The objective for this paper was to present and discuss the use of odds ratios and prevalence ratios using real data with a complex sampling design. We carried out a cross-sectional study using data obtained from a two-stage stratified cluster sample from a study conducted in 2001-2002 (n = 1,958). Odds ratios and prevalence ratios were obtained by unconditional logistic regression and Poisson regression, respectively, for later comparison using the Stata statistical package (v. 7.0). Confidence intervals and design effects were considered in the evaluation of the precision of estimates. Two outcomes of a cross-sectional study with different prevalences were evaluated: vaccination against influenza (66.1%) and self-referred lung disease (6.9%). In the high-prevalence scenario, using prevalence ratios the estimates were more conservative and we found narrower confidence intervals. In the low-prevalence scenario, we found no important numeric differences between the estimates and standard errors obtained using the two techniques. A design effect greater than one indicates that the sample design has increased the variance of the estimate. However, it is the researcher's task to choose which technique and measure to use for each data set, since this choice must remain within the scope of epidemiology.
Cross-sectional studies; Odds ratio; Prevalence ratios
