Clinical Attachment Loss is Cross-Sectionally Associated with Elevated Glucose among Adults without Diabetes

Abstract

Aim:

We investigate whether periodontal measures are cross-sectionally associated with prediabetes and cardiometabolic biomarkers among non-diabetic younger adults.

Materials and Methods:

N=1071 participants (mean age=32.2 years[SE=0.3]; 73% female) from the Oral Infections, Glucose Intolerance and Insulin Resistance Study were enrolled. Full-mouth clinical attachment loss (fm-CAL), probing depth (fm-PD), and bleeding on probing were ascertained. Interproximal CAL (i-CAL) and probing depths (i-PD) served as our primary exposures. Glucose, HbA1c, insulin, and insulin resistance (HOMA-IR) outcomes were assessed from fasting blood. Prediabetes was defined per American Diabetes Association guidelines. Prediabetes prevalence ratios (PR[95%CI]) and mean[SE] cardiometabolic biomarkers were regressed on periodontal variables via multivariable robust variance Poisson regression or multivariable linear regression.

Results:

Prevalence of prediabetes was 12.5%. Fully-adjusted prediabetes PR in Tertiles 3 vs 1 of mean i-CAL was 2.42(1.77,3.08). Fully-adjusted fasting glucose estimates across i-CAL tertiles were 83.29[0.43], 84.31[0.37], 86.48[0.46]; p for trend<0.01. Greater percent of sites with i-PD≥3mm showed elevated natural-log-HOMA-IR after adjustment (0–12% of sites=0.33[0.03], 13–26% of sites=0.39[0.03], ≥27% of sites=0.42[0.03]; p for trend=0.04).

Conclusion:

i-CAL (vs. fm-CAL) was associated with elevated fasting glucose and prediabetes whereas i-PD (vs. fm-PD) was associated with insulin resistance. Future studies are needed to examine periodontal disease and incident prediabetes.

1. INTRODUCTION

Prediabetes is a growing public health problem.¹ Approximately 35% of adults in the United States (~88 million) are estimated to have prediabetes.² While prediabetes is a strong predictor of future diabetes development, it is reversible, as approximately 13% of individuals with prediabetes revert to normal glycemic levels.³ Risk factor management in the prediabetic state could reduce the burden of diabetes, which is a leading cause of morbidity and mortality in the US.^4,5

Periodontitis is a chronic inflammatory disease characterized by microbial dysbiosis of the subgingival biofilm and a breach of the homeostatic defenses at the dento-gingival niche^6,7. It is also hypothesized as a potential risk factor for incident diabetes.^8–11 Constituents of the periodontal microbiome have been identified to associate with impaired glucose regulation and insulin resistance,^12–14 and earlier randomized controlled trials demonstrated that periodontal treatment improves glycemic control in individuals with diabetes.^15–17 However, previous literature has largely focused on incident diabetes in diabetes-free participants, or glycemic changes among participants with diabetes and periodontitis. Less is known about the relationship between periodontitis and prediabetes.

Previous studies have provided mixed findings regarding the relationship between periodontitis and prediabetes, with several studies indicating higher prediabetes incidence or prevalence among those with versus without periodontitis,^18,19 while others observed inverse or no associations.^20–22 Although many prior studies assessing the relationship between periodontitis and diabetes risk have enrolled populations with mean age >40 years with higher risk for diabetes,^8,23–27 few studies have enrolled younger individuals free of prevalent cardiometabolic disease. Furthermore, many studies use categorical definitions of periodontitis that do not represent periodontitis as a continuum, nor do they contrast current versus cumulative measures of disease.

Presently, we investigate the cross-sectional association between several characterizations of periodontitis and the prevalence of prediabetes and cardiometabolic biomarkers among young-to-middle-aged non-diabetic adults enrolled in the Oral Infections, Glucose Intolerance, and Insulin Resistance Study (ORIGINS). We hypothesize that periodontitis and poorer periodontal health measures will be associated with i) a higher prevalence of prediabetes, ii) elevated fasting plasma glucose and hemoglobin A1c (HbA1c), and iii) greater fasting insulin levels and insulin resistance.

2. METHODS

2.1. Study Background and Design

ORIGINS is an ongoing prospective cohort study examining features of periodontitis in relation to impaired glucose metabolism among a racially diverse population without diabetes, conducted at the Columbia University Irving Medical Center (CUIMC) in New York City, NY.^12,28 Participants who were i) age 20–55 years old; ii) did not have diabetes mellitus (Type 1 or 2) via no self-report of physician diagnosis; iii) fasting plasma glucose <126 mg/dL and hemoglobin A1c <6.5%; and iv) reported no history of cardiovascular disease or chronic inflammatory conditions were enrolled into the study.^12,28 Participant enrollment occurred across two study waves. Wave 1 participants were recruited between February 2011 to May 2013 from an employee union at the CUIMC via direct mail, through recruitment flyers, and advertisements in local publications. Wave 2 participants were recruited between January 2016 to January 2020 from any employee and student from the CUIMC via recruitment flyers and advertisements in local publications.^{12,13,28–31} From a baseline cohort of n=1100 (Wave 1: n=300, Wave 2: n=800), we excluded 29 participants who were missing periodontal data, cardiometabolic data, or key demographic measures, resulting in a final sample of n=1071 (Supplemental Figure 1). Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were followed in constructing this cross-sectional analysis (Supplemental Materials).

2.2. Ethics Statement

All participants in ORIGINS provided written informed consent prior to enrollment. The Institutional Review Boards at Columbia University and the University of Minnesota approved study protocols and procedures.

2.3. Periodontal Measures

Assessment of periodontal status in ORIGINS have been described previously.^12,28 Briefly, trained and calibrated dental examiners conducted full-mouth periodontal assessments.²⁸ Examiners used Hu-Friedy UNC-15 probes to measure clinical attachment loss (CAL), probing depth (PD), and bleeding on probing (BOP) at six sites per tooth (mesiobuccal, midbuccal, distobuccal, mesiolingual, midlingual, and distolingual).²⁸ Periodontitis was defined using the Centers for Disease Control and Prevention and American Academy of Periodontology classification (CDC/AAP).³² We considered CAL to reflect cumulative periodontal tissue destruction, probing depth and BOP to characterize current periodontal tissue inflammation, and the CDC/AAP system as a combination of both.³³

We first dichotomized CDC/AAP periodontal categories into the categories of healthy/mild disease and moderate/severe disease. In a sensitivity analysis, we additionally classified periodontitis as healthy vs mild/moderate/severe. Then we computed mean full-mouth CAL (fm-CAL), mean full-mouth PD (fm-PD), and the percent of sites with BOP for each participant across all available probing sites (up to 192 possible sites), and divided each full-mouth measure into tertiles. Mean interproximal CAL (i-CAL) and interproximal PD (i-PD) were also measured and categorized into tertiles. Lastly, we computed the extent and severity of fm-CAL and fm-PD using the following categorizations: i) the percent of probed full-mouth sites with CAL ≥3mm (%sites-fm-CAL; 0%, 1–2%, 3–9%, 10% or more) and ii) the percent of probed full-mouth sites with PD ≥3mm (%sites-fm-PD; 0–12%, 13–26%, 27% or more). The same categorizations were applied for interproximal CAL (%sites-i-CAL) and interproximal PD (%sites-i-PD), respectively. A cutpoint of 3mm was selected as an indicator of nascent disease at a given site due to the lack of severe periodontal disease (i.e., very few participants had CAL>3mm) in this young, generally periodontally healthy cohort.

2.4. Cardiometabolic Laboratory Measures

Trained examiners collected participant blood samples after overnight fast.^12,28 Fasting plasma glucose and glycated hemoglobin A1c (HbA1c) levels were derived from blood samples using standard methods (Cobra Integra 400 Plus; Roche Diagnostics, Indianapolis, IN, USA).²⁸ High-sensitivity insulin was ascertained from fasting blood samples and measured using a Millipore insulin testing kit during Wave 1 (Millipore; Billerica, MA, USA) and a Mercodia insulin testing kit during Wave 2 (Mercodia Ultrasensitive Insulin ELISA; Uppsala, Sweden). Test-retest assessments for these kits found a correlation of 0.98 (p<0.001) among n=14 ORIGINS participants. We used fasting blood glucose and high-sensitivity insulin measures to calculate the Homeostasis Model Assessment for Insulin Resistance (HOMA-IR).³⁴ Prevalent prediabetes was determined per the American Diabetes Association definition of fasting plasma glucose between 100–125 mg/dL or HbA1c between 5.7–6.4%.³⁵

2.5. Covariables

Sociodemographic factors including age, sex, race/ethnicity (as a social construct), education level, and annual income were ascertained via self-report during study visits. Behavioral risk factors of smoking history, smoking pack-years, and physical activity (metabolic equivalent of task units) were also obtained from self-report. Systolic and diastolic blood pressures were measured in triplicate, with the average taken from the last two readings.²⁸ Serum lipid profile components of total cholesterol and high-density lipoprotein (HDL) were ascertained through standard methods (Cobras Integra 400 Plus) while low-density lipoprotein (LDL) was estimated using the Friedewald Equation. Body mass index (BMI) was ascertained from height and weight measures during study visits.²⁸ Lastly, healthy diet composition was measured from a validated food frequency questionnaire and quantified via the Alternative Healthy Eating Index (AHEI).³⁶ Details of this metric have been described in prior work within ORIGINS.²⁹ In short, AHEI is a composite diet score derived from 11 dietary elements (intake of fruits, vegetables, nuts, red meat, sugar-sweetened beverages, omega-3 fatty acids, polyunsaturated fats, trans fats, alcohol, whole grains, and refined grains).^29,36 Scores range from 16.4 to 83.0, with higher scores corresponding to healthier quality diets.^29,36

2.6. Statistical Analyses

We first assessed participant characteristics across tertiles of mean i-CAL and CDC/AAP periodontal disease classification (healthy/mild vs moderate/severe disease). Next, we used robust variance Poisson regression to estimate prevalence ratios and 95% confidence intervals (PR [95% CI]) of prediabetes across periodontal variables (reference=healthy/mild disease or lowest tertile/category).³⁷ Then we employed linear regression to compute mean estimates and standard errors [SE] of fasting plasma glucose, HbA1c, natural-log (LN) insulin, and LN-HOMA-IR across periodontal variables. Lastly, we assessed obesity as an effect measure modifier in the associations between periodontitis, fm-CAL, i-CAL and our cardiometabolic outcomes through stratification of obesity status (BMI ≥30 kg/m²) and interaction testing using a continuous BMI variable.

We used multivariable modeling to adjust for study wave (Model 1 = wave), demographics (Model 2 = Model 1 + age, race/ethnicity, sex, education, income), health behaviors (Model 3 = Model 2 + smoking status, BMI, AHEI), and cardiovascular risk factors (Model 4 = Model 3 + HDL, mean systolic and diastolic pressures). Linear trend tests were conducted for all models.

To account for missing observations of key covariables (education, income, smoking status, BMI, physical activity, AHEI, systolic blood pressure, diastolic blood pressure), we performed multiple imputations via chained equations assuming the data were missing at random.³⁸ Imputations were conducted via fully conditional specification through a three phase process in SAS. First, we separately generated imputation datasets from 10 replications using ‘PROC MI’. Next, we ran our regression models within each of the imputed replications. Lastly, we pooled the generated model estimates and variances from each replication with ‘PROC MIANALYZE’. Data preparation and analyses were performed in PC-SAS (Version 9.4, SAS Institute Inc., Cary, NC). Figures were generated in RStudio (Version 4.2.2, The R Foundation for Statistical Computing Platform).

2.7. DATA AVAILABILITY

Data used in this study are available upon request to the investigators.

3. RESULTS

3.1. Participant Characteristics

Among n=1071 participants, average age at baseline was 32.3 years (SE=0.3) while 72.5% of participants were female (reflective of the source population which is predominately female). The distribution of participants by self-identified race/ethnicity is as follows: 15.2% Black, 33.1% Hispanic, 27.2% Non-Hispanic White, and Other race/ethnicity: 24.5%.

Table 1 presents participant characteristics for the entire sample and by tertiles of mean i-CAL. Overall, 36.1% of participants had moderate/severe periodontitis while prediabetes prevalence was 12.5%. Compared to the lowest i-CAL tertile, participants in the highest tertile were more likely to have higher incomes, lower education, higher prevalence of overweight or obese BMI, higher blood pressures, and be current or former smokers. Furthermore, participants in the highest i-CAL tertile were more likely to have moderate/severe periodontal disease, greater mean probing depth, and greater interproximal probing depths and CAL compared to those in the lowest i-CAL tertile. Lastly, prediabetes prevalence and glucose measures tended to be greater among participants in i-CAL Tertile 3 vs Tertile 1. Supplemental Table 1 presents participant characteristics by study wave and CDC/AAP periodontal classification.

Table 1:

Participant characteristics by tertiles of mean interproximal clinical attachment loss; ORIGINS (n=1071, 2011–2020)

	Mean [Standard Error] or % (n)
		Mean Interproximal Clinical Attachment Loss (i-CAL)
	All	Tertile 1	Tertile 2	Tertile 3	p-value
	N=1071	n=355	n=359	n=357
Age, years	32.3 [0.3]	27.3 [0.4]	33.9 [0.5]	35.8 [0.5]	<0.01
Sex					0.37
Male	27.5% (294)	24.8% (88)	29.2% (105)	28.3% (101)
Female	72.5% (777)	75.2% (355)	70.8% (254)	71.7% (256)
Race/ethnicity					<0.01
Black	15.2% (163)	10.7% (38)	17.3% (62)	17.7% (63)
Hispanic	33.1% (355)	18.0% (64)	34.0% (122)	47.3% (169)
Non-Hispanic White	27.2% (291)	40.3% (143)	22.0% (79)	19.3% (69)
Other self-identified race/ethnicity	24.5% (262)	31.0% (110)	26.7% (96)	15.7% (56)
Income					<0.01
< $34,999	35.1% (376)	45.1% (160)	35.9% (129)	24.4% (87)
$35,000 to $64,999	34.0% (364)	27.6% (98)	33.2% (119)	41.2% (147)
≥ $65,000	28.6% (306)	24.5% (87)	27.6% (99)	33.6% (120)
Missing	2.3% (25)	2.8% (10)	3.3% (12)	0.8% (3)
Education					<0.01
High school completion	25.2% (270)	9.3% (33)	26.7% (96)	39.5% (141)
College/vocational training completion	47.7% (511)	59.4% (211)	43.2% (155)	40.6% (145)
Advanced degree	25.4% (272)	29.9% (106)	27.3% (98)	19.1% (68)
Missing	1.7% (18)	1.4% (5)	2.8% (10)	0.8% (3)
Physical activity level					<0.01
None	31.9% (342)	23.7% (84)	37.6% (135)	34.5% (123)
Low	8.5% (91)	11.0% (39)	4.7% (17)	9.8% (35)
Moderate	13.5% (145)	12.7% (45)	14.2% (51)	13.7% (49)
High	37.2% (398)	44.8% (159)	30.4% (109)	36.4% (130)
Missing	8.9% (95)	7.9% (28)	13.1% (47)	5.6% (20)
Body mass index (kg/m2)	25.9 [0.2]	24.2 [0.3]	26.2 [0.3]	27.3 [0.3]	<0.01
Body mass index (categorical)					<0.01
Normal	52.9% (567)	67.0% (238)	51.3% (184)	40.6% (145)
Overweight	27.5% (295)	20.6% (73)	27.3% (98)	34.7% (124)
Obese	18.2% (195)	10.7% (38)	20.0% (72)	23.88% (85)
Missing	1.3% (14)	1.7% (6)	1.4% (5)	0.8% (3)
Blood pressure, mmHg
Systolic	116.9 [0.4]	113.8 [0.6]	118.2 [0.7]	118.5 [0.7]	<0.01
Diastolic	73.0 [0.3]	69.8 [0.4]	73.6 [0.5]	75.6 [0.5]	<0.01
Missing	0.4% (4)	0.0% (3)	0.0% (1)	0.0% (0)
Cholesterol, mg/dL
Total	173.4 [1.0]	169.7 [1.7]	176.4 [1.8]	174.0 [1.7]	0.02
HDL	60.4 [0.5]	61.7 [0.8]	60.5 [0.9]	58.9 [0.9]	0.07
LDL	96.9 [0.9]	92.7 [1.5]	99.3 [1.6]	98.7 [1.5]	<0.01
Missing LDL	0.6% (6)	0%	0%	0%
Total cholesterol/HDL ratio	3.1 [0.0]	2.9 [0.0]	3.1 [0.1]	3.2 [0.1]	<0.01
Smoking History					<0.01
Current	7.2% (77)	3.9% (14)	7.88% (28)	9.8% (35)
Former	7.7% (83)	6.2% (22)	7.2% (26)	9.8% (35)
Never	81.6% (874)	85.6% (304)	79.9% (287)	79.3% (283)
Missing	3.5% (37)	4.2% (15)	5.0% (18)	1.1% (4)
Smoking intensity
Pack-years	0.5 [0.1]	0.2 [0.1]	0.5 [0.1]	0.9 [0.2]	<0.01
Missing	3.0% (32)	2.5% (9)	3.9% (14)	2.5% (9)
Alternative Healthy Eating Index	46.6 [0.4]	47.7 [0.7]	45.0 [0.6]	47.0 [0.7]	<0.01
Missing	4.7% (50)	2.8% (10)	2.8% (10)	8.4% (30)
Periodontal status (CDC/AAP)					<0.01
Healthy/mild disease	63.9% (684)	85.1% (302)	69.1% (248)	37.5% (134)
Moderate/severe disease	36.1% (387)	14.9% (53)	30.9% (111)	62.5% (223)
Mean full-mouth probing depth, mm	2.0 [0.0]	1.9 [0.0]	1.9 [0.0]	2.2 [0.0]	<0.01
Median (IQR)	1.9 (1.8, 2.2)	1.9 (1.8, 2.0)	1.9 (1.7, 21)	2.3 (2.0, 2.5)
Mean interproximal probing depth, mm	2.2 [0.0]	2.1 [0.0]	2.1 [0.0]	2.5 [0.0]	<0.01
Median (IQR)	2.2 (2.0, 2.5)	2.1 (2.0, 2.2)	2.1 (1.9, 2.4)	2.4 (2.2, 2.7)
Mean full-mouth clinical attachment loss, mm	0.8 [0.0]	0.1 [0.0]	0.6 [0.0]	1.7 [0.0]	<0.01
Median (IQR)	0.5 (0.2, 1.4)	0.1 (0.1, 0.2)	0.5 (0.3, 0.8)	1.6 (1.4, 1.9)
Mean interproximal attachment loss, mm	0.8 [0.0]	0.1 [0.0]	0.5 [0.0]	1.8 [0.0]	<0.01
Median (IQR)	0.5 (0.1, 1.4)	0.1 (0.0, 0.1)	0.5 (0.3, 0.7)	1.7 (1.4, 1.9)
Mean % of sites with bleeding on probing	30.0% [0.0]	0.3 [0.0]	0.3 [0.0]	0.4 [0.0]	<0.01
Median (IQR)	30.0% (20.0%, 50.0%)	30.0% (20.0%, 40.0%)	30.0% (20.0%, 40.0%)	40.0% (20.0%, 60.0%)
Prediabetes status					<0.01
Healthy	87.5% (937)	96.3% (342)	87.5% (314)	78.7% (281)
Prediabetic	12.5% (134)	3.7% (13)	12.5% (45)	21.3% (76)
Fasting plasma glucose, mg/dL	84.7 [0.2]	82.4 [0.4]	85.0 [0.4]	86.7 [0.4]	<0.01
Median (IQR)	84.0 (80.0, 89.0)	82.0 (78.0, 86.0)	84.0 (80.0, 90.0)	86.0 (81.0, 91.0)
HbA1c, %	5.3 [0.0]	5.2 [0.0]	5.3 [0.0]	5.4 [0.4]	<0.01
Median (IQR)	5.3 (5.1, 5.5)	5.2 (5.0, 5.3)	5.3 (5.1, 5.5)	5.4 (5.2, 5.6)
Insulin, uIU/mL	8.4 [0.2]	1.8 [0.0]	1.9 [0.0]	2.1 [0.0]	<0.01
Median (IQR)	6.8 (4.8, 10.1)	1.8 (1.5, 2.2)	1.9 (1.5, 2.3)	2.1 (1.7, 2.4)
Insulin resistance, HOMA-IR	1.8 [0.0]	0.2 [0.0]	0.4 [0.0]	0.5 [0.0]	<0.01
Median (IQR)	1.4 (1.0, 2.2)	0.2 (-0.1, 0.6)	0.3 (0.0, 0.8)	0.5 (0.1, 0.9)

Abbreviations: CDC/AAP = Centers for Disease Control and Prevention/ American Academy of Periodontology periodontitis classification; HDL = high density cholesterol; HOMA-IR = Homeostasis Model Assessment for Insulin Resistance; IQR = interquartile range; LDL = low density cholesterol; ORIGINS = Oral Infections Glucose Intolerance and Insulin Resistance Study. P-values were computed using a Chi-squared testing for categorical variables and ANOVA for continuous variables to assess statistical difference across tertiles of mean i-CAL. P-value <0.05 indicates statistical significance.

3.2. Periodontitis and Cardiometabolic Outcomes

Compared to periodontally healthy/mild disease, moderate/severe periodontitis was associated with a 12% greater prediabetes prevalence in a full covariable-adjusted model, although the confidence interval included 1 (1.12 [0.79, 1.45]; Table 2). Estimated glucose, HbA1c, insulin, and HOMA-IR were similar across periodontitis groups in multivariable modeling. Sensitivity analyses were consistent when periodontitis was categorized as healthy vs mild/moderate/severe disease (data not shown).

Table 2:

Association between periodontitis and cardiometabolic measures; ORIGINS (2011–2020)

	Periodontal Status (CDC/AAP classification)
	Healthy/Mild Disease	Moderate/Severe Disease	Linear Trend
	n=684	n=387
Prediabetes (PR, 95% CI)	n with prediabetes = 59	n with prediabetes = 75
Model 1	Ref.	2.09 (1.51, 2.89)	<0.01
Model 2	Ref.	1.18 (0.83, 1.53)	0.40
Model 3	Ref.	1.16 (0.84, 1.49)	0.44
Model 4	Ref.	1.12 (0.79, 1.45)	0.57
Fasting plasma glucose, mg/dL, mean (SE)
Model 1	83.81 (0.29)	86.26 (0.39)	<0.01
Model 2	84.50 (0.28)	85.03 (0.38)	0.29
Model 3	84.62 (0.27)	84.83 (0.37)	0.66
Model 4	84.63 (0.27)	84.81 (0.37)	0.71
HbA1c, %, mean (SE)
Model 1	5.25 (0.01)	5.25 (0.01)	<0.01
Model 2	5.35 (0.02)	5.35 (0.02)	0.50
Model 3	5.28 (0.01)	5.29 (0.02)	0.66
Model 4	5.28 (0.01)	5.28 (0.01)	0.71
LN-Insulin, uIU/mL, mean (SE)
Model 1	1.92 (0.02)	2.02 (0.03)	0.01
Model 2	1.94 (0.02)	1.98 (0.03)	0.22
Model 3	1.95 (0.02)	1.96 (0.03)	0.75
Model 4	1.95 (0.02)	1.96 (0.02)	0.81
LN-HOMA-IR, mean (SE)
Model 1	0.34 (0.02)	0.34 (0.02)	<0.01
Model 2	0.46 (0.03)	0.46 (0.03)	0.20
Model 3	0.38 (0.02)	0.39 (0.03)	0.74
Model 4	0.37 (0.02)	0.37 (0.02)	0.81

Data are derived from multiple imputation analysis to account for missing covariates (10 imputations of n=1071 analytical sample via fully conditional specification).

Model 1: adjusts for study wave; Model 2: Model 1 + adjustments for age, race/ethnicity, sex, education, income; Model 3: Model 2 + adjustments for smoking status, body mass index, Alternative Healthy Eating Index. Model 4: Model 3 + adjustments for high density cholesterol, mean systolic and diastolic blood pressures. P-value <0.05 indicates statistically significant linear trend.

Abbreviations: PR = prevalence ratio; SE = standard error; CI = confidence interval; HOMA-IR = Homeostasis Model Assessment for Insulin Resistance; ORIGINS = Oral Infections Glucose Intolerance and Insulin Resistance Study; CDC/AAP = Centers for Disease Control and Prevention/American Academy of Periodontology periodontitis classification; LN = natural log.

3.3. Clinical Attachment Loss and Cardiometabolic Outcomes

Displayed in Table 3 and Figure 1, fully-adjusted prediabetes PRs increased across mean i-CAL tertiles (Tertile 2: 1.74 [1.13, 2.34], Tertile 3: 2.42 [1.77, 3.08]; p for trend=0.01). Additionally shown in Figure 2, fasting plasma glucose levels increased across tertiles of i-CAL in multivariable modeling (Tertile 1: 83.29 [0.43] mg/dl; Tertile 2: 84.31 [0.37] mg/dL; Tertile 3: 86.48 [0.46] mg/dL; p for trend=0.01). HbA1c similarly elevated across tertiles (p for trend<0.01), while no significant trends for insulin and HOMA-IR were found. Glucose and HbA1c were consistently higher as the %sites-i-CAL≥3mm became greater (Supplemental Table 2).

Table 3:

Association between tertiles of mean interproximal clinical attachment loss and cardiometabolic measures; ORIGINS (2011–2020)

	Mean Interproximal Clinical Attachment Loss (i-CAL)
	Tertile 1 n=355	Tertile 2 n=359	Tertile 3 n=357	Linear Trend
Mean i-CAL (min, max), mm	0.08 (0.00, 0.20)	0.53 (0.21, 1.10)	1.78 (1.11, 7.67)
*Prediabetes	n=13	n=45	n=76
Model 1	Ref.	3.54 (1.94, 6.46)	7.12 (3.87, 13.12)	<0.01
Model 2	Ref.	1.94 (1.32, 2.56)	2.88 (2.22, 3.53)	<0.01
Model 3	Ref.	1.80 (1.18, 2.41)	2.70 (2.05, 3.34)	<0.01
Model 4	Ref.	1.74 (1.13, 2.34)	2.42 (1.77, 3.08)	0.01
**Fasting plasma glucose, mg/dL (SE)
Model 1	81.58 (0.43)	84.58 (0.40)	87.91 (0.48)	<0.01
Model 2	83.09 (0.44)	84.36 (0.38)	86.63 (0.48)	<0.01
Model 3	83.23 (0.43)	84.33 (0.37)	86.52 (0.47)	<0.01
Model 4	83.29 (0.43)	84.31 (0.37)	86.48 (0.46)	<0.01
**HbA1c, % (SE)
Model 1	5.14 (0.02)	5.30 (0.02)	5.42 (0.02)	<0.01
Model 2	5.22 (0.02)	5.29 (0.02)	5.35 (0.02)	<0.01
Model 3	5.22 (0.02)	5.29 (0.02)	5.35 (0.02)	<0.01
Model 4	5.23 (0.02)	5.28 (0.02)	5.35 (0.02)	<0.01
**LN-Insulin, uIU/mL (SE)
Model 1	1.90 (0.03)	1.99 (0.03)	1.98 (0.04)	0.11
Model 2	1.97 (0.03)	1.97 (0.03)	1.92 (0.04)	0.39
Model 3	2.00 (0.03)	2.00 (0.03)	1.90 (0.03)	0.07
Model 4	2.00 (0.03)	1.97 (0.03)	1.90 (0.03)	0.06
**LN-HOMA-IR (SE)
Model 1	0.30 (0.03)	0.41 (0.03)	0.44 (0.04)	<0.01
Model 2	0.38 (0.04)	0.40 (0.03)	0.37 (0.04)	0.88
Model 3	0.41 (0.03)	0.40 (0.03)	0.35 (0.03)	0.30
Model 4	0.41 (0.03)	0.39 (0.03)	0.35 (0.03)	0.28

Data are derived from multiple imputation analysis to account for missing covariates (10 imputations of n=1071 analytical sample via fully conditional specification).

^*Prediabetes models present prevalence ratios for Tertiles 2 and 3 versus Tertile 1 of mean i-CAL.

^**Fasting glucose, HbA1c, LN-Insulin, and LN-HOMA-IR models present mean(SE) of the respective outcomes across tertiles of mean i-CAL.

Model 1: adjusts for study wave; Model 2: Model 1 + adjustments for age, race/ethnicity, sex, education, income; Model 3: Model 2 + adjustments for smoking status, body mass index, Alternative Healthy Eating Index. Model 4: Model 3 + adjustments for high density cholesterol, mean systolic and diastolic blood pressures. P-value <0.05 indicates statistically significant linear trend.

Abbreviations: PR = prevalence ratio; SE = standard error; CI = confidence interval; CAL = attachment loss; HOMA-IR = Homeostasis Model Assessment for Insulin Resistance; ORIGINS = Oral Infections Glucose Intolerance and Insulin Resistance Study; LN = natural log.

Figure 1:

Multivariable Prediabetes Prevalence Comparison of Full-Mouth versus Interproximal Periodontal Measures

Fully-adjusted prevalence ratios computed from robust variance Poisson regression with study wave, age, race/ethnicity, sex, education, income, smoking status, body mass index, Alternative Healthy Eating Index, high density cholesterol, systolic blood pressure, and diastolic blood pressure as covariables. Abbreviations: CI = Confidence Interval

Figure 2:

Multivariable Estimated Cardiometabolic Biomarkers across Periodontal Measures

Multivariable estimates of cardiometabolic biomarkers computed from linear regression with study wave, age, race/ethnicity, sex, education, income, smoking status, body mass index, Alternative Healthy Eating Index, high density cholesterol, systolic blood pressure, and diastolic blood pressure as covariables. Biomarkers are listed as follows: A = Fasting plasma glucose (mg/dL); B = HbA1c (%), C = LN-Insulin; D = LN-HOMA-IR. Abbreviations: LN = Natural Log; HOMA-IR = Homeostasis Model Assessment for Insulin Resistance. Linear trend p-value is listed within each figure (p-value < 0.05 indicates significance). Asterisks represent p-values for two-sample t-tests of estimated biomarkers (Tertile 1 set as reference). * indicates p-value <0.05; ** indicates p-value < 0.01.

The associations between tertiles of mean fm-CAL and cardiometabolic outcomes are shown in Supplemental Table 3. Compared to Tertile 1, the fully-adjusted prediabetes PRs in Tertiles 2 and 3 were 1.34 (0.79, 1.88) and 1.98 (1.39, 2.56), respectively (also presented in Figure 1). Estimated glucose was higher in Tertile 3 than in Tertile 1 in multivariable modeling (86.46 [0.46] mg/dL vs 82.24 [0.42] mg/dL; p for trend<0.01; also shown in Figure 2). HbA1c exhibited comparable elevations across fm-CAL tertiles (p for trend<0.01), while no notable trends for insulin and HOMA-IR were found. Lastly, %sites-fm-CAL≥3mm was not associated with prediabetes prevalence or cardiometabolic biomarkers (Supplemental Table 4).

3.4. Probing Depth and Cardiometabolic Outcomes

After full covariable adjustment, participants in the third i-PD tertile had higher estimated LN-insulin (1.97 [0.03] uIU/mL) and LN-HOMA-IR (0.40 [0.03]) compared to those in the first Tertile (LN-insulin=1.91 [0.03] uIU/mL; LN-HOMA-IR=0.35 [0.03]; Table 4 and Figure 2). PRs for prediabetes, fasting glucose, and HbA1c were generally similar across i-PD tertiles (Figure 1). As for %sites-i-PD≥3mm, there were linear increases in LN-insulin (1.89 [0.03] uIU/mL; 1.96 [0.03] uIU/mL; 1.99 [0.02] uIU/mL; p for trend=0.02) as well as LN-HOMA-IR (0.33 [0.03]; 0.39 [0.03]; 0.42 [0.03]; p for trend=0.04) after full multivariable adjustment (Table 5). Consistent with mean i-PD, glucose and HbA1c levels were similar across categories of %sites-i-PD≥3mm.

Table 4:

Association between tertiles of mean interproximal probing depth and cardiometabolic measures; ORIGINS (2011–2020)

	Mean Interproximal Probing Depth (i-PD)
	Tertile 1 n=352	Tertile 2 n=362	Tertile 3 n=357	Linear Trend
Mean i-PD (min, max), mm	1.84 (1.23, 2.07)	2.20 (2.07, 2.35)	2.67 (2.35, 6.63)
*Prediabetes (PR, 95% CI)	n=40	n=33	n=61
Model 1	Ref.	0.76 (0.49, 1.17)	1.21 (0.73, 2.03)	0.51
Model 2	Ref.	0.82 (0.39, 1.24)	0.94 (0.47, 1.42)	0.82
Model 3	Ref.	0.83 (0.42, 1.25)	0.81 (0.35, 1.27)	0.43
Model 4	Ref.	0.86 (0.44, 1.29)	0.81 (0.35, 1.27)	0.43
**Fasting plasma glucose, mg/dL (SE)
Model 1	84.47 (0.44)	83.86 (0.41)	85.77 (0.47)	0.13
Model 2	84.95 (0.42)	84.26 (0.38)	84.89 (0.45)	0.79
Model 3	85.13 (0.41)	84.47 (0.37)	84.49 (0.44)	0.28
Model 4	85.15 (0.41)	84.58 (0.37)	84.37 (0.44)	0.21
**HbA1c, % (SE)
Model 1	5.30 (0.02)	5.24 (0.02)	5.32 (0.02)	0.79
Model 2	5.31 (0.02)	5.26 (0.02)	5.29 (0.02)	0.42
Model 3	5.31 (0.02)	5.26 (0.02)	5.28 (0.02)	0.21
Model 4	5.31 (0.02)	5.27 (0.02)	5.28 (0.02)	0.19
**LN-Insulin, uIU/mL (SE)
Model 1	1.87 (0.03)	1.93 (0.03)	2.06 (0.03)	<0.01
Model 2	1.87 (0.03)	1.95 (0.03)	2.04 (0.03)	<0.01
Model 3	1.90 (0.03)	1.97 (0.03)	1.99 (0.03)	0.05
Model 4	1.91 (0.03)	1.98 (0.02)	1.97 (0.03)	0.12
**LN-HOMA-IR (SE)
Model 1	0.30 (0.03)	0.36 (0.03)	0.50 (0.04)	<0.01
Model 2	0.31 (0.03)	0.37 (0.03)	0.47 (0.04)	<0.01
Model 3	0.34 (0.03)	0.40 (0.03)	0.42 (0.03)	0.10
Model 4	0.35 (0.03)	0.41 (0.03)	0.40 (0.03)	0.20

Data are derived from multiple imputation analysis to account for missing covariates (10 imputations of n=1071 analytical sample via fully conditional specification).

^*Prediabetes models present prevalence ratios for Tertiles 2 and 3 versus Tertile 1 of mean i-PD.

^**Fasting glucose, HbA1c, LN-Insulin, and LN-HOMA-IR models present mean(SE) of the respective outcomes across tertiles of mean i-PD.

Model 1: adjusts for study wave; Model 2: Model 1 + adjustments for age, race/ethnicity, sex, education, income; Model 3: Model 2 + adjustments for smoking status, body mass index, Alternative Healthy Eating Index. Model 4: Model 3 + adjustments for high density cholesterol, mean systolic and diastolic blood pressures. P-value <0.05 indicates statistically significant linear trend.

Abbreviations: PR = prevalence ratio; SE = standard error; CI = confidence interval; HOMA-IR = Homeostasis Model Assessment for Insulin Resistance; ORIGINS = Oral Infections Glucose Intolerance and Insulin Resistance Study; LN = natural log.

Table 5:

Association between percent of sites with interproximal probing depth ≥3mm and cardiometabolic measures, ORIGINS (n=1071, 2011–2020)

	Percent of Sites with Interproximal Probing Depth (%sites-i-PD) ≥3mm
	0–12%	13–26%	27% or more	Linear Trend
	n=280	n=321	n=470
Mean i-PD (min, max), mm	1.86 (1.23, 2.21)	2.09 (1.64, 2.41)	2.57 (1.99, 6.63)
*Prediabetes (PR, 95% CI)	N=21	N=39	N=74
Model 1	Ref.	1.57 (0.94, 2.61)	1.79 (1.07, 2.99)	0.03
Model 2	Ref.	1.43 (0.94, 1.91)	1.32 (0.81, 1.83)	0.41
Model 3	Ref.	1.31 (0.82, 1.79)	1.15 (0.66, 1.64)	0.76
Model 4	Ref.	1.43 (0.94, 1.93)	1.14 (0.63, 1.65)	0.83
**Fasting plasma glucose, mg/dL (SE)
Model 1	84.30 (0.48)	84.05 (0.44)	85.37 (0.39)	0.10
Model 2	84.84 (0.46)	84.31 (0.41)	84.87 (0.36)	0.94
Model 3	85.10 (0.44)	84.43 (0.40)	84.64 (0.36)	0.46
Model 4	85.07 (0.44)	84.51 (0.40)	84.60 (0.35)	0.45
**HbA1c, % (SE)
Model 1	5.20 (0.02)	5.30 (0.02)	5.32 (0.02)	<0.01
Model 2	5.23 (0.02)	5.31 (0.02)	5.30 (0.02)	<0.01
Model 3	5.23 (0.02)	5.32 (0.02)	5.30 (0.01)	0.02
Model 4	5.23 (0.02)	5.32 (0.02)	5.29 (0.01)	0.02
**LN-Insulin, uIU/mL (SE)
Model 1	1.84 (0.03)	1.93 (0.03)	2.04 (0.03)	<0.01
Model 2	1.86 (0.03)	1.94 (0.03)	2.03 (0.03)	<0.01
Model 3	1.89 (0.03)	1.95 (0.03)	1.99 (0.02)	0.02
Model 4	1.89 (0.03)	1.96 (0.03)	1.99 (0.02)	0.02
**LN-HOMA-IR (SE)
Model 1	0.27 (0.04)	0.35 (0.03)	0.48 (0.03)	<0.01
Model 2	0.29 (0.04)	0.36 (0.03)	0.46 (0.03)	<0.01
Model 3	0.33 (0.03)	0.38 (0.03)	0.42 (0.03)	0.04
Model 4	0.33 (0.03)	0.39 (0.03)	0.42 (0.03)	0.04

Data are derived from multiple imputation analysis to account for missing covariates (10 imputations of n=1071 analytical sample via fully conditional specification).

^*Prediabetes models present prevalence ratios for 13–26% of sites and 27% of sites or more versus 0–12% of sites for the exposure %sites-i-PD.

^**Fasting glucose, HbA1c, LN-Insulin, and LN-HOMA-IR models present mean(SE) of the respective outcomes across categories of %sites-i-PD.

Model 1: adjusts for study wave; Model 2: Model 1 + adjustments for age, race/ethnicity, sex, education, income; Model 3: Model 2 + adjustments for smoking status, body mass index, Alternative Healthy Eating Index. Model 4: Model 3 + adjustments for high density cholesterol, mean systolic and diastolic blood pressures. P-value <0.05 indicates statistically significant linear trend.

Abbreviations: PR = prevalence ratio; SE = standard error; CI = confidence interval; HOMA-IR = Homeostasis Model Assessment for Insulin Resistance; ORIGINS = Oral Infections Glucose Intolerance and Insulin Resistance Study; LN = natural log.

Prevalence of prediabetes, glucose, and HbA1c were comparable by fm-PD tertiles, while estimated insulin and HOMA-IR were slightly higher in Tertiles 2 and 3 vs Tertile 1, although the trend was not significant (Supplemental Table 5 and Figure 2). %sites-fm-PD≥3mm was not associated with prediabetes or estimated glucose (Supplemental Table 6). However, there was a significant elevation in LN-insulin across the three increasing categories of %sites-fm-PD (1.90 [0.03] uIU/mL; 1.98 [0.02] uIU/mL; 1.98 [0.03] uIU/mL; p for trend=0.03) and a non-significant increase of LN-HOMA-IR from the first category (0–12% of sites=0.33 [0.03]) to the latter two categories (13–26% & 27% or more=0.41 [0.03] for both; p trend=0.05) after multivariable adjustment.

3.5. Bleeding on Probing and Cardiometabolic Outcomes

Higher BOP was not associated with prevalence of prediabetes, fasting glucose, insulin, or HOMA-IR, although greater BOP was significantly associated with a linear decrease in HbA1c across BOP tertiles (Supplemental Table 7).

3.6. Obesity Effect Measure Modification

Despite comparable estimates by periodontal disease status, participants with obesity had higher prediabetes PRs and cardiometabolic biomarkers than non-obese participants (Supplemental Table 8). Greater mean fm-CAL (Supplemental Table 9) and mean i-CAL (Supplemental Table 10) was associated with higher prediabetes prevalence, glucose, HbA1c, and insulin among individuals without obesity. Overall, there was no evidence of effect measure modification between mean CAL measures and BMI.

4. DISCUSSION

We cross-sectionally examined the association between periodontal disease, prediabetes, and cardiometabolic biomarkers among non-diabetic young to middle-age adults from a racially/ethnically diverse population-based setting. Clinical attachment loss was associated with prediabetes, fasting glucose, and HbA1c whereas probing depth was related to insulin and insulin resistance. Compared to full-mouth periodontal measures, associations were generally stronger for interproximal-based measures, which are more representative indicators of frank periodontitis rather than traumatic recessions at buccal surfaces.

It is well established that diabetes and hyperglycemia are strong predictors of periodontal disease.¹⁵ Moreover, there is a growing amount of evidence suggesting that periodontal disease is associated with cardiovascular and cardiometabolic outcomes,^9,39,40 including the prevalence and incidence of diabetes as shown in various large cohort studies.^8,9 Limited research^19,21 has explored the relationship between periodontal disease and biomarkers of diabetes risk, especially among younger and relatively healthier adults.

Our current findings generally support the hypothesis that periodontal disease may be associated with early cardiometabolic abnormalities and are consistent with a limited number of prior studies. A 2012 study from the National Health and Nutrition Examination Survey (NHANES) showed positive dose-response relationships between mean fm-PD (but not fm-CAL) and insulin and HOMA-IR.⁴¹ Another NHANES study from 2014 found associations between both mean fm-PD and fm-CAL, and prediabetes defined by an oral glucose tolerance test (an indicator of insulin resistance) after multivariable adjustment.¹⁸ A more recent NHANES study showed moderate and severe CDC/AAP-defined periodontitis were associated with 46% (OR=1.46[1.29,1.65]) and 62% (OR=1.62[1.31,2.01]) greater odds of prediabetes compared to those without periodontitis, respectively.¹⁹ A strong proof of concept in support of these observational studies comes from a multicenter randomized controlled trial of patients with established periodontitis. In this study, non-surgical periodontal therapy lowered HbA1c levels from 5.9% to 5.4% among patients with prediabetes compared to patients without prediabetes who observed no change in HbA1c throughout a 15-month follow-up period⁴²—demonstrating potential causal links between the periodontal treatment and glucose control. Our analysis, along with findings from previous literature, may suggest that cumulative measures of periodontal disease such as CAL are likely predictors of impaired fasting glucose, while more active inflammation and disease activity, such as probing depth, may more strongly associate with insulin resistance.

Our findings support prior work in ORIGINS related to the oral microbiome and cardiometabolic risk. In a subset of n=300 participants, elevated levels of Aggregatibacter actinomycetemcomitans, Porphyromonas gingivalis, Treponema denticola, and Tannerella forsythia (assessed via whole genomic DNA probes) were positively associated with prediabetes,²⁸ while compositional shifts in major phyla were related to insulin resistance and inflammation.¹² A longitudinal study extended these findings by reporting greater baseline microbial dysbiosis being associated with higher levels of glucose change in the same sample.¹³

We did not detect interactions between obesity and periodontal measures for the prediabetes outcome, which was potentially attributable to insufficient statistical power. A recent investigation in the Hispanic Community Health Study/Study of Latinos Study reported that in the full sample, moderate/severe periodontal disease (vs. none/mild) was not associated with risk of incident prediabetes after covariable adjustment.²¹ Interestingly, when stratified by obesity status, the association became strongly positive among participants with obesity compared to those who were non-obese, which is contradictory to our current findings in which the trend was for positive associations among the overweight/normal BMI group.²¹ While obesity may be a potential effect measure modifier, sufficiently powered future research studies with better measures of adiposity are necessary to further inform this potential.⁹

It is biologically plausible that periodontitis could contribute to cardiometabolic risk generally, and prediabetes specifically. Chronic inflammation, which is common in both periodontitis and prediabetes is a commonly cited mechanism linking these conditions.⁴³ Indeed, periodontal pathogens express virulence factors that may enhance local and systemic inflammation and several prior studies have reported periodontitis to be associated with increased biomarkers of systemic inflammation.^7,44 Multiple randomized controlled trials have demonstrated that periodontal therapy to effectively reduce systemic inflammation, consistent with an improvement in clinical periodontal measures.^15–17 Another potential mechanism that more directly links periodontal status to prediabetes risk includes pancreatic infection by oral pathogens. Human tissue and mouse models have demonstrated that circulating bacteria of periodontal origin may exhibit pancreatic tropism,^45,46 possibly triggering localized pancreatic inflammation and beta-cell damage.

There are multiple strengths to our study. First, our study population consisted of a racially and ethnically diverse cohort within New York City. Second, we cross-sectionally evaluated biomarkers of diabetes risk among a relatively young adult cohort free of diabetes or cardiovascular disease which might confound the relationship between periodontal disease and cardiometabolic health. Third, trained and validated examiners conducted comprehensive full-mouth periodontal assessments which allowed for the comparison of associations based on full-mouth vs. interproximal periodontal measures. Fourth, we included healthy diet composition as a potential confounder in addition to conventional biological and sociodemographic covariables, a unique aspect of our modeling strategy. Lastly, we employed multiple imputations to account for missing covariables and retain sufficient statistical power.

Our study does have notable limitations. First, our cross-sectional analyses lack the ability to inform temporality, raising the potential for reverse causality, as adverse cardiometabolic profile may lead to periodontal disease.¹⁵ Second, although our multivariable modeling approach is robust, the possibility of residual confounding due to known and unknown confounders exists. An example would be accelerated biological aging, as those with advanced biological age would be at a greater risk of periodontal disease and prediabetes/diabetes.^47,48 Third, the risk of misclassification of periodontitis is possible given the younger age of our study cohort and lower extent and severity of periodontal disease. A more nationally representative cohort of adults without Type 2 diabetes would help capture the true distributions of periodontitis and prediabetes. Thus, our findings cannot be generalizable to the broader US or global population. Lastly, the use of different insulin assays in each study wave may have led to measurement error, potentially biasing insulin resistance estimates towards the null.

5. CONCLUSIONS

Our findings suggest that among younger adults without diabetes, CAL is positively associated with prevalent prediabetes, HbA1c, and glucose, while PD is associated with elevated insulin and insulin resistance levels. Future investigations should more directly assess the underlying role of the oral microbiome on glucose intolerance and insulin resistance in longitudinal cohorts. Interventional studies of periodontal therapies among people with evident oral dysbiosis and gingival inflammation prior to advanced periodontitis should also be considered. If future studies deem these associations to be causal, the population health benefit could be substantial.

Clinical Relevance.

Scientific Rationale for Study:

Prediabetes is a highly prevalent condition that leads to the onset of diabetes. Periodontitis has emerged as a potential risk factor for diabetes, but its relationship with prediabetes is not well studied in young adults.

Principal Findings:

We found clinical attachment loss was associated with prediabetes and elevated blood glucose levels, while probing depth was associated with higher insulin resistance.

Practical Implications:

Oral health interventions may serve as viable approaches in preventing prediabetes and cardiometabolic disorders.