Multiple regression analysis of cytogenetic human data

Biomonitoring studies on cytogenetic outcomes in humans should be considered as epidemiological studies, rather than randomized trials. Under this light the emphasis given to the achievement of a significant p-value should be reduced, since this measure suffers from major limitations. The use of a point estimate (and its corresponding confidence interval) to measure the association between exposure and effect offers several advantages, including the adjustment for confounding, and the evaluation of possible interaction between factors. In most instances the use of multivariate statistical methods allows an efficient analysis of these studies, even in presence of a small sample size and several covariates. In this paper we re-analyzed four biomonitoring studies by using multivariate methods to estimate relative risks through statistical modeling. The use of multiple regression techniques allowed the computation of point estimates of association and their confidence intervals for each covariate evaluated by the studies considered; the estimate of the effect of confounding variables such as smoking habits, age and gender; and the presence of interaction between covariates. Measures of association estimated through univariate and multivariate statistical approaches are compared. The advantages of the latter technique are discussed.