To incorporate repeated observations across multiple HPV types in addition to repeated observations across multiple visits, a mixed effects model of Pijk (which assumes a common OR across the types) can be defined as (1.6)logitPijk=μ+αi+γk+δZij+ηWij where the type-specific baseline prevalence rate represented by γ k, like αi, is assumed be a random effect. This model assumes that correlations among repeated infections of the same type are higher than correlations between different types of HPV, consistent with empiric observations. We can further expand model (1.6) to allow HPV type-specific odds ratios (details are not provided here). Mixed effects models for binary outcome are in general more computationally intensive than GEE and can sometimes have difficulty in convergence (i.e., the models can not be fit and no result is obtained) especially with small datasets.
