Association of tooth loss with liver cancer incidence and chronic liver disease mortality in a rural Chinese population

Background Tooth loss has been reported to be associated with the risk of liver cancer in several prior studies in economically advantaged countries. Whether this relationship is also evident in economically disadvantaged populations is not known. Methods We analyzed data from the Nutrition Intervention Trials, two randomized placebo-controlled trials of vitamin/mineral supplementation in Linxian, China. Participants who reported having lost permanent teeth were examined to determine the number of teeth remaining. In the 30-year follow-up period, 329 liver cancers were diagnosed and 368 chronic liver disease deaths occurred. Tooth loss was categorized based on loess smoothed age-specific predicted quartiles. Cox proportional hazards regression was used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the two outcomes. Results Overall, persons in the highest quartile of age-specific tooth loss had an increased risk of liver cancer (HR = 1.27, 95%CI: 0.96, 1.67) which was not statistically significant. Results varied by sex and body mass index (BMI), however. Women in the highest quartile of age-specific tooth loss had a significantly increased risk (HR = 1.64, 95%CI: 1.04, 2.59), while men did not (HR = 1.08, 95%CI = 0.75, 1.57), and persons with a BMI > 23.0 kg/m2 (HR = 1.71, 95%CI: 1.00, 2.92) had a significantly increased risk, while persons with a BMI <23.0 kg/m2 did not (HR = 1.14, 95%CI: 0.82, 1.5). No relationships with chronic liver disease mortality were observed. Conclusions In a rural, economically disadvantaged population, persons with the highest levels of age-specific tooth loss had an increased risk of liver cancer. The results, which were stronger among women and persons with greater BMI, suggest that periodontal disease may increase risk of liver cancer.


Introduction
Liver cancer is the sixth most commonly occurring cancer in the world and the second leading cause of cancer mortality [1]. All major liver cancer risk factors cause chronic inflammation, which may progress to chronic liver disease and eventually liver cancer. Liver disease, particularly its most severe form, cirrhosis, is itself a major source of mortality [2]. The great majority of liver cancers occur in either Africa or Asia, with China alone accounting for half of all cases worldwide [3]. In China, major risk factors include consumption of aflatoxin-contaminated food and chronic hepatitis B and C virus (HBV/HCV) infection [4]. A factor that may be related to risk of liver cancer is periodontal disease, a common condition common among adults [5], with rising prevalence in China [6], that results from infection of the tissue surrounding and supporting the teeth. Severe disease, called periodontitis, includes destruction of connective tissue, bone loss, and tooth loss. Periodontitis may have systemic health effects as evidenced by studies of cardiovascular disease [7], and stroke [8], and has been associated with cancers of the mouth [9], esophagus [10,11], and gastrointestinal tract [11,12].
Using tooth loss as a proxy variable for periodontal disease, three prior studies have prospectively evaluated the association between tooth loss and risk of liver cancer [13][14][15]. A cohort study of male Finnish smokers [15] and a hospital case-control study in Japan [14] both reported increased risks of liver cancer with higher levels of tooth loss. In contrast, a survey of octogenarians in Fukuoka, Japan found no association between tooth loss and liver cancer [13]. As the results of these studies varied, and two of the studies relied on self-report of tooth loss rather than on dental examination, we sought to evaluate the relationship of tooth loss and liver cancer in a rural, economically disadvantaged population in which tooth loss was determined by dental exam.

Study population
The Nutrition Intervention Trials were two randomized placebo-controlled trials of vitamin/ mineral supplementation in Linxian, China, a rural county with high rates of esophageal and gastric cancers. Both the General Population Trial [16] and the Dysplasia Trial [17], have been previously described [18]. Briefly, the General Population Trial [16] consisted of 29,584 persons aged 40-69 at baseline who were randomized to intervention groups receiving daily multivitamin/mineral supplementation starting on March 1986. The Dysplasia Trial [17] consisted of 3,318 persons aged 40-69 with cytological evidence of esophageal dysplasia who were randomized to intervention with multivitamin/mineral supplements starting in May 1985. Data from the two trials were pooled and participants were excluded from the analysis if they did not provide questionnaire information about permanent teeth loss (n = 126), if they did not receive an examination after reporting any permanent teeth loss (n = 80), or if their recorded censoring date occurred before the start of intervention (n = 7). The final analytic population consisted of 32,689 participants.
The study protocols were approved by the Institutional Review Board of the National Cancer Institute. The study was registered as ClinicalTrials.gov number NCT00342654.
Population subsample with HBV/HCV status. Blood samples were collected from participants at baseline in both trials. Serum markers to determine HBV and HCV infection status were analyzed in two previous nested case-controls studies in the Linxian trials population [21,22]. Antibody to hepatitis C virus (anti-HCV) was analyzed using the ORTHO HCV version 3.0 enzyme-linked immunosorbent assay (ELISA) Test System from Ortho-Clinical Diagnostics, Raritan, NJ; hepatitis B surface antigen (HBsAg) was analyzed by enzyme immunoassay using the Bio-Rad Genetic Systems HBsAg EIA 3.0 kit of Bio-Rad Laboratories, Hercules, CA; and antibody to hepatitis B core antigen (anti-HBc) was analyzed by ELISA using the HBc (recombinant) ORTHO ELISA Test System of Ortho-Clinical Diagnostics.

Statistical analysis
Follow up time began at the start of intervention (March 1, 1986 for General Population Trial and May 1, 1985 for Dysplasia Trial) and ended with the first diagnosis of liver cancer, death (from any cause), or March 31, 2016, resulting in 650,235 person-years of follow up. The frequency and proportion of covariates were calculated overall and within categories of age-specific tooth loss. Cox proportional hazards regression was used to estimate the hazard ratio (HR) and 95% confidence intervals (CIs) for the two outcomes, liver cancer incidence and chronic liver disease mortality. For conventional tooth loss, three methods to were compared to control for age: adjustment for age at randomization and age 2 as continuous variables; stratification using 1-year age groups; and stratification using 5-year age groups. No meaningful difference was seen among these methods, thus age at randomization and age 2 were used in all models for conventional tooth loss. All p-values were two-sided with a significance level of α = 0.05.
Measures of association were estimated with adjustment for age at randomization and age 2 using loess smoothed age-specific predicted cut-offs, sex, BMI, education, cigarette smoking, alcohol use, and study trial. Tests for trend were constructed using either the number of teeth lost or the quartile of age-specific tooth loss as continuous variables. The proportional hazards assumption was tested using an interaction between tooth loss and log (follow-up time), as a continuous variable, in models that included adjustment for confounding. No violations of the assumptions were observed. The following sensitivity analyses were performed to assess the robustness of the findings: (1) Restricted to participants ! 50 years old at randomization, as periodontal disease is the main cause of tooth loss in older populations; (2) Restricted to persons who never smoked to address concerns of confounding by cigarette smoking [23]; (3) Excluded the first two years of follow-up to reduce the influence of preclinical, undiagnosed liver disease causing tooth loss; (4) Restricted to participants with known HBV/HCV status (n = 1,644); (5) Excluded the participants in the Dysplasia Trial. In addition, stratified analysis by sex and BMI (< 23.0 kg/m 2 vs. ! 23.0 kg/m 2 ) were conducted, based on recent animal studies of sex differences in inflammatory markers [24,25] and studies of the contribution of metabolic syndrome to the periodontal-hepatic relationship [26,27]. All analyses were performed using SAS 9.3 (SAS Institute Inc., Cary, NC).

Results
Demographic characteristics of the study population by age-specific tooth loss quartile are shown in Table 1. The majority of the study population had a BMI of 18.5-22.99 kg/m 2 (65.7%), had less than a primary school education (71.4%), were non-smokers (70.1%) and did not consume alcohol (77.0%). Table 2 shows the adjusted association of tooth loss with incident liver cancer and chronic liver disease mortality. In regard to liver cancer, there was no association with ever/never loss  Table 3 displays the results of analyses stratified on sex. Although there was no indication of interaction by sex (p interaction = 0.67), the stratified analyses found women with the greatest number of teeth lost had a significantly increased risk of liver cancer (HR = 1.78, 95%CI: 1.00, Tooth loss, liver cancer incidence, and chronic liver disease mortality in rural China 3.18) while the men did not (HR = 0.99, 95%CI: 0.61, 1.59). Similarly, women in the highest quartile of tooth loss had a significantly increased risk of liver cancer (HR = 1.64, 95%CI: 1.04, 2.59) while the men did not (HR = 1.08, 95%CI: .075, 1.57). There were no associations with chronic liver disease mortality among either men or women. Shown in Table 4 are the results of the analysis stratified on BMI. Among participants with a BMI ! 23.0 kg/m 2 , persons in the highest age-specific quartile of tooth loss had a significantly increased risk of liver cancer (HR = 1.71, 95%CI: 1.00, 2.92). Restricting the analysis to persons with a BMI ! 25.0 kg/m 2 further strengthened the relationship (data not shown), however the estimates lacked precision as only 9.7% of the cohort had a BMI in this range. While the test for interaction with BMI was not significant (p interaction = 0.28), the association between tooth loss and liver cancer was not evident among persons with a BMI < 23.0 kg/m 2 (HR = 1.14, 95%CI: 0.82, 1.57). Similarly, there was no association between chronic liver disease mortality in either strata of BMI.
The results did not change when the analysis was restricted to persons ! 50 years of age or restricted to persons who never smoked (data not shown). The results also did not change when the first two years of follow-up were excluded or in a subset analysis of persons with known HBV/HCV status (data not shown). Similarly, exclusion of the persons in the Dysplasia Trial did not affect the results (data no shown). Estimated using cox-proportional hazards regression adjusted for age (age (at randomization) and age2 or loess smoothed age-specific predicted quartiles) gender, body mass index, body mass index, education, cigarette smoking, alcohol use, and trial membership b p-value of test for trend using either the number of teeth loss or the quartile of age-specific tooth loss as a continuous covariate c Age-specific quartiles of tooth loss were estimated using loess regression https://doi.org/10.1371/journal.pone.0203926.t003 Tooth loss, liver cancer incidence, and chronic liver disease mortality in rural China

Discussion
In this prospective study in a rural Chinese population, persons in the highest quartile of agespecific tooth loss had an increased risk of liver cancer. The increased risk was more notable, and statistically significant, among women and among persons with a BMI > 23.0 kg/m 2 . Three previous prospective studies have examined risk of liver cancer with tooth loss, one among a population of male Finnish smokers [15], one among patients at an oncology hospital in Japan [14], and one among community-dwelling octogenarians in Japan [13]. The study of Finnish smokers found significant associations between liver cancer and loss of 11-31 teeth (HR = 1.42, 95%CI: 1.01, 1.98), as well as loss of all teeth (HR = 1.45, 95%CI: 1.00, 2.10). The Japanese study in an oncology hospital found a significant association between liver cancer and loss of 12-23 teeth (HR = 1.74, 95%CI: 1.04, 2.89), but no association with loss of a greater number of teeth. The study of community-dwelling octogenarians found no significant association between liver cancer mortality and tooth loss (HR = 1.07, 95%CI: 0.98-1.17). No studies have prospectively evaluated the association of tooth loss with chronic liver disease mortality, however, several studies have found increased periodontal disease among persons with cirrhosis [28], and one study found decreased mortality among persons with cirrhosis treated for periodontal disease [29]. Tooth loss, liver cancer incidence, and chronic liver disease mortality in rural China

Table 4. Association of tooth loss to incident liver cancer and chronic liver disease (CLD) mortality by body mass index in the Nutrition
Why the associations between tooth loss and liver cancer varied among the studies is not certain, but several differences exist among the studies. The lack of association reported in the study of Japanese octogenarians may have been due to it being a small study of only 697 participants with a limited number of cancer deaths. In addition, liver cancer is more commonly diagnosed at ages younger than 80 years, so persons who live to be octogenarians are likely to be a healthier group of people who are at lower risk of cancer, overall. In the Japanese casecontrol study, the association seen with loss of 12-23 teeth, but not with the loss of a greater number of teeth, is curious, but as no further stratified analysis was presented, it is not clear whether the results differed by sex or other characteristics. In the study of male Finnish smokers, tooth loss of all levels was associated with a significantly increased risk of liver cancer. The lack of association among the men in the current study, in contrast to the Finnish study, could be due to a number of variables given that Finland, in contrast with Linxian, China is a highincome country with an economically-advantaged population. Nevertheless, in the current study, there was an association between the highest level of tooth loss and liver cancer among women. The association among women, but not among men, may be related to women having a higher prevalence of tooth loss (78.5% vs 72.1%) and being more likely to be in the highest quartile of age-specific tooth loss (31.2% vs 19.9%). Women were also more likely to have a BMI > 23.0 kg/m 2 , a factor that was also associated with higher risk of liver cancer associated with tooth loss.
Tooth loss can be the result of periodontal disease, but can also be the result of trauma or caries. A comprehensive oral health examination among~600 participants of the cohorts, however, found that high levels of moderate to severe periodontal disease were present, while caries were less common [19]. These results suggest that the primary cause of tooth loss in the population was periodontal disease. Periodontal disease has been shown to be a risk factor for several systemic conditions, including cardiovascular disease [7], and stroke [8], and has been associated with cancers of the mouth [9], esophagus [10,11], and gastrointestinal tract [11,12]. These relationships may stem from the establishment of a systemic inflammatory condition through mediators such as histamine, cytokines and proteases [30], or alternatively, periodontal disease may serve as a marker for an immune system deficient in the ability to clear infection. Additionally, tooth loss may result from an oral flora that produces carcinogenic byproducts.
Strengths of this study include its large sample size, prospective design, and extensive length of follow-up (up to 30 years). In addition, number of teeth lost, the primary exposure of interest, was determined by physical examination rather than self-report. Large variation in the primary exposure allowed robust estimation of a dose-response relationship. Methods for identification of cancer cases and causes of mortality assured essentially complete ascertainment of both outcomes. This population was highly homogeneous regarding occupation and socioeconomic status. We controlled for age using several methods, including categorizations based on loess smoothed age-specific predicted cut-offs. Further, we controlled for important risk factors for liver cancer including BMI, alcohol use, and cigarette smoking.
A limitation of this study was a lack of complete ascertainment of HBV/HCV status for the study population. The subset analysis of persons who were tested for HBV and HCV indicated, however, that the results were not affected by adjustment for viral status. The study also lacked information on exposure to aflatoxin B 1 , a known liver cancer risk factor. Previous research, however, reported low levels of aflatoxin B 1 exposure in Linxian [31].
This study is the first to evaluate the effects of tooth loss on liver cancer incidence and chronic liver disease mortality in a rural, economically disadvantaged population. The increased risk of liver cancer with the highest levels of tooth loss, a relationship particularly evident among women, suggests that periodontal disease may be a risk factor for liver cancer.
Further study of this relationship is clearly warranted as the identification of modifiable risk factors for liver cancer may have the potential to decrease the burden of the world's second greatest contributor to cancer mortality.