Grading the evidence for the association between periodontal disease and chronic non-communicable diseases: an umbrella review of meta-analyses of observational studies Authors: Haozhu Liu, Xiao Bai, Yuanfang Wang, Dongqiu Dai Published: 2026 ABSTRACT AbstractBackgroundSubstantial but inconsistent evidence has implicated periodontal disease (PD) as a potential risk factor for multiple chronic non-communicable diseases (NCDs), but the overall quality and credibility of this evidence have not been systematically evaluated and quantified in aggregate. We thus systematically assessed the associations between PD and the risk of chronic NCDs.MethodsFollowing PRISMA guidelines, we searched PubMed, Embase, Cochrane, and Web of Science for meta-analyses of observational studies. Methodological quality was assessed with AMSTAR 2. Summary effects were recalculated using random-effects models, with heterogeneity and bias evaluated via I2 statistics, 95% prediction intervals, and Egger's and excess significance tests. Evidence was graded into five credibility levels.ResultsThirty-eight meta-analyses with 58 associations across nine disease categories were included. Eighty-one percent of associations were significant (P < 0.05), and 25.9% remained significant at P < 10⁻⁶. However, only 11 associations (18.9%) had 95% prediction intervals excluding the null value, indicating limited robustness and substantial between-study heterogeneity for most associations. Four associations (6.9%) provided highly suggestive evidence (Class II), namely Helicobacter pylori infection, urogenital cancer, total cancer, and oral cancer. Three of these four Class II associations had 95% prediction intervals excluding the null value, whereas total cancer did not (95% PI, 0.98–1.32). None of the four Class II associations was supported by a representative meta-analysis rated as high quality by AMSTAR 2. In exploratory subgroup analysis, the oral cancer association weakened and was downgraded from Class II to Class IV when restricted to objective clinical/radiographic PD definitions. Methodological quality was limited, with 26 of 38 included meta-analyses (68.4%) rated as low or critically low quality. The overall evidence base should be interpreted cautiously.ConclusionsThis umbrella review found statistically highly suggestive evidence linking PD to an increased risk of a limited number of chronic NCDs, namely Helicobacter pylori infection, urogenital cancer, total cancer, and oral cancer. However, total cancer did not have a 95% prediction interval excluding the null value, and none of the four Class II associations was supported by a high-quality representative meta-analysis; moreover, the oral cancer association weakened and was downgraded to Class IV under objective clinical/radiographic PD definitions. These findings should therefore be interpreted cautiously, and PD should be regarded as a statistically associated factor across a limited set of chronic NCDs, with insufficient evidence to support causal or modifiable interpretation.Trial registrationPROSPERO CRD420251009702.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12903-026-09402-3.Keywords: Periodontal diseases, Noncommunicable Diseases, Umbrella review, Evidence grading Introduction Periodontal diseases are common inflammatory conditions of the supporting tissues of the teeth, with gingivitis and periodontitis representing the two principal clinical entities. Gingivitis is a reversible plaque-induced inflammatory condition confined to the gingiva, whereas periodontitis is a destructive inflammatory disease characterized by clinical attachment loss, periodontal pocket formation, alveolar bone loss, and, in severe cases, tooth loss [1, 2]. The 2018 EFP/AAP classification scheme provides the current operational standard for staging and grading periodontitis in clinical and research settings [3]. Periodontal diseases are therefore not only a major cause of oral morbidity but also a potentially important source of systemic inflammatory burden. According to the Global Burden of Disease Study 2019, an estimated 1.1 billion prevalent cases of severe periodontitis existed globally, with an age-standardized prevalence of 9.8% among individuals aged 15 years and older [4]. The burden of periodontal disease remains substantial worldwide and accounts for a considerable proportion of oral disease [5]. Beyond local tissue destruction, PD has been hypothesized to contribute to chronic non-communicable diseases (NCDs) through several pathways, including chronic low-grade inflammation, immune dysregulation, and microbiota dysbiosis [6–9]. However, the epidemiological evidence remains inconsistent. For example, studies on diabetes, inflammatory bowel disease, and gastrointestinal cancer have produced discordant findings, with some reporting positive associations and others finding no significant relationship [10–15]. These inconsistencies may reflect divergent diagnostic criteria for PD, differences in study populations and follow-up periods, and inadequate control of confounding factors. As a result, the overall strength and credibility of the association between PD and chronic NCDs remain uncertain when judged from primary studies alone. To address this uncertainty, multiple systematic reviews and meta-analyses have examined the relationship between PD and specific chronic NCDs, including psychiatric disorders [16], respiratory diseases [17], cardiovascular diseases [18], cancers [19, 20], and autoimmune diseases [21]. However, most previous meta-analyses have focused on a single disease category and therefore do not provide an integrated view of the broader PD–NCD evidence base. Several umbrella reviews have also addressed parts of this topic, but important gaps remain. A 2022 umbrella review in Nature Communications evaluated oral health more broadly and examined multiple oral conditions in relation to systemic NCDs, rather than focusing specifically on PD as the central exposure [22]. A 2024 umbrella review in BMC Oral Health focused specifically on cardiovascular outcomes, without covering other major systems such as cancers, metabolic disorders, respiratory diseases, or gastrointestinal diseases [23]. A more recent umbrella review published in Evidence-Based Dentistry in 2025 focused on the bidirectional association between periodontitis and NCDs and assessed certainty using the GRADE approach, but its search window was restricted to January 2021 through July 2024 and it did not apply our Class I–V credibility framework based on I2 and 95% prediction intervals [24]. In addition, a 2024/2025 umbrella review in Brain Sciences focused only on stroke-related outcomes [25]. Therefore, to our knowledge, no previous umbrella review has simultaneously centered PD as the principal exposure, covered multiple major chronic NCD systems in a single framework, and applied a strict credibility-grading strategy incorporating heterogeneity together with 95% prediction intervals. From a clinical perspective, this topic is relevant not only to dentists but also to physicians involved in the management of chronic NCDs. Dentists are often the first clinicians to detect periodontal inflammation and tissue destruction, whereas physicians in cardiology, endocrinology, oncology, gastroenterology, respiratory medicine, neurology, and obstetrics may encounter patients in whom periodontal status is overlooked despite its possible systemic relevance. A comprehensive umbrella review may therefore help clinicians distinguish which PD–NCD associations are supported by more credible evidence and which remain uncertain, thereby informing interdisciplinary referral, risk communication, and future research priorities. Accordingly, the present study was designed to provide a comprehensive, periodontal disease-centered umbrella review across nine disease categories using a standardized Class I–V evidence-grading strategy. Specifically, we aimed to critically appraise the methodological quality of the included meta-analyses and to assess the strength and credibility of the available evidence regarding the association between PD and chronic NCDs. Methods The protocol was registered with PROSPERO and the study followed Preferred Reporting Items of Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Search strategy A comprehensive literature search was conducted across PubMed, Embase, Cochrane and Web of Science from database inception to September 28, 2025, to identify systematic reviews and meta-analyses examining associations between PD and multiple chronic NCDs (Table S1). In addition to the indexed databases, grey literature was searched in OpenGrey (https://opengrey.eu/). We merged keywords and subject headings in accordance with the thesaurus of each database and applied exploded subject headings, using a search syntax based on the combination of terms related to periodontal disease and chronic non-communicable diseases. Manual searches of reference lists from retrieved articles were performed to ensure inclusivity. Selection and criteria of studies Two authors independently reviewed titles or abstracts of the literature and assessed full-text articles for eligibility. Inclusion criteria focused on meta-analyses of observational studies investigating PD and chronic conditions (e.g., diabetes, malignancies) in eligible populations without a prespecified age restriction. However, we anticipated that most retained meta-analyses would involve adult or predominantly adult populations, because many of the chronic NCD outcomes of interest are primarily investigated in adults. Exclusion criteria included meta-analyses addressing outcomes outside the scope of this review (including those centered on mortality, recurrence rate, progression-free survival, disease-free survival, and overall survival), non-English articles, or studies lacking quantitative synthesis. These outcomes were excluded because the present umbrella review was designed to address disease occurrence/incidence rather than prognosis after diagnosis. Prognostic outcomes involve different study populations, time origins, and clinical interpretations, and their inclusion would have introduced substantial conceptual and methodological heterogeneity into an incidence-focused synthesis. In addition, the available prognostic meta-analyses were limited and unevenly distributed across disease categories, which could have led to an unbalanced synthesis and biased interpretation. Throughout this review, PD was used as an umbrella exposure term encompassing periodontal disease, periodontitis, gingivitis, or related periodontal conditions as defined in the included meta-analyses; Table 1 retained the exposure terminology reported by each representative meta-analysis. Overlapping meta-analyses examining the same outcome were resolved by selecting the study with the largest dataset. This rule was prioritized because it maximized primary-study coverage and statistical information for recalculation, while alternative considerations such as methodological quality, recency, and comprehensiveness were assessed separately and acknowledged as limitations. Inter-reviewer agreement for title/abstract screening and full-text eligibility assessment was quantified using Cohen’s kappa coefficient and percentage agreement. Disagreements were resolved through discussion with a third author. Table 1Evidence grading for associations between periodontal disease and risk of chronic non-communicable diseasesFirst author, YearExposureOutcomeRandom-EffectsRandomI2,%95% PIEggerLSOEP value*CEEffect metricsummary effect size (95% CI)P valueP valueWang, 2024 [19]PDOGC1.31 (1.16–1.49)2.16E-0597.5(0.78–2.21)0.083Yes112.677 < 0.0001ⅢHRWang, 2024 [19]PDLiver cancer1.09 (1.04–1.13)2.07E-040(0.82–1.44)NAYes11.197NPⅢHRWang, 2024 [19]PDEsophageal cancer1.4 (1.17–1.67)2.66E-0462.5(0.84–2.33)NAYes43.529 < 0.0001ⅢHRWang, 2024 [19]PDColorectal cancer1.21 (1.05–1.39)7.21E-0389(0.76–1.91)0.316Yes52.23 < 0.0001ⅣHRWang, 2024 [19]PDGastric cancer1.13 (1.01–1.26)3.20E-0275.3(0.83–1.53)NANo20.4117.42E-01ⅣHRWang, 2024 [19]PDPancreatic cancer1.35 (1.00–1.82)4.80E-0297.3(0.46–3.99)0.584Yes48.826NPⅣHRWang, 2024 [19]PDBiliary tract cancer1.13 (0.9–1.42)3.04E-0187.9(0.09–13.77)NANo10.197.10E-01VHRMoradi, 2023 [26]PDHelicobacter pylori Infection2.46 (2.01–3.01)2.37E-1848.7(1.24–4.88)0.133Yes1712.245 < 0.0001ⅡORStöhr, 2021 [27]PDDiabetes mellitus1.26 (1.12–1.41)9.20E-0570.6(0.92–1.70)0.192Yes63.148 < 0.0001ⅢRRDibello, 2024 [16]PDCognitive Impairment1.27 (1.11–1.44)4.50E-0434.6(0.93–1.73)0.036No40.9531.69E-01ⅢRRDibello, 2024 [16]PDCognitive decline3.01 (1.56–5.79)1.00E-0382.4(0.32–27.82)NAYes44.727NPⅣRRDibello, 2024 [16]PDDementia1.22 (1.04–1.42)1.20E-0297.6(0.71–2.1)NAYes67.473NPⅣRRDibello, 2024 [16]PDDepression1.07 (0.95–1.21)2.47E-0173.2(0.73–1.57)NAYes21.402 < 0.0001VRRLarvin, 2021 [28]PDCardiovascular disease1.19 (1.14–1.23)8.94E-2093.9(1.02–1.38)0.072Yes1711.191.19E-01ⅢRRLi, 2022 [29]PDUrogenital cancer1.24 (1.17–1.31)3.32E-1423.1(1.11–1.39)0.471Yes75.583.38E-11ⅡHRZhang, 2022 [30]PDPreterm birth1.58 (1.38–1.81)1.60E-118.9(1.26–1.98)0.06Yes82.3592.32E-01ⅢORZhang, 2022 [30]PDLow birth weight2.39 (1.70–3.38)6.65E-0781.8(0.65–8.76)0.006No100.844.02E-01ⅢORZhang, 2022 [30]PDSmall for gestational age1.62 (0.86–3.06)1.36E-0157.3(0.13–20.26)NAYes12.502NPVORWang, 2022 [31]PDTotal cancer1.14 (1.08–1.20)2.13E-0766.2(0.98–1.32)0.381Yes76.792 < 0.0001ⅡHRQin, 2021 [18]PDMyocardial infarction1.16 (1.06–1.27)1.00E-0384.8(0.88–1.54)0.017Yes51.539 < 0.0001ⅣORWei, 2021 [32]PDProstate cancer1.17 (1.06–1.29)1.10E-035.2(1.01–1.35)NAYes31.3452.40E-02ⅣRRShao, 2018 [33]PDBreast cancer1.22 (1.06–1.40)4.00E-0351.4(0.88–1.70)0.095Yes32.143.47E-03ⅣRRWu, 2020 [34]PDHematopoietic and lymphatic cancers1.16 (1.04–1.29)8.00E-0322.7(0.92–1.46)NAYes20.9969.60E-02ⅣRRXie, 2018 [35]PDBladder cancer1.09 (0.95–1.26)2.27E-010(0.8–1.48)NANo00.41NPVHRMa, 2024 [36]PDOral cancer2.94 (2.13–4.07)6.76E-1152.4(1.22–7.12)0.703Yes75.2651.60E-02ⅡORMartos, 2025 [37]PDGestational diabetes mellitus1.84 (1.26–2.69)1.74E-0341.1(0.68–4.97)0.693No54.7244.07E-03ⅣORBai, 2023 [38]PDHead and neck cancer2.20 (1.19–4.06)1.16E-0294.5(0.21–23.32) †NANo44.945NPⅣORRodriguez, 2022 [39]PDAcute coronary syndrome1.48 (1.35–1.63)5.30E-1682.2(0.95–2.29) < 0.0001Yes3338.946NPⅢORVerma, 2023 [40]PDLung cancer1.52 (1.26–1.83)1.30E-0582.3(0.83–2.76)NAYes53.153.47E-03ⅢRRLv, 2020 [41]PDARMD1.35 (1.07–1.70)1.10E-0280.4(0.62–2.94)NANo23.867NPⅣRRRomandini, 2024 [42]PeriodontitisDental caries1.63 (1.32–2.01)5.99E-0683(0.81–3.28)NAYes55.903NPⅢORUngprasert, 2017 [43]PeriodontitisPsoriasis1.54 (1.35–1.75)1.42E-1021.4(1.12–2.10)NAYes32.572 < 0.0001ⅢRRWu, 2020 [72]PeriodontitisType 2 diabetes mellitus4.04 (2.48–6.59)2.17E-080(0.17–96.05) †NAYes32.416 < 0.0001ⅣORZhang, 2025 [21]PeriodontitisCrohn's disease1.34 (1.04–1.73)2.50E-0265.9(0.64–2.80)NANo21.22.65E-01ⅣRRZhang, 2025 [21]PeriodontitisInflammatory bowel disease1.15 (0.98–1.35)8.90E-0282.6(0.72–1.84)0.386No41.4843.39E-01VRRZhang, 2025 [21]PeriodontitisUlcerative colitis1.00 (0.75–1.34)9.89E-0175.9(0.45–2.24)NAYes31.3186.08E-01VRRFilho, 2020 [17]PeriodontitisAsthma3.54 (2.47–5.07)4.92E-120(0.35–36.27) †NAYes32.985 < 0.0001ⅣORFilho, 2020 [17]PeriodontitisPneumonia3.21 (1.99–5.17)1.59E-060(0.15–70.56) †NAYes31.8292.28E-04ⅣORFilho, 2020 [17]PeriodontitisCOPD1.78 (1.04–3.04)3.40E-0237.8(0.27–11.71)NANo21.1361.71E-01ⅣORMachado, 2020 [44]PeriodontitisPolycystic ovarian syndrome1.46 (1.29–1.66)7.29E-090(0.63–3.37)NAYes21.995 < 0.0001ⅣRRStacchi, 2016 [45]PeriodontitisPeri-implantitis0.23 (0.11–0.47)6.66E-050(0.00–25.9)NAYes10.9529.58E-01ⅣRRStacchi, 2016 [45]PeriodontitisImplant loss0.37 (0.10–1.40)1.44E-010(0.00–2148.96) †NANo00.404NPVRRLarvin, 2023 [46]PeriodontitisOsteoporosis1.40 (1.12–1.76)3.00E-0393.3(0.5–3.92)NAYes42.35 < 0.0001ⅣRRLarvin, 2023 [46]PeriodontitisRheumatoid arthritis1.28 (1.06–1.54)1.00E-0276.5(0.58–2.84)NAYes41.285 < 0.0001ⅣRRLarvin, 2023 [46]PeriodontitisLiver disease1.46 (0.94–2.26)9.20E-0285(0.01–250.97) †NAYes30.8683.86E-03VRRMa, 2020 [47]PeriodontitisMelanoma1.21 (1.03–1.42)1.70E-020(0.44–3.34)NAYes10.9221.50E-02ⅣHRWang, 2019 [48]PeriodontitisPeripheral vascular disease1.60 (1.41–1.82)2.40E-1380.5(1.05–2.46) < 0.0001Yes2116.2461.51E-03ⅢORFarook, 2021 [49]PeriodontitisErectile dysfunction2.55 (1.70–3.84)6.69E-0697.8(0.65–9.96)NAYes65.488 < 0.0001ⅢORChen, 2025 [50]PeriodontitisHeart failure1.66 (1.26–2.18)3.08E-0411.6(0.18–14.84)NAYes21.3434.71E-04ⅢORXu, 2023 [51]PeriodontitisNon-alcoholic fatty liver disease1.04 (0.97–1.12)2.81E-0158.5(0.87–1.24)NAYes31.3843.86E-03VORDewan, 2024 [52]PeriodontitisStroke1.81 (1.38–2.38)1.95E-0574.1(0.78–4.22) < 0.0001Yes1110.5894.51E-03ⅢHRKhodadadi, 2022 [53]PeriodontitisObstructive sleep apnea2.17 (1.66–2.83)1.25E-0842.6(1.15–4.08)0.065Yes73.542 < 0.0001ⅢORKaliamoorthy, 2022 [54]PeriodontitisAlzheimer's disease1.79 (1.15–2.78)1.00E-0225(0.04–86.13) †NAYes12.163NPⅣORUrbina, 2023 [55]PeriodontitisPostoperative complications4.72 (1.12–19.82)3.40E-0256.8(0.00- 1.89 × 10⁷)†NAYes20.9188.75E-03ⅣORSantos, 2023 [56]Severe periodontitisMetabolic syndrome1.55 (1.33–1.79)1.18E-0857.5(1.00–2.39)NAYes68.451NPⅢORSantos, 2023 [56]Moderate periodontitisMetabolic syndrome1.30 (1.13–1.49)1.58E-0451.1(0.91–1.86)NAYes40.455.00E-01ⅢORTsimpiris, 2023 [57]Chronic periodontitisHelicobacter pylori Infection in mouth1.87 (0.86–4.10)1.16E-0165.8(0.08–42.96)NAYes21.732 < 0.0001VORDewan, 2024 [52]GingivitisStroke1.93 (0.92–4.06)8.40E-0273.5(0.19–19.65)NAYes32.562.20E-02VHRAbbreviations: PD periodontal disease, OGC overall gastrointestinal cancers, ARMD age-related macular degeneration, COPD chronic obstructive pulmonary disease, CI confidence interval, PI prediction interval, LS largest study with significant effect (P ≤ 0.05); O, observed number of statistically significant component studies, E expected number of statistically significant component studies, CE class of evidence, NP not pertinent, NA not assessable*P value of the excess significance test. NP was assigned when O ≤ E, indicating that the observed number of statistically significant studies did not exceed the expected number. For evidence grading, NP was treated as satisfying the no-excess-significance criterion when that criterion was relevant†PI was considered interpretively uninformative because the association was based on fewer than five component studies or because the PI width exceeded 100-foldAll statistical tests were two-sided. Summary estimates were recalculated within each association using the native metric reported in the retained representative meta-analysis; no standardization to a common metric was performed across different associations, and ORs, RRs, and HRs should not be directly compared across rows or outcomes using different metrics. Egger’s regression test was interpreted only for associations including at least 10 eligible effect estimates. Associations with fewer than 10 eligible effect estimates were treated as not assessable (NA) for the small-study-effects criterion rather than as showing absence of small-study effects, and therefore could not satisfy the Class II “no small-study effects” requirement. PD was used as an umbrella exposure term encompassing periodontal disease, periodontitis, gingivitis, or related periodontal conditions as defined in the included meta-analyses; the Exposure column retains the terminology reported by each representative meta-analysis Evidence grading for associations between periodontal disease and risk of chronic non-communicable diseases Abbreviations: PD periodontal disease, OGC overall gastrointestinal cancers, ARMD age-related macular degeneration, COPD chronic obstructive pulmonary disease, CI confidence interval, PI prediction interval, LS largest study with significant effect (P ≤ 0.05); O, observed number of statistically significant component studies, E expected number of statistically significant component studies, CE class of evidence, NP not pertinent, NA not assessable *P value of the excess significance test. NP was assigned when O ≤ E, indicating that the observed number of statistically significant studies did not exceed the expected number. For evidence grading, NP was treated as satisfying the no-excess-significance criterion when that criterion was relevant †PI was considered interpretively uninformative because the association was based on fewer than five component studies or because the PI width exceeded 100-fold All statistical tests were two-sided. Summary estimates were recalculated within each association using the native metric reported in the retained representative meta-analysis; no standardization to a common metric was performed across different associations, and ORs, RRs, and HRs should not be directly compared across rows or outcomes using different metrics. Egger’s regression test was interpreted only for associations including at least 10 eligible effect estimates. Associations with fewer than 10 eligible effect estimates were treated as not assessable (NA) for the small-study-effects criterion rather than as showing absence of small-study effects, and therefore could not satisfy the Class II “no small-study effects” requirement. PD was used as an umbrella exposure term encompassing periodontal disease, periodontitis, gingivitis, or related periodontal conditions as defined in the included meta-analyses; the Exposure column retains the terminology reported by each representative meta-analysis Data extraction and quality assessment Two authors independently extracted data using a standardized template, capturing author details, publication year, study design, population characteristics, exposure definitions, outcome, and relative effect measures (odds ratios, hazard ratios, or relative risks with 95% CIs). Data-extraction agreement was assessed using percentage agreement across prespecified key extraction fields, including reference, outcome, exposure terminology, number of cases, summary effect estimate, 95% CI, and I2. Because the extracted variables included both textual/categorical fields and numerical values, percentage agreement was considered more appropriate than a single overall Cohen’s kappa coefficient. Text fields were compared after harmonizing prespecified terminology variants, and numerical fields were considered concordant if they were identical after rounding to two decimals or differed by ≤ 0.01. Discrepancies were resolved by consensus, with consultation with a third author when necessary. Methodological quality of included meta-analyses was evaluated using the 16-item AMSTAR 2 checklist, categorizing reviews as high, moderate, low, or critically low quality [58] (Table S2). Inter-reviewer agreement for the overall AMSTAR 2 rating was quantified using Cohen’s kappa coefficient and percentage agreement. Disagreements in study selection, data extraction, and quality assessment were resolved by discussion with a third author. Statistical analysis Summary association estimates were recalculated using random-effects models to account for between-study heterogeneity. For each association, the recalculation was performed using the original relative effect metric retained in the representative meta-analysis (OR, RR, or HR). No cross-metric transformation (e.g., using methods such as Chinn’s conversion) was performed to standardize ORs, RRs, and HRs to a single common scale across different associations. Therefore, the summary estimates reported in Table 1 should be interpreted as association-specific pooled relative measures rather than directly comparable effect sizes across rows or outcomes using different metrics [59, 60]. Heterogeneity was quantified via the I2 statistic (≥ 50%: significant; ≥ 75%: high) [61, 62]. The 95% prediction interval (PI) was calculated to estimate the plausible range of future relative effect estimates [63]. PIs were flagged as interpretively uninformative when an association was based on fewer than five component studies or when the PI width exceeded 100-fold, because such values may mainly reflect sparse evidence or unstable estimation. The > 100-fold PI-width threshold was an author-defined exploratory interpretive flag intended to identify extremely unstable prediction intervals; it was not a formal evidence-grading criterion. Small-study effects were assessed using Egger’s regression asymmetry test (P < 0.10) [64] only for associations including at least 10 eligible effect estimates, because the test is generally considered underpowered in smaller meta-analyses. For associations with fewer than 10 eligible effect estimates, small-study effects were considered not formally assessable rather than interpreted as absent. For OSA, the retained meta-analysis included 9 unique primary studies contributing 10 eligible effect estimates; therefore, Egger’s test was considered assessable and was calculated using the 10 effect estimates. For associations with formally assessable or clinically important evidence signals, funnel plots were also visually inspected as a complementary assessment of small-study effects and publication bias. The corresponding funnel plots for the four current Class II associations, namely Helicobacter pylori infection, urogenital cancer, total cancer, and oral cancer, are provided in Figures S1–S4. Excess significance bias was evaluated by comparing the observed number (O) of statistically significant studies with the expected number (E) based on the statistical power of individual studies. When O > E and the P value for the excess significance test was < 0.10, excess significance was considered positive [65, 66]. When O ≤ E, the test was marked as not pertinent (NP), indicating no excess of observed statistically significant findings; for evidence grading, NP was treated as satisfying the no-excess-significance condition when that criterion was relevant. Kappa values were interpreted as near-perfect (0.81–1.00), substantial (0.61–0.80), moderate (0.41–0.60), fair (0.21–0.40), and poor (0–0.20) [67]. All analyses were conducted in Stata 16.0. Assessment of evidence credibility Associations were classified into five credibility tiers using the established criteria from previous umbrella reviews [68–70] (Table S3). Each association was assigned to the highest evidence class for which all required criteria were satisfied. A criterion-by-criterion example for an association that did not satisfy Class II requirements is provided in Table S8. Class I (convincing evidence): P < 1 × 10⁻⁶ under the random-effects model, > 1000 cases, I2 < 50%, 95% PI excluding the null value, no small-study effects, and no excess significance bias. Class I (convincing evidence): P < 1 × 10⁻⁶ under the random-effects model, > 1000 cases, I2 < 50%, 95% PI excluding the null value, no small-study effects, and no excess significance bias. Class II (highly suggestive evidence): P < 1 × 10⁻⁶ under the random-effects model, > 1000 cases, and no small-study effects. Absence of excess significance bias and a 95% PI excluding the null value were not part of the Class II definition; therefore, Class II evidence should not be interpreted as necessarily robust, reproducible, transportable across populations, causal, or clinically actionable. Class II (highly suggestive evidence): P < 1 × 10⁻⁶ under the random-effects model, > 1000 cases, and no small-study effects. Absence of excess significance bias and a 95% PI excluding the null value were not part of the Class II definition; therefore, Class II evidence should not be interpreted as necessarily robust, reproducible, transportable across populations, causal, or clinically actionable. Class III (suggestive evidence): P < 0.001 under the random-effects model and > 1000 cases. Class III (suggestive evidence): P < 0.001 under the random-effects model and > 1000 cases. Class IV (weak evidence): P < 0.05 under the random-effects model. Class IV (weak evidence): P < 0.05 under the random-effects model. Class V (non-significant evidence): P ≥ 0.05 under the random-effects model. Class V (non-significant evidence): P ≥ 0.05 under the random-effects model. For associations with fewer than 10 eligible effect estimates, the small-study-effects component was considered not assessable; therefore, absence of small-study effects was interpreted only when Egger’s test was formally assessable. For Class I, NP in the excess-significance column indicated O ≤ E and was treated as satisfying the no-excess-significance criterion. Sensitivity analyses Because only one representative meta-analysis was retained for each association, conventional sensitivity analysis based on excluding low-quality meta-analyses was not fully applicable, as removal of the selected meta-analysis would eliminate the corresponding association from the umbrella review rather than generate a revised pooled estimate. Therefore, we conducted an additional quality-based sensitivity assessment focused on associations classified as highly suggestive evidence (Class II). For Class II associations whose representative meta-analysis was rated as low or critically low quality by AMSTAR 2, we searched for alternative eligible meta-analyses addressing the same association and appraised them using AMSTAR 2. When a higher-quality alternative meta-analysis was available, we used it for replacement sensitivity analysis and reassessed the credibility of evidence. Associations without any higher-quality eligible alternative meta-analysis were considered not assessable in this replacement sensitivity analysis. Subgroup analysis by PD definition To further explore heterogeneity arising from differences in the definition of periodontal disease (PD) exposure, we conducted exploratory, hypothesis-generating subgroup analyses for selected associations. Associations were selected when the retained meta-analysis had a clinically important or credibility-relevant signal and when the PD definitions of the primary studies could be reliably traced and categorized from the source articles. Primary studies were classified as objective clinical/radiographic PD when the exposure was defined using clinical periodontal examination or radiographic criteria, such as probing depth, clinical attachment loss, bleeding on probing, alveolar bone loss, or standardized periodontal case definitions. Studies were classified as non-objective/proxy-based or mixed/unclear PD when exposure was based on self-reported periodontal disease, administrative diagnosis, tooth-loss proxies, combined proxy and objective measures, or other indirect/non-standard definitions. Summary estimates were recalculated within each subgroup, and the credibility of evidence was reassessed using the same umbrella-review criteria as in the main analysis. Because these subgroup analyses were selectively feasible only for associations with retrievable study-level exposure definitions, they should be interpreted as exploratory rather than confirmatory. Study-level exposure classifications for the four outcomes included in these subgroup analyses are provided in Table S9. Overlap assessment To quantify the degree of overlap of primary studies across the included meta-analyses, we calculated the corrected covered area (CCA) [71]. For this purpose, we constructed a citation matrix of retained meta-analyses against their included primary studies. The CCA was calculated as (N − r)/(rc − r), where N is the total number of study occurrences across reviews, r is the number of unique primary studies, and c is the number of retained reviews. Consistent with previous methodological recommendations, CCA values were interpreted as slight (0–5%), moderate (6–10%), high (11–15%), or very high (> 15%) overlap. Because overlapping meta-analyses addressing the same association had already been handled by retaining the review with the largest dataset, the CCA was intended to assess residual overlap at the umbrella-review level. Results Literature searches In total, 4,159 records were initially identified. After removal of duplicates, 2,545 records underwent title/abstract screening, and 266 full-text articles were assessed for eligibility. Inter-rater agreement was substantial at the title/abstract screening stage (Cohen’s κ = 0.805; percentage agreement = 96.46%, 2,455/2,545) and near-perfect at the full-text assessment stage (Cohen’s κ = 0.984; percentage agreement = 99.62%, 265/266). Finally, 38 meta-analyses were included in this umbrella review [16–19, 21, 26–57, 72]. The literature screening process is outlined in Fig. 1. A detailed list of the 228 publications excluded from the analysis, including the justification for their removal, is available in Table S4. Literature search flow diagram Characteristics of included articles A total of 38 unique meta-analyses examining 58 associations between PD and chronic NCD risk were included. These 58 associations covered outcomes across nine clinical categories: cancer (N = 17, 29.3%), cardiovascular and cerebrovascular diseases (N = 7, 12.1%), metabolic disorders (N = 7, 12.1%), musculoskeletal diseases (N = 3, 5.2%), neuropsychiatric diseases (N = 5, 8.6%), gastrointestinal diseases (N = 6, 10.3%), perinatal outcomes (N = 3, 5.2%), respiratory diseases (N = 4, 6.9%), and other diseases (N = 6, 10.3%) (Fig. 2). The retained meta-analyses were published between 2016 and 2025 and included between 3 and 46 primary studies per association (mediaN = 6). The total sample size across associations was 123,680,092 participants, with 1,483,017 reported cases. Case counts per association ranged from 13 to 256,608, and 45 associations (77.6%) reported more than 1,000 cases. Descriptive characteristics are detailed in Table S5. Distribution of chronic non-communicable diseases associated with periodontal disease Methodological quality assessment We evaluated the methodological rigor of the included meta-analyses using the AMSTAR 2 tool. Twelve studies (31.6%) were rated as high quality, sixteen (42.1%) were rated as low quality, and the remaining ten (26.3%) were rated as critically low quality (Fig. 3). A total of 94.7% of the included meta-analyses had a non-critical weakness related to item 10. Among the studies rated as low or critically low quality, one or more critical flaws were identified most frequently in item 2 (50.0%) and item 7 (26.9%). The detailed item-level evaluations are shown in Table S2. For data extraction, the initial percentage agreement across prespecified key extraction fields was 98.7% (458/464 comparison cells). Most remaining discrepancies reflected terminology variants in outcome names, such as abbreviations versus full disease names, and all discrepancies were resolved by consensus, with consultation with a third author when necessary. Inter-reviewer agreement for the overall AMSTAR 2 rating was substantial, with a Cohen’s kappa of 0.758 and a percentage agreement of 84.2% (32/38 studies). Because AMSTAR 2 assesses methodological quality whereas the umbrella-review evidence classes assess statistical credibility, no direct one-to-one correspondence between AMSTAR 2 ratings and evidence classes was expected. This distinction was further explored in the AMSTAR 2–based sensitivity analysis. Risk of Bias analysis of included systematic reviews and meta-analyses using AMSTAR-2 Sensitivity analysis by AMSTAR 2 quality Among the 4 associations classified as highly suggestive evidence (Class II) after revision of the small-study-effects assessment, none was supported by a representative meta-analysis rated as high quality by AMSTAR 2. Specifically, oral cancer and urogenital cancer were supported by low-quality representative meta-analyses, whereas Helicobacter pylori infection and total cancer were supported by critically low-quality representative meta-analyses. For these associations, candidate alternative meta-analyses were assessed using AMSTAR 2. A moderate-quality alternative meta-analysis was identified only for Helicobacter pylori infection, but reassessment was not feasible because the corresponding quantitative synthesis did not meet the umbrella-review grading requirements. No higher-quality eligible alternative meta-analyses were identified for the remaining Class II associations. These results are presented in Table S6. Subgroup analysis by PD definition To address heterogeneity in exposure definition, we performed exploratory subgroup analyses according to whether PD was defined by objective clinical/radiographic criteria or by non-objective definitions. Among the selected associations with retrievable study-level PD definitions, the magnitude and credibility of associations varied by exposure definition (Table S7). The study-level exposure definitions and harmonized classifications underlying these exploratory subgroup analyses are reported in Table S9. For oral cancer, the association was weaker under objective PD definitions (OR 2.57, 95% CI 1.49–4.44; Class IV) than under non-objective or proxy-based definitions (OR 3.21, 95% CI 2.18–4.71; Class II), indicating that this association should be interpreted more cautiously when restricted to objectively assessed PD. For total cancer, Class II evidence was retained only in the non-objective subgroup, whereas the objective PD subgroup was downgraded to Class IV. In contrast, for urogenital cancer, the objective PD subgroup retained Class II evidence, whereas the non-objective subgroup became non-significant. Because obstructive sleep apnea (OSA) was specifically highlighted by the reviewer, we additionally re-examined the study-level PD definitions of the retained OSA meta-analysis. The retained OSA meta-analysis comprised 9 unique primary studies contributing 10 eligible effect estimates; 9 effect estimates used objective clinical/radiographic definitions and 1 used a non-objective administrative/history-based definition. The pooled estimate among the objectively defined studies remained significant (OR 2.45, 95% CI 1.72–3.47), but it no longer met Class II criteria because the subgroup included fewer than 1,000 cases. Only one non-objective study was available, precluding a meaningful subgroup meta-analysis for that category. Overlap of primary studies across included meta-analyses We assessed overlap of primary studies across the retained meta-analyses using the corrected covered area (CCA). The overall CCA was approximately 0.41%, indicating slight overlap at the umbrella-review level after applying our prespecified rule of retaining only the meta-analysis with the largest dataset for overlapping reviews addressing the same association. However, the pairwise CCA matrix showed that localized overlap remained present among several conceptually related reviews (Fig. 4). The highest pairwise CCA values were mainly observed within cancer-related outcome clusters, including Li, 2022 with Ma, 2020 (37.5%), Li, 2022 with Wei, 2021 (35.7%), Ma, 2020 with Wei, 2021 (30.8%), Ma, 2020 with Wang, 2024 (30.4%), and Ma, 2020 with Wang, 2022 (26.5%). A high pairwise overlap was also observed between Larvin, 2023 and Stöhr, 2021 (25.8%). Therefore, although overlap was limited overall, cancer-cluster associations with high pairwise overlap should be interpreted as one related evidence stream rather than as fully independent signals. Cancer-related review pairs with high or very high pairwise CCA values are summarized in Table S10. Pairwise corrected covered area (CCA) overlap matrix across the retained meta-analyses. Values are shown as percentages. Green indicates slight overlap (0–5%), yellow indicates moderate overlap (6–10%), orange indicates high overlap (11–15%), and red indicates very high overlap (> 15%) Summary association estimates and study heterogeneity Because ORs, RRs, and HRs were preserved on their original scales, effect-size magnitudes should not be directly compared across rows using different metrics. Of the 58 associations examined in this analysis, 47 (81.0%) demonstrated statistical significance at the P < 0.05 threshold using a random-effects meta-analysis approach (Table 1). Among these, 28 (48.3%) maintained significance at a more stringent P value of 10 − 3. When applying an even stricter threshold of P < 10 − 6, 15 associations (25.9%) continued to show statistically robust results. Additionally, when evaluating the largest study for each association, 46 of the 58 (79.3%) demonstrated significance at P < 0.05. Among the examined associations, 37 (63.8%) demonstrated substantial heterogeneity (I2 > 50%), with 22 (37.9%) showing notably high heterogeneity levels (I2 > 75%). Evaluation of between-study variability through 95% prediction intervals revealed that only 11 associations (18.9%) excluded the null value (Table 1). Several associations had mathematically valid but interpretively uninformative prediction intervals because of sparse evidence or an author-defined exploratory PI-width flag, including postoperative complications, implant loss, liver disease, T2DM, HNC, Alzheimer’s disease, pneumonia, and asthma. Four associations were classified as highly suggestive evidence (Class II), namely Helicobacter pylori infection, urogenital cancer, total cancer, and oral cancer. Three of these four Class II associations had 95% prediction intervals excluding the null value, whereas total cancer did not (95% PI, 0.98–1.32). Cancer outcomes PD was significantly associated with an elevated risk of multiple cancer types. The strongest associations were observed for head and neck cancer (HNC) (OR, 2.20; 95% CI, 1.19–4.06; I2 = 94.50%) and oral cancer (OR, 2.94; 95% CI, 2.13–4.07; I2 = 52.40%). Increased risks were also identified for esophageal cancer (HR, 1.40; 95% CI, 1.17–1.67; I2 = 62.50%), lung cancer (RR, 1.52; 95% CI, 1.26–1.83; I2 = 82.30%), and pancreatic cancer (HR, 1.35; 95% CI, 1.00–1.82; I2 = 97.30%). Moderately elevated risks were found for oral and gastrointestinal cancer (OGC) (HR, 1.31; 95% CI, 1.16–1.49; I2 = 97.50%), colorectal cancer (CRC) (HR, 1.21; 95% CI, 1.05–1.39; I2 = 89.00%), urogenital cancer (HR, 1.24; 95% CI, 1.17–1.31; I2 = 23.10%), breast cancer (RR, 1.22; 95% CI, 1.06–1.40; I2 = 51.40%), and melanoma (HR, 1.21; 95% CI, 1.03–1.42; I2 = 0.00%). Smaller but significant effects were noted for gastric cancer (HR, 1.13; 95% CI, 1.01–1.26; I2 = 75.30%), liver cancer (HR, 1.09; 95% CI, 1.04–1.13; I2 = 0.00%), prostate cancer (RR, 1.17; 95% CI, 1.06–1.29; I2 = 5.20%), hematopoietic and lymphatic cancers (HLC) (RR, 1.16; 95% CI, 1.04–1.29; I2 = 22.70%), and total cancer (HR, 1.14; 95% CI, 1.08–1.20; I2 = 66.20%). No significant associations were detected for biliary tract cancer (BTC) (HR, 1.13; 95% CI, 0.90–1.42; I2 = 87.90%) or bladder cancer (HR, 1.09; 95% CI, 0.95–1.26; I2 = 0.00%). Overall, of the 17 cancer associations examined, only oral cancer, urogenital cancer, and total cancer met Class II criteria, whereas most other cancer associations provided Class III or weaker evidence. Cardiovascular and cerebrovascular diseases PD or periodontitis demonstrated significant associations with increased risks of several cardiovascular and cerebrovascular conditions. The strongest associations were observed for stroke (HR, 1.81; 95% CI, 1.38–2.38; I2 = 74.10%), heart failure (HF) (OR, 1.66; 95% CI, 1.26–2.18; I2 = 11.6%), and peripheral vascular disease (PVD) (OR, 1.60; 95% CI, 1.41–1.82; I2 = 80.5%). A notably elevated risk was also found for acute coronary syndrome (ACS) (OR, 1.48; 95% CI, 1.35–1.63; I2 = 82.2%). Additionally, PD was associated with moderate risk increases in overall cardiovascular disease (CVD) (RR, 1.19; 95% CI, 1.14–1.23; I2 = 93.9%) and myocardial infarction (MI) (OR, 1.16; 95% CI, 1.06–1.27; I2 = 84.8%). In contrast, no significant association was detected between gingivitis and stroke (HR, 1.93; 95% CI, 0.92–4.06; I2 = 73.50%). Overall, cardiovascular and cerebrovascular associations were supported mainly by Class III or weaker evidence, with no association reaching Class II. Metabolic or endocrine disorders PD or periodontitis was significantly associated with an elevated risk of various metabolic conditions. The most substantial risk increase was observed for type 2 diabetes mellitus (T2DM) (OR, 4.04; 95% CI, 2.48–6.59; I2 = 0.00%), followed by gestational diabetes mellitus (GDM) (OR, 1.84; 95% CI, 1.26–2.69; I2 = 41.10%). PD was also associated with diabetes mellitus (DM) (RR, 1.26; 95% CI, 1.12–1.41; I2 = 70.60%) and polycystic ovarian syndrome (PCOS) (RR, 1.46; 95% CI, 1.29–1.66; I2 = 0.00%). Regarding metabolic syndrome, severe periodontitis showed a stronger association (OR, 1.55; 95% CI, 1.33–1.79; I2 = 57.50%) than moderate periodontitis (OR, 1.30; 95% CI, 1.13–1.49; I2 = 51.10%). In contrast, no significant association was detected between periodontitis and non-alcoholic fatty liver disease (NAFLD) (OR, 1.04; 95% CI, 0.97–1.12; I2 = 58.50%). Overall, metabolic and endocrine associations were supported by Class III or weaker evidence, with no association reaching Class II. Musculoskeletal diseases Periodontitis showed significant positive associations with both rheumatoid arthritis (RA) (RR, 1.28; 95% CI, 1.06–1.54; I2 = 76.50%) and osteoporosis (RR, 1.40; 95% CI, 1.12–1.76; I2 = 93.30%). In contrast, no significant relationship was detected between periodontitis and implant loss (RR, 0.37; 95% CI, 0.10–1.40; I2 = 0.00%). Overall, musculoskeletal associations provided weak or non-significant evidence, with no association reaching Class II or Class III. Neurological and psychiatric diseases PD or periodontitis was significantly associated with elevated risks of several neurological and psychiatric conditions. The strongest association was observed with cognitive decline (RR, 3.01; 95% CI, 1.56–5.79; I2 = 82.40%), followed by Alzheimer's disease (OR, 1.79; 95% CI, 1.15–2.78; I2 = 25.00%). Significant but more modest risks were identified for cognitive impairment (RR, 1.27; 95% CI, 1.11–1.44; I2 = 34.60%) and dementia (RR, 1.22; 95% CI, 1.04–1.42; I2 = 97.60%). However, no significant association was found between periodontitis and depression (RR, 1.07; 95% CI, 0.95–1.21; I2 = 73.20%). Overall, neurological and psychiatric associations were supported mainly by Class III or weaker evidence, with no association reaching Class II. Gastrointestinal diseases Additionally, periodontitis demonstrated distinct associations with gastrointestinal conditions. A robust positive association was identified with Helicobacter pylori (H. pylori) infection (OR, 2.46; 95% CI, 2.01–3.01; I2 = 48.70%), followed by a significant link to Crohn’s disease (CD) (RR, 1.34; 95% CI, 1.04–1.73; I2 = 65.90%). In contrast, no statistically significant associations were found between periodontitis and inflammatory bowel disease (IBD) (RR, 1.15; 95% CI, 0.98–1.35; I2 = 82.60%), ulcerative colitis (UC) (RR, 1.00; 95% CI, 0.75–1.34; I2 = 75.90%), oral H. pylori infection in chronic periodontitis (OR, 1.87; 95% CI, 0.86–4.10; I2 = 65.80%), or liver disease (RR, 1.46; 95% CI, 0.94–2.26; I2 = 85.00%). These findings indicate that the evidence was stronger for gastric H. pylori infection than for oral H. pylori colonization in the context of chronic periodontitis. Overall, among gastrointestinal associations, only Helicobacter pylori infection met Class II criteria; the remaining gastrointestinal associations provided Class III or weaker evidence. Perinatal outcomes PD was significantly linked to adverse perinatal outcomes, demonstrating a notably elevated risk for low birth weight (LBW) (OR, 2.39; 95% CI, 1.70–3.38; I2 = 81.80%) and preterm birth (PTB) (OR, 1.58; 95% CI, 1.38–1.81; I2 = 8.90%). In contrast, no significant association was observed between PD and small for gestational age (SGA) (OR, 1.62; 95% CI, 0.86–3.06; I2 = 57.30%). Overall, perinatal associations were supported by Class III or weaker evidence, with no association reaching Class II. Respiratory diseases Periodontitis exhibited significant associations with several respiratory conditions, demonstrating the strongest association for asthma (OR, 3.54; 95% CI, 2.47–5.07; I2 = 0.00%) and pneumonia (OR, 3.21; 95% CI, 1.99–5.17; I2 = 0.00%). It was also significantly associated with obstructive sleep apnea (OSA) (OR, 2.17; 95% CI, 1.66–2.83; I2 = 42.60%) and chronic obstructive pulmonary disease (COPD) (OR, 1.78; 95% CI, 1.04–3.04; I2 = 37.80%). Overall, respiratory associations were supported by Class III or weaker evidence, with no association reaching Class II after applying the small-study-effects assessability criterion. Other diseases Periodontitis demonstrated significant associations with a spectrum of systemic and oral conditions. The largest summary estimate was observed for postoperative complications (OR, 4.72; 95% CI, 1.12–19.82; I2 = 56.80%); however, the wide confidence interval and extremely wide prediction interval indicate substantial imprecision and instability, and this association should therefore be interpreted with caution. Significant positive associations were also identified for erectile dysfunction (OR, 2.55; 95% CI, 1.70–3.84; I2 = 97.80%), dental caries (OR, 1.63; 95% CI, 1.32–2.01; I2 = 83.00%), psoriasis (RR, 1.54; 95% CI, 1.35–1.75; I2 = 21.40%), and age-related macular degeneration (ARMD) (RR, 1.35; 95% CI, 1.07–1.70; I2 = 80.40%). Overall, associations in the other-diseases category provided Class III or weaker evidence, with no association reaching Class II. Small-study effects Egger’s regression test was considered formally assessable only for the 19 associations (32.8%) that included at least 10 eligible effect estimates. Among these assessable associations, small-study effects were identified in 10 associations (10/19, 52.6%) (Table 1). For the remaining 39 associations with fewer than 10 eligible effect estimates, publication bias could not be reliably assessed using Egger’s test and was therefore treated as not assessable rather than interpreted as evidence for or against small-study effects. Excess significance The findings indicated that 33 associations (56.9%) exhibited excess significance bias (Table 1). When O ≤ E, the test was marked as not pertinent (NP), indicating that the observed number of statistically significant studies did not exceed the expected number; for evidence grading, NP was treated as satisfying the no-excess-significance criterion when that criterion was relevant. Evidence grading Despite thorough investigation, no convincing evidence was identified. Four associations (6.9%) presented highly suggestive evidence, namely Helicobacter pylori infection, urogenital cancer, total cancer, and oral cancer. Three of these four Class II associations had 95% prediction intervals excluding the null value; total cancer did not. In addition, none of the four Class II associations was supported by a representative meta-analysis rated as high quality by AMSTAR 2. Nineteen associations (32.8%) exhibited suggestive evidence. An additional 24 associations (41.4%) showed weak but statistically significant evidence. For the remaining 11 associations (19.0%), evidence was non-significant (Table 1). Discussion This umbrella review provides an integrated, credibility-based assessment of the reported associations between PD and NCDs. Although many PD–NCD associations were statistically significant, the overall evidence base remained limited: no association reached Class I evidence, only four reached Class II evidence, and only 11 of 58 associations had 95% prediction intervals excluding the null value. Therefore, the observed associations, including those with the strongest statistical support, should be interpreted as observational signals of varying credibility rather than evidence of causality. Among the identified associations, the relationship between PD and oral cancer was the most pronounced. Several mechanisms involving periodontal pathogens, chronic inflammation, immune dysregulation, and oral microbial dysbiosis have been proposed to explain a possible link between PD and oral cancer [73–76]. However, these mechanisms should be regarded as biologically plausible but hypothesis-generating rather than confirmatory. In the main analysis, the retained meta-analysis by Ma et al. reported an approximately threefold higher risk of oral cancer among individuals with PD [36]. Nevertheless, this finding should still be interpreted cautiously. Importantly, our exploratory subgroup analysis showed that the oral cancer association weakened when PD was restricted to objective clinical or radiographic definitions (OR 2.57; Class IV) compared with non-objective or proxy-based definitions (OR 3.21; Class II). This pattern suggests that non-objective or proxy-based PD definitions may overestimate the apparent magnitude and credibility of the association. Accordingly, oral cancer remains an important statistical signal, but its credibility is more limited when the analysis is restricted to objectively assessed PD. The present findings may justify increased clinical awareness of oral mucosal abnormalities in patients with objectively documented periodontitis, but they do not by themselves support strong screening recommendations or causal inference. This umbrella review also identified a statistically highly suggestive association between periodontitis and gastric Helicobacter pylori infection. However, this finding should be clearly distinguished from the separate association between chronic periodontitis and oral H. pylori infection, which was not statistically significant in our analysis. Therefore, although oral–gastric transmission and periodontal reservoir hypotheses remain biologically plausible, they were not directly confirmed by the present umbrella-review evidence [77–79]. These findings should be interpreted as hypothesis-generating observational evidence, and any benefit of adjunctive periodontal treatment for improving eradication or preventing reinfection still requires confirmation in prospective interventional studies. Several mechanisms have been proposed to explain possible PD–NCD links, including systemic inflammation, microbial dissemination, endothelial dysfunction, and immune dysregulation [80–84]. While these mechanisms are biologically plausible, they should be regarded as hypothesis-generating rather than conclusive evidence. Confounding and reverse causality remain important alternative explanations, particularly in neuropsychiatric or function-limiting disorders, where reduced self-care ability, deteriorating oral hygiene, dietary changes, medication effects, xerostomia, or reduced access to dental care may exacerbate periodontal conditions. Therefore, some reported PD–NCD associations may partly reflect shared risk factors or underlying disease processes that worsen periodontal status, rather than a unidirectional causal effect of PD on systemic disease incidence. A particularly important finding of this umbrella review was that 10 of 19 assessable associations (52.6%) showed small-study effects and 33 of 58 examined associations (56.9%) showed excess significance bias. These findings suggest that the broader PD–NCD meta-analytic literature may be vulnerable to inflation of positive findings, rather than being affected by isolated problems in only a few individual meta-analyses. In practical terms, the observed number of statistically significant studies may exceed the number expected based on study-level power, raising concern that positive associations may be overrepresented and that effect magnitudes may be overestimated. However, excess-significance testing has methodological limitations and depends on assumptions about study-level power, the selected plausible effect size, between-study heterogeneity, and the completeness of the evidence base. Therefore, positive excess-significance results should be interpreted as warning signals for possible bias or selective reporting, rather than definitive proof of publication bias. Accordingly, even when an association fulfills statistical criteria for Class II or Class III evidence, it should not automatically be interpreted as stable or unbiased evidence. The current PD–NCD evidence base should therefore be regarded as suggestive but provisional until better-powered, prospectively registered, and methodologically rigorous studies become available. Several limitations should be considered when interpreting these findings. First, the methodological quality of the included evidence was generally low: more than two-thirds of the meta-analyses were rated as low or critically low quality by AMSTAR 2. Because AMSTAR 2 and the Class I–V framework assess different dimensions, a statistically suggestive association is not necessarily methodologically robust. In a quality-based sensitivity check, none of the four Class II associations was supported by a high-quality representative meta-analysis: oral cancer and urogenital cancer relied on low-quality reviews, and H. pylori infection and total cancer on critically low-quality reviews; only one association had a higher-quality alternative, and reassessment was not feasible even there. Thus, even the statistically strongest associations in this review remain methodologically fragile. Second, heterogeneity was substantial (I2 > 50% in 63.8% of associations), and only 11 of 58 associations had prediction intervals excluding the null value, indicating limited robustness and transportability. Several prediction intervals were mathematically valid but interpretively uninformative given their extreme width and sparse underlying evidence. Third, although overall CCA indicated only slight overlap across meta-analyses, pairwise CCA revealed localized clusters of high overlap, concentrated among cancer-related reviews sharing large cohort studies. Related cancer associations should therefore be interpreted as evidence clusters rather than fully independent signals. Fourth, subgroup analyses by PD definition were exploratory and feasible only where study-level exposure data were retrievable, introducing possible selection bias. In addition, periodontal disease definitions have evolved over time, particularly before and after the adoption of the 2018 EFP/AAP classification system. This temporal heterogeneity in exposure definitions may affect subgroup findings and limit comparability between older and newer primary studies. Importantly, the oral cancer association weakened and was downgraded from Class II to Class IV when restricted to objective clinical/radiographic PD definitions, suggesting that exposure definition may materially affect the credibility of this signal. Additional limitations include: effect estimates (ORs, RRs, HRs) were preserved on their original scales and are not directly comparable across outcomes; the review addressed incidence rather than prognosis-related endpoints (e.g., mortality, recurrence, survival); no formal Bonferroni or false-discovery-rate correction was applied, which should be considered given the large number of association tests. Instead, as commonly used in umbrella reviews, we applied stricter credibility thresholds, including P < 10⁻⁶, number of cases, heterogeneity, prediction intervals, small-study effects, and excess-significance testing, to reduce overinterpretation of nominally significant associations. However, this credibility-grading approach is not equivalent to formal multiplicity control, and chance findings due to multiple testing cannot be fully excluded; only English-language publications were included; and although no age restriction was applied a priori, most evidence derived from adult populations, precluding age-stratified sensitivity analyses. Finally, because this review included only meta-analyses of observational studies, the findings should not be interpreted as establishing PD as a causal or modifiable risk factor for chronic NCDs, but rather as associations of varying credibility to guide future etiological research and hypothesis generation. Future research should prioritize large prospective cohort studies using standardized and objective periodontitis definitions, such as the 2018 EFP/AAP classification [3]. Mendelian randomization studies may also help clarify whether some of the observed associations reflect causal relationships rather than residual confounding. In addition, more randomized controlled trials are needed to determine whether periodontal therapy can reduce the incidence of associated chronic NCDs or improve clinically relevant outcomes after disease onset. Such evidence will be essential before stronger causal or practice-level conclusions can be justified. Conclusions This umbrella review identified statistically highly suggestive evidence that PD is associated with an increased risk of a limited number of chronic NCDs, namely oral cancer, Helicobacter pylori infection, urogenital cancer, and total cancer. However, total cancer did not have a 95% prediction interval excluding the null value, and none of the four Class II associations was supported by a representative meta-analysis rated as high quality by AMSTAR 2. In addition, the oral cancer association weakened and was downgraded to Class IV when restricted to objective clinical/radiographic PD definitions, indicating that exposure definition materially affects the credibility of this signal. Because only a minority of associations had prediction intervals excluding the null value and many representative meta-analyses were of low or critically low methodological quality, the overall robustness of the evidence remains limited. These findings relate to disease occurrence rather than prognosis after diagnosis and should not be interpreted as evidence of causality. Whether treatment of PD can reduce the incidence or improve prognosis of chronic NCDs remains to be established in future randomized trials and other causal-inference studies. At present, PD should be regarded as a statistically associated factor across a limited set of chronic NCDs, with insufficient evidence to support causal or modifiable interpretation. Supplementary Information Additional file 1: The PRISMA 2020 checklist of this review. Additional file 2: Tables S1–S10, Figures S1-S4.