We use data from Strickler et al(8) to illustrate the gain in power with GEE compared with standard logistic regression. Based on 535 women with a CD4+ count <200 cells/mm3 and 458 women with ≥500 cells/mm3, Figure 2 shows the power to detect OR=2.0 for the association of HPV prevalence and CD4+ count with (i) only one observation (e.g., one visit with one HPV type), J=1 (i.e., standard logistic regression), versus (ii) J=5 or =10 (based on GEE) per subject under various assumptions regarding the prevalence rate of HPV infection. In our example, the correlation between repeated observations across HPV types and visits were assumed to be either 0.3 (moderate) or 0.6 (high). As demonstrated in Figure 2, the power when J=5 is substantially larger than the power without repeated observations (J=1). The power when J=10 is only slightly higher than the power with J=5, indicating that while there is a benefit to statistical power with a certain number of repeated observations, there is a diminishing impact on power as the number of repeats increases. Further, the increase in power
