A multilevel logistic regression model for the analyses of concurrent Human papillomavirus (HPV) infections

Human Papillomavirus (HPV) is a sexually transmitted infection that causes cervical cancer. A nested three-level logistic regression model was introduced in order to investigate whether, in the IARC HPV prevalence surveys, co-infection with different HPV types occurs more or less frequently than expected if the infections are independent from one another. Two random effects, at individual and study-area level, were specified, while the fixed-effect covariates at individual level were age and lifetime number of sexual partners. The Best Linear Unbiased Predictors (BLUP) technique was used to estimate the random components. The predictions of the random effects at individual level are particularly important because they can be considered as a synthetic estimate of all those residual sources of individual variability, i.e., unmeasured risk factors due to sexual habits, that otherwise could not be accounted for. Individual probabilities of being positive for each HPV type are thus estimated, and the expected vs observed number of infections are compared, given the positivity for a different HPV type. Few positive associations (HPV58 with 33 being the strongest) were found in this analyses. However, the majority of HPV types, particularly the two most oncogenic types, HPV16 and 18, that are also included in the prophylactic vaccine, were not associated with one another.