Lifestyle interventions can reduce the risk of Barrett’s esophagus: a systematic review and meta‐analysis of 62 studies involving 250,157 participants

Abstract Background Barrett's esophagus (BE) is a well‐established risk factor for esophageal adenocarcinoma. Our objective was to investigate the effectiveness of lifestyle interventions on BE risk. Methods We searched PubMed, Embase, and Web of Science up to 30 September 2020. The summary relative risks (RRs) and 95% confidence intervals (CIs) for the highest versus lowest categories of exposure were assessed. Analyses of subgroup, dose–response, sensitivity, and publication bias were conducted. Results Sixty‐two studies were included that involved more than 250,157 participants and 22,608 cases. Seven lifestyle factors were investigated: smoking, alcohol, body mass index (BMI), physical activity, sleep time, medication, and diet. We observed statistically significant increased BE risks for smoking (RR = 1.35, 95% CI = 1.16–1.57), alcohol intake (RR = 1.23, 95% CI = 1.13–1.34), body fatness (RR = 1.08, 95% CI = 1.03–1.13), less sleep time (RR = 1.76, 95% CI = 1.24–2.49), and proton pump inhibitors use (RR = 1.64, 95% CI = 1.17–2.29). Reduced risks of BE were found for aspirin (RR = 0.70, 95% CI = 0.58–0.84) and the intake of vitamin C (RR = 0.59, 95% CI = 0.44–0.80), folate (RR = 0.47, 95% CI = 0.31–0.71), and fiber (RR = 0.95, 95% CI = 0.93–0.97). The quality of most included studies was high and the subgroup analysis according to the quality score showed significant results (p < 0.05). There was no publication bias for smoking and alcohol. Although the analysis suggested significant evidence of publication bias for BMI, sensitivity analysis showed that the changes in the recalculated RRs were not significant. Conclusions The large meta‐analysis revealed that lifestyle modifications could reduce the risks of BE and, consequently, esophageal adenocarcinoma.


| INTRODUCTION
Esophageal cancer is a highly lethal cancer with 572,000 new cases and 509,000 deaths occurring worldwide in 2018. 1 Considering the increasing trend in the incidence of and the high fatality of esophageal cancer, finding novel strategies to prevent the development of this cancer is an urgent need. Barrett's esophagus (BE) is considered a well-established risk factor 2 and the only known precursor for esophageal adenocarcinoma. 3 Esophageal adenocarcinoma is estimated to be at least 10 times as high among patients with BE as it was in the general population. 4 Recently, an increasing number of studies have focused on lifestyle and modifiable risk factors for BE. [5][6][7][8][9] However, the associations between several factors and BE risk are inconsistent, including alcohol, 10,11 BMI, 12,13 and nonsteroidal anti-inflammatory drug (NSAID) use. 14,15 Although smoking has been systematically evaluated, 16 the population included was limited to males. Furthermore, the included studies of that meta-analysis were published up to 2011. Numerous high-quality studies have appeared during the last approximately 5 years, and an updated meta-analysis may clarify the impact of the recent studies. To the best of our knowledge, no other main lifestyle factors have been assessed systematically with respect to BE risk.
Thus, given the large burden of esophageal adenocarcinoma worldwide and the controversial evidence of BE, we conducted a large systematic review and meta-analysis with the following objectives: (1) to provide an update based on more-sufficient evidence and a quantitative synthesis of the eligible data on the associations between lifestyle factors and BE risk; (2) to conduct dose-response analyses to further evaluate potential dose-response associations, where possible; and (3) to perform subgroup analyses to further explore the associations by study design, geographic area, publication year, sample size, quality score and adjustments, including smoking, alcohol, BMI, and reflux, where possible.

| Selection criteria
The selection criteria were independently judged by two reviewers (ZZ and ZY), which were as follows: we selected the highest quality studies, the largest samples, and the most recent studies for the studies reporting the similar data; given the varied diagnostic criteria of BE, 17 both of the American College of Gastroenterology clinical guidelines 18 and the British Society of Gastroenterology guidelines 19 on the diagnosis of BE met the eligibility criteria; narrative, systematic reviews, and meta-analysis were excluded if they did not include original data; editorials, letters, comments, case reports, and conference abstracts were excluded; studies in which only the abstract could be obtained were excluded; esophagitis, esophageal cancer, gastrointestinal stromal tumors, and other tumors of the esophagus were excluded; study populations of other comorbidities (e.g., inflammatory bowel disease, adenomas, polyps, and diverticulitis) were excluded; the language of included studies was limited to English; and studies were limited to those involving humans.

| Study quality and data extraction
The study quality of cohort studies and case-control studies was assessed using the Newcastle-Ottawa Scale (NOS). 20 The NOS range is 0-9 stars, and a high-quality study includes 7 or more stars. The NOS is judged on three factors, including the elucidation of the exposure or outcomes of interest, the selection of the study populations, and the comparability of the populations. An 11-item checklist recommended by the Agency for Healthcare Research and Quality (AHRQ) was used to assess the methodological quality of cross-sectional studies. The range of AHRQ is 0-11 scores. Low quality is 0-3, moderate quality is 4-7, and a high-quality study ranges from 8 to 11. Two reviewers (ZZ and ZY) independently assessed the study quality, and discrepancies in interpretation were resolved by a consensus decision made by the third reviewer (CZ).
A data extraction sheet was generated for each study. Detailed information included the first author, publication year, country, study type, study period, study population, assessment method, type of exposure measured, exposure categories, adjusted RR (95% CI), adjusted variables, and quality score. SPSS 22.0 (Chicago, Illinois, USA) was used to collect and extract data. RevMan5.3 (The Cochrane Collaboration, Oxford, UK) software was used for the synthesis and analysis of data based on relative risks (RRs) and 95% confidence intervals (95% CIs).

| Statistical analysis
We conducted this meta-analysis for the risk of BE and smoking, BMI, physical activity, sleep duration, medications, and dietary factors. Medications included aspirin, NSAIDs, PPIs, and statins. Dietary factors included alcohol, vitamin C, folate, selenium, total meat, and white meat. Stratified analysis was not performed for alcohol. Beer, wine, and spirits were included. Fruits, vegetables, fat, red meat, and processed meat were excluded because we have previously analyzed these issues. 21 Vitamin D and calcium were excluded due to the limited studies. A random-effects model was used to pool the RRs and 95% CIs if there was heterogeneity among studies, and a fixed-effects model was used if there was no heterogeneity. The method described by Greenland et al 22 was used for the nonlinear dose-response analysis. Studies that reported at least three quantitative exposure categories for RRs with their corresponding 95% CIs were included for dose-response analysis.
Heterogeneity among studies was detected using I 2 statistics (I 2 < 50% was considered low heterogeneity, and I 2 > 50% was considered to indicate substantial heterogeneity) 23 and Q statistics (p < 0.1 was considered representative of significant heterogeneity). Gastroesophageal reflux disease (GERD) is a well-established risk factor for the development of BE. 24 The data of non-GERD patients as the control group were preferred for summary estimates to eliminate possible heterogeneity. Additionally, a subgroup analysis was conducted to further explore the sources of heterogeneity by study design, geographic area, publication year, sample size, quality score, and adjustments (smoking, alcohol, BMI, and reflux symptom), where possible.
Publication bias was assessed using funnel plots and Egger's test (p < 0.1 was considered significant publication bias). 25 Sensitivity analyses were conducted by removing one study at a time to investigate the influence of a specific study on the pooled risk estimate. Figure S1 shows the flowchart of the search strategy for selecting the eligible studies. In total, 5712 studies were initially identified; 5079 studies were excluded for duplication and 633 studies were selected for further consideration after excluding the duplicates deriving from individually combination of search terms. Of those, 529 studies were excluded after reviewing the titles and abstracts, and 56 studies were included after reviewing the full-text article. Finally, 62 studies met the eligibility criteria after including 6 studies from the reference review.

| Literature selection, study characteristics, and quality scores
The 62 selected studies were conducted in 16 countries or regions worldwide and involved more than 250,157 participants and 22,608 cases. These included studies provided 128 separate estimates to the associations of lifestyle factors and BE risk. More detailed information on these studies has been listed in Table 1.

| Current versus never
Thirty studies that involved 225,250 participants were included and a random-effects model yielded a positive association (RR = 1.35, 95% CI = 1.16-1.57) ( Figure 1). Additionally, the association was unchanged by the separate evaluations based on the study design, with 1.45 (1.14-1.83) for cohort studies and 1.25 (1.05-1.49) for case-control studies ( Figure 1, Table 2).

| Former versus never
Eleven studies met the criteria, and a significant increased BE risk was observed (RR = 1.37, 95% CI = 1.16-1.62) ( Figure S2A). The changes in the recalculated RRs were not significant, with a range from 1. 29

| Highest versus lowest category
Four studies were included in the analysis for the highest to lowest number of cigarettes/day ( Figure S2B), and a fixedeffects model yielded a significantly positive association (RR = 1.36, 95% CI = 1.02-1.81) without heterogeneity

| Dose-response analysis
We conducted a dose-response analysis to further explore the association between pack-years of smoking and BE risk. Six studies were included, and the results of 1.10 (1.05-1.14) indicated that BE risk increases by 10% for each 10-year increment. We further checked for nonlinearity of the doseresponse association, and the evidence suggested that the best-fitting model was a nonlinear model (P nonlinearity < 0.01, Figure S3A).

| Heterogeneity
There was significant heterogeneity (p < 0.01, I 2 = 71%), but subgroup analyses (Table 2) for highest versus lowest exposure suggested that the positive association was stable by all of the confounders (study design, geographic area, publication year, sample size, quality score, alcohol, BMI, and reflux symptom).

| Publication bias
A funnel plot, Begg's test, and Egger's test were used to explore the publication bias. Indeed, Egger's test indicated evidence of publication bias (p < 0.1), but the funnel plot provided a visible result of no publication bias observed in Figure S3A. Additionally, the changes in the recalculated RRs were not significant (Figure 2), with a range from 1.32 (1.14-1. 53

| Heterogeneity
There was no significant heterogeneity (p < 0.10, I 2 = 29%) and subgroup analyses also suggested that the positive association was stable by all of the confounders (Table 3).

| Publication bias
The funnel plot ( Figure S4B) and Egger's test (p = 0.34) suggested no significant evidence of publication bias. Additionally, the sensitivity analysis also showed that the changes in the recalculated RRs were not significant, with a range from 1.

| Highest versus lowest category
Twenty-two studies that involved 211,607 participants were included, and a random-effects model yielded a positive association (RR = 1.08, 95% CI = 1.03-1.13) (Figure 3). The association was unchanged by the separate evaluations based on study design (Table 4).

| Dose-response analysis
Twelve studies were included, and the results of 1.08 (1.07-1.10) indicated that the BE risk increases by 8% for each 5 kg/m 2 increase in BMI. We further checked for nonlinearity of the dose-response association, and the evidence suggested that the best-fitting model was a nonlinear model (P nonlinearity < 0.01, Figure S3B).

| Heterogeneity
There was significant heterogeneity (p < 0.01, I 2 = 63%), but subgroup analyses (Table 4) suggested that the positive association was stable by all of the confounders.

| Publication bias
Indeed, the funnel plot ( Figure S4C) and Egger's test (p < 0.1) suggested significant evidence of publication bias. Five studies investigated BE risk and aspects of waistto-hip ratio (WHR), waist circumference, hip circumference, waist-to-thigh ratio, and visceral adiposity. These investigations suggested that high WHR, 35,63,85,86 waist circumference, 35 waist-to-thigh ratio, 35 and visceral adiposity 87 are associated with the presence of BE, whereas hip circumference (gluteofemoral obesity) may decrease BE risk. 85

| Physical activity
Four studies were included and involved 25,491 participants. A fixed-effects model yielded a null F I G U R E 1 Forest plot of smoking (current vs. never) and Barrett's esophagus risk. The results demonstrated that smoking is associated with Barrett's esophagus risk association (RR = 1.06, 95% CI = 0.83-1.37) without heterogeneity (p = 0.77, I 2 = 0%) ( Figure 4A). Additionally, the changes in the recalculated RRs were not significant, with a range from 0.97 (0.69-1.36) to 1.14 (0.83-1.56).

| Aspirin
Six studies were included that involved 3,742 participants. A fixed-effects model yielded an inversed

| Dietary factors
Significant inverse associations were observed between BE risks and the highest versus lowest intakes of vitamin C (4 studies involving 2,241 participants, RR = 0.59, 95%  Figure 6C). No associations were detected between BE risks and the highest versus lowest intakes of total meat (4 studies involving 122,861 participants, RR = 0.86, 95% CI = 0.72-1.02, Figure 6D), white meat (2 studies involving 121,328 participants, RR = 0.93, 95% CI = 0.78-1.11, Figure 6E), and selenium (3 studies involving 2,009 participants, RR = 0.92, 95% CI = 0.68-1.25, Figure 6F). Our previous study 21 has investigated BE risk and the intake of fruits, vegetables, fat, red meat, and processed meat, and the results demonstrated that vegetable intake was significantly associated with a decreased risk of BE, and there were no associations between the intake of fruits, fat, red meat or processed meat and BE risk. Other studies investigated the associations between the intake of vitamin B6, 42 vitamin B12, 42 calcium, 40 tea, 88 and coffee. 88 Most of studies reported nonsignificant associations, but one study showed a decreased risk of BE with calcium intake. 40

| DISCUSSION
This large systematic analysis is the first to comprehensively explore the influence of lifestyle interventions on the risk of BE. Our analyses demonstrated a significantly increased BE risk associated with smoking, alcohol intake, high BMI, less sleep time, and PPI use. Inversed associations were observed T A B L E 3 Subgroup analyses of alcohol intake (highest vs. lowest category) and Barrett's esophagus risk with aspirin use, vitamin C intake, and dietary fiber intake. No associations were found for physical activity, NSAID use, folate, total meat, white meat, and selenium. Additionally, the results of detailed subgroup analyses and dose-response analyses were consistent with the original analyses. Our analyses for smoking revealed a statistically significant 35% increased risk of BE for the current versus never smoking. When former versus never smoking was further analyzed, this increased to a 37% increased risk. In addition, the positive associations were supported by the detailed subgroup analyses. Furthermore, the dose-response analysis indicated that the BE risk increases by 10% for each 10 packyear increment. Additionally, the analysis for the highest to lowest number of cigarettes/day and BE risk also showed a significantly positive association. Taken together, all the analyses suggested that smoking (including current, past, longer pack-years, and more number of cigarettes/day) may be an independent risk factor for BE development. Nevertheless, long-term smoking cessation may diminish this risk, 31 which suggested a feasible option for smoking cessation as a risk modification strategy. In addition, because smoking is also a well-established risk factor for esophageal cancer, 89,90 it would be beneficial to quit smoking whenever possible, to reduce the risks of BE and esophageal adenocarcinoma.
A significant 23% increased risk was observed for highest versus lowest alcohol intake and BE risk, which was supported by the results of detailed subgroup analyses. Moreover, the International Agency for Research on Cancer (IARC, http://monog raphs.iarc.fr/ENG/Class ifica tion/Class ifica tions Group Order.pdf) and the World Cancer Research Fund International (WCRF, http://wcrf.org/int/resea rch-wefund/conti nuous -updat e-proje ct-findi ngs-repor ts/summa rygloba l-evide nce-cancer) have classified alcohol as a Group 1 carcinogen for esophageal cancer. Thus, a decreased intake of F I G U R E 3 Forest plot of BMI (highest vs. lowest category) and Barrett's esophagus risk. The results demonstrated that high BMI is associated with Barrett's esophagus risk alcohol is advisable to reduce the risk of BE and esophageal adenocarcinoma.
Although Qumseya et al 91 conducted a meta-analysis to the association between obesity and BE, there were no detailed subgroup analyses, sensitivity analysis, and doseresponse analysis. In our pooled analyses, body fatness was indicated by the BMI. The analyses for highest versus lowest BMI demonstrated a statistically increased risk of BE, which was supported by the results of detailed subgroup analyses. The dose-response analysis indicated that the risk was 8% for increase of per 5 kg/m 2 . Our results revealed that BMI may be an independent, strong predictor of BE. Other measures, including WHR, waist circumference, waist-to-thigh ratio, and visceral adiposity were also associated with increased BE risk. It remains unclear how high body fatness increases BE risk. Abdominal obesity may increase the abdominal pressure, subsequently inducing relaxation of the lower esophageal sphincter, which results in an increased risk of gastroesophageal reflux disease and thus BE. 13,92 Additionally, the continuous update report published in 2016 of WCRF on esophageal cancer has judged the evidence for the role of body fatness to be "convincing" (http://wcrf.org/ int/resea rch-we-fund/conti nuous -updat e-proje ct-findi ngsrepor ts/oesop hagea l-cancer). Thus, keeping the weight as low as possible within the healthy range is helpful to reduce the risks of BE and esophageal adenocarcinoma.
Physical activity is often considered an inverse factor for esophageal cancer. 93 However, our analysis revealed no association between BE risk and the highest versus lowest category of physical activity. Given the limited included studies, more studies are necessary to further verify this association. Although Lam et al 94 95 reported that short sleep time may be correlated with the severity of GERD. Nevertheless, due to the limited studies and the unclear mechanism, further studies would be helpful to clarify this association. It is surprising that aspirin use was protective against BE with a 30% lower risk, but that nonaspirin NSAID use was not. The results were consistent with the results of the largest study 65 to date that addressed aspirin/NSAID effect on BE. The exact mechanisms of this difference are still unclear. It is possible that the cases (more likely to be obese and having GERD) may take nonaspirin NSAIDs as a substitute for aspirin due to milder on the stomach. Analyses for PPI use yielded a statistically significant 64% increased risk of BE. PPIs are used to eradicate H pylori infection in combination with antibiotics, and positive H pylori infection is associated with a reduced risk of BE. 96,97 In contrast, the positive association may result from the fact that, in routine care, more BE patients take antacid medications such as PPIs to alleviate GERD symptoms, compared with controls. 8, 72 We did not obtain the details from each study, and the positive association may thus be caused by confounding. Although the included studies for statins were limited, the present study suggested that statins may prevent BE development.
A 41% decreased risk was observed for the highest versus lowest intake of vitamin C and BE risk. Protective associations were also observed with the intake of folate and dietary fiber. Analyses for the highest versus lowest intakes of total meat, white meat, and selenium yielded nonsignificant risks of BE. Systematic analyses could not be conducted for other common dietary factors, such as other vitamins, calcium, tea, and coffee, because of the limited studies, and further studies are required to further validate our findings and to reveal these uncertain conclusions.
Our study has several strengths. The first strength was that our systematic analysis was based on the main modifiable lifestyle factors, the substantial sample size and the quantitative synthesis of the eligible data, which provided sufficient robust and reliable evidence and increased the statistical power of our findings. Second, we performed detailed subgroup analyses and dose-response analyses to further detect the associations rather than simply conducting categorical comparisons. These independent analyses provided accurate evaluations and strengthened the conclusion. Third, we broadly and systematically searched three large databases to identify studies published from inception through 30 September 2020, and the reference lists of the included studies were also searched manually to identify additional The results demonstrated that the intake of vitamin C, folate, and dietary fiber may reduce the Barrett's esophagus risk and there is no association between total meat and white meat and the risk of Barrett's esophagus literature. Two reviewers selected the studies and extracted the data independently and in duplicate, which increased the validity of our analyses. Last, 62 included studies were identified from 16 countries or regions in the Americas, Europe, Asia, and Australia, which increased the generalizability of our results.
However, the limitations of the present meta-analysis should be taken into consideration. First, the diagnostic criteria of BE may vary among the included studies. 17 The updated American College of Gastroenterology (ACG) clinical guidelines recommend that intestinal metaplasia (IM) is required for the diagnosis of BE 98 because IM is the only type of esophageal columnar epithelium that clearly predisposes to malignancy. 18 However, this contrasts with the current British Society of Gastroenterology guidelines for BE diagnosis, which stated that IM is not necessary for the diagnosis. 19 Second, although most of studies were adjusted for major confounders, information on some other confounders (e.g., hot drinks, H pylori infection, and hiatal hernia) could still not be obtained in several studies. Thus, our results should be considered carefully due to possible confounding. Third, the ranges of the highest to lowest category varied in the included studies, which influenced the accuracy of the results to some extent, and we cannot thoroughly exclude the potential bias. Nevertheless, to reduce the bias to a large extent, the pooled results for the highest compared with lowest category were adopted and pooled, and the results were further verified by dose-response analyses, which yielded results similar to the original analyses. Finally, the language of the included studies was limited to English, which may lead to potential selection bias.

| CONCLUSIONS
This large systematic analysis demonstrated that smoking, alcohol intake, high BMI, and less sleep time are associated with BE risk. There are statistically significant reduced risks of BE with aspirin use and the intake of vitamin C, folate, and dietary fiber. Our findings strengthen our understanding of the potential mechanisms of BE development and highlight an awareness that lifestyle interventions may reduce the risks of BE and, consequently, esophageal adenocarcinoma.