Periodontitis and Systemic Disease: The Impact of Covariate Selection

Most evidence implicating periodontitis as a causal agent in systemic disease comes from observational research, where confounding is ubiquitous. There are substantial methodological differences among observational studies, of which an important one is the choice of covariates used to adjust for confounding. The present study assesses the impact of covariate selection on the association between periodontitis and systemic disease, using cardiovascular disease and cognitive function as examples. Data were taken from the National Health and Nutrition Examination Survey, where periodontal status was assessed by full-mouth examination. Nineteen covariates were available for selection, including sociodemographic factors, health behaviors, and physiologic measures. Multiverse analysis and specification curves were used to visualize how the distribution of estimated associations between periodontitis and systemic disease vary across 393,216 models, depending on the selection of covariates. The findings show that estimates of the association between periodontitis and systemic disease are sensitive to the choice of covariates included within models, in some cases spanning positive and negative directions (i.e., Janus effect). Depending on the covariates included in the adjustment set, the odds ratio for the association of severe periodontitis with cardiovascular disease ranged from 0.81 to 1.51, while the linear regression coefficient for the association with cognitive function ranged from −1.56 to 0.50. Our study makes 2 important contributions. First, we introduce the idea of multiverse analysis to periodontal research and show that it can be a valuable tool for understanding challenges around model selection and misspecification. Second, we use rigorous quantitative methods to highlight the importance of careful covariate selection in determining our understanding of the relationship between periodontitis and systemic disease.