Wei, Lin and Weissfeld(26) developed an approach that can be used to simultaneously analyze time to first incident detection of several types of HPV either in the same or different clinical visits, taking into account possible correlations between the types. While each HPV type is allowed to have its own baseline hazard function, a common (overall) exposure effect can be assumed in the following WLW model: (1.8)λk(tik)=λ0k(tik)eβZi(tik)+γWi(tik) where λ0k(t) is the baseline hazard function for the kth type of event and exp(β) is the common hazard ratio across oncogenic HPV types. Similar to the parameter estimate in GEE, β has a population-averaged interpretation. Model (1.8) is in essence a Cox model stratified by HPV type and it is estimated under an independent working correlation, with a variance estimator that is robust to possible correlations between events.
