Tooth loss is associated with increased risk of esophageal cancer: evidence from a meta-analysis with dose-response analysis

Epidemiological studies have revealed the association between tooth loss and the risk of esophageal cancer (EC); however, consistent results were not obtained from different single studies. Therefore, we conducted the present meta-analysis to evaluate the association between tooth loss and EC. We conducted electronic searches of PubMed until to February 10, 2015 to identify relevant observational studies that examined the association between tooth loss and the risk of EC. Study selection and data extraction from eligible studies were independently performed by two authors. The meta-analysis was conducted using Stata 12.0 software. Finally eight eligible publications with ten studies involving 3 cohort studies, 5 case-control studies, and 1 cross-sectional study were yielded. Meta-analysis identified tooth loss increased risk of EC 1.30 times (Relative risk = 1.30, 95% confidence interval = 1.06–1.60, I2 = 13.5%). Dose-response analysis showed linear relationship between tooth loss and risk of EC (RR = 1.01, 95%CI = 1.00–1.03; P for non-linearity test was 0.45). Subgroup analysis proved similar results and publication bias was not detected. In conclusion, tooth loss could be considered to be a significant and dependent risk factor for EC based on the current evidence.


Results
Study selection and characteristics. From the 155 records initially identified, 8 articles involving 3 cohort studies 17,27,28 , 5 case-control studies [29][30][31][32] , and 1 cross-sectional studies 33 were selected for the present meta-analysis. A detailed flow chart of the selection process is shown in Fig. 1.
Among the selected studies, the study of Guha et al. 31 involved 2 multicentric case-control studies from central Europe (including Romania, Poland, and Russia) and Latin America (including Cuba, Argentina, and Brazil), while the other 8 were single center studies. All the cases under study were confirmed to have EC through histological, pathological or cytological means, and 6 articles of 7 studies were ESCC 17,27,29,[31][32][33] . All the studies clearly defined the reference group of on the basis of tooth loss, with the major characteristics presented in Table 1. All the studies reported adjusted point estimates and 95% CIs. The adjusted covariates are shown in Table 2.
Tooth loss and risk of EC. Among the eight included studies, two showed a significantly positive association between tooth loss and the risk of EC 29,30 , while the other six were negative 17,27,28,[31][32][33] . No substantial heterogeneity among these trials was observed (I 2 = 13.5%, p = 0.32), and the overall result based on the fixed effect model showed that tooth loss increased risk of EC by 1.30 times (RR = 1.30, 95%CI = 1.06-1.60; Fig. 2).   Table 2. Adjustments in studies included in the meta-analysis Five studies provided sufficient data of dose-response relationship. Our dose-repose meta-analysis showed an increased risk of EC corresponding to every 1 tooth loss increment (RR = 1.01, 95%CI = 1.00-1.03). No evidence of nonlinear relationship was detected (P = 0.45; Fig. 3). Table 3 shows the results of subgroup analyses. The results of all subgroup analyses showed significantly increased risk in case-control studies and in adjustments for smoking and/or alcohol.
Publication bias. Visual inspection of the funnel plot did not identify any substantial asymmetry (Fig. 4), which indicated no publication bias existed.

Discussion
The present meta-analysis found evidence of an association between tooth loss and the risk of developing EC, with the loss of tooth significantly increasing the risk by 30%. And the relationship between tooth loss and the risk of developing EC is linearly dependent. Subgroup analyses revealed similar results of pooled estimations, and publication bias was not observed. In fact, this significant association could be explained. Periodontal disease is a chronic inflammatory disease, which contributing to constant low-grade systemic inflammation with elevated levels of circulating inflammatory markers 4,34 and is associated with HNC 35 and EC 36,37 . Periodontal disease is the major cause of tooth loss 38,39 , and tooth loss is also a maker of systemic inflammation 40,41 . The link between inflammation and cancer has long been recognized [42][43][44][45] , and it is likely that the progression of periodontal disease to tooth loss also represents a progression in the breakdown of normal cell-cycle control and potential carcinogenesis. In addition, the progression of tooth loss destroys the normal periodontal tissue, allowing oral microorganisms to accumulate deep into the oral tissue, thereby facilitating their growth. Oral microorganisms produce greater amounts of nitrosamine 46 , which is significantly associated with the development of cancer 47,48 . It is not difficult to envisage that during normal physiological behavior such as swallowing and drinking, oral microorganisms as well as the produced nitrosamine are passed into the esophagus from the oral cavity along with food and drinks; therefore, an association between tooth loss and EC seems plausible.
Subgroup analyses provided varied results of cross-sectional study design, which could be partly due to the smaller number of included studies or the limitations of this design; however, the pooled results from case-control studies and cohort studies presented a significantly increased risk of EC associated with tooth loss (Table 3). On the other hand, according to the Hill's criteria, which are widely accepted for determining causality 49 , our meta-analysis    indicates that tooth loss is a marker rather than a causal factor for EC. In our meta-analysis, all included studies did not show consistency in results, and pooled RRs only showed a weak association. In addition, only 3 cohort studies among the 10 included studies involved to temporality, although their pooled result showed a significant association. Furthermore, there is no evidence whether control of tooth loss can prevent EC up to now; therefore, the specificity of the association cannot be evaluated. Moreover, there is lack of experimental research. In summary, these factors suggest that that tooth loss is a marker rather than a causal factor for EC. The heterogeneity was none to mild among the studies examining the association between tooth loss and the risk of EC 23 . This mild heterogeneity can be attributed to the differences in the characteristics of different populations, definitions of the reference and tooth loss group(s), and adjustment for confounding factors. Our subgroup analyses provide evidence that the study design, definition of the reference group, reported effect size, pathological type, and adjustment of origin was not the source of the heterogeneity (Table 3). Hence, the mild heterogeneity might be the inherent shortage of meta-analysis or due to the statistic heterogeneity.
Our meta-analysis also has some strengths and implications. To the best of our knowledge, this study is the first meta-analysis on this topic. We have searched relevant published studies via electronic and manual searching and collected all published studies that met the inclusion criteria; no publication bias has been detected. In addition, all included studies provided adjusted effect sizes. Nine studies have adjusted for smoking, while 7 have simultaneously adjusted for smoking and alcohol (Table 2). For smoking and alcohol are the well accepted risk factor of EC 9-16 , we conducted subgroup analysis to investigate the smoking adjusted studies and both smoking and alcohol adjusted studies, the results were both significantly and similar. Thereby ensuring the accuracy of the results obtained and indicating that tooth loss is independent of the conventional risk factors for EC. Furthermore, accumulating evidence and the large sample size of included studies provide the statistical power to present precise and reliable risk estimates relative to a single study. The results of the included studies are not consistent, and subgroup analysis revealed that the sample size could influence the result. Moreover, we could not identify a causal relationship between tooth loss and the risk of EC; the actions, such as toothbrushing 32,50 are suggested to prevent tooth loss and this may decrease the risk of developing EC. Relevant animal studies and interventional studies are necessary to further explore the association between tooth loss and EC. Finally, people who cessation smoking, mild drinking, daily use of dental floss, and seeks periodontal therapy are all important ways of countering the increased risk of EC in these subjects.
Despite these strengths, several limitations should be acknowledged in our meta-analysis. First, the definition of the reference group and tooth loss used varied among studies. Nowadays, no international unification index for evaluating tooth loss in relevant studies is available, which could result in heterogeneity and increase the difficulty of performing meta-analysis, even resulting in failure to meet the criteria for meta-analysis. Second, substantial heterogeneity was observed among studies, although it was mild and very commonly found in meta-analysis of observational studies and is understandable, which cannot be ignored. We have investigated heterogeneity using subgroup analyses; however, the real source(s) of heterogeneity could not be identified. Third, subgroup analyses revealed that the results are significantly inconsistent with limited statistical power owing to a relatively small number of included studies. We used Hill's criteria to argue causality and have discussed how the results could be influenced by the sample size; greater numbers of relevant studies addressing this topic are required for enlarging the sample size. Furthermore, while almost all included studies have adjusted for smoking and alcohol consumption, other unmeasured factors, such as diabetes, gastroesophageal reflux, and socioeconomic factors may confound the interpretation of an association between tooth loss and EC and potentially introduce bias. Only a few included studies have adjusted for these factors (Table 2). Fourth, although no evidence of publication bias was observed, the statistical power of our study is limited by a relatively small number of included studies; thus, it is difficult to be certain that there is no publication bias. These limitations of the present meta-analysis may affect the accuracy of our results. Finally, for lacking of relevant assessment tool of cross-sectional study, we could not assess the risk of bias of included studies 51 .
In summary, the present meta-analysis indicates that tooth loss is probably a significant and dependent risk factor for EC, suggesting that people who have lost teeth should pay attention to the symptoms and avoid other classical risk factors for EC. Moreover, according to the Hill's criteria for causal inference, tooth loss is a marker rather than causality for EC.