Tooth Erosion and Eating Disorders: A Systematic Review and Meta-Analysis

Background Eating disorders are associated with the highest rates of morbidity and mortality of any mental disorders among adolescents. The failure to recognize their early signs can compromise a patient's recovery and long-term prognosis. Tooth erosion has been reported as an oral manifestation that might help in the early detection of eating disorders. Objectives The aim of this systematic review and meta-analysis was to search for scientific evidence regarding the following clinical question: Do eating disorders increase the risk of tooth erosion? Methods An electronic search addressing eating disorders and tooth erosion was conducted in eight databases. Two independent reviewers selected studies, abstracted information and assessed its quality. Data were abstracted for meta-analysis comparing tooth erosion in control patients (without eating disorders) vs. patients with eating disorders; and patients with eating disorder risk behavior vs. patients without such risk behavior. Combined odds ratios (ORs) and a 95% confidence interval (CI) were obtained. Results Twenty-three papers were included in the qualitative synthesis and assessed by a modified version of the Newcastle-Ottawa Scale. Fourteen papers were included in the meta-analysis. Patients with eating disorders had more risk of tooth erosion (OR = 12.4, 95%CI = 4.1–37.5). Patients with eating disorders who self-induced vomiting had more risk of tooth erosion than those patients who did not self-induce vomiting (OR = 19.6, 95%CI = 5.6–68.8). Patients with risk behavior of eating disorder had more risk of tooth erosion than patients without such risk behavior (Summary OR = 11.6, 95%CI = 3.2–41.7). Conclusion The scientific evidence suggests a causal relationship between tooth erosion and eating disorders and purging practices. Nevertheless, there is a lack of scientific evidence to fulfill the basic criteria of causation between the risk behavior for eating disorders and tooth erosion.


Introduction
The incidence of eating disorders (EDs) has increased over the past decade, both in males and females [1]. These conditions are associated with significant functional impairment and serious physical and psychological consequences due to an excessive preoccupation with body weight or shape. The mortality and morbidity rates associated with EDs are among the highest of any mental disorders [2].
Medical complications from EDs may affect any organ and be life-threatening [3]. Tooth erosion (TE) has been considered an oral manifestation of EDs associated with vomiting practices [3][4][5][6]. TE is a complex and multifactorial condition characterized by a progressive and irreversible loss of tooth structure due to a chemical process without bacterial involvement. It is clinically detectable as thinner enamel with chamfered ridges, cupped cusp tips and grooved incisal edges, sometimes with dentine exposure [7,8].
Dentists usually monitor their patients on a regular basis, sometimes throughout their childhood and adolescence. Therefore, they may be the first health professionals to suspect EDs, due to their oral implications, contributing to the patient's early referral to specific treatment [3,9]. Nevertheless, even with the increasing prevalence of EDs, the causal effect between these EDs and TE has not been thoroughly discussed in literature.
Hill's criteria of causation [10] must be considered in contemporary epidemiology and consist of nine items: strength of association, consistency, specificity, temporality, dose response, experimental evidence, biological plausibility, coherence, and analogy [11].
An evaluation of Hill's criteria of causation applied to the possible causal relationship between EDs and TE suggests specificity between both conditions (patients who suffer from EDs may present TE), temporality (the cause -EDs -occur before the consequence -TE), biological plausibility (vomiting practices related to the EDs causes an acid attack to tooth enamel), coherence (one cause is specific to one effect), and analogy (other diseases or exposures, such as acidic food or gastroesophageal reflux can cause TE).
Nevertheless, all the points raised before have to be systematically discussed and analyzed before drawing inferential causal conclusions. Furthermore, it is important to evaluate the strength of evidence of such an association and of the dose-response relationship, and search for experimental evidence. The aim of the present systematic review and meta-analysis was to search for scientific evidence of the following clinical question: Do EDs increase the risk of TE?

Search Strategy
The inclusion criteria for this systematic review were: epidemiological studies (cross-sectional, case-control, cohort and clinical trials) concerning etiological factors and/or the prevalence of TE and its association to any type of EDs (bulimia, anorexia, bingeeating, dysmorphic body disorder, vomiting, hyperphagia) in humans (Checklist S1 presents the PRISMA checklist for systematic reviews).
The exclusion criteria were: unrelated epidemiological studies (other outcome rather than TE), reviews, studies reporting vomiting habits not related to eating disorders, case reports/case series/letters to the editor, laboratorial studies (in vitro studies, extracted teeth, fossils), studies reporting dental treatment, dental materials, knowledge concerning TE, epidemiological studies that did not associate EDs with TE, studies with self-report of TE and infeasibility of extracting data.
The electronic search retrieved 1094 abstracts and titles ( Figure 1). Reference Manager Software (Reference Manager, Thomson Reuters, version 12.0.3) was used to organize the studies. After the duplicate references were removed, a total of 822 abstracts and titles were read and analyzed by two independent and calibrated reviewers (APH and PADO). As a calibration exercise, the reviewers thoroughly discussed the criteria and applied them to a sample of 20% of the retrieved studies to determine inter-examiner agreement. After adequate agreement was achieved (Kappa: 0.72 to 0.77) all the studies were read by the reviewers independently. Disagreements were resolved by consensus and by the supervision of the gold standard (CCM). If relevant data was missing or if the paper was not available, the primary authors were contacted for additional information/article request. Systematic reviews, theoretical reviews and additional articles of potential relevance were also manually searched. Grey literature was searched from BBO, which retrieved thesis and monographies, and from MEDLINE, which retrieved abstracts recently presented in congress. During the abstracts' analysis all studies addressing risk factors for TE were selected, even when EDs were not mentioned, in order to search for relevant data not reported in the abstract. The exclusion criteria for abstract and title selection are detailed in Figure 1. Among the 822 records screened, 159 were selected for full text analysis (Table S1 presents excluded studies from full text analysis).

Data Extraction
Data extraction was conducted by two independent reviewers (APH and PADO) and supervised by the gold standard (CCM). The data analysis is described in Tables 1 and 2. The main analyzed outcome was that TE and EDs were extracted as categorical variables based on authors' descriptions. Extraction was based on non-exposure to risk factor vs. exposure, as follows: -Control (no EDs) vs. anorexia -Control (no EDs) vs. bulimia -Control (no EDs) vs. bulimia with self-induced vomiting (SIV) -Control (no EDs) vs. bulimia without SIV -Control (no EDs) vs. any type of EDs -EDs without SIV vs. EDs with SIV -Control (no ED risk behavior) vs. ED risk behavior Not all studies described all variables. Confounders were extracted and described as if they were evaluated in multivariate analysis.

Methodological Quality Assessment
The quality of the studies was peer-reviewed by APH and PADO using a modified version of the Newcastle-Ottawa Scale for observational studies. Disagreements were resolved by consensus. No cohort study was selected during the analyses as they did not fit the inclusion criteria. No clinical trials were found by electronic and manual search.
A system of points (stars) was given to the eligible categories: sequence generation entries, allocation concealment, blinding, incomplete outcome data, and sample losses. The scale scores varied depending on the study design: for cross-sectional studies it ranged from 0 (lowest grade) to 6 (highest grade) and for case-control studies it ranged from 0 to 10. Studies with scores above the median were classified as high quality studies [12] : .3 for cross-sectional studies and .5 for case-control studies.
Each cross-sectional study could be awarded a maximum of one point for each numbered item, except for the 'Comparability' criteria, in which a maximum of two stars could be scored. When referring to the case-control studies, one star could be awarded for each numbered item, except for the items 'control for confounders' and 'diagnosis of tooth erosion', in which a maximum of two stars could be scored (Tables 3 and 4).

Statistical Methods and Data Synthesis
The Comprehensive Meta-Analysis program, Version 2 [13] was used for meta-analysis. Heterogeneity between studies was evaluated using I 2 statistics [14]. Meta-analysis was conducted when I 2 was below 50% and the sensitivity test was conducted when heterogeneity ranged from moderate to high, in order to exclude studies that would increase the heterogeneity. Random effect model was used when heterogeneity was high and fixed effect model for low heterogeneity (0.0%) [14,15]. Risk measures, 95% confidence interval (CI) and p-value were described in forest plots, and summary risk measures were calculated. Publication bias was not evaluated as there were not enough studies to be grouped in a funnel plot [16,17].

Studies Characteristics
The study selection process is presented in Figure 1. After full text analysis, 23 studies were included in the qualitative synthesis of this systematic review (13 case-control and 10 cross-sectional) and 14 were included in the quantitative synthesis (meta-analysis). A summary of included studies with details including the studies' outcomes as reported and quality appraisal scores is shown in Table 1 and 2.
In general, for case-control studies, cases were recruited from reference centers of EDs and controls were recruited from universities and dental services [4,[18][19][20][21][22][23][24]. Data for comparison of TE experience among patients suffering from bulimia and anorexia patients was provided by 2 studies [23,24].
Other studies [25][26][27] compared TE in bulimics with selfinduced vomiting and non-vomiting groups. There was only one case-control study [28] that selected patients based on outcome (with TE vs. patients without TE). Three cross-sectional studies         [5,6,29] evaluated TE in adolescents with ED risk behavior, and the others evaluated adolescents already diagnosed with EDs. Quality assessment. The quality of cross-sectional studies ranged from 2 to 6 (maximum: 6) and 3 to 9 (maximum: 10) for case-control studies (Tables 1, 2, 3

Data Synthesis
Comparison of vomiting practices. Figure 2 shows metaanalysis of three cross-sectional studies [25][26][27] of patients with EDs who SIV vs. patients with EDs that did not SIV and its association with TE. There is a significant association between patients who SIV and increased risk of TE (Summary OR = 19.6, 95%CI = 5. 6-68.8).
ED risk behavior and its association with TE. The metaanalysis and sensitivity analysis of two cross-sectional studies [6,29] showed a significant association between adolescents with ED risk behavior and TE (Summary OR = 11.6, 95% CI = 3.2-41.7) (Figure 4).

Assessment of Methodological Quality
Among the cross-sectional studies, the major shortcomings involved the data collection process and comparability issues on the basis of the study design or analysis. One study did not mention whether the patients suffering from EDs were diagnosed by clinicians or if validated criteria were used to evaluate the EDs. It also did not mention if the patients were referred from hospitals [25]. Only a few studies reported the criteria used to identify TE or if the examiners had been trained and calibrated [6,9,25,36].
Another shortcoming refers to the external validity of the studies analyzed. During the analysis it was observed that the majority of the studies were conducted among female samples; from the 23 studies included in the meta-analysis, nine included male participants [4,5,[25][26][27][28][29]32,35]. Also, the studies had a broad age range, varying from 12 [5] to 83 [28] years.
The absence of controlling for confounders was also observed. Only two studies reported adjusted analysis for confounders, and one controlled the association between TE and EDs for dietary habits and oral hygiene [6]. The other study controlled salivary   Table 4. Cross-sectional studies included in this systematic review presented according to their quality score. Tooth Erosion and Eating Disorders: Systematic Review  Tooth Erosion and Eating Disorders: Systematic Review factors and the duration of oral manifestations and EDs [25]. The studies did not clearly report whether there were sample losses, with only one study including a written description of this aspect [24].
With respect to the case-control studies, one study [28] selected the cases by the presence of TE instead of the manifestation of EDs. Only two studies did not report whether a baseline examination was conducted to ensure that the controls were not suffering from EDs at the beginning of the study [33,34]. Regarding the oral examinations, four studies did not mention the criteria used for TE diagnosis [18,30,33,34]. Details on the blinding process were also fairly reported as only four studies reported that the examiners were blinded concerning the ED status of the participants during the oral examinations [6,20,22,32].

Strength of Evidence
The evidence found in the meta-analysis of the case-control studies indicates that patients suffering from different types of EDs, anorexia, bulimia and bulimia with SIV have a greater risk of exhibiting TE in comparison to individuals without such exposure ( Figure 3). It was also observed that EDs were significantly associated with the severity of TE (p,0.01) [35]. Only one study did not show this association as TE was not detected in any patient [9]. However, the study pointed out some limitations, such as the small number of participants, which limited the statistical power of the analysis performed [9]. Bulimia and anorexia were significantly associated with TE ( Figure 3). However, the authors did not specify whether the disorders involved vomiting practices. The association of bulimia, anorexia and TE was also highlighted by other studies [19,34].
There was a statistical association of bulimia with SIV and TE, but not for bulimia without SIV (Figure 3). This reinforces the hypothesis that purging techniques are crucial cofactors for the occurrence of TE, which results from a chronically acidic oral environment [4]. This can be reinforced by Figure 2, which shows that patients suffering from EDs who SIV had a significant greater risk of TE when compared to patients who did not SIV. In this case, EDs were not specified by the authors. Such an association was also observed in other studies [28,36].
Nevertheless, analysis from Figures 2 and 3 involved a small number of studies. For example, in Figure 3, the subgroup of bulimia presented four studies and all other subgroups were composed of fewer studies. Also, Figure 2 included only three studies. The small number of studies analyzed can decrease the statistical power of the tests. In this research it can be observed that some results (Figures 2, 3 and 4) present large confidence intervals, decreasing the statistical power and the precision of the estimated population effect size [37]. One possible explanation for the low statistical power resulting from such intervals may rely on the small sample of some studies.
Furthermore, the methodological quality can influence the meta-analysis interpretation [12]. The studies included in Figures 2 and 3 were very heterogeneous. Figure 3 included studies with quality ranging from 5 to 9 points; Figure 2 included two studies that scored 3 points [26,27] and one that scored 5 points [25] (Tables 2 and 4). It is well known that the type of quality assessment scale used affects the analysis and the conclusions of meta-analytic studies, therefore the use of other quality scales could have resulted in different outcomes [12].
Several studies [4,[19][20][21][31][32][33][34][35] which were conducted in different settings and using different methods showed similar findings related to the possible causal relationship between EDs and TE. Specificity, temporality, and dose-response relationships were also observed. Some studies observed that the years of exposure to EDs and vomiting practices influenced the risk of TE [35,36]. Nevertheless, only cross-sectional and case-control studies were included in this systematic review. No cohort study was found. This highlights the need of prospective cohort studies, which could provide the highest strength of association to confirm this scientific evidence.
The association between the ED risk behavior and the presence of TE was investigated by three studies [5,6,29]. One of the studies [5] did not enter the meta-analysis, as it was excluded during the sensitivity analysis. When analyzing the forest plot it can be observed that although the studies' meta-analysis showed a significant association between these variables, the confidence intervals were large, decreasing the statistical power [37]. Moreover, the studies' quality assessment revealed a high degree of heterogeneity; one study scored 3 points [29] and the other scored 6 points [6] (Tables 2 and 4). The methodological heterogeneity evaluated by the quality assessment scale can influence the meta-analysis interpretation [12] and the strength of evidence. Inferential causal conclusions associating the ED risk behavior and TE cannot be drawn due to the studies' cross-sectional design and the limited number of studies which were evaluated in the meta-analysis (Figure 4). The consistency related to the possible causal relationship between ED risk behavior and TE is weak. There are only 3 studies analyzing this issue and two of them were conducted in the same place [5,29]. Furthermore, by analyzing the literature it is still not possible to draw any conclusion related to the specificity, temporality, dose-response, experimental evidence, biological plausibility, coherence or analogy between both conditions considering only three cross-sectional studies. Up to the present time, there is still a lack of solid evidence for an increased TE risk due to EDs, once there is no low risk of bias reporting a large effect.
In the present systematic review it was not possible to evaluate the publication bias as there were not enough studies to be grouped in a funnel plot [17]. However, there seems to be a tendency to report a positive association between EDs and TE.

Conclusions
Purging practices seem to increase the risk of TE. Nevertheless, there is a significant lack of scientific evidence to fulfill the basic criteria of causation between both conditions. Moreover, the present systematic review does reveal that to date there is no solid evidence in support of the postulated causal role of EDs in the occurrence of TE.
It is important to conduct prospective cohort studies in this area in order to investigate such evidence. Special attention should be given to studies on ED risk behavior, as the main goal should be to detect subclinical cases and avoid the onset of EDs and further comorbidities.

Supporting Information
Table S1 List of titles selected for full text analysis and the reasons for exclusion.