It is common in HPV studies to use multivariable logistic regression to cross-sectionally analyze the factors associated with “any HPV”, “any oncogenic HPV type”, or an individual HPV type, at a given study visit. In such analyses, each woman contributes a single outcome. Notationally, let Yi=1 represent the detection of HPV16 infection for the ith person at the selected study visit, and 0 otherwise. We model Pi = P(Yi = 1) by (1.1)logitPi=β0+βZi+γWi where Zi is our exposure variable of interest (e.g., a vector of indicators of 12 separate CD4+ / HIV RNA strata) and Wi is the vector of other adjusted variables, including age, race, lifetime of number of male sex, etc. for the ith person. The primary parameter of interest is β , defined as the log odds ratio (OR) of HPV16 infection between women in a certain CD4+ / HIV RNA stratum and HIV-negative women.
