Resumen de Reexamining the Association Between Smoking and Periodontitis in the Dunedin Study With an Enhanced Analytical Approach
Background: Smoking is a major risk factor for periodontal disease. Conventional oral epidemiology approaches have found strong, consistent associations between chronic smoking and periodontal attachment loss (AL) through ages 26, 32, and 38 years, but those statistical methods disregarded the data�s hierarchical structure. This study reexamines the association using hierarchical modeling to: 1) overcome the limitations of an earlier approach (trajectory analysis) to the data and 2) determine the robustness of the earlier inferences.
Methods: Periodontal examinations were conducted at ages 26, 32, and 38 years in the Dunedin Multidisciplinary Health and Development Study. The number of participants examined at those three ages were 913, 918, and 913, respectively. A generalized linear mixed model with a quasi-binomial approach was used to examine associations between chronic smoking and periodontal AL.
Results: At ages 26, 32, and 38, smokers had 3.5%, 12.8%, and 23.2% greater AL than non-smokers. Regular cannabis use was associated with greater AL after age 32, but not at age 26. Males had more AL than females. Participants with high plaque scores had consistently greater AL; those who were of persistently low socioeconomic status had higher AL at ages 32 and 38, but not at age 26. The amount of AL in anterior teeth was less than in premolars and molars. Gingival bleeding was associated with higher AL at ages 26, 32, and 38.
Conclusion: The smoking�periodontitis association is observable with hierarchical modeling, providing strong evidence that chronic smoking is a risk factor for periodontitis.
Periodontitis is a chronic polymicrobial disease that occurs when the host response to usually commensal organisms in the plaque biofilm becomes destructive.1 Smoking remains the most important risk factor,2-4 with evidence suggesting that its periodontal effects occur regardless of what is smoked.5-7 Smoking is thought to exert its effect through affecting neutrophil function, causing shifts to a more pathogenic microflora, and causing sustained peripheral vasoconstriction.8 From a life-course perspective, the effect of chronic smoking is an example of the risk accumulation model, whereby its effects are cumulative over time.9 Moreover, marked social gradients in the occurrence of smoking contribute to socioeconomic gradients in periodontitis,10,11 but the extent of that contribution is currently unclear. Day-to-day self-care and plaque control are also important in the condition�s occurrence.
Although previous papers have reported a strong association between smoking and periodontal disease through ages 26, 32, and 38 years in a birth cohort,5-7 the approaches adopted for those data analyses do have limitations. A large proportion of that earlier work considered the use of hypothesis testing, logistic regression, and growth trajectory modeling to identify the risk factors. One of the main concerns is that the response variable for these analyses was the prevalence, the extent, or the severity of periodontitis, yet there is considerable variation in case definitions for it in the literature. This applies no less to defining incident cases, with incident attachment loss (AL) being defined as a change in AL of ?2, ?4, ?5, or even ?6 mm, depending on the sample and the inter- and intra-examiner reliability. Such variation will inevitably lead to considerable heterogeneity in analytic outcomes. Another fundamental problem is that those methods do not account for the natural hierarchical structure of the periodontal data. Thus, while they can be used to identify person-level risk factors, the site-specific nature of the AL is unable to be taken into account.
The aim of this study is to reexamine the periodontal effects of smoking and the impact of the other putative risk factors through early to middle adulthood cross-sectionally, using a more informative approach. Multilevel modeling is recognized as the most appropriate method for analyzing data of this type, because it accounts for the hierarchical structure of the data and allows the simultaneous examination of covariates� effects on different levels, for each person, tooth, and site.12 However, multilevel analysis carries the assumption that the random effects and measurement errors are normally distributed. This assumption is unlikely to be satisfied with AL data. To address this, a generalized linear mixed model (GLMM) with a quasi-binomial approach was used here as an extension of the traditional multilevel modeling method for data analysis.
