Studies included for the meta-analysis if they referred to observational studies including cross-sectional, case–control, cohort and nested case–control studies. The article should estimate the association between PD and MI and provide the odds ratio (OR) or relative risk (RR) with the corresponding 95% confidence intervals (CIs).

Pubmed, EMBASE and the Cochrane Library Databases were searched to identify relevant studies through July 2016. The medical subject headings (MeSH terms) as well as free text terms were used in the search strategy. The terms applied were as follows: ("periodontal disease" OR "periodontitis" OR "periodontal" OR "periodontal attachment loss" OR "periodontal pocket" OR "alveolar bone loss") AND ("myocardial infarction" OR "acute myocardial infarction" OR "acute coronary syndrome" OR "cardiovascular disease" OR "coronary heart disease" OR "unstable angina") (Additional file 1: Table S1). The search was restricted to studies conducted on human subjects and published in English and Chinese. The bibliographies of the articles found were hand searched for possible additional studies.

Two reviewers (S. Xu and X. Liu) screened the titles and abstracts of the search results and articles were excluded if they did not meet the above-mentioned inclusion criteria.

Newcastle-Ottawa Quality Assessment Scale was used to assess the quality of the observational studies, including case–control and cohort studies [27] and the Agency for Healthcare Research and Quality (AHRQ) for cross-sectional studies [28]. For case–control and cohort studies, the quality was ranked as followed: low quality = 0-4; moderate quality = 5-7; high quality = 8-9.

To rank the cross-sectional studies, the AHRQ was quantified in the following way. An item would be scored ‘0’, if it was answered ‘NO’ or ‘UNCLEAR’. If it was ‘YES’, then this item was scored ‘1’. Article quality was ranked as follows: low quality = 0-3; moderate quality = 4-7; high quality = 8-11 [29].

PD was defined to include self-report diagnosis, or any measure of disease according to clinical, radiographic and microbiological assessment (including pocket probing depth, attachment loss, bleeding on probing, plaque index, gingival index, X-ray and microbiological results). At the very beginning, all related OR/RR based on different PD assessments were extracted. The one selected for final meta-analysis was according to the frequency of each OR/RR in all included studies. The order in this study was clinical attachment loss (CAL), pockets deep and periodontal bone loss.

The OR was used in these studies, while RR and hazard ratio (HR) were considered equivalent. For one study that reported stratified OR/RR in different population, we considered each subgroup analysis as an independent study. Before pooling the data, ORs were transformed into their natural logarithm to stabilize their variance, normalize the distribution and then statistically pooled [30].

The heterogeneity among studies was evaluated by Cochrane Q test and quantified as I² metric. For the Q statistic, a P-value <0.1 was considered statistically significant heterogeneity. For I² statistic, a value of zero was considered no heterogeneity, 25%-49% low, 50%-74% moderate, and above 75% high, respectively. To explore possible reasons for heterogeneity and test the robustness of the association, subgroup analyses were performed based on study design, methodological quality, gender, locations and PD exposure assessment parameters. When obvious heterogeneity across studies exists, the random-effect model was used to calculate the pooled estimates. Otherwise, the fixed-effect model was applied.

Publication bias was assessed by both Begg’s rank correlation test and Egger’s linear regression test. Funnel plots were also generated to assess the publication bias.

STATA statistical software, version 11.0 (STATA Corporation, College Station, TX, USA) was used for analysis. P values were 2-sided and P < 0.05 was considered statistically significantly otherwise indicated.

Initially, a total of 2558 articles were identified from PubMed (2041 articles), EMBASE (415 articles) and the Cochrane Library Database (102 articles) search. Of these citations, after evaluating the titles and abstracts, 373 studies were chosen for further review by reading the full text. Among them, 353 literatures were excluded because 286 articles were irrelevant, 14 had no OR/RR values, 28 had no clearly MI endpoint, 3 were not treated PD as exposure and 22 had no original data. Finally 22 studies from 20 articles representing 129,630 participants were included in our meta-analysis [6‐24, 31]. See Fig. 1 for the flow chart of the study selection process.

The characteristics of each study were demonstrated in Table 1, with adjusted covariates of each study presented. Four prospective cohort studies, 6 cross-sectional and 12 case–control studies evaluated the association between PD and MI. Case–control studies included a total of 2912 cases of MI and 3291 controls [6‐15, 24]. All studies were published from 1996 to 2016. The reported age ranged from 23 to 83 years. The enrolled years of patients varied from 1982 to 2014. Among 22 studies, 8 were carried out primarily in America [8, 9, 16, 18, 21‐23], 5 studies were conducted in Asian countries [12, 13, 20, 31] and 9 studies were from European countries [6, 7, 10, 11, 14, 15, 17, 19, 24]. Sixteen articles reported the OR values between PD and MI, 2 articles gave the HR and 2 articles presented with RR values. In 2 of the studies, OR/RR was less than one [20, 24]. Only two studies did not perform the adjustments for potential confounders [10, 19]. Fourteen studies showed PD was associated with MI [6‐17, 31], while 8 studies demonstrated PD was not related to MI [18‐24, 31]. PD exposure parameters varied among studies including clinical attachment loss (CAL), probing depth, deep pockets, and periodontal bone loss. The quality of the studies was high overall, without remarkable limitations identified (Additional file 2: Table S2; Additional file 3: Table S3).

The OR/RR from 22 studies were pooled and analyzed as presented in Fig. 2. Overall, the PD had statistically significant correlation with MI risk (OR = 2.02, 95% CI 1.59-2.57). A severe statistical heterogeneity among studies was revealed (Q = 152.77, P < 0.001, I² = 86.3%). Sensitivity analysis showed two studies contributed most to the heterogeneity [6, 10]. Although the sensitivity remains marked, the Q value and I² decreased to 70.68, and 73.1%, respectively. In the subgroup analysis by type of study (Fig. 3), cohort studies showed marginally increased risk of MI (RR 1.18, 95% CI 0.98-1.42; I² = 64.8%, n = 4) [16, 21‐23], while PD increased the risk of MI in both cross-sectional studies (OR 1.71, 95% CI 1.07-2.73; I² = 16.7%, n = 6) [17‐20, 31] and case–control studies (OR 2.93, 95% CI 1.95-4.39; I² = 90.2%, n = 12) [6‐15, 24].

By removing each study at a time, sensitivity analysis was conducted to determine the influence of each study on the pooled OR. By omitting one study in turn, the pooled OR and 95% CI was not modified significantly (Fig. 4). This sensitivity analysis indicated that the results of this meta-analysis were stable and reliable.

Additionally, publication bias was observed from Egger’s Test (P = 0.013), Begg’s Test (P = 0.195), and Funnel plot (Fig. 5). Trim and fill algorithm analysis was performed and the analysis showed “no trimming performed; data unchanged”.

To examine the stability of the studies, subgroup analyses were carried out based on several factors. Patients with PD had increased risk of MI across most subgroups by location, PD exposure parameters and study quality (Table 2). In the gender-specific subgroup, 3 female pooled studies showed an increased risk of MI (OR = 1.64, 95% CI 1.20-2.25) [8, 18, 23], while the pooled results in the 4 male studies were marginally increased (OR = 1.18, 95% CI 0.96-1.44) [9, 21, 22, 31]. Overall, subgroup analyses by various factors demonstrated that patients with PD have a higher risk to develop MI.

In this study, we identified a statistical positive correlation between PD and MI. Patients with PD have approximately a 1-fold increase in risk for MI. Given the high prevalence of PD in population [32], this increase might mean a profound public health impact.

Several meta-analyses have reported that subjects with PD have higher odds and higher risks of developing cardiovascular diseases [33‐38]. Teeuw et al pooled 25 studies on treatment of periodontitis and demonstrated intervention of PD improved the atherosclerotic profile [39], but the direct evidence of reduction in the risk of cardiovascular events associated with periodontitis treatment remains lacking. Further, a meta-analysis reported that patients with periodontitis compared to controls have increased arterial stiffness [39]. However, the only two interventional studies contradicted with each other on the effects of periodontal treatment on arterial stiffness [40]. Therefore, PD appears to be associated with increased risk of CVD, but the causal relation is still not well established.

Myocardial infarction is one of the most life-threating cardiovascular diseases. Some investigations mentioned the relationship between PD and MI, while dispute exists. In 1996, Joshipural KJ’s group followed up a cohort including 44119 males for 6 years and found no overall association between PD and fatal or non-fatal MI in this population [21]. This negative result was further echoed by some other prospective cohort studies [22, 23]. However, a large body of evidence showed there was an association between PD and MI. Very recently a large case–control study indicated that the risk of MI was significantly increased in patients with PD even after adjustment for confounding factors [15]. These disputes required further systemic analyses on the relation between MI and PD. However, to the best of our knowledge, no meta-analyses specified the relation between PD and MI.

In this study, we have showed an overall increased risk for MI in PD patients. The subgroup meta-analyses by study design showed that PD was associated with MI in both cross-sectional and case–control studies, while the relation in the cohort studies was marginal. Likewise, the subgroup analyses by country, PD exposure, participant number and study quality consistently showed an apparent increased risk of MI in PD subjects. Moreover, a marginal but non-significant increase of risk was present in male population. These non-significant findings exist in cohort studies and male participants might be attributed to the small number of studies included in the subgroup meta-analyses. Intriguingly, Rydén L group in their case–control study specified that the risk of a first MI was significantly increased in patients with PD. [15] However, most of the studies did not clarify the MI occurrence which presumably might be first ever MI. Renvert S et al also reported the relation between PD and recurrent MI [10]. First and recurrent MI differentiation would better illustrate the predication value of PD. Regarding limited data, the PD predication efficacy comparison between first and recurrent MI remains unknown.

Although not fully illustrated, several mechanisms were speculated to be involved in such a relation between PD and MI. The periodontal pathogen burden and its by-products might not only contribute to the development of atherosclerosis but also the rupture of atherosclerotic plaques and occurrence of MI. Dozens of literature has demonstrated that periodontal bacteria are detected in from atherosclerotic plaques. Ohki T et al. further suggested that three species of periodontal bacteria were present in the thrombi of patients with acute MI [41]. This finding suggested that periodontal bacteria might play a role in plaque inflammation and instability. Meanwhile, periodontal bacteria and the lipopolysaccharide from them could trigger inflammatory factor release, such as tumor necrosis factor α, interleukins, whose roles are well established in atherogenesis and rupture of developed lesions [42]. Further, C-reactive protein (CRP), a non-specific acute-phase factor participating in the systemic response to inflammation, helps to improve the risk prediction of cardiovascular events. Patients with PD also have increased serum levels of CRP relative to unaffected subjects [43]. The level of infection with periodontal pathogens is positively correlated with CRP level [44]. These evidence suggests that CRP might underlie the relation between PD and MI. In addition, platelet aggregation and thromboembolic events could be activated by periodontal pathogens themselves and many other cytokines [38].

Heterogeneity among studies was revealed in this analysis, although the relation between PD and MI has similar tendency in the overall and subgroup analysis. The significant heterogeneity was also reported in meta-analysis on PD and cardiovascular diseases [45, 46]. By nature, the diagnosis of PD was quite variable and included a series of parameters, such as attachment loss, bone loss pocket, probing depth, bleeding on probing, plaque index, gingival index, X-ray and microbiological results, teeth loss and the Russell periodontal index. Different PD assessment resulted in discrepancy when analyzing the relation between PD and MI. Renvert S’ group compared 5 PD parameters and their combination, founding that comprehensive PD data should be used in studies of associations between periodontitis and heart diseases [47]. Many other risk factors including age, smoking, and diabetes are common in both PD and MI. These factors could be taken into consideration as confounding factor. Different confounding factors were adjusted as demonstrated in Table1. This different adjustment could also affect the adjusted OR/RR when investigating the relation between PD and MI. Moreover, study design, study quality, number of participants and location were examined to find the sources of heterogeneity. The subgroup analysis showed these factors did not appreciably change the heterogeneity among studies. Further, sensitivity analyses showed no alterations to the pooled OR by omitting each study. Together, the significant heterogeneity should be considered when interpreting the results in this study, although the strong relation between PD and MI are observed.

Most of the studies that exclude for the meta-analysis in this study also supported a positive association between PD and MI. A population-based study showed periodontitis was associated with an increased risk of major adverse cardiac events including MI occurrence [48]. Other studies also indicated that PD may be associated with MI in Turkish and Iran population [49, 50]. Vice versa, there was higher prevalence of edentulousness and advanced PD in the hospitalized MI group than that in the non-MI group [51]. A case–control study also showed MI patients exhibited an unfavorable dental chronic infection compared with healthy patients. These data from the other angle suggests an association between PD and MI.

Several limitation of this meta-analysis should be noted. First, limited by the nature of observational studies, the causality between PD and MI cannot be inferred from our findings. Second, as aforementioned, definitions for PD in each study included in the analysis were not identical. Although we set rules when extracting data, this discrepancy in the assessments for PD seemed not to be eliminated nor even decreased obviously. Additionally, the differences in the study design, population and criteria to measure the exposure may weaken the evidence and lead to considerable heterogeneity. Although further analysis were performed, the sources of the heterogeneity remained not clear. The meta-regression analysis might further explain the source of heterogeneity. For the paucity of the data, these analyses were not conducted. Finally, the concrete mechanisms to explain the PD and MI relation still need further investigation, though possible mechanisms were discussed above. Further prospective studies are warranted to establish the association between PD and MI, while it might also be helpful to confirm the relation with randomized, controlled interventional trials including medication to control the pain and inflammation and surgery to remove the inflamed tissue and reduce the bone damage.