This paper shows that marginal and mixed effects models are appropriate, efficient methods for the analysis of HPV natural history data. While it has been common to use standard logistic regression for the analysis of the prevalent (cross-sectional) detection of HPV, and standard Cox models for time to incident detection (or clearance) of HPV, these common approaches do not make use of all the available data. That is, standard logistic and Cox regression models can only consider a single HPV type, and logistic regression can also only consider a single visit or a single time period. Marginal and mixed effects models, in contrast, can simultaneously address multiple oncogenic HPV types as well as multiple visits over time. As shown in paper, they consequently have greater statistical power in the analysis of HPV natural history data. In addition, these models allow comparison of exposure-disease associations between HPV types. Thus, marginal and mixed effects models could have many useful applications to HPV research, both in randomized clinical trials and epidemiologic studies.
