Regression models in periodontal epidemiology: purpose, approach and interpretation

Abstract

AbstractStatistical models, including those using regression analyses, are ubiquitous in periodontal research. However, many investigators may not have a thorough understanding of the concepts underlying the complex statistical models used in much of the current literature. This review aims to demystify biostatical regression models and familiarize the reader with the regression models most commonly encountered in the periodontal literature. Complex models can be intuitively understood in terms of stratified analyses. Therefore, we discuss the use of stratified analyses and emphasize their natural connection to regression models using examples from the periodontal research literature. We critically review various types of regression models, as well as the fundamental assumptions unique to each. An introduction to the fundamentals of model building describes the basics of variable selection procedures and variable management, including discussions regarding the assumptions inherent in common practices. Lastly, modeling approaches to assess confounding and effect modification are discussed by use of examples from the periodontal epidemiological literature.