The risk of tooth loss in patients with diabetes: A systematic review and meta‐analysis

Abstract Aim The aim of this systematic review was to comprehensively and critically summarize and synthesize the risk of losing teeth among with diabetes mellitus (DM) compared to those without DM, as established in observational studies. Materials and methods MEDLINE‐PubMed and Cochrane databases were searched through a period from their inception through October 2020 to identify eligible studies. Papers that primarily evaluate the number of teeth in DM patients compared to non‐DM individuals were included. A descriptive analysis of the selected studies was conducted, and when feasible, a meta‐analysis was performed. The quality of the studies was assessed. Results A total of 1087 references were generated, and screening of the papers resulted in 10 eligible publications. A descriptive analysis demonstrated that six of these studies indicate a significantly higher risk of tooth loss in DM patients. This was confirmed by the meta‐analysis risk ratio of 1.63 95% CI (1.33; 2.00, p < 0.00001). Subgroup analysis illustrates that this is irrespective of the risk‐of‐bias assessment. The higher risk of tooth loss in DM patients was also higher when only DM type II patients or studies with a cross‐sectional design were considered. Patients with a poor DM control status presented a significantly increased risk of tooth loss. When the data were separated by the world continent where the study was performed, Asia and South America had numerically higher risks and a 95% CI that did not overlap with Europe and North America. Conclusion There is moderate certainty for a small but significantly higher risk of tooth loss in DM patients as compared to those without DM.


| INTRODUC TI ON
Tooth loss considerably affects oral health-related quality of life (OHRQoL), causing chewing difficulty, poor dietary intake and functional disorders. 1 A predominant reason for tooth loss is periodontitis, which is an inflammation of periodontal tissues. Damage from periodontal disease can lead to loosening of teeth and, in a final stage, to tooth loss. 2,3 The manifestation and progression are influenced by a wide variety of determinants and factors that have been linked with general health. Notably, the association between periodontitis and diabetes mellitus (DM) has been highlighted in the literature. Periodontal disease is considered the sixth complication of DM. 4 Another primary cause of tooth loss is dental caries. Its development of which is presumably enhanced in DM patients. 5,6 Due to the ageing population, DM is a growing public health problem, and it likely contributes to a greater demand for health care. 7 The negative effects of elevated blood sugars on the immune system result in an increased susceptibility to infections. 8 The risk for development and progression of periodontitis is increased approximately threefold in DM patients as compared to non-diabetic individuals (non-DM). 9,10 Furthermore, DM is associated with increased severity of periodontal disease. 11 The increased risk of dental caries in DM patients can likely be explained by decreased salivary flow rates 12 and expanded levels of glucose in the saliva. 13 The American Diabetes Association and International Diabetes Federation have published DM care guidelines, 7,14 of which the main goal is prevention and treatment of DM complications, thereby optimizing quality of life (QoL). 14 Periodontal pocket depth and clinical attachment loss are commonly utilized to define a patient with periodontitis. 15 However, these outcome measurements are surrogate endpoints of disease. A true endpoint (e.g., tooth loss) would directly assess patients' experience on the onset of periodontitis.
Moreover, tooth loss also affects QoL. 1 A recent systematic review (SR) and meta-analysis assesses predictors of tooth loss, including DM, in periodontitis patients. 16 However, no SR with a specific focus on the risk of tooth loss in DM patients has yet been performed. In the light of the increasingly available evidence, the aim of this SR is to comprehensively and critically summarize and synthesize the available scientific evidence emerging from observational studies on the number of teeth among DM patients as compared to non-DM patients.

| ME THODS AND MATERIAL S
The preparation and presentation of this SR is in accordance with the Cochrane Handbook for Systematic Reviews 17 and the guideline for Meta-Analysis of Observational Studies in Epidemiology (MOOSE). 18 A protocol was developed a priori following the initial discussion between the members of the research team. This study is registered at the ACTA University Ethical Committee by number 2021-71228.

| Focused question
A precise review question was formulated utilizing the population, exposure, comparison, outcomes and study (PECOS) framework as follows 19 : -Is there a higher risk, loosing teeth among patients with DM compared to those without DM, as it was established in observational studies?
-Due do a potential link between DM and both caries and periodontitis, it is hypothesized that DM patients are at higher risk, loosing teeth. for appropriate papers that answer the focused question. Table 1 provides details regarding the search approach employed. For the search, no limitation was applied on language or date of publication.  ○ Clinically determined number of teeth (no radiographs).

| Search strategy
○ Number of missing teeth or number of teeth present as an absolute number of teeth or as a population mean.
○ Tooth loss presented as cross-sectional data for an individual over the lifetime until the moment of assessment (not for the duration of a specific period).
Any disagreement between the two reviewers about the eligibility of studies was resolved after additional discussion. If disagreement persisted, a third reviewer, G.A.W., was consulted, whose judgement was considered to be decisive. Thereafter, the selected full-text papers that fulfilled all eligibility criteria were identified and included in this SR for data extraction and estimation of the risk of bias. At this stage, the reasons for exclusion were recorded (see online Appendix S1). The application of the ROBINS-E tool consists of the following steps:

| Methodological quality assessment
-Step I: framing the review question, describing potential confounders, co-interventions and exposure and outcome measurement accuracy information. - Step II: describing each eligible study, including specific confounders and co-interventions for each study. - Step III: determining risk-of-bias consideration through seven items regarding the strengths and limitations of studies.
Quality was assigned as low risk of bias, moderate risk of bias, serious risk of bias, critical risk of bias or no information with the following domains: bias due to confounding, bias in selection, bias in classification, bias due to departures from intended exposures, bias due to missing data, bias in measurement of outcomes and bias in selection of reported results.
The judgements within each domain are carried forward to an overall risk of bias. A study was classified as having a low risk of bias when all domains were judged to be at low risk of bias.
Moderate risk of bias was assigned when, for one or more domains, the study was judged not to be higher than moderate risk of bias. A study was classified as having serious risk of bias when, for one or more domains at the most, serious risk of bias was scored. An overall critical risk of bias was scored when at least one domain was judged to be at critical risk of bias. No information was assigned if the study was judged to be at serious or critical risk of bias and there was a lack of information in one or more key domains. 20-22

| Data extraction
For those papers that provided insufficient data to be included in the analysis, the first or corresponding authors were contacted by email to query whether additional data could be provided. When an included study provided multiple age groups of individuals 18 years and older, data were merged so that these were considered as one group. If a DM group was specified in the categories of prediabetes and DM, then the prediabetic data were excluded. When DM types I and II are presented separately in the original included papers, these groups were merged for the overall analysis. If possible, a subgroup analysis on DM types I and II was performed if the original group data allowed for separation of these two groups.

| Assessment of clinical and methodological heterogeneity
The factors utilized to assess the clinical heterogeneity of the outcomes of the various studies are as follows: -Characteristics of participants: age, gender and continent.
-Evaluable number of teeth.
-DM type: I or II.
-Method of assessment: professionally diagnosed or self-reported DM. 23 Factors employed to assess the methodological heterogeneity were study design details and the total number of evaluated teeth, reference point (28 or 32).
When clinical or methodological heterogeneity was presented across studies, sources of heterogeneity were investigated with subgroup or sensitivity analyses. 17 As the total number of evaluable teeth (28 or 32) has a direct influence on the relative ratio of the missing teeth to the total number of teeth, this was defined a priori as a reason for subgroup analy- Sensitivity analyses were undertaken to evaluate the effect of excluding studies based on specific aspects in the domain of clinical or methodological heterogeneity. The testing for publication bias per outcome was utilized as proposed by Egger et al. 27 If the meta-analysis involved a sufficient number of trials to make visual inspection of the funnel plot meaningful (a minimum of 10 trials), then these plots were employed as tools to assess publication bias.
The presence of asymmetry in the inverted funnel is suggestive of publication bias. 17  considering and reporting each of these factors. An overall rating of confidence in effect estimates was considered critical for the final recommendation. 32 Any disagreement between the two reviewers was resolved after additional discussion. If a disagreement persisted, then the judgement of a third reviewer (G.A.W.) was decisive.

| Search and selection process
Searching the MEDLINE-PubMed and Cochrane databases resulted in 1087 unique papers, as Figure 1 illustrates.
The first screening of the titles and abstracts resulted in 27 papers for which the full papers were obtained. In the second phase, after full-text reading and contact with the corresponding authors, 16 studies were excluded the reasons for which are presented in online Appendix S1. Three papers do not provide necessary data regarding the overall number of missing teeth, and after contacting the authors, this information could not be retrieved (Wiener et al 2017, 33 Kapp et al 2007, 34 Jung et al 2010). 35 41 concern the same study population.

| Assessment of clinical heterogeneity
Considerable heterogeneity was observed among the 10 included studies. Characteristics of study design, study population and diagnostic as well as assessment methods are presented in The female gender is more prevalent in seven studies (I, II, IV, VI, VII, VIII and X), and two studies include more males (V and IX).
One case-control study makes an effort to match the gender dis- reports DM based on both clinical assessments and self-reports.
In one paper, it was unclear how the DM status had been assessed (X). 46 In total, three studies specifically focus on DM type II (I, 42 III 40 and VIII). 48 One paper differentiates between types I and II (VII). 44 For the overall calculations, data from these groups were merged, while for the subgroup analysis, the original group data were em-

| Assessment of methodological heterogeneity
Eight of the included observational studies utilize a cross-sectional design (I, 42 IV, 47

| Methodological quality assessment
A summary of the methodological quality and potential risk-of-bias scores is presented in Table 3. Detailed quality assessment for each included study is provided in online Appendix S2.
Based on a summary of the bias assessment domains, the estimated potential risk of bias is low for two studies: II 43

| Study results
From the included studies, the overall DM population consisted of 5699 patients and the non-DM controls of 23.579 individuals. The overall prevalence of DM in the included cross-sectional studies is 16.8%.

| Statistical heterogeneity
Considerable heterogeneity was observed in the meta-analyses; for details, see Tables 5 and 6.
This implies a variation between studies due to heterogeneity rather than chance. To explore heterogeneity, a subgroup analysis was performed to attempt to explain the variation in effects.
Subgroup analysis on the evaluated number of teeth, either 28 or 32, revealed an overlap for the 95% CI and with the overall 95% CI.

| Publication bias
Testing for publication bias was possible for the overall analysis, which is presented in Appendix S4. The funnel plot reveals that almost all outcomes are located at the top of the funnel, suggesting that no studies concerning small populations were included.
Furthermore, the distribution is asymmetrical around the overall value. Consequently, it is presumed that a potential risk for publication bias may exist. to the magnitude of the risk for tooth loss. In summary, this SR is based on 10 observational studies ( Figure 1) and the potential risk of bias was estimated as low to serious (Table 3 and Appendix S2).

| Evidence profile
Because data from studies were derived from different populations and world continents, the findings are considered to be generalizable. Based on the heterogeneity between the included studies, data were judged to be rather inconsistent (see Table 2). The data were considered to be rather precise, because all selected studies focused on tooth loss as a primary outcome and because the majority reveal an overlap in the overall 95% CI (see Figure 2, Tables 5 and 6 and online Appendix S3). As publication bias may be present and the funnel plots indicate that outcomes could be overestimated, the presence of reporting bias is likely. The interpretation of the overall RR being 1.63 is that it concerns a small effect. 51 Considering all GRADE aspects, the evidence profile that emerges from this review is that the strength is moderate.

| DISCUSS ION
The present review summarizes the available body of dental and

| Selection choices made
The selection process of the included papers of this SR deviates from the traditional Cochrane approach. 17 However, the foundation is based on similar principles. A two-step approach was utilized: first, screening of titles and abstracts was performed; second, more specific inclusion criteria were implemented to ensure that the only studies included presented data about tooth loss among DM patients and non-DM individuals as the primary outcome. The reviewers are aware that there may be additional information available where data on diabetic status and number of teeth are retrieved from reported demographic data and presented as an interesting result. [54][55][56] Inclusion of these data may introduce a reporting bias that affects the conclusion drawn 57 ; therefore, it was specifically prespecified that primary outcomes from the study protocol should be included in the final data presentation. The inclusion of reported outcomes should not be based on a selection of results that were not the primary focus of the study. 58 From a statistical perspective, the sample size of the included studies should have been driven by the primary outcome, which positively affects the power. Consequently, for the present SR, only papers with tooth loss and DM as the primary focus of the original study were sought, and these two aspects had to be mentioned as the aim in the abstract or title. With this approach, it was considered that the most reliable and valid estimation of the RR was obtained.

| Diabetes mellitus comorbidities
For this SR, only DM without reported comorbidities was considered. Papers on participants with other systemic diseases were excluded 59,60 to avoid bias in the observed association between DM and tooth loss. However, DM has many risk factors, such as age, overweight and obesity, inactivity, habitual smoking, food intake, socio-economic status, family history of DM, geographical region and blood pressure. 61 The included papers did not adjust for these factors. Only in one paper (V 50 ) was smoking specifically mentioned: none of the DM patients reported being smokers, and only nonsmoking non-DM individuals were considered as a control group. A TA B L E 4 A descriptive summary of statistical significance levels of the difference between DM patients compared to non-DM with regard to number of teeth range of predictors for tooth loss in periodontitis patients has been reported. A recent SR assesses the consistency and magnitude of different predictors, concluding that age, non-compliance, smoking, DM, teeth with bone loss, high probing pocket depth, mobility and molars, especially with furcation involvement, demonstrate a higher risk of tooth loss. 16 Considering the above, there appears to be an overlap of potential causal components for tooth loss in diabetics and periodontitis with the following factors: age, smoking habit and diabetic status. In future studies, it is recommended to include these factors in the analysis. Because the eligible studies of the present review did not report or take these into consideration, the reported outcome allows only for the interpretation of an unadjusted effect size. From the obtained observational data, it is also not possible to make causality claims. As stated earlier, geographical region, gender, type of DM and type of assessment may interfere in the DM and tooth loss association.

| Reporting bias
The main origin of publication bias is failure to publish negative outcomes or null findings. Additionally, it is more difficult to publish papers in which no differences between groups are found. 29,62 The consequences are that this may lead to overestimation of exposure as deducted based on the meta-analyses. 63 The present funnel plot (see online Appendix S4) illustrates that almost all outcomes were located at the top of the funnel, suggesting that relatively few small studies were included. The usage of a strict inclusion criteria may explain this specific distribution. It is recognized that studies with small sample sizes that fail to establish a difference between groups either have not been published or have difficulties in being published in impact factor journals. 62

| Type of diabetes
As prediabetes may be reversible, 64 data from these participants were not considered, as only one study (II 43 ) was available.
Gestational diabetes consists of high blood glucose only during pregnancy 65 and was consequently not analysed in the present review.
Type I diabetes can develop at any age but occurs most frequently in children and adolescents. However, type II DM is more common in adults and accounts for approximately 90% of all diabetes cases. 66 Three of the included studies specifically focus on DM type II (I, 42 III 40 and VIII 48 ). Only one paper (VII 44 ) differentiates between types I and II. It was therefore not possible to perform a subgroup analysis to compare types I and II in this dataset. Analysis focused on DM type II, for which a RR of 1.56 for the risk of tooth loss was found.
However, the relationship between DM type II and tooth loss is complicated by the fact that the disease onset generally occurs in middle and late ages, coinciding with the time that periodontitis becomes more prevalent. 44 Nevertheless, studies focusing on type I DM patients also indicate an increased risk of periodontitis compared to TA B L E 5 Overview (sub) analysis overall and evaluable number of teeth (28/32)  Considerable heterogeneity was observed in the outcomes of most sub-analyses; however, sub-analysis on diabetes type II did not provide an explanation for the high level of heterogeneity.
Only the subgroup analysis on diabetic status being either poorly or well-controlled revealed a low level of statistical heterogeneity (0%-23%). This could indicate that diabetic control is an aspect that contributes to heterogeneity among study outcomes. However, this sub-analysis was based on only two studies that had similar populations and study designs. Because this study's meta-analyses indicated a heterogeneity in the outcome, the reader should exercise caution in utilizing the RR as the exact measure of the risk for tooth loss. see Figure 2 and studies with either 28 or 32 evaluable teeth were not separated (see Table 6 as well as online Appendices S3-1 and S3-5). In the cases in which wisdom teeth are included in the evaluation, prophylactic removal should be considered as a reason for extraction. This aspect was not analysed in the selected studies that evaluate 32 teeth.

| Gender
Seven of the included papers feature more females than male participants, while DM type II is more common in males than females. 78 Females generally have a greater knowledge and more positive attitude than males towards oral health behaviour. 79 This is associated with a reduced risk for the progression and severity of periodontitis. 80 The skewed gender distribution towards females could cause underestimation of the outcome for this SR. TA B L E 7 GRADE evidence profile for the number of teeth and risk ratio among DM as compared to non-DM

#10 #10
Risk of bias (Table 3, Appendix S2) Low to serious Consistency (Table 2 • The language restriction to English resulted in three potential studies that had to be excluded. Two were in Spanish, 82,83 and one was in Hungarian. 84 Based on the information provided in the English abstract, it appears that in these three studies, tooth loss was greater among DM patients as compared to non-DM individuals. These results corroborate the present findings.
• Caries and periodontitis are the predominant reasons for tooth loss. None of the included studies provided details that could help discern what the indications for extraction had been.
• Factors such as differentiation between DM types I and II, type of assessment (self-report or professional), gender and age may have

| CON CLUS ION
There is moderate certainty evidence for a small but significant higher risk of tooth loss in DM patients as compared to those without DM.
Subgroup analysis showed that this was also higher if only DM type II was considered. If the data were separated by the world continent where the study was performed, analysis showed that the magnitude of the risk was particularly higher in Asia and South America.

| Scientific rationale for the study
Diabetes mellitus (DM) is a chronic inflammatory disease. Evidence supports an increased risk for periodontal diseases and incidence/ severity of caries in DM patients. Both are primary sources of tooth loss. It has not been systematically being reviewed whether DM is associated with a higher risk of tooth loss compared to non-DM individuals.

| Principal findings
Diabetes mellitus patients have a significantly higher risk of tooth loss than in non-DM individuals.

| Practical implications
Diabetes mellitus patients shall get attention on oral disease prevention by the dental care practitioners. They are at increased risk of tooth loss, which in particular applies to DM patients from Asia and South America.

ACK N OWLED G EM ENTS
The

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflicts of interest.

E TH I C A L A PPROVA L
Ethical approval was not required. This study is registered at the ACTA University Ethical Committee by number 2021-71228.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data derived from public domain resources. The data that support the findings (the seven included studies) of this study are available from search databases PubMed/Medline or Cochrane-CENTRAL.
These data were derived from resources available in original papers that are published in the public domain. Some first or corresponding authors of inculded papers were contacted for additional data.