Evidence of past dental visits and incidence of head and neck cancers: a systematic review and meta-analysis

We identified all the published studies using an extensive search of the PubMed, CINAHL, and the Cochrane database from the inception of relevant database until August 2017. The following search terms were used with Boolean operators to combine searches: (“oral cancers” OR “cancers of head and neck” OR “cancer of tongue” OR “cancer of oropharynx” OR “cancer of hypopharynx” OR “cancer of esophagus OR cancers of the UADT” AND “dental visits” OR “visits the dentist” OR “dental check-up” OR “dental examination” OR “dental treatment” OR “dental care” OR “oral hygiene” OR “periodontitis”) with no limitations on year of publication and language (Table 1). A health librarian reviewed and provided input on the search strategy. Additional search strategies included (i) a hand search of the reference lists of included studies, (ii) the use of the “related citations” feature in PubMed, (iii) an ongoing surveillance of the literature while updating the manuscript, and (iv) authors were contacted for the articles for which full text was not available. EndNote software was used to remove the duplicates for the same type of article in more than one database. Alerts with search strategies were created in the databases to maintain an ongoing surveillance of the literature.

To be eligible for inclusion, the paper had to report a primary study with any population, one or both genders specified, participants of any age, incidence of any cancer subsite of HN, and UADT reported as the health outcome, frequency of dental visits/check-ups prior to the diagnosis of one of these cancers (assessed as the exposure), and availability of sufficient data to estimate the measure of association, i.e., unadjusted odds ratio (OR) along with its corresponding 95% confidence interval (CI). Where a single study was described in several publications, the study which reported the incidence data most comprehensively was included in the analysis. Gray literature, such as unindexed or unpublished conference proceedings, pre-prints, and state of art reports, were not included due to limited resources to access the same.

Two reviewers (B.G. and N.K.) independently screened the title and abstract of the identified citations. Full texts of citations judged as potentially eligible were acquired by at least one of the two reviewers. Thereafter, both the reviewers used a standardized and pilot-tested form to independently screen every full text for eligibility. Disagreements were resolved by consensus among the authors. Data extraction from individual studies included information on first author’s last name, year of publication, region of study, study design, number of cases and controls (or number of participants and events), population characteristics (gender and age), exposure definition (frequency or reason of dental visits which were defined as never/only when in pain, every 6 months or less and every 6–12 months, less than once a year and more than once a year, 1–2 in a year, 3–5 in a year and > 5 in a year, ≥ once every 5 years versus < once every 5 years, never versus yes, no regular/special dental care versus regular/special dental care), definition of cancer site, its subsite and its diagnostic/confirmation criteria, method of selection of controls (hospital/population based), adjusted covariates in the regression model, and risk measure in each reviewed article.

Studies were assessed for methodological quality using the quality assessment tool for quantitative studies developed by the Effective Public Health Practice Project (EPHPP) [16]. This tool has six components (selection bias, study design, confounders, blinding, data collection method, withdrawals, and dropouts). Based on this criterion, a global rating as strong, moderate, or weak was assigned to each study based on the no weak rating, one weak rating, and two or more weak ratings for any of the above mentioned six components, using the criteria described in the EPHPP dictionary itself.

A random effects model was employed in all meta-analysis procedures which produces results that generalize to a range of populations and to different study designs in addition to accounting for heterogeneity between studies. Forest-plots were used to demonstrate the effect of each study and the summary effect size. For effect size estimates, standard errors of its logarithm were calculated from the reported or estimated CIs, assuming that the effect size was log-normally distributed. The logarithms of the effect sizes and their corresponding standard errors formed the data points for random effects meta-analysis. For each analysis, heterogeneity was assessed using by Cochran’s Q statistic (measure of weighted square deviations), with N-1 degrees of freedom (where N is the number of studies), results of statistical test based on Q statistic, between studies variance (T2), and ratio of the true heterogeneity to total observed variation (I²). We conducted sensitivity analysis by dropping one study at a time and examining its influence on the summary effect estimates. To investigate publication bias, funnel plots were constructed, plotting the logarithmically transformed ORs against the standard error of the associated log OR. The distribution of study risk estimates across the funnel plot was examined visually and Egger’s test for small study effects was performed to assess the degree of asymmetry. Comprehensive meta-analysis software was used for all analyses [17]. Unadjusted effect estimates were used in the meta-analysis as the confounding variables used in the multivariate regression model varied significantly between studies. However, unadjusted OR could not be computed for two studies due to limited data on number of cases and controls [18, 19].

The classification of exposure variable (history and frequency of past dental visits prior to the diagnosis of cancer for cases and disease/date of interview for the hospital or the population based controls) differed between studies, with few studies reporting more than two categories. In the second situation, a single effect estimate was computed by comparing the frequency of dental visits in highest category versus the pooled data from the other categories.

Studies that evaluated cancers of the HN/UADT and oral cavity were pooled in two separate evaluations. Some studies presented data on both HNCs as well as OCs and were included in both evaluations. Two different meta-analyses were conducted based on the health outcome as reported by the authors in the articles: one for HNCs/UADTCs and the other meta-analysis was conducted based on OCs exclusively. To report the pooled effect of frequency of dental visits on incidence of cancers, the exposure variable was categorized as yes/regular/frequent (subjects in highest category of dental visits in each study) and never/irregular/not frequent (other dental visits categories of each study). A third meta-analysis was executed by pooling the data from all the studies irrespective of the cancers with the frequency of dental visits as subgroup. Three studies could not be included in any of the other subgroups and shared individual identity for subgroups.

Overall, this systematic review includes 32 case-control and 6 other design studies not limited to cross-sectional, observational, case-series, and screening. Fourteen of these studies were conducted in Europe [20‐33]. There were nine studies conducted in North America [1, 2, 4, 18, 34‐38], five in South America [39‐43] nine in Asia [44‐52], and a single study in Australia [53]. For the meta-analysis, 16 case-control studies had hospital-based controls [2, 20, 23, 27, 28, 30, 36, 39, 41, 43‐45, 48, 50, 52, 54, 55], seven had population-based controls [18, 24, 29, 34, 35, 47, 56], and two studies had both hospital- and population-based controls [22, 40]. There was one study on OCs conducted in India, and this did not provide the source, nor describe the type of controls [49].

Among the studies included in the meta-analysis, 11 were based on HNCs and 13 studies included OCs as the health outcome (details of same available in Tables 2 and 3). Only two studies included both HNCs and cancers of the esophagus [22, 24]. The studies with HNCs/UADTCs represented a total sample size of 38,552 including 17,313 cases and 21,239 controls. The studies with OCs represented a total sample size of 22,542 including 10,982 cases and 22,542 controls. All the studies included both males and females with one exception which included only females, this being conducted in China with OCs as the health outcome [44]. The overall age group of the study participants ranged from < 20 to ≥ 80 years. Dental visits as never/only when in pain or less than once a year and more than once a year were reported in 11 studies. One study reported frequency of dental visits as 1–2 in a year, 3–5 in a year, and > 5 in a year. Never, only when in pain, and regular dental visits were reported in one study. Never, every 6 months or less, and every 6–12 months dental visits were reported in three studies. Dental visits as never, < every 5 years, every 2–5 years, and at least every year were reported in three studies. Never, ≥ once every 5 years versus < once every 5 years dental visits were reported in two studies. Dental visits as never versus yes were reported in four studies. No regular/special dental care versus regular/special dental care was used in one study.

Majority of the studies (29) were of strong quality. Moderate quality was assessed for seven studies and three studies were assessed as weak (Table 4).

The odds of never/irregular/not frequent dental visits as compared to yes/regular/frequent dental visits were greater among the cases as compared to controls. Never/irregular/not frequent dental visits increased the risk of cancer significantly. Under the random effects model, the overall pooled estimate risk for cancer was (OR 2.24; 95% CI 1.89, 2.65, P < 0.001). The test for heterogeneity produced Tau square of 0.00, Q = 38.25, I² = 34.63%, test for overall effect z = 8.81, (P < 0.001). The highest risk estimates observed were (OR 11.89; 95% CI 3.33, 42.51, P < 0.001) in a study conducted in Poland from 1997 to 2000. However, the wide CIs indicate the small sample size of the study. We did not find any statistically significant results for three studies (Fig. 2).

The odds of never/irregular/not frequent dental visits as compared to yes/regular/frequent dental visits were higher among the cases as compared to controls indicating a statistically significant increase in the cancer. Under the random effects model, the overall pooled estimate risk was (OR 1.93; 95% CI 1.47 to 2.52, P < 0.001). The test for heterogeneity produced Tau square of 0.00, Q = 15.96, I² = 24.83%, test for overall effect z = 4.76, (P < 0.001). Highest risk was found in a study conducted in Taiwan on 287 cases and 296 controls (OR 6.47; 95% CI 3.78 to 11.09, P < 0.001) (Fig. 3). We also observed that yes/regular/frequent dental visits showed protective effect and decreased the incidence of OC by 52%. The overall pooled estimate risk for cancer was (OR 0.48; CI 0.38 to 0.60, P < 0.001), data not shown.

The studies were divided into various subgroups according to reported frequency of dental visits in respective studies as follows: (Never/≤ 6 months, > 6 months); (Never, < once a year, ≥ once a year); (Never, yes); (Special dental care); (only in pain, no visits); (1–2 visits a year, 3–5 visits a year). Figure 4 illustrates the subgroup analysis by frequency of past dental visits in these studies. The overall pooled estimate risk was (OR 2.01; 95% CI 1.76 to 2.30, P < 0.001). The test for heterogeneity produced Tau square of 0.00, Q = 36.33, I² = 31.76%, test for overall effect z = 9.24, (P < 0.001).

The symmetrical funnel plot in Fig. 5a–c by visual inspection for frequency of dental visits and incidence of HNCs/UADTCs and OC indicates that there was no publication bias in our meta-analysis. The Egger’s regression intercept was − 0.9849, standard error = 0.8461, 95% CI − 2.73 to 0.76, t = 1.16, df = 24, and P = 0.2558. Publication bias for studies conducted on OCs shows Egger’s regression intercept = − 1.3239, standard error = 1.7850, 95% CI − 5.25 to 2.60, t = 0.7416, df = 11, and P = 0.474. For all the studies included in meta-analysis and grouped by the frequency of dental visits: the Egger’s regression intercept was − 0.889, standard error = 0.8946, 95% CI − 2.73 to 0.9941, t = 0.9940, df = 24, and P = 0.330. Results of meta-regression analysis indicate that none of the included subgroups for history of dental visits influenced the effect estimate (Table 5).

Due to limited resources, this review searched and included studies from three large commonly used databases only. Greater availability of resources and access may have allowed a wider search yielding more studies (published and unpublished) via other databases, web resources, and gray literature. There is also a possibility of having excluded some studies which reported/published insufficient data to meet the inclusion criteria of this review when these studies may have had adequate unpublished data. However, it was beyond the scope of this review to trace this data.

Though most of the studies were of strong quality, they differed in nature of population, study setting, cancer case definition, method of diagnosis of cancer, selection and type of controls, history of frequency of dental visits, method of data collection, and its analysis. Also, the positive association between never/irregular dental visits and incidence of cancer could be understated in our pooled estimate as this meta-analysis is conducted on case-control studies where recall and selection bias remains a concern. Another potential limitation is the varied distribution of frequency of dental visits among the studies. Unfortunately, not all dentists consider it mandatory to undertake a comprehensive examination of the oral and oropharyngeal soft tissues at the time of visits made for issues involving strictly dental problems, such as toothache or the need for dental restorations.

Please refer to Table 6.