Gingivitis and periodontitis are among the most common inflammatory conditions in human adults(Reference Eke, Dye and Wei1). The inflammatory process in gingivitis is restricted to the superficial periodontal tissue and does not lead to periodontal attachment loss, making gingivitis lesions reversible in nature once the cause(s) is/are removed. Periodontitis, on the other hand, is characterised by more profound inflammation that leads to breakdown of the tooth-supporting apparatus and may lead to tooth loss eventually. The pathogenesis of periodontal disease is not yet fully understood, but there is substantial evidence that most of the periodontal tissue destruction is caused by the host immune response to the bacterial challenge of periodontal pathogens(2). Pro-inflammatory mediators, such as TNF-α, IL-1, IL-6, IL-8 and PGE2, are key players in this process(2). To date, only a few of the established periodontal disease predictors can be modified through improvement of lifestyle factors. Prevention of periodontal disease at the population level requires better understanding of modifiable risk factors for periodontal disease(Reference Joshipura and Andriankaja3).
The relationship between periodontitis and systemic inflammation is complex, and current evidence proposes that the relationship is bidirectional. In fact, systemic inflammation seems to pathophysiologically elucidate several of the reported associations between periodontal disease and multiple cardiometabolic conditions(Reference Joshipura and Andriankaja3). Much of the published work in the last three decades has focused on the potential effect of periodontitis on elevated systemic inflammatory biomarkers(Reference Loos and Van Dyke4–Reference Gocke, Holtfreter and Meisel9). The reverse was only reported in humans recently, that is, baseline systemic inflammation measures were positively correlated with periodontal disease progression(Reference Pink, Kocher and Meisel10). The impact of diet on modifying the risk of periodontitis is strongly biologically plausible through modulation of systemic inflammation(Reference Chapple11). However, the literature relating diet and periodontal disease in humans is relatively scarce(Reference Joshipura and Andriankaja3).
An empirical dietary inflammatory pattern (EDIP) was recently developed by Tabung et al. using reduced rank regression (RRR)(Reference Tabung, Smith-Warner and Chavarro12). Dietary pattern analyses are practical platforms to study the overall impact of diet on disease outcomes beyond the specific effects of certain foods or nutrients. Dietary patterns can be generally classified into either a priori or a posteriori indices. While a priori index is hypothesis-oriented and is based on the current scientific evidence regarding the relationship between diet and diseases (e.g. Alternate Healthy Eating Index), a posteriori method on the other hand is data-driven and is based on statistical exploratory methods (e.g. principal component analysis-driven dietary pattern). RRR is an innovative approach in nutrition epidemiology in that it is a posteriori in nature, but it incorporates prior knowledge about diseases and their pathways(Reference Hoffmann, Schulze and Schienkiewitz13). When applied, information about food or nutrients intake is used by RRR to maximally explain the variability in response variables, which can be disease mediators (e.g. inflammatory biomarkers). An advantage of RRR over other methods of dietary patterns is that it is based on biological principles and mechanisms of disease development, which could bolster evidence of causality between diet and an outcome of interest(Reference Willett14).
The aim of the current study was to test the hypothesis that the inflammatory potential of diet could modify the risk of periodontitis, by prospectively evaluating the association between EDIP and incidence of periodontal disease in the Health Professionals Follow-up Study (HPFS).
Methods
Study population
The HPFS is an ongoing cohort study that enrolled 51 529 male health professionals (dentists, pharmacists, optometrists, osteopathic physicians, podiatrists and veterinarians) who answered and returned the baseline mailed questionnaire in 1986 when they were 40–75 years old. Study participants provided thorough medical and dental history in addition to lifestyle behaviour and body measurements (e.g. height and weight) through biennial questionnaires. Data about participants’ diet were collected through semi-quantitative FFQ every 4 years, starting at baseline. The adequacy, reproducibility and validity of the FFQ in assessing diet have been previously reported(Reference Willett14,Reference Rimm, Giovannucci and Stampfer15) .
We excluded participants who only responded to the baseline questionnaire (n 3309) and those who had missing periodontal data (n 1117). Participants who reported periodontitis at baseline (n 8333) and those who were edentulous (n 485) were excluded because they were not at risk for incident periodontal disease. In addition, we excluded participants who reported myocardial infarction (n 1486), coronary artery surgery (n 671), diabetes (n 809) or cancer (n 1317) at baseline because those events could strongly modify dietary habits. We also excluded participants with missing data on BMI (n 902), physical activity (n 128) or age (n 36) at baseline. Participants with energetic intake outside the plausible range (3347–17 573 kJ/d (800–4200 kcal/d)) and those who left 70 or more out of the 131 FFQ items blank were also excluded (n 1033). Our analysis consists of 34 940 men at baseline. The study was approved by the institutional review boards of the Brigham and Women’s Hospital and the Harvard T.H. Chan School of Public Health.
Outcome assessment
The endpoint in our study was self-reported incidence of periodontal disease, defined as answering ‘yes’ to the question ‘have you been professionally diagnosed with periodontal disease with bone loss?’, which was asked biennially in the mailed questionnaire. The self-reported data for assessing periodontal disease were shown to be valid in a subsample of the HPFS when compared with bitewing radiographs(Reference Joshipura, Pitiphat and Douglass16,Reference Joshipura, Douglass and Garcia17) . There were no secondary endpoints in our study.
Main exposures assessment
The main exposure in the current study is RRR-derived EDIP, which was developed in the Nurses’ Health Study and was validated in the Nurses’ Health Study II and the HPFS(Reference Tabung, Smith-Warner and Chavarro12). In summary, the index was derived in a subsample of Nurses’ Health Study (n 5239) for whom dietary data and plasma concentrations of IL-6, C-reactive protein, TNF-α receptor 2 were available. Blood was collected in 1989 and 1990. Diet was measured by averaging the two FFQ that were close to the blood collection, that is, 1986 and 1990 cycles. The 131 FFQ items were grouped into thirty-nine pre-defined food groups based on nutrient composition and on culinary use. Those groups were used in the RRR to explain as much variability in IL-6, C-reactive protein and TNF-α receptor 2 as possible. The first factor explained most of the variation, and it was retained. Then, stepwise linear regression models were fitted to identify food groups most predictive of that factor. The analysis resulted in eighteen food groups and beverages. EDIP scores are a weighted average of those food groups. Negative scores indicated anti-inflammatory diets, while positive scores indicated proinflammatory diets. Fish (other than dark-meat fish), tomatoes, processed meats, high-energy beverages, other vegetables (i.e. vegetables other than leafy green vegetables and dark yellow vegetables), red meats, low-energy beverages, refined grains and organ meats were associated with higher concentrations of the inflammatory biomarkers, whereas pizza, wine, leafy green vegetables, dark yellow vegetables (comprising carrots, yellow squash and yams), beer, coffee, fruit juice, snacks and tea were inversely associated with the biomarkers. The construct validity of EDIP was evaluated in independent subsamples from Nurses’ Health Study II and HPFS(Reference Tabung, Smith-Warner and Chavarro12), and in other cohorts(Reference Aroke, Folefac and Shi18,Reference Tabung, Giovannucci and Giulianini19) . For our analysis, we calculated EDIP scores for each FFQ cycle (i.e. every 4 years) until the start of each 2-year follow-up interval. The cumulative average of the EDIP scores was calculated, to better represent the long-term dietary intake and to minimise within-person variability and measurement error. EDIP scores were divided into quintiles.
Covariates assessment
We controlled for confounding by adjusting the analysis for important risk factors of periodontal disease, that may also modify the exposure, that is, age, smoking, BMI, physical activity and alcohol consumption(Reference Joshipura and Andriankaja3,Reference Pitiphat, Merchant and Rimm20–Reference Jimenez, Hu and Marino25) . We controlled for smoking by using the Comprehensive Smoking Index, which is an algorithm that takes into account the updated information on each questionnaire cycle regarding: duration of smoking in years, smoking intensity (calculated as number of cigarettes smoked per d), time since smoking cessation in years and a specific biological half-life of smoking effect on the disease, estimated to be 1·5 years for periodontal disease(Reference Dietrich and Hoffmann26,Reference Leffondre, Abrahamowicz and Siemiatycki27) . BMI was updated biennially, and values were categorised as follows (<18·5, 18·5–24·9, 25–29·9 and ≥30 kg/m2). We used updated BMI, as a positive association between updated BMI and periodontal disease was reported previously in this cohort(Reference Jimenez, Hu and Marino25). Self-reported physical activity data were quantified using the metabolic equivalent of task to calculate metabolic equivalent of task hours on each cycle. Metabolic equivalent of task-h for each participant was added, and data were categorised into quintiles. We controlled for physical activity using the updated measures for each follow-up. We used the cumulative average of alcohol intake estimated from the FFQ and classified the data into: 0, 0·1–4·9, 5–14·9, 15–29 and ≥30 g/d. Cumulative average energetic intake was estimated from the FFQ, and EDIP scores were adjusted for total energy intake using the residual method. Information about self-reported diagnosis of diabetes during the follow-up was updated at each questionnaire cycle. Self-reported diabetes in this cohort showed good validity(Reference Hu, Leitzmann and Stampfer28).
Statistical analysis
Baseline descriptive statistics for the study cohort by quintiles of EDIP were calculated, using means for continuous data and percentages for categorical data. The hazard ratio of periodontal disease was calculated by comparing each of the higher quintiles of EDIP to the lowest, by fitting Cox proportional hazard models with age in months as the underlying timescale. Person-time was calculated from the return of the baseline questionnaire until incidence of periodontal disease, mortality, last available response or end of follow-up (which was 31 January 2010), whichever came first. The models were adjusted for the confounding variables. We conducted the test for linear trend by assigning each individual the median value of their EDIP quintile. All missing exposure or covariate data during follow-up were handled by carrying forward values from the last cycle.
We compared the results with and without including BMI in the models, to evaluate any potential mediation by adiposity. We also stratified the models by updated BMI categories (18·5–29·9 and ≥30 kg/m2). Underweight men (BMI < 18·5 kg/m2) were excluded from the stratified analysis as they only contributed <0·5 % of the total person-time. To investigate the association within other covariates besides BMI, we also stratified the analysis by occupation (dentist v. non-dentist), age (≥65 v. <65 years), physical activity (above the median v. below), smoking at baseline (never, former, current), alcohol consumption (0, 0·1–4·9 and ≥5 g/d) and diabetes over the follow-up. The stratified analysis models were fully adjusted except for the stratification variable. To test for the statistical significance of interactions, we created indicator variables for the following: being a dentist, updated obesity status (BMI ≥ 30 kg/m2), updated binary physical activity level, updated binary age and updated diabetes. We fitted adjusted models that included interaction terms between the dietary patterns (using the median values of quintiles) and the indicator variables for stratifying factors. For smoking, we used continuous Comprehensive Smoking Index value, instead of an indicator variable. We did the same for alcohol consumption, where we used the continuous intake of alcohol (g/d), instead of an indicator variable. The P–value for each interaction was calculated using a Wald test (one df test). All the analysis was performed using SAS for UNIX statistical software (version 9.4; SAS Institute).
Results
The number of reported new cases of periodontitis was 3738 over the 24 years of the study follow-up (747 517 person-years). Table 1 shows the distribution of age-adjusted baseline characteristics by quintiles of EDIP. Participants with higher scores of EDIP tended to be less physically active, were more likely to be never smokers and consumed less alcohol than those with lower scores of EDIP. Number of teeth was similar across EDIP quintiles. The mean intake of proinflammatory food groups increased with higher EDIP, while the mean intake of anti-inflammatory food groups decreased.
Table 1. Age-standardised characteristics of the Health Professionals Follow-up Study (HPFS) study population in1986 (baseline) by quintile of the empirical dietary inflammatory pattern (EDIP)*
(Mean values and standard deviations; numbers and percentages)
MET, metabolic equivalent of task.
* Values are standardised to the age distribution of the study population.
EDIP scores showed no overall association with the risk of periodontal disease in our analysis; the hazard ratio (HR) in the highest quintile of EDIP was 1·01 compared with the lowest quintile (95 % CI 0·90, 1·12, P-value for trend = 0·80) (Table 2). Adjusting for BMI in the models did not significantly change the association (HR 0·99, 95 % CI 0·89, 1·10, P-value for trend = 0·97). The results were similar among subgroups defined by age, physical activity level, diabetes and profession (Table 3). We detected a marginally significant effect modification by obesity (defined as BMI ≥ 30 kg/m2 v. non-obese) (P-value for interaction = 0·06) (Table 3). There was a modest elevated risk of periodontal disease for obese individuals comparing the highest quintile of EDIP with the lowest quintile (HR 1·27, 95 % CI 0·94, 1·73, P-value for trend = 0·07). We performed a secondary analysis evaluating the association among non-smokers by excluding current smokers at baseline. Participants who reported being ‘current smokers’ at baseline on average contributed more than 50 % of their person-time being current smokers during the follow-up, while those who reported being ‘former’ or ‘never’ smokers only contributed <2 % of the person-time being current smokers. Also, there is evidence that the periodontal status and response to treatment for former smoker are closer to never smokers than to current smokers and seem to get similar to never smokers several years after quitting(Reference Warnakulasuriya, Dietrich and Bornstein29). Smoking is a very strong risk factor of periodontitis in this cohort (online Supplementary Table S1). After exclusion of current smokers at baseline, the number of new cases of periodontitis was 3199, over 655 151 person-years. We stratified this secondary analysis among non-smokers by the other periodontitis risk factors (Table 4). Obese men in the highest quintile of EDIP had 39 % more risk of periodontitis than those in the lowest quintile of EDIP (HR 1·39, 95 % CI 0·98, 1·96, P-value for trend = 0·03) (P-value for interaction by obesity = 0·07). The association was similar among all the other risk factors subgroups. We also examined the joint associations between EDIP and BMI on periodontitis, overall and among non-smokers (Figs. 1 and 2). Obese individuals with the highest EDIP scores had significantly higher risk of periodontal disease.
Table 2. Relating quintiles of empirical dietary inflammatory pattern (EDIP) scores and incidence of periodontitis
(Hazard ratios (HR) and 95 % confidence intervals)
* Model 1: age adjusted.
† Model 2: adjusted for age, smoking (Comprehensive Smoking Index), physical activity (metabolic equivalent of task (MET) quintiles), alcohol (g/d: 0, 0.1–4.9, 5–14.9, 15–29, 30+), occupation (dentist v. non-dentist) and race (White/Black/Asian/Other).
‡ Model 3: model 2 and adjusted for BMI (<18.5, 18.5–24.9, 25–29.9, 30+ kg/m2).
Table 3. Multi-variate association between quintiles of empirical dietary inflammatory pattern (EDIP) and periodontal disease within subgroups*
(Hazard ratios (HR) and 95 % confidence intervals)
* Models adjusted for age, smoking (Comprehensive Smoking Index), physical activity (metabolic equivalent of task (MET) quintiles), alcohol (g/d: 0, 0.1–4.9, 5–14.9, 15–29, 30+), occupation (dentist v. non-dentist), race (White/Black/Asian/Other) and BMI (<18.5, 18.5–24.9, 25–29.9, 30+ kg/m2), except for the stratified variable.
† P-value when each quintile was assigned the median value and treated as a continuous variable.
‡ P-value for the interaction term between an indicator variable for the stratifying term and the continuous variable for dietary pattern (that was used for test of trend).
Table 4. Multi-variate association among non-smokers (excluding current smokers at baseline) between quintiles of empirical dietary inflammatory pattern (EDIP) and periodontal disease within subgroups*
(Hazard ratios (HR) and 95 % confidence intervals)
* Models adjusted for age, smoking (Comprehensive Smoking Index), physical activity (metabolic equivalent of task (MET) quintiles), alcohol (g/d: 0, 0.1–4.9, 5–14.9, 15–29, 30+), occupation (dentist v. non-dentist), race (White/Black/Asian/Other) and BMI (<18.5, 18.5–24.9, 25–29.9, 30+ kg/m2), except for the stratified variable.
† P-value when each quintile was assigned the median value and treated as a continuous variable.
‡ P-value for the interaction term between an indicator variable for the stratifying term and the continuous variable for dietary pattern (that was used for test of trend).
Fig. 1. Multi-variable hazard ratio (HR) for periodontal disease by joint classification of empirical dietary inflammatory pattern (EDIP) and BMI categories.
Fig. 2. Multi-variable hazard ratio (HR) for periodontal disease by joint classification of empirical dietary inflammatory pattern (EDIP) and BMI categories, excluding current smokers at baseline. Adjusted for age, smoking (Comprehensive Smoking Index), physical activity (metabolic equivalent of task (MET) quintiles), alcohol (g/d: 0, 0·1–4·9, 5–14·9, 15–29, 30+), occupation (dentist v. non-dentist) and race (White/Black/Asian/Other). *P-value < 0·05.
Results among obese non-smokers remained similar through multiple sensitivity analyses (online Supplementary Table S2). First, we re-did the BMI-stratified analysis adjusting for continuous BMI (model B), and the results remained similar (HR 1·39, 95 % CI 0·99, 1·97, P-value for trend = 0·03). Then, we evaluated if diabetes could mediate the relationship between inflammatory diet and periodontitis incidence. The relationship between diabetes and periodontal disease is complex and often described as ‘bidirectional’(Reference D’Aiuto, Gable and Syed30,Reference Taylor31) . In addition, proinflammatory diet is reported to be associated with increased risk of diabetes(Reference Schulze, Hoffmann and Manson32); hence, diabetes could be a mediator. We re-did the analysis adjusting for incident diabetes during follow-up (model C), and the association was slightly attenuated (HR 1·35, 95 % CI 0·96, 1·91, P-value for trend = 0·05). In addition, we conducted a separate analysis where we censored individuals when they reported diabetes diagnosis during follow-up (model D), and the results remained significant (HR 1·38, 95 % CI 0·96, 1·99, P-value for trend = 0·03). We also did the analysis excluding those who had less than seventeen teeth at baseline (model E), and the association was similar (HR 1·41, 95 % CI 0·99, 2·01, P-value for trend = 0·03). In addition, we evaluated potential confounding by nutritional supplements, including multi-vitamins, vitamin D, fish oil, cod liver and dehydroepiandrosterone (model F), and the results were similar. We also adjusted for medications that could downgrade systemic inflammation, including aspirin, acetaminophen, nonsteroidal anti-inflammatory drugs and lipid lowering drugs(Reference Andriankaja, Jimenez and Munoz-Torres33), and the results remained similar. Furthermore, we included medications and nutritional supplements in the models (model H), and that did not significantly change the results. Also, we conducted the analysis excluding participants who reported signs of cognitive impairment during follow-up (model I)(Reference Fondell, Townsend and Unger34). Cognitive impairment was defined by answering ‘yes’ to any of the following questions: ‘do you have trouble remembering things from one second to the next?’, ‘do you have any difficulty in understanding or following spoken instructions?’, ‘do you have more trouble than usual following a group conversation or a plot in a TV programme due to your memory?’ or ‘do you have trouble finding you way around familiar streets?’. The association remained similar (HR 1·42, 95 % CI 0·97, 2·07, P-value for trend <0·05). We also did the analysis excluding those with BMI ≥ 35 kg/m2 (model J), and that attenuated the results (HR 1·28, 95 % CI 0·88, 1·86, P-value for trend = 0·11). We also evaluated the association after censoring individuals who had major health condition, MI, stroke, cancer or coronary artery surgery during the follow-up, as those events may independently modify the exposure and/or the risk of the outcome (model K), and the results were attenuated (HR 1·33, 95 % CI 0·93, 1·91, P-value for trend = 0·10). We evaluated the association among former smokers at baseline only (model L, HR 1·51, 95 % CI 0·95, 2·39, P-value for trend = 0·05), and among never smokers at baseline only (model M, HR 1·26, 95 % CI 0·73, 2·19, P-value for trend = 0·31), and the association was attenuated but remained in the same direction.
Discussion
We used EDIP as an RRR-derived dietary index in a large cohort of men with 24 years of follow-up to examine the empirical cumulative inflammatory impact of diet on incidence of periodontal disease. The results showed no overall association between EDIP scores and periodontitis. Subgroup analyses suggested a small non-significant elevated risk of periodontitis among obese men with higher EDIP scores. We performed a secondary analysis among non-smokers during the follow-up and observed a statistically significant association between EDIP and incidence of periodontitis among obese non-smokers.
EDIP scores were weighted averages of the pro- and anti-inflammatory food groups that maximally explained the variability in IL-6, C-reactive protein and TNF-α receptor 2 inflammatory mediators. To our knowledge, this is the first study that primarily investigated the association between the systemic inflammatory impact of diet and incidence of periodontitis. Although a few studies have prospectively investigated the relationship between some of the food groups that composed of EDIP and periodontitis as an outcome(Reference Pitiphat, Merchant and Rimm20,Reference Ng, Kaye and Garcia35–Reference Merchant, Pitiphat and Franz38) , comparing our RRR pattern approach with the ‘single food/nutrient’ approach may not be appropriate, due to the distinct methodological differences between the two methods. For instance, alcohol is an anti-inflammatory component of EDIP and hence could be viewed as being protective against periodontal disease. However, consumption of alcohol has been associated with increased risk of periodontitis, most likely through other mechanisms other than systemic inflammation(Reference Pitiphat, Merchant and Rimm20,Reference Wagner, Haas and Oppermann36) , and we addressed this issue by controlling for the cumulative alcohol intake in the adjusted models. A finding that is more analogous to the current study is the positive association we observed in another analysis between the principal component analysis-derived Western dietary pattern, which was high in processed meat, red meat, butter, high-fat dairy products, eggs and refined grains, and periodontal disease that was limited to obese(Reference Alhassani, Hu and Li39). Although the Western pattern was not derived using inflammatory mediators, it has repeatedly been associated with systemic inflammation(Reference Barbaresko, Koch and Schulze40–Reference Lopez-Garcia, Schulze and Fung42); hence, we previously hypothesised that the apparent impact in obese could be due to modification of systemic inflammation.
Our current results show no indication of an overall association between EDIP and periodontitis, suggesting that the inflammatory aspects of diet may not have a significant impact on periodontitis risk. Nevertheless, this null association between EDIP and periodontitis in the current study could alternatively be attributable to methodological factors. First, inherent to dietary pattern analysis is the potential dilution of some components’ effects, as foods and nutrients that compose the pattern do have the biological potential to either embellish or abolish each other’s impact(Reference Hu43). For example, Ng et al. recently reported that higher consumption of coffee was associated with a lower risk of periodontal bone loss over the follow-up period of their study(Reference Ng, Kaye and Garcia35). The influence of coffee as an anti-inflammatory component of EDIP could have been weakened by other components of the index either through systemic inflammation or through other mechanisms. Second, the incidence of periodontitis in our study was defined by answering ‘yes’ to the question ‘have you been professionally diagnosed with periodontal disease with bone loss?’ in the biennial questionnaire. Self-reported periodontal disease was evaluated in a subsample of the HPFS and showed appropriate validity. The positive predictive value among the dentists in HPFS was 0·76, and the negative predictive value was 0·74. For non-dentists, the positive predictive value was 0·83 and the negative predictive value was 0·69, making the method a suitable proxy and a valid ‘endpoint’ in this cohort(Reference Joshipura, Pitiphat and Douglass16,Reference Joshipura, Douglass and Garcia17) . Clinically, periodontal tissue breakdown is assessed based on a continuum of measures in addition to many signs and symptoms that determine periodontitis case diagnosis at an individual-patient level; the extent, rate and risk of future disease are evaluated to diagnose periodontitis stage and grade(Reference Caton, Armitage and Berglundh44). However, there is a lack of consensus of what determines a periodontitis case at the population level(Reference Kingman, Susin and Albandar45,Reference Holtfreter, Albandar and Dietrich46) . Although several associations with periodontal disease have been documented in this cohort(Reference Jimenez, Hu and Marino25,Reference Merchant, Pitiphat and Franz38,Reference Jimenez, Giovannucci and Krall Kaye47,Reference Jimenez, Hu and Marino48) , it is possible that the case ascertainment method in the present study may not have been sensitive to detect a potential impact of inflammatory diet on periodontal tissue.
Our results suggest that BMI, mainly obesity, may act as an effect modifier, rather than a mediator or a confounder. Obese individuals in the highest quintile of EDIP had a higher risk of periodontitis compared with the lowest quintile, but the association was only statistically significant after exclusions of current smokers at baseline. Smoking is the most important environmental risk factor of periodontitis(Reference Johannsen, Susin and Gustafsson49), with the population attributable risk of periodontitis due to smoking estimated to be up to and even more than 50 %, depending on the periodontitis case definition and participants’ age(Reference Hyman and Reid50–Reference Tomar and Asma52), our secondary analysis among non-smokers suggests that the prominent deleterious effect of smoking on periodontal health could mask other risk factors, and hence the observed relationship in our study in obese individuals was definite only among non-smokers. Recently, Jauhiainen et al. prospectively investigated the association between diet quality, using the Baltic Sea Diet Score and the Recommended Finnish Diet Score, and periodontal disease over 11 years of follow-up(Reference Jauhiainen, Ylostalo and Knuuttila53). They found a stronger impact of poor diet among the non-smokers.
The longitudinal prospective association between systemic inflammation and periodontal disease progression has been reported recently by Pink et al. (Reference Pink, Kocher and Meisel10). They found a positive association between baseline measures of fibrinogen and leucocytes as markers of systemic inflammation and periodontal tissue loss over the 11-year follow-up period of the study(Reference Pink, Kocher and Meisel10). As obesity is associated with ‘metainflammation’, a state of low-grade chronic systemic inflammation orchestrated by host cells in response to excessive energy and nutrients intake(Reference Gregor and Hotamisligil54), it is plausible that the harmful inflammatory impact of diet on the periodontium is mainly through exacerbation of this ‘baseline’ metainflammation state; hence, the observed association in our study was limited to obese individuals and was not observed in other subgroups. Another potential mechanism by which proinflammatory diet may attribute to the pathogenesis of periodontal disease is mediated by diabetes. An RRR-derived inflammatory dietary pattern has been previously associated with increased risk of diabetes(Reference Schulze, Hoffmann and Manson32). Also, the association between EDIP and periodontitis among obese non-smokers was attenuated when we adjusted for incidence diabetes in the model, which suggests that among those who reported diabetes during follow-up, diabetes could have acted as a mediator in the relationship between EDIP and periodontitis. However, in a separate analysis where we censored men if they reported diabetes diagnosis during follow-up, the observed association between EDIP and periodontitis remained significant, which insinuates that the relationship could be related to other inflammatory mechanisms other than diabetes. On the other hand, as obese are at higher risk of diabetes(Reference Koh-Banerjee, Wang and Hu55), we cannot eliminate the possibility that undiagnosed pre-diabetes insulin resistance and glucose intolerance may have mediated the association between EDIP and periodontitis in the obese–non-smokers subgroup. There is a strong evidence that insulin resistance does exaggerate the adverse metabolic effects of diet(Reference Liu, Willett and Stampfer56–Reference Tabung, Wang and Fung59). In addition, pre-diabetes and insulin resistance have been associated with increased periodontal disease(Reference D’Aiuto, Gable and Syed30,Reference Timonen, Saxlin and Knuuttila60–Reference Islam, Seo and Lee63) .
Our study has several strengths. The study population is a large cohort of highly educated and motivated participants, which minimises information bias. The prospective panel design of the study with the long follow-up time, and updated measures, does support better understanding of temporality and may aid in establishing causality. Our study however has several limitations. First, the study is observational in nature; hence, confounding cannot be ruled out. Another limitation is the self-reported data. However, the validity of self-administered FFQ and the self-reported periodontal disease have been evaluated. FFQ data showed reasonable correlations with diet records, and self-reported periodontal disease had acceptable positive and negative predictive values compared with intraoral radiographs(Reference Willett14–Reference Joshipura, Douglass and Garcia17). It is expected however that some degree of misclassification occurred, which we assumed would be non-differential, and could have attenuated the results towards the null(Reference Rothman, Greenland and Lash64). In addition, we used the cumulative average of EDIP scores, which in addition to the energy adjustment, mitigate the issue of measurement error. Furthermore, the study population is composed of health professional men, mainly of Caucasian descent; hence, the generalisability of the results may be limited. However, the homogeneity of the cohort improves the internal validity of the study as confounding by socio-economic and educational factors was inherently minimised. Furthermore, some EDIP components may appear contrary to the prevailing knowledge, for example, the positive association of tomatoes and the inverse association of pizza with concentrations of inflammatory markers. EDIP scores were developed using an empirical approach in which the combination of foods that maximally predicted concentrations of inflammatory biomarkers (IL-6, C-reactive protein and TNF-α receptor 2) were selected in an unbiased and unsupervised manner. Fresh tomatoes have a low content of bioavailable lycopene which is a major anti-inflammatory nutrient, whereas cooked tomato paste (e.g. in pizza) contains 2–5 times higher concentrations of bioavailable lycopene(Reference Gartner, Stahl and Sies65,Reference Marcotorchino, Romier and Gouranton66) . Also, if fresh tomatoes are incorporated in salads that include sources of fats like olive oil or avocado, this would make the lycopene more bioavailable. However, a limitation of FFQ is that they do not generally assess the way foods are prepared, combined or eaten. Also, not all EDIP components are universally confirmed as pro- or anti-inflammatory. For instance, tomatoes are positively associated with inflammation in EDIP. Yet, studies have found either no association(Reference Blum, Monir and Khazim67,Reference Markovits, Ben Amotz and Levy68) or inverse association between tomatoes and inflammatory markers(Reference Li, Chang and Huang69,Reference Mohri, Takahashi and Sakai70) .
In conclusion, we observed no overall association between EDIP scores and the risk of self-reported periodontal disease assessed using questionnaires in this population over the study period. However, only among obese non-smokers, those with higher EDIP scores had a significantly higher risk of periodontal disease compared with those with lower scores. Findings of the current study suggest a potential role of diet in modifying the risk of periodontitis, through systemic inflammation, in obese non-smokers. Future research could focus on using clinical periodontal measures (such as probing depth and periodontal attachment loss) to explore if inflammatory dietary pattern could influence the risk in other subgroups and to explore the specific components of diets that are more germane to periodontal health.
Acknowledgements
The authors thank the participants and staff of the Health Professionals Follow-Up Study.
Supported by the National Institutes of Health (NIH) research grant UM1 CA167552, Health Professionals Follow-Up Study infrastructure grant.
All authors have contributed significantly in the manuscript. A. A. contributed in conception and design of the study, data analysis, and interpretation of data; and drafting the article. F. H. contributed in conception and design of the study and interpretation of data; and revising the article critically for important intellectual content. B. R. contributed in conception and design of the study, data analysis and interpretation of data; and revising the article critically for important intellectual content. F. T. contributed in data analysis and interpretation of data; and revising the article critically for important intellectual content. W. W. contributed in conception and design of the study, acquisition of data and interpretation of data; and revising the article critically for important intellectual content. K. J. contributed in conception and design of the study and interpretation of data; and revising the article critically for important intellectual content. All authors have read and approved the final version of the manuscript.
The authors had no conflicts of interest to disclose.
Supplementary material
For supplementary material referred to in this article, please visit https://doi.org/10.1017/S0007114520005231
