Purpose: Previous models have evaluated the effectiveness and cost-effectiveness of HPV vaccination and Cervical Cancer screening based on few natural history parameter sets, despite considerable uncertainty surrounding these parameters. More research is needed to better characterize and understand the natural history of HPV. The objective of this study is to develop a framework to identify all plausible natural history parameters of a detailed cohort model of HPV infection, consistent with empirical epidemiological trends.

Methods: We developed a 3 stages process: 1) sampling, 2) fitting, and 3) sensitivity analysis. Prior uniform probability distributions were defined for each natural history parameter using minimum and maximum parameter values derived from the literature. Numerous combinations of parameters were drawn from prior distributions using Latin Hypercube Sampling (LHS). Model predictions on age/type-specific incidence and prevalence of HPV infection, CIN and cervical cancer as well as genotype distributions were compared to empirical North-American longitudinal and cross-sectional data using different GOF algorithms. Sensitivity analyses of the quality of fits (GOF) to the different parameters values were performed by univariate and multivariate analysis.

Results: Preliminary analysis suggests that, GOF variability due to natural history parameters is important. Given model assumptions, the most influential parameters are clearance and natural immunity. The LHS-GOF parameterization process effectively and objectively finds input parameter sets that best fit North-American HPV data. Identification of parameter sets with equal GOF selects different natural history scenarios that are equally plausible.

Conclusions: This work is original because we focus on understanding the variation and sensitivity of the GOF results to parameter assumptions to answer important parameterization questions such as: 1) What is the overall GOF variability due to uncertainty in natural history parameters (prior distributions)?; 2) Which input parameters explain most of the fitting imprecision (i.e. GOF variability)?; 3) Which parameter sets best fit current empirical North-American HPV epidemiological trends?; 4) How sensitive are 1-3 to GOF measures used, 5) What important data are needed. The 3-stages parameterization process (LHS + GOF + SA) permits to effectively explore a wide range of parameter values and to interpret the results within a statistical framework necessary to explore natural history assumptions. Such research is needed due to the structural complexity of HPV models and high degree of uncertainty in the natural history of HPV.