Assessment of the association between periodontal disease and total cancer incidence and mortality: a meta-analysis

Methodology

This meta-analysis was reported according to the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) guidelines. The protocol was registered in PROSPERO (http://www.crd.york.ac.uk/PROSPERO) before the research was begun (CRD 42020183497).

Criteria for the inclusion of studies

The following criteria were applied: (1) studies published in English or Chinese; (2) studies on human subjects; (3) observational studies (including cohort and case-control studies)with a prospective or retrospective design; (4) studies with periodontal disease as the exposure and, the incidence or mortality associated with breast or prostate cancer, lung& bronchial or colorectal cancer, or total cancers as the outcome; and (5) studies providing relative risks (RRs), hazard ratios (HRs), or odds ratios (ORs) and corresponding 95% confidence intervals (CIs).

Literature search and selection

Databases: PubMed, Embase, Web of Science, China National Knowledge Infrastructure [CNKI] and Wanfang.

Search terms: (<Neoplasm= OR <Carcinoma= OR <cancer= OR <tumor= OR <tumour=) and (<periodontal diseases= OR <gum disease= OR <periodontitis= OR <gingivitis= OR <periodontal= OR <periodontium= OR <periodontal attachment loss= OR <periodontal pocket= OR <alveolar bone loss= OR <tooth mobility=).

Time: by to July 2020.

First, two authors (Kaili Wang and Zheng Zhang) independently retrieved studies from the databases using specific search terms. Subsequently, a hand search was performed by reading abstracts and key words, and studies that were eligible for the inclusion criteria were selected. Finally, a consensus regarding the final studies to be included was reached via a discussion or by consulting a third author (Zuomin Wang).

Data extraction

Kaili Wang sorted out the information for the subsequent classification of the included studies: first author, publication date, country/region, study design type, numbers of samples, age range of the study population, diagnostic criteria for periodontal disease and cancer, cancer type and outcomes, follow-up period, whether adjustments were performed for smoking, adjusted variables, and multiple adjusted HRs and corresponding 95% CIs. Next, Zheng Zhang checked the information and supplemented it if necessary.

Quality assessment

We evaluated the quality of the included studies using the Newcastle-Ottawa scale. This evaluation system includes three aspects: selection, comparability and exposure. In the selection part, we evaluated the following four aspects: whether the exposed cohort could basically represent the average of the community population, whether the non-exposed cohort was from the same source as the exposure cohort, whether the selection of the exposed cohort was reliable based on records, and whether no outcome events occurred in the study subjects before the study began. In the comparability part, we assessed whether the exposed and non-exposed cohorts were matched in terms of age and other confounding factors. In the exposure part, we assessed the following three aspects: whether there were reliable data or records, such as examination or medical records, for the evaluation of outcome events; whether the follow-up period was equal to or longer than 5 years; and the completeness of follow-up through factors, such as the loss to follow-up rate.

Based on the scores, the studies were defined as follows: <4 points, low quality; 4–6 points, moderate quality; and 7–9 points, high quality.

Classification of studies

First, the included studies were classified according to the study scope: breast, prostate, lung & bronchial, colon & Rectum, and total cancers. Next, according to the specific content and purpose of the studies, we divided them based on two aspects: cancer occurrence and cancer mortality. Of course, som studies involved both incidence and mortality.

Statistical analysis

We used the HR with the 95% CI to assess the association between periodontal disease and the risk of cancer incidence and mortality. The data extracted from the studies were analyzed using STATA software (version 12.0; Stata Corp, College Station, TX, USA). Heterogeneity was tested through the Q test (statistical significance was set at P < 0.1) and I² test (I ² ≥ 50% indicated significant heterogeneity) and was represented using a forest plot.

Subgroup analyses were conducted to explore the factors influencing heterogeneity. Subgroup analyses were stratified by study population, study design, dental status, follow-up period, and whether or not smoking was adjusted for. A sensitivity analysis was performed to test the stability of the results.

Begg’s and Egger’s tests were used for assessing publication bias. Begg’s test, also known as the rank correlation test, tests the correlation between effects and sample size. Egger’s test, also known as the linear regression method, establishes a regression equation based on the standard normal deviate and precision of each included study. The larger the intercept of the regression line, the higher the bias. A P-value of > 0.05 was considered to indicate no publication bias..

Search and selection results

The process of study selection is shown in Fig. 1. First, 3,324 records were collected from PubMed, 5,681 from Embase, 5,435 from Web of Science, 1,095 from Wanfang, and 1,576 from CNKI. Subsequently, 9,860 publications remained after the duplicates were removed, of which 72 were deemed eligible based on their abstracts and key words. Finally, 27 publications were included in the meta-analysis after 45 studies were excluded for several reasons. Duplicates were removed for the following reasons: review/commentary/letter (n = 13), repeat publication (n = 7), no available data (n = 22), and no full text (n = 3).

Characteristics of the included studies

Table 1 describes the studies included in the review. The articles were published between 2003 and 2020. Twelve studies were conducted in North America and six in China. Six studies were retrospective, and the others were prospective. Periodontal disease was diagnosed through a dental examination in 18 studies (Hujoel et al., 2003; Ahn, Segers & Hayes, 2012; Wen et al., 2014; Chrysanthakopoulos, 2016; Chung et al., 2016; Mai et al., 2016; Dizdar et al., 2017; Han, 2017; Lee et al., 2017; Sfreddo et al., 2017; Chou et al., 2018; Heikkilä et al., 2018; Hu et al., 2018; Michaud et al., 2018; Güven et al., 2019; Lu et al., 2019; Tai et al., 2019; Huang et al., 2020), among which the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) was used in five studies (Wen et al., 2014; Chung et al., 2016; Chou et al., 2018; Hu et al., 2018; Tai et al., 2019), and the definition criteria of periodontal pocket and attachment loss were used in 13 studies (Hujoel et al., 2003; Ahn, Segers & Hayes, 2012; Chrysanthakopoulos, 2016; Mai et al., 2016; Dizdar et al., 2017; Han, 2017; Lee et al., 2017; Sfreddo et al., 2017; Heikkilä et al., 2018; Michaud et al., 2018; Güven et al., 2019; Lu et al., 2019; Huang et al., 2020). The diagnosis was self-reported by patients based on questionnaires and tooth loss history in the remaining nine studies (Michaud et al., 2008; Arora et al., 2010; Mai et al., 2014; Freudenheim et al., 2016; Michaud et al., 2016; Momen-Heravi et al., 2017; Nwizu et al., 2017; Lee et al., 2018; Yoon et al., 2019), among which dentists validated the diagnosis in three studies (Michaud et al., 2016; Momen-Heravi et al., 2017; Lee et al., 2018). Six studies analyzed periodontal disease and cancer mortality (Hujoel et al., 2003; Ahn, Segers & Hayes, 2012; Mai et al., 2014; Heikkilä et al., 2018; Lu et al., 2019; Huang et al., 2020), and 22 studies analyzed periodontal disease and cancer incidence (Michaud et al., 2008; Arora et al., 2010; Wen et al., 2014; Chrysanthakopoulos, 2016; Chung et al., 2016; Freudenheim et al., 2016; Mai et al., 2016; Michaud et al., 2016; Dizdar et al., 2017; Han, 2017; Lee et al., 2017; Momen-Heravi et al., 2017; Nwizu et al., 2017; Sfreddo et al., 2017; Chou et al., 2018; Hu et al., 2018; Lee et al., 2018; Michaud et al., 2018; Güven et al., 2019; Lu et al., 2019; Tai et al., 2019; Yoon et al., 2019). Cancer diagnosis was based on ICD-9/10-CM in 15 studies (Hujoel et al., 2003; Arora et al., 2010; Ahn, Segers & Hayes, 2012; Mai et al., 2014; Wen et al., 2014; Chung et al., 2016; Lee et al., 2017; Momen-Heravi et al., 2017; Nwizu et al., 2017; Sfreddo et al., 2017; Chou et al., 2018; Heikkilä et al., 2018; Hu et al., 2018; Tai et al., 2019; Yoon et al., 2019), medical records and histological examination in 11 studies (Michaud et al., 2008; Chrysanthakopoulos, 2016; Freudenheim et al., 2016; Mai et al., 2016; Michaud et al., 2016; Dizdar et al., 2017; Lee et al., 2018; Michaud et al., 2018; Güven et al., 2019; Lu et al., 2019; Huang et al., 2020), and patient self-report in 1 study (Han, 2017). All the participants were aged above 18 years, and the follow-up period was more than 10 years in 15 studies (Hujoel et al., 2003; Michaud et al., 2008; Arora et al., 2010; Ahn, Segers & Hayes, 2012; Wen et al., 2014; Mai et al., 2016; Michaud et al., 2016; Dizdar et al., 2017; Lee et al., 2017; Momen-Heravi et al., 2017; Chou et al., 2018; Heikkilä et al., 2018; Michaud et al., 2018; Tai et al., 2019; Huang et al., 2020) and less than 10 years in 12 studies (Mai et al., 2014; Chrysanthakopoulos, 2016; Chung et al., 2016; Freudenheim et al., 2016; Han, 2017; Nwizu et al., 2017; Sfreddo et al., 2017; Hu et al., 2018; Lee et al., 2018; Güven et al., 2019; Lu et al., 2019; Yoon et al., 2019). Most studies collected information on whether the participants smoked, including how often and how much they smoked and the pack-years. Regarding smoking, eight studies (Wen et al., 2014; Chung et al., 2016; Dizdar et al., 2017; Chou et al., 2018; Heikkilä et al., 2018; Hu et al., 2018; Güven et al., 2019; Tai et al., 2019) did not perform adjustments for smoking; however, the remaining studies included smokers and took smoking into account when calculating the HRs.

Quality assessment

Table S1 shows the quality assessment of the included studies. Twenty-four studies scored seven or more stars, indicating that 88.9% of the studies were of good quality.

Association between periodontitis and cancer incidence

The relationship between periodontal disease and cancer incidence is shown in Fig. 2. The pooled estimate based on 11 studies revealed a modest association between periodontal disease and breast cancer incidence (HR = 1.26, 95% CI [1.11–1.43]), with greater heterogeneity (I² = 75.8%, P_Q = 0.000). Similarly, we found a modest association between periodontal disease and prostate (HR = 1.26, 95% CI [1.03–1.54], I² = 72.9%, P_Q = 0.001), lung & bronchus (HR = 1.30, 95% CI [1.13–1.50], I² = 63.7%, P_Q = 0.000), colon & rectum (HR = 1.18, 95% CI [1.03–1.36], I² = 79.0%, P_Q = 0.000), and total (HR = 1.14, 95% CI [1.08–1.20], I² = 67.4%, P_Q = 0.000)cancer incidence.

Next, we explored the source of heterogeneity through subgroup analyses with I² >50% and P < 0.1.

Sensitivity analyses of the association between periodontitis and cancer incidence

As shown in Fig. 3, the pooled HRs were the same, and the 95%CI overlapped with the previous effect size. This means that our results are very reliable.

Subgroup analyses of the association between periodontitis and cancer incidence

The subgroup analysis results are presented in Table 2.

The following were the observations regarding the 11 studies that were conducted on periodontal disease and breast cancer incidence. Four studies were from North America and showed an HR of 1.12 (I² = 0%, 95% CI [1.05–1.19], P = 0.001). Two studies were retrospective and showed an HR of 1.52 (I² = 0%, 95% CI [1.17–1.98], P = 0.002). The pooled HR in four studies in which the periodontal disease diagnosis was self-reported was 1.15 (I² = 0%, 95% CI [1.09–1.22], P = 0.001). Six studies with follow-up periods of more than 10 years showed an HR of 1.11( I² = 14.4%, 95% CI [1.03–1.19], P = 0.005). The pooled HR in seven studies in which smoking was adjusted for was 1.13 (I² = 7.7%, 95% CI [1.07–1.21], P = 0.001). The CIs were all on the right side. The results of the subgroup analyses were consistent with those of the full dataset analysis. We can state with certainty that periodontal disease increases the risk of breast cancer incidence.

The following were the observations in the seven studies on periodontal disease and prostate cancer incidence. Three studies were from North America and showed an HR of 1.03 (I² = 36.9%, 95% CI [0.89–1.18], P = 0.728). Six studies with follow-up periods of more than 10 years showed an HR of 1.16 (I² = 60.8%, 95% CI [0.97–1.38], P = 0.100).We found that periodontal disease increased the risk of prostate cancer incidence under certain conditions in the European people, dental examination diagnosis of periodontal disease, follow-up period of less than 10 years. However, the association was not affected by smoking.

Among the 14 studies on periodontal disease and lung and bronchial cancer incidence, eight studies were from North America and showed an HR of 1.34 (I² = 46.9, 95% CI [1.24–1.46], P = 0.001), and eight studies in which periodontal disease was diagnosed through dental examination showed an HR of 1.33 (I² = 75.7, 95% CI [0.94–1.88], P = 0.000). Similar to prostrate cancer, we found that periodontal disease increased the risk of lung & bronchus cancer incidence under certain conditions in the North American population, self-reported diagnosis of periodontal disease , follow-up period of more than 10 years, and adjustments for smoking, but the association was not affected by study design type.

Among the 13 studies on periodontal disease and colorectal cancer incidence, seven studies were from North America and showed an HR of 1.08 (I² = 18.2, 95% CI [0.98–1.19], P = 0.134), and five studies in which periodontal disease was diagnosed by dental examination showed an HR of 1.17 (I² = 26.4, 95% CI [0.92–1.48], P = 0.203). We found that periodontal disease increased the risk of colorectal cancer incidence under certain conditions in retrospective study, self-reported diagnosis of periodontal disease , follow-up period of less than 10 years, adjustments for smoking, but the association was not affected by study design type.

Among the 12 studies on periodontal disease and total cancers incidence, three studies were from Asia and showed an HR of 1.09 (I² = 92.6, 95% CI [0.94–1.27], P = 0.243), and eight studies with adjustments for smoking showed an HR of 1.13 (I² = 0, 95%CI [1.09–1.17], P = 0.001). We concluded that periodontal disease increased the risk of total cancer incidence under certain conditions in the North American and European population, prospective study design, but the association was not affected by the diagnostic method for periodontal disease, follow-up period, adjustment for smoking.

Association between periodontitis and cancer mortality

We also analyzed the relationship between periodontal disease and cancer mortality (Fig. 4). A relationship between periodontal disease and total cancer mortality was observed in four studies, and the HR ranged from 1.24 to 1.58; the pooled HR was 1.40, with no heterogeneity (I² = 0, P_Q = 0.718), indicating that the result is reliable. Although the results of current meta-analysis suggest that periodontal disease may increase the risk of breast (HR = 1.25, 95%CI [0.83–1.88], I² = 0.0%, P_Q = 0.809), prostate (HR = 1.78, 95%CI [0.92–3.46], I² = 0.0%, P_Q = 0.962), lung & bronchus(HR = 1.48, 95%CI [0.93–2.35], I² = 50.8%, P_Q = 0.154), colon & rectum (HR = 1.66, 95%CI [0.44–6.27], I² = 76.7%, P_Q = 0.083) cancer mortality, the overall estimate was not significant because of high levels of uncertainty or moderate heterogeneity. However, we did not perform a subgroup analysis because there were only two studies.

Publication bias in meta-analysis

Finally, we performed Begg’s and Egger’s tests to evaluate publication bias. The results of the two tests are shown in Table 3. We conducted Begg’s test in the case of missing items; the P values of the test results after a continuity correction based on Kendall’s score in the remaining studies ranged from 0.133−1.000. All the P values were greater than 0.05, indicating that there was no publication bias. We also conducted Egger’s test in the case of missing items, the P values of the results of assessing whether the intercept of the remaining items was ranged from 0.147−0.860. All the P values were greater than 0.05, indicating no publication bias. The results of Begg’s and Egger’s tests were consistent, indicating that there is no publication bias in this meta-analysis.