Clinical studies of periodontal diseases often involve obtaining observations for variables measured at multiple sites within each patient studied. These observations are thus inherently dependent, and the use of standard statistical methods to measure associations among the variables is inappropriate since they ordinarily assume independence among observations. Methods are suggested for extending standard statistical methods to the situation where data are dependent. For example, if the sensitivity and specificity of a diagnostic test is of interest, the usual binomial distribution, which assumes independence, can be replaced with the correlated binomial, and these characteristics can still be estimated in the presence of correlated data. In situations where the interrelationships of several variables to a response would ordinarily be investigated using multiple linear or logistic regression, the presence of correlated data such as that arising from clinical studies of periodontal disease requires that adjustments be made to account for the correlation. One of the most promising methods for taking such correlation into account is the use of generalized estimating equation (GEE) methodology, which is described in this paper and employed in two examples. The two examples involve measuring the association of levels of aspartate aminotransferase (AST) with tissue destruction, and the evaluation of the effects of metronidazole and surgery on attachment level in different groups of chronic periodontitis patients. Several statistical approaches and their conclusions are compared for each of the examples. GEE methodology allows the extension of standard regression models to situations involving dependent data while maintaining the ease of interpretation associated with regression results. Copyright © 1991, Wiley Blackwell. All rights reserved
