| To Register SMDM Homepage | |
|
|
|
||||
Purpose: The multi-step model of cervical cancer pathogenesis involves, as the first step, infection with high-risk types of human papillomavirus (HPV). Many women with transient HPV will develop cervical abnormalities, although low-grade (LG) lesions have a high-rate of spontaneous regression. Persistence of high-risk HPV types is a prerequisite for the development of high-grade lesions (HG) and cancer. Our objective was to use primary longitudinal data on HPV infection, and subsequent detection of LG and HG lesions to inform underlying transition probabilities in a natural history model of cervical cancer.
Methods: We developed a first-order Monte Carlo simulation model of underlying HPV and cervical disease to calibrate to outcomes in a study of 2,400 high-risk Brazilian women who received HPV testing and cytology screening at enrollment, 8 months, 12 months, and every year thereafter for 6 years. We simulated the underlying disease process for an initial cohort of 13-year-old girls, who were assigned a study entry age and screening schedule based on primary data from the study. We assumed that clearance of HPV depended on age and HPV type, and progression to LG and HG lesions depended on age, duration of HPV infection, and HPV type (high-risk or low-risk). Model outputs consisted of “results” from the screening tests, which depended on their sensitivities and specificities. Hazard ratios for detection of LG and HG lesions were compared with those from the analysis of the primary data.
Results: We identified a set of model parameters that describe the underlying transitions among HPV and cervical disease states that calibrate well to estimated hazard ratios for the association of low-risk and high-risk HPV status at enrollment with detection of LG and HG cervical lesions. Within 24 months, hazard ratios for association between high-risk HPV status and detection of HG lesions was 5.97 and of LG lesions was 5.08 (6.25 and 5.39 in the Brazil study, respectively). When allowing for time-varying HPV status, however, model derived hazard ratios were significantly lower than the primary data, suggesting a re-parameterization of the way that HPV is modeled over time.
Conclusions: Leveraging primary data from longitudinal studies together with model simulation methods provides unique opportunities for parameterizing the unobservable and transient nature of HPV infection and its role in the development of cervical cancer.
See more of Poster Session - Public Health; Methodological Advances
See more of The 26th Annual Meeting of the Society for Medical Decision Making (October 17-20, 2004)