Dietary factors and risk of mortality among patients with esophageal cancer: a systematic review

Background The effects of dietary factors on prognosis of esophageal cancer remain unclear. This systematic review and meta-analysis aimed to assess the association between dietary intake and the risk of mortality among patients with esophageal cancer. Methods Six electronic databases (PubMed, Web of Science, OVID, ProQuest, CNKI and Wanfang) were searched for studies published up to Oct. 2019 that examined the association between dietary intake and all-cause mortality, esophageal cancer-specific mortality and esophageal cancer recurrence. The pooled hazard ratio (HR) with 95% confidence interval (CI) were derived by comparing the highest with the lowest categories of each dietary item and by using random effect models. Results A total of 15 cohort studies were included in this study and all reported pre-diagnosis dietary exposure; two focused on dietary folate, 12 on alcohol consumption and three on other dietary components (sugary beverages, phytochemicals and preserved vegetables). When comparing the highest with the lowest categories, dietary folate intake was associated with a reduced risk of esophageal cancer-specific mortality in patients with esophageal squamous cell carcinoma (HR: 0.41, 95% CI: 0.25–0.69), with low heterogeneity (I2 = 0%, P = 0.788). When comparing the highest with the lowest categories of alcohol consumption, alcohol consumption was associated with an increased risk of all-cause mortality in patients with esophageal squamous cell carcinoma (HR: 1.29, 95% CI: 1.07–1.55; heterogeneity: I2 = 53%, P = 0.030), but this increased risk was not significant in patients with esophageal adenocarcinoma (HR = 1.05, 95% CI: 0.84–1.32). Conclusions This review with pre-diagnostic dietary exposure showed that dietary folate intake was associated with a reduced risk of mortality of esophageal squamous cell carcinoma, whereas alcohol consumption was associated with an increased risk. More studies are needed to investigate effect of dietary factors, especially post-diagnosis dietary consumption, on esophageal cancer prognosis.


Background
Esophageal cancer (EC) is one of the most malignant tumors worldwide, ranking seventh in cancer incidence and sixth in cancer mortality in 2018 [1]. Esophageal adenocarcinoma (EAC) and esophageal squamous cell carcinoma (ESCC) are two main histological types. EAC is the main histological type in developed countries whereas ESCC predominates in eastern Asia and Africa [2]. The prognosis of esophageal cancer is poor; the 5year survival rate of EC in the United States is 19%, in Europe 12.4%, and in China 20.9% [3][4][5].
The prognosis of esophageal cancer is influenced by many different factors [6]. Noteworthy, increasing evidence is highlighting the pivotal effects of nutritional factors on cancer prognosis and survival: natural ingredients such as lycopene and beta-carotene in the plant food could inhibit EC109 cell viability [7], dietary interventions could improve diet quality in cancer survivors [8], and nutritional support could improve esophageal cancer prognosis by improving treatment compliance, reducing toxicity and enhancing the immune response [9]. However, most studies focused on the roles of perioperative nutrition support in postoperative complications [10,11], and the results from epidemiological studies on the associations between dietary factors and EC prognosis are inconclusive [12][13][14][15]. A meta-analysis showed that pre-diagnosis alcohol drinking increased risk of death in ESCC rather than in EAC [16], however, some new evidence has emerged after this study published and the pooled results are needed to be updated. Therefore, by summarizing the results of observational studies, this systematic review and meta-analysis was conducted to evaluate the association between food and its components intake and risk of mortality among patients with esophageal cancer, with addressing the difference in histology and the difference between allcause mortality and cancer-specific mortality.

Literature search
Literatures published up to October. 2019 were systematically searched through four English databases (PubMed, Web of Science, OVID, ProQuest) and two Chinese databases (CNKI and Wanfang). PRISMA statement for conducting and reporting meta-analysis of observational studies was followed [17]. The search strategy was as follows: [ Two researchers (LP Sun & LB Yan) independently conducted the literature retrieval, identified potential studies, extracted information from the included papers, and assessed the quality of included studies. Discrepancies were settled down by group discussion with other two professionals (X LIU & CX Zhang). The titles and abstracts of initially identified papers were firstly reviewed, and then the full texts of the selected papers were reviewed to determine eligibility. To avoid omission of literatures, backward and forward citation tracking in both Web of Science and Scopus were also used to identify articles.

Study selection criteria
Studies meeting the following conditions would be included: (i) study design was cohort study with esophageal cancer patients; (ii) the consumption of food and/or its components but not supplementations was accessible; (iii) prognostic outcomes included all-cause mortality, esophageal cancer-specific mortality or esophageal cancer recurrence; (iv) the hazard ratio (HR) or relative risk with 95% confidence interval (95% CI) were reported or could be calculated. When there were papers from the same study or covering the same population, only the most comprehensive or latest data was selected. Case reports, cross-sectional studies, editorials, abstracts, reviews, articles without full text, duplicated studies, animal studies and vitro studies were excluded.

Data extraction
Data and information were extracted from the included studies, including name of the first author, publication year, country, sources of patients, sample size, histological type, follow-up duration, stage/grade grouping, dietary exposure, dietary assessments, outcomes, comparison method, effect size, confounders and covariates.

Quality assessment
Quality of the included papers was evaluated according to the Newcastle-Ottawa Scale (NOS) Criteria for nonrandomized studies [18]. A maximum total 9 points were assigned to each study, with a maximum of 4 for selection, 2 for comparability and 3 for outcome. Studies were considered to be of high quality (> 6), median quality (4~6) or low quality (≤3, [19]).

Statistical analyses
Meta-analysis was conducted to estimate the pooled HR with 95% CI by comparing the highest with the lowest categories of dietary intake in each selected item. A random effect model was selected in the meta-analysis [20]. Heterogeneity between studies was measured using Q and I 2 statistics. Sensitivity analysis was conducted by excluding the study one by one from the pooled results, by excluding the studies not reporting adjusted effects, and by excluding the studies not collecting information of stage and severity of esophageal cancer. The metaanalysis with crude HRs obtained from univariate analysis was also conducted. Publication bias was tested by funnel plot with Begg's tests [21]. A power calculator used to estimate statistical power of meta-analyses [22]. Statistical analysis was completed in Stata 15.1 (Stata Corporation, College Station, TX) and R software (version 3.5.3). Figure 1 shows the flowchart of literature selection. After removal of 4215 duplicates, 3795 potentially eligible records were left, of which 3677 were excluded after reviewing the titles and abstracts. Finally, 15 eligible cohort studies [12][13][14][15][23][24][25][26][27][28][29][30][31][32][33] containing 6826 esophageal cancer patients were identified through full text review of 118 studies. The most common reasons for exclusion were study designs, lack of data on dietary exposure and no outcomes of interest.
For study quality of 15 included studies, average NOS score was 6.65, ranging from 4 to 8; 12 studies were high quality (NOS score ≥ 7) and three studies were median quality.

Dietary folate intake
Only two studies [24,25] reported effects of dietary folate intake on esophageal cancer-specific mortality in ESCC. These two studies provided adjusted risk estimates from multivariate analysis. Categorization of folate intake was different in these two studies. The highest vs. the lowest in one article [24] was ≥95.5 μg/day vs. < 30.0 μg/day, while in another article [25] was > 300μg/day vs. < 230μg/day. When pooled these two studies (Fig. 2), the overall HR was 0.41 (95% CI: 0.25-0.69) with low statistical heterogeneity (P for heterogeneity = 0.79, I 2 = 0%). The funnel plot did not reveal asymmetry (Fig. 3) and the corresponding Begg's test did not show publication bias (P = 1.00). Power calculation for random effect model was 96.5%.

Other dietary components
Only one study on sugary beverages [12], one on phytochemicals [13] and one on preserved vegetables [30] were found. Miles et al. [12] studied effects of sugary beverages intake on prognosis of EC and EAC. They found that soft drinks and fruit juices intake would worsen prognosis of EC patients; however, when the study population was restricted to patients with EAC, no significant association between sugary beverages and allcause mortality was found. Petrick et al. [13] studied effects of consumption of total dietary flavonoids, dietary flavonoid subclasses (anthocyanidins, flavan-3-ols, flavanones, flavones, flavonols and isoflavones) and lignans on all-cause mortality in ESCC and EAC, respectively; however, only lignans was found to reduce all-cause mortality of ESCC by 42% (HR = 0.58, 95% CI: 0.37-0.92). As indicated in a study done by Shi et al. [30], in ESCC patients, when comparing with patients consuming preserved vegetables < 1 time/week, those who consumed preserved vegetables ≥1time/week had a 1.58-fold (95% CI: 1.01-2.47) risk of all-cause mortality.
The pooled results of the association between alcohol consumption and risk of death among different subtypes of esophageal cancer are shown in Fig. 4. When comparing the highest with lowest consumption of alcohol, the pooled HR was 1.48 (95% CI: 1.19-1.84) with low statistical heterogeneity (P for heterogeneity = 0.35, I 2 = 4.2%) in EC and 1.29 (95% CI: 1.07-1.55) with moderate statistical heterogeneity (P for heterogeneity = 0.03, I 2 = 53.0%) in ESCC, whereas no association was found in EAC (HR = 1.05, 95% CI: 0.84-1.32). The funnel plots did not show  Patients with a present or past history of alcohol consumption were referred to as drinkers significant asymmetry for any types of esophageal cancer (Fig. 5), and the corresponding Begg's tests did not show publication bias (all P > 0.30). Power calculation for these three random effect models was all equal to 100.0%. The sensitivity analysis was only performed on alcohol consumption. The analysis was repeated consecutively by removing one study from the pooled results each time, and significant change was observed only after the exclusion of Huang's study [27] in EC and ESCC, respectively ( Supplementary Fig. S1); the analysis was also conducted by excluding 3 studies [23][24][25] not reporting adjusted effect size, and the pooled results were 1.26 (95% CI: 1.01-1.60) for ESCC and 1.01(95% CI: 0.70-1.47) for EAC (Table 2). When the unadjusted effect was considered, the pooled HR was 1.27 (95% CI: 1.06-1.53) for ESCC. Only one study [23] on EAC provided crude effect; hence, the pooled analysis was not conducted ( Table 2). When the studies which did not include information of stage and severity of esophageal cancer were excluded, the pooled HR was 1.26 (95% CI: 1.01-1.60) for ESCC and 1.01(95% CI: 0.70-1.47) for EAC.

Discussion
As far as we know, there lacks of systematic review with quantitative analysis to evaluate ordinary dietary behavior and prognosis of esophageal cancer. The results from meta-analysis displayed that pre-diagnostic dietary folate intake was significantly related to a decreased risk of esophageal cancer-specific mortality in ESCC and prediagnostic alcohol consumption was associated with an increased risk of all-cause mortality in EC and in ESCC.
Alcohol is one of the major determinant factors for developing esophageal cancer [34], but effects of it on esophageal cancer prognosis remain controversial [16]. The pooled results in our study showed that prediagnostic alcohol consumption could increase risk of mortality among EC and ESCC patients by 48 and 29% respectively; however, this effect was not found in EAC patients. Without adjustments, confounding factors could lead to misinterpretation of the association between independent variables and dependent variables [35]. Hence, we pooled crude HRs obtained from univariate analysis and adjusted HRs obtained from multivariate analysis, respectively; the association remained the same, further suggesting the strong effects of alcohol intake on ESCC mortality and indicating that our results was stable and robust. Similar results were obtained when the studies not including information of stage and severity of esophageal cancer were excluded. The findings from our study are consistent with the results from a previous meta-analysis by Fahey et al. [16], though the contrast groups were different between two studies. To examine the robustness of our results, we repeated the analysis by excluding the study one by one from the pooled results, and no significant change was observed for ESCC and EAC with the exception of excluding the study by Huang et al. [27], indicating the results our study obtained were stable. Omitting the study by Huang et al. [27] altered the positive association between alcohol consumption and risk of death in EC and ESCC to no association. This may be due to that the sample size of this study was the largest, therefore the weight of the study was relatively large when calculating the pooled results, indicating that more studies with large scale samples are needed.
Our results of meta-analysis also indicated that intake of dietary folate was associated with 59% reduced risk of esophageal cancer-specific mortality among ESCC patients.  This finding is consistent with findings on other cancers [36,37]. Folate metabolites have become diagnostic and therapeutic targets for several types of cancers in recent years [38,39]. The possible mechanism maybe due to that folate deficiency indirectly affects DNA and RNA methylation, thereby alters the expression of tumor suppressor genes and proto-oncogenes [40,41]. However, only two studies focused on effects of dietary folate intake on EC mortality, and all were carried out in China with small sample size. Thus, replication of our results in other large studies in different countries is warranted. The evidence for other dietary components is limited. Lignans was found to reduce risk of all-cause mortality of ESCC by 42% [12], while positive association between the other two factors, sugary beverages and preserved vegetables, and esophageal cancer survival was revealed [13,30]. Sugar in food can lead to awful disease progression through increased inflammation. The inflammation can be caused by oxidative stress, which ultimately accelerates DNA damage and elevates levels of interleukin-cytokines and other pro-inflammatory molecules [42]. N-nitroso compounds widely existing in processed foods (eg. preserved vegetables) may play an important role in tumor progression. N-nitroso compounds give rise to excessive expression of cyclinE 1, cyclinD 1, transform growth factor α and epidermal growth factor receptor in esophageal tissues, thus enhance cancer progression [43]. Anti-cancer effects of phytochemicals like lignans are mainly through estrogen/anti-estrogen activity, anti-proliferation or apoptosis, prevention of oxidation, induction of cell cycle arrest, regulation of changes in host immune system, anti-inflammatory activity and cell signal transduction [44].
This study has some strengths. First, the literature retrieval was from multiple databases and the selection was determined by independent reviewers, which was helpful to avoid literature omission. Second, we included studies according to the strict inclusion and exclusion criteria, which may augment the validity of our findings. Third, sensitivity analyses yielded similar results, indicating the stability of our findings. There were also some limitations in this study. There were only 15 studies included; most of them focused on alcohol consumption and only five focused on other dietary components; pooling results from limited evidence may influence the stability of our results, though the power of random effect models in our study is fairly strong and sensitivity analyses yield similar results. Consuming large amounts of fruits and vegetables after diagnosis could reduce the mortality of cancer and diets before diagnosis may reflect the changes in taste or appetite attributable to cancer [45]. Besides, cancer patients may follow the doctor's advice to change their diets in order to obtain better prognosis. However, dietary behaviors in this systematic review and metaanalysis were all pre-diagnostic, the roles of postdiagnostic dietary intake in prognosis of esophageal cancer are unclear.

Conclusion
In summary, this review with limited evidence suggested that folate intake was associated with a reduced risk of esophageal cancer-specific mortality for ESCC, whereas alcohol consumption was associated with increased risk of mortality for ESCC. More studies are needed to investigate effect of dietary factors, especially post-diagnosis dietary consumption, on esophageal cancer prognosis.