Relationship Between Periodontal Disease and C-Reactive Protein Among Adults in the Atherosclerosis Risk in Communities Study FREE

C-REACTIVE protein (CRP) is a liver-produced, acute-phase reactant that serves as a systemic marker of inflammation. Levels of CRP can be used to monitor patients with overwhelming infections, and elevated CRP levels have been demonstrated in persons with ischemia and myocardial infarction.¹ When monitoring a patient's acute-phase response, a serum CRP concentration exceeding 10 mg/L is generally regarded as the threshold indicative of significant inflammatory disease.²

Concentrations of CRP in the general population are usually low (<10 mg/L in 98% of people).³ Nonetheless, when measured using a high-sensitivity enzyme-linked immunosorbent assay (ELISA) method, a doubling of normal CRP concentration (ie, increasing within the range 3.0-10.0 mg/L) may be sufficient to indicate dysregulation of proinflammatory mechanisms among otherwise healthy individuals.⁴ The relevance of CRP concentrations in the upper range of normal (referred to as "high-normal" by Tracy et al⁵) has been highlighted in several population cohort studies^1,5- 9 in which elevated CRP concentration has been a predictor of subsequent coronary heart disease (CHD). These findings are consistent with clinical studies in which CRP levels were prognostic markers for myocardial infarction among individuals with unstable angina¹⁰ and for recurrent infarction.¹¹

It remains unclear whether the predictive capacity of high-normal CRP concentrations is indicative of an underlying etiologic mechanism, in which inflammation contributes to atherogenesis or instability of atherogenic plaques, or whether elevated CRP concentration is merely a marker of atherosclerosis or other vascular damage. Furthermore, cardiovascular disease and acute-phase inflammatory response share many common risk factors, including older age, low socioeconomic status, cigarette smoking, high body mass index (BMI; calculated as weight in kilograms divided by the square of height in meters), and diabetes mellitus.^{1,3,9,12- 16} Some studies⁹ have found that adjustment for such factors statistically accounts for the association between high-normal CRP values and cardiovascular disease. Others studies^7,14,17 have shown a persistent independent effect of CRP on cardiovascular disease after controlling for those factors, although in some cases only among population subgroups. Using CRP data obtained at baseline from the Atherosclerosis Risk in Communities (ARIC) study, Folsom et al¹⁸ found that CRP concentration was a moderately strong marker of risk of subsequent CHD after controlling for demographics and traditional CHD risk factors. As expected, other markers of inflammation attenuated the association of CRP concentration with CHD incidence, indicating that inflammation in general, rather than CRP level specifically, was associated with increased CHD risk.

Discrepancies between studies may reflect difficulty in identifying relevant factors, such as latent infections, that contribute to acute-phase response in ostensibly healthy individuals. Periodontal disease has been shown to elicit a systemic inflammatory response in animal experimental studies^19- 20 and epidemiologic studies.^13,21- 22 Periodontal disease is a chronic infection capable of causing an ulcerated lesion at the dental-epithelial junction that is estimated to range from 8 to 20 cm².²³ Key periodontal pathogens are highly invasive in connective tissue and are capable of evading cellular immune defense mechanisms.²⁴ Clinically, this destructive process manifests as deepening of the epithelial attachment around the teeth, loss of periodontal attachment, and, ultimately, loosening of the tooth.

The purpose of this study is to investigate the associations among periodontal disease, CRP levels, and established risk factors for elevated CRP levels in middle-aged men and women selected as a probability sample of the residents of 4 communities in the United States.

We undertook this cross-sectional study using data from ARIC, an ongoing cohort study of a sample of community-dwelling adults in 4 US localities: Forsyth County, NC (Winston-Salem); Jackson, Miss; Washington County, Md (Hagerstown); and suburbs of Minneapolis, Minn. Baseline examinations, tests, and interviews were conducted during 1987-1989, and participants subsequently were contacted on an annual basis. Survivors were invited to take part in 3 follow-up data collections at intervals of approximately 3 years. This study uses data from the third follow-up visit, conducted between January 1, 1996, and December 31, 1998, when eligible participants had oral examinations in addition to the battery of examinations, tests, and interviews conducted for the parent ARIC study. Participants were not eligible for the dental component if they had no natural teeth, had medical contraindications to periodontal probing (eg, susceptibility to subacute bacterial endocarditis), or did not provide signed consent for the oral examinations.

Clinically measurable periodontal disease was used as the main exposure variable indicative of an oral infectious burden. Oral examinations were a modification of the protocol used for the Third National Health and Nutrition Examination Survey (NHANES III)²⁵ and included measurement, in millimeters, of periodontal pocket depth made at up to 6 sites per tooth around all remaining teeth (range, 6-192 sites per participant). Examinations were conducted by 4 dental hygienists, all of whom underwent the same 2-day training and calibration process, during which they were matched with a "gold standard examiner" and with each other. The average percentage agreement for all examiners with the gold standard examiner for pocket depth was 92.5%, and the intraclass correlation statistic was 0.91. During on-site examination of study participants, each examiner had a written manual that described in detail decision points involved with each index used. During the 3-year course of the examinations, a quality assurance program was in place to maintain standardization of examiners and to prevent drift. Each year, the examiners were calibrated with the gold standard examiner, and reliability was recalculated for all examiners. The intraclass correlation statistic was never less than 0.80.

Serum CRP levels were used as the outcome variable for this study, representing systemic, acute-phase response. Fasting blood samples were taken from all participants, and serum samples were frozen for subsequent analysis as previously described.^26- 27 Serum concentrations of CRP were quantified using a commercially available ELISA (VIRGO CRP 150 kit; Hemagen Diagnostics Inc, Waltham, Mass). The assay was specific for CRP, with no interference from any other normal or abnormal plasma components. Each ELISA plate was used to analyze serum samples from up to 72 individuals and was standardized using replicate standard concentrations of 0.5, 1.0, 2.0, 5.0, 10.0, 15.0, 20.0, 35.0, and 50.0 mg/L according to the World Health Organization First International Reference Standard (85/506) for CRP. The range of detection for CRP was 0.5 to 50.0 mg/L. All CRP values less than 0.5 mg/L were imputed to 0.25 mg/L, whereas values greater than the upper threshold of detection were truncated to 50.0 mg/L.

For assessment of intraplate reliability, replicate standard concentrations of 5, 10, 15, 20, and 35 mg/L within each plate were used to compute intraplate coefficients of variation (CVs). Observed CVs ranged from 0% to 37.3%, with 25.5% of plates having intraplate CVs of 10.0% or more. Interplate reliability among the 99 plates used was assessed for the same standard concentrations, and interplate CVs ranged from 0% to 10.6%. To control for possible effects of relatively high intraplate variability in statistical analyses, we used an indicator variable to flag participants whose CRP measurements derived from a plate that had an intraplate CV greater than 10.0%.

We selected covariates that previously had been found to mediate or modify the association between periodontal disease and acute-phase response (Table 1). Although previous studies had not evaluated associations with dental visits, we believed that it was important to attempt to control for the possibility that relatively frequent dental visits may reduce the infectious burden around teeth with periodontal pockets. Because linear regression models use zero as a reference value for covariates even if zero is biologically implausible (as is the case for BMI), it was preferable to express BMI using a continuous measure that had a meaningful central value and scale. We therefore created a centered, scaled measure of BMI (25 was subtracted from each individual's observed BMI value, then divided by 5). Hence, the parameter estimate for the centered, scaled variable indicates the effect of a difference in BMI of 5 compared with a reference category of 25. In addition to the covariates in Table 1, 2 study design variables were considered in multivariable modeling: study center and the indicator variable flagging CRP measurements from plates with an intraplate CV greater than 10.0%.

Periodontal disease was quantified using extent scores that represent the percentage of probed sites that have pocket depth above a specified threshold. Pocket depth was selected as the exposure variable describing periodontal disease because it provides a better proxy indicator for existing periodontal infection than other markers of periodontal disease experience such as attachment loss or recession, which reflect existing and historical disease activity. For this analysis, we computed the extent of periodontal pocketing³⁰ at a threshold of 4 mm or more and categorized individuals into 3 levels: 0% to less than 10%, 10% to less than 30%, and 30% or more of sites. The extent score was used in preference to other variables that summarize periodontal pocketing because it has the desirable property of capturing the intensity of an individual's response to challenge from periodontal pathogens while allowing for the fact that individuals differ substantially in the subgingival surface area of periodontium, due primarily to varying levels of tooth loss in this older adult population.

The initial analysis compared mean CRP values and prevalence of elevated CRP levels (≥10 mg/L) among periodontal disease subgroups. Stratified and multivariable analyses were then undertaken using the covariates listed in Table 1. For statistical evaluation of the main hypotheses, we constructed a least squares multivariable regression model, with CRP concentration as the dependent variable. To assess associations using only the clinical threshold for elevated CRP values, we also constructed a multivariable logistic regression model to assess factors associated with odds of elevated CRP concentration (≥10 mg/L).

The approach for constructing both multivariable models was identical. First, we included all covariates (Table 1) together with the previously described study design variables. Then we excluded nonsignificant variables (P>.05), provided that their removal did not produce a change of more than 20% in the parameter estimate for periodontal disease. This latter criterion was used to ensure that excluded variables did not substantially confound the main association of interest between periodontal disease and CRP concentration. Finally, the reduced model was evaluated for interactions between periodontal disease and each of the remaining explanatory variables. Hierarchical principles were observed when evaluating interaction terms, that is, the component first-order terms were retained and the additional (type III) effect of the interaction term was assessed.

When a statistically significant interaction involved a continuous variable, a nested interaction term was used to aid in interpretation of results. For example, after identifying a statistically significant interaction between dichotomized periodontal status and the continuous measure of BMI, the first-order term for BMI and the periodontal × BMI interaction was removed and replaced with 2 terms representing BMI nested within periodontal status. The nested effects permitted us to quantify the impact of BMI on CRP concentration within each stratum of periodontal disease. Statistical significance was set at P<.05.

The study sample comprised 5552 people aged 52 to 75 years (Table 2). Three quarters of the participants had little or no periodontal pocketing of 4 mm or more, although 4.6% of people had extensive periodontal disease, defined as greater than 30% of periodontal sites with pockets of 4 mm or more. The prevalence of elevated serum CRP levels (≥10 mg/L) was 16.6%. Mean (SE) CRP concentration was 5.83 (0.11) mg/L, and the distribution was skewed with a median of 2.70 mg/L and an undetectable level (<0.5 mg/L) observed in 15.6% of participants (data not tabulated).

Mean (SE) CRP concentration was 7.6 (0.65) mg/L in participants who had 30% or more of periodontal sites with pockets of 4 mm or more, which was approximately one-third greater than the concentrations for people with 0% to less than 10% of periodontal sites affected (5.8 [0.13] mg/L; P = .001) or 10% to less than 30% of periodontal sites affected (5.4 [0.25] mg/L; P = .001). Owing to the virtual equivalence in mean CRP values between the latter 2 groups (P = .17), they were combined for subsequent analyses, creating a reference group that represented 95% of the participants. The percentage of participants with CRP levels of 10 mg/L or greater was also greater (P = .02) by approximately one third among those with the most extensive periodontal pocketing (22.0% prevalence) compared with those with less extensive periodontal disease (16.3% prevalence) (Table 3). In addition, there were statistically significant variations in mean serum CRP levels among categories of all but 2 of the covariates listed in Table 1: age (P = .48) and aspirin use (P = .11).

The approximate one-third increase in mean CRP between periodontal disease groups persisted within most strata of covariates (Table 3). Exceptions occurred for the following subgroups, where extensive periodontal pocketing was associated with lowered CRP levels: people with a BMI of 30 or greater, diabetic patients, and those with more than a 1-year history of arthritis. In addition, among participants whose last dental visit was 6 to 12 months before the study, mean CRP levels were equivalent for the 2 periodontal disease groups. These stratified findings were similar when the percentage of people with CRP levels of 10 mg/L or greater was contrasted between periodontal disease groups (Table 3). However, among current smokers, the prevalence of elevated CRP levels did not differ between periodontal disease groups.

After the first step in constructing the least squares multivariable model, extensive periodontal pocketing was significantly associated with mean CRP levels together with BMI (used as a continuous variable), age, sex, diabetes mellitus, cigarette use, and use of nonsteroidal anti-inflammatory drugs. The reduced model was evaluated for interactions between periodontal disease and each other variable, yielding one statistically significant interaction between BMI and periodontal disease (P<.01). Results for the final model are presented in Table 4, where the interaction is depicted as a nested effect of BMI (expressed as a scaled, centered continuous variable) within periodontal disease. At the normative BMI value of 25, extensive periodontal pocketing was associated with a predicted increase of 2.73 mg/L, which was greater than the elevation attributable to cigarette smoking. Figure 1 depicts the interaction between BMI and periodontal disease. Based on the regression model, when BMI equaled 20 there was a predicted 2-fold difference in mean CRP levels between high and low periodontal pocket groups (7.5 vs 3.6 mg/L), but the difference decreased with increasing BMI and was negligible when BMI equaled 35 (9.0 vs 9.3 mg/L).

The indicator variable, flagging participants whose CRP assay came from an ELISA plate with a CV of 10% or more, was not statistically significant in this model (P = .13), and it did not change parameter estimates for periodontal disease and the nested BMI terms by more than 0.5%. Finally, we investigated waist-to-hip ratio as an alternative marker of adiposity. It explained much less of the variance in CRP concentration than BMI, and its addition to the model in Table 4 did not substantially change our interpretation of the effects of periodontal disease.

One influential observation was noted in the final multivariable model. The participant was a 71-year-old woman who was a current smoker and whose BMI was 14. She had 4 periodontal pockets of 4 mm or more among the 6 sites measured on her only tooth (extent = 66.7%), and her serum CRP level was 32.5 mg/L. When this woman was removed from the multivariable model, first-order terms and the interaction remained statistically significant, although the parameter estimate for BMI in the extensive periodontal disease group increased from 0.60 to 0.81, achieving a statistically significant slope (P = .05).

Our overall interpretations from the model in Table 4 were confirmed by using a multivariable logistic regression model in which the odds of elevated serum CRP levels (≥10 mg/L) were the dependent variable (results not tabulated). The odds ratio for elevated CRP levels associated with extensive periodontal disease was dependent on BMI: at a BMI of 20, the odds ratio was 2.5 (95% confidence interval, 1.8-3.5), whereas at a BMI of 35, the odds ratio was 1.1 (95% confidence interval, 0.9-1.4). Removal of the previously noted participant who had influential values reduced the odds ratio at a BMI of 20 to 2.3 (95% confidence interval, 1.6-3.2).

The results of this study demonstrate elevated concentrations of the acute-phase reactant serum CRP in individuals with extensive periodontal disease. This association was modified by the individual's weight: the effect of periodontal disease, after adjustment for established risk factors for elevated serum CRP levels, was most pronounced in lean individuals (eg, BMI of 20), in whom extensive periodontal disease was associated with a predicted 2-fold increase in mean CRP levels and more than twice the odds of elevated CRP levels (≥10 mg/L). In contrast, periodontal disease was not associated with increased levels of CRP in obese individuals (eg, BMI of 35). For an individual with normative BMI of 25, extensive periodontal pocketing was associated with a predicted increase of 2.7 mg/L, which is within the conventional reference range but which is probably indicative of dysregulation of proinflammatory mechanisms.⁴ These results confirm those of previous cross-sectional studies^13,21- 22 of periodontal-CRP associations and further suggest that periodontal disease and adiposity compete for common proinflammatory pathways that elicit an acute-phase response.

The hepatic stress induced by either periodontal disease or obesity can result in a 2-fold increase in CRP concentration, an elevation that is substantially below the 0.5- to 3.0-log increases often associated with acute trauma or infection. This suggests that either of these chronic, low-grade stressors can act to trigger a hepatic response that fulfills a low-level threshold response, one that increases CRP values from approximately 3.6 to between 7.5 and 9.0 mg/L. This increase remains below what would be considered a clinically significant elevation that is indicative of an acute insult (ie, >10 mg/L). These data are consistent with the concept that there is 2-stage hepatic activation of the CRP response, one that induces a mild CRP elevation that can be elicited by low-level stressors such as periodontal disease (ie, a 1.8-fold increase in individuals with BMI <30) or obesity and a second dynamic response that results in a 3- to 1000-fold increase in CRP concentration above the 10 mg/L cutoff point. The fact that periodontal disease progression is episodic, with periods of acute exacerbation, is consistent with the 2-fold increase in the prevalence of clinically elevated CRP levels (>10 mg/L) seen in these results.

Thus, it seems that periodontal disease may serve as either a chronic or an acute stimulus of hepatic CRP. In the presence of obesity, the chronic added burden of periodontal disease would not seem to be sufficient to enhance CRP levels above the background elevation elicited by obesity. However, the presence of periodontal disease in nonobese individuals seems to result in mild CRP elevations consistent with those modest increases associated with greater cardiovascular event risk among systemically healthy individuals. A positive association between BMI and CRP has been observed in other epidemiologic studies,^5,16 and the primary mechanism is thought to be production by adipocytes of tumor necrosis factor α (TNF-α), which, in turn, stimulates CRP synthesis in the liver.³¹ In one study,³² weight loss among diabetic patients was accompanied by a reduction in serum TNF-α levels. Similarly, in response to periodontal infection, activated monocytes in the periodontal tissues produce proinflammatory cytokines, including interleukins, that initiate destruction of connective tissues.¹⁹ It is possible, therefore, that adiposity and periodontal disease serve as low-grade, proinflammatory stressors mediated by TNF-α. Verification of this interpretation calls for studies that measure TNF-α levels, which were not available for the present analysis. In principle, this question should also be addressed using a direct measure of adiposity, rather than BMI, although, as noted in the "Results" section, waist-to-hip ratio accounted for less variation in CRP levels than BMI. The latter effect was consistent with results reported by Hak et al.³¹

Several caveats need to be considered when interpreting the main findings from this study. At first appearance, the overall CRP levels in this ARIC sample are strikingly high. For example, 16.6% of ARIC participants had elevated CRP levels (≥10 mg/L) compared with only 7.3% of those in the NHANES III of the US adult population.¹³ In part, this may reflect the higher sensitivity of the ELISA method used to quantify CRP in this study compared with the latex-enhanced nephelometry method used in the NHANES III. However, it probably also represents important features of the ARIC sample, most notably older age and greater BMI: ARIC participants were aged 52 to 75 years, and 74% of them had a BMI of 25 or greater, whereas among NHANES III³³ adults 20 years and older, 55% had a BMI of 25 or greater. Nonetheless, there is considerable variability in population levels of serum CRP, even in studies using comparable ELISA methods, ranging from 0.02 to 3.25 mg/L among US blood donors⁴ to 0.05 to 27.06 mg/L among Welsh men.⁹ Consequently, we focused on relative differences among subgroups within this study rather than on absolute values of CRP.

We defined extensive periodontal disease as pocketing of 4 mm or more in greater than 30% of the periodontal sites measured, which is higher than the threshold of greater than 10% of sites used in the NHANES III.¹³ The discrepancy is due partly to different clinical protocols: the NHANES III periodontal assessment was conducted at only 2 sites per tooth in up to 14 teeth per individual, whereas the ARIC protocol assessed pocket depth at 6 sites on all teeth. Partial recording systems are known to underestimate the true amount of periodontal disease,³⁴ and, hence, our ARIC examination protocol yielded a wider range of extent scores, permitting us to identify a severely diseased group using a higher threshold. Nonetheless, in a separate analysis, ARIC participants had a higher prevalence of periodontal disease than relevant age-specific estimates from NHANES III, even when ARIC extent scores were recalculated using the same, partial recording system used for NHANES III (data not tabulated). For these reasons, we again advise readers to focus on relative differences among periodontal disease groups within this study rather than on absolute prevalence. Indeed, we emphasize the similarities in results between this study, where the association of CRP concentration with periodontal disease was most apparent in people who were not overweight, and the NHANES III, where the "effect" of periodontal disease was most apparent in individuals who did not have any established risk factors for elevated CRP levels.

The relatively high CVs for replicate standards in one quarter of the ELISA plates raise concerns about reliable quantification of CRP in these data, and hence possible misclassification of the outcome variable. However, we have no reason to believe that the misclassification would be associated with periodontal status; hence, this problem should not bias the main association reported herein. In addition, inclusion in the analyses of the variable flagging plates with high CVs did not alter the interpretation of the results in Table 4.

Our findings are consistent with those of previous population studies and suggest that extensive periodontal disease, which affected 1 in 20 ARIC participants, is an important marker for elevation of normal levels of CRP. An implication of the observed interactions among periodontal disease, BMI, and CRP concentration is that medical and dental diagnoses may be necessary when evaluating sources of acute-phase response in some patients. For example, if one of the objectives of a weight reduction program is to reduce a proinflammatory state indexed by serum CRP level, it would be valuable to evaluate the patient's periodontal status. In obese patients who have extensive periodontal disease, weight loss may be insufficient to reduce acute-phase reactant levels. Currently, there is only limited evidence that periodontal therapy could reduce individuals' systemic inflammatory burden. Ebersole et al²¹ reported reductions in a related acute-phase reactant, haptoglobin, after scaling and root planing in periodontal patients. Recently, we have begun to study the effects of periodontal therapy on serum CRP levels in patients with periodontal disease, and the initial results are encouraging (J. Elter, DMD, PhD, unpublished observations, 2002). However, there is insufficient evidence to recommend periodontal therapy to control high-normal values of CRP, and it has not been demonstrated that reducing CRP levels improves health. Instead, the findings from this study highlight the benefits for systemic health of maintaining periodontal health.

Corresponding author and reprints: James D. Beck, PhD, Department of Dental Ecology, University of North Carolina, CB7450, Chapel Hill, NC 27599-7450 (e-mail: jim_beck@dentistry.unc.edu).

Accepted for publication August 15, 2002.

This study was supported by grants DE00427 and DE13079 from the National Institute of Dental and Craniofacial Research and by ARIC contracts supported by the National Heart, Lung, and Blood Institute (Bethesda, Md), and by grant RR-00046 from the General Clinical Research Centers Program of the Division of Research Resources of the National Institutes of Health (Bethesda).

This study was presented in part at the annual meeting of the International Association of Dental Research, San Diego, Calif, March 8, 2002.

We thank the staff and participants in the ARIC study for their important contributions.