ characteristics. To reduce the influence of heterogeneity, a random effects model was used. All stratified analyses showed decreased risk ratios among former smokers compared with current smokers. In analyses restricted to higher-quality studies, the decrease in risk ratio was greater between current and former smokers, indicating the robustness of the findings. The restriction of studies to those published in the English language with full text might have resulted in the exclusion of some small or low-quality studies. However, the results were unlikely affected by any such exclusion because no publication bias was detected. Finally, biases of observational studies cannot be avoided, but the results from cohort studies revealed similar results as those from case-control studies, which indicate robustness. Moreover, the risk reductions seen after smoking cessation are biologically plausible.

To the best of our knowledge, this is the first systematic review and meta-analysis estimating the influence of smoking cessation on the risk of esophageal cancer by histological type. Yet, the decreased risk of ESCC among former smokers compared with current smokers is consistent with the results of a meta-analysis of 15 Japanese studies, showing a risk ratio of esophageal cancer of 3.73 (95% CI = 2.16 to 6.43) in current smokers and 2.21 (95% CI = 1.60 to 3.06) in former smokers (82). Although the histologic type of cancer was not provided in that study, the vast majority of patients with esophageal cancer in Japan have ESCC. Regarding EAC, our results are similar to a meta-analysis that included studies published before January 2010, implying a risk ratio of esophageal and gastric cardia adenocarcinoma combined of 2.32 (95% CI = 1.96 to 2.75) in current smokers and 1.62 (95% CI = 1.40 to 1.87) in former smokers (83). However, due to the limited statistical power, separate analysis of EAC (excluding cardia cancer) was not possible in that study (83). A pooled analysis of 12 studies suggested benefits of smoking cessation for EAC after 10 years (odds ratio = 0.71, 95% CI = 0.56 to 0.89) (14). However, the risk of EAC following smoking cessation of 10 to 20 years or more than 20 years was not assessed in that study (14). The present meta-analysis showed that any benefit from smoking cessation became evident only more than 20 years after smoking cessation, which is in agreement with opinions from two literature review articles (16,84).

It is interesting to note that smoking cessation seems to have a stronger influence on ESSC risk in Western populations than in Asian populations. The incidence rate of ESCC is highest in Asian countries globally, with an estimated 80% of all global ESCC cases occurring in Asia, and China alone contributed to more than half of these cases (85). Despite the much higher prevalence of tobacco smoking in men than women in Asian populations (eg, 25.5 times higher in men than women in China), the sex difference in the incidence rate of ESCC is less marked (eg, 2.8 times higher in men than women in China) (86,87). Thus, the high risk of ESCC in Asian populations is likely to be attributable to other risk factors, for example, dietary factors (including hot food and beverage, red and processed meat, low vegetable and fruit intake, etc.) tobacco smoke pollution, household air pollution and other sources of polycyclic aromatic hydrocarbons, and genetic factors (88–94). The high baseline ESCC risk posed by risk factors other than tobacco might have neutralized the risk reduction related to smoking cessation in Asian populations. Furthermore, a large proportion of Asian studies did not adjust for confounders, for example, alcohol consumption, which might have led to an overestimation of the risk of ESCC in former smokers in the study.

In conclusion, this comprehensive systematic review and meta-analysis of 52 studies from different regions globally suggests that smoking cessation is associated with a rapid and strong reduction in the risk of ESCC. The benefits of smoking cessation on ESCC were stronger in Western populations than in Asian populations. Any reduction of EAC risk following smoking cessation is limited and slow. The preventive effects of smoking cessation on esophageal cancer shown in this study can help guide future health policy and clinical practice.

Funding

This work was supported by the Swedish Research Council (grant number 521-2014-2536) and the Swedish Cancer Society (grant number CAN 2015/460).

Notes

The study sponsors had no role in the study design, the data collection, analysis, or interpretation, the writing of the report, or the decision to submit the manuscript for publication.

All authors have no conflicts of interest to declare.