Cervical cancer models overview
Despite substantial reductions in cervical cancer incidence since the widespread adoption of cytology (Pap-based) screening, over 12,000 women develop and 4,000 women die from cervical cancer in the United States (US) each year. The discovery of human papillomavirus (HPV) as the necessary cause of cervical cancer has led to a surge of new evidence and emergence of innovative HPV-based technologies for both primary and secondary prevention that have allowed for significant changes in the approach to cervical cancer control over the past decade. These new opportunities for improving cervical cancer control in the US also pose challenges for decision-making.
Five cervical cancer modeling groups joined CISNET in 2015 to begin collaborative modeling to address these challenges in cervical cancer control efforts. The collaborative group includes four independent teams who have been at the frontier of modeling cervical cancer screening and HPV vaccination over the last decade: Jane Kim (Harvard T.H. Chan School of Public Health), Shalini Kulasingam (University of Minnesota), Marjolein van Ballegooijen (Erasmus Medical Center, the Netherlands), and Karen Canfell (Cancer Council New South Wales CCNSW], Australia). An additional team led by Ruanne Barnabas (University of Washington [UW]), one of the original architects of HPV transmission modeling, will play a focused role in addressing key questions for HIV-positive women.
HPV Infection: All models include the natural history of HPV infection to cervical cancer; the process of HPV acquisition and transmission is explicitly simulated in four independent "dynamic" models (Harvard, UW, Erasmus MC, CCNSW) which allow for interaction among individuals in the population. The process of cervical carcinogenesis is simulated in greater detail in four independent "static" models (Harvard, UMN, Erasmus MC, CCNSW), which do not capture interactions among individuals. In all models, the effect of HPV natural immunity is modeled as a reduction in the probability of future type-specific infection and is able to flexibly vary by duration and pattern of waning.
HPV Classification: One key model difference includes the grouping of HPV types in the models. For example, the UMN, UW and Erasmus MC (MISCAN) models pool all non-16/18 high-risk HPV types into a single category, whereas, in anticipation of the next-generation vaccine analyses, the CCNSW (Policy1) model additionally stratifies and pools the five additional high-risk types included in the 9-valent (9v) vaccine, and the HSPH model stratifies by each individual high-risk type included in the 9v vaccine. There are also differences in how cervical intraepithelial neoplasia (CIN) states are grouped; for example, the UMN, MISCAN, and Policy1 static models incorporate HPV, CIN1, 2 and 3 as separate states, whereas the HSPH model combines HPV and CIN1 into a single state. Cancer groupings vary by staging according to the US Surveillance, Epidemiology, and End Results (SEER) Program (i.e., local, regional, distant; HSPH, UMN, a Policy1 option) or the International Federation of Gynecology and Obstetrics (FIGO) (I to IV; MISCAN, a CCNSW POLICY1 option, UW). Models also differ by conditional transition probabilities (e.g., age, duration of HPV infection or health state), calibration parameters (e.g., natural history, test characteristics), and cycle length (e.g., monthly, annual). These differences may also lead to variations in outcomes, especially if health-related events or interventions occur in intervals shorter than one year.
HPV Vaccination: HPV vaccination in all models, dynamic or static, is modeled as a reduction in the incidence of vaccine-type HPV infections, the extent of which depends on uptake by age, number and timing of doses, vaccine efficacy, and duration of vaccine protection (i.e., waning).
Cervical Cancer Screening: Screening is used to detect the presence of low- and high-grade precancerous lesions, which can be treated and removed before progressing to cancer, as well as for early detection of invasive cancer. The efficacy of screening depends on coverage by age, interval, screening and diagnostic test characteristics, treatment efficacy, and compliance to follow-up visits. All models use histologic classifications (i.e., CIN) to represent the "true" precancer health states.
HIV: HIV infection dynamics are modeled concurrently with HPV in the UW model and the STDSIM model (developed by Erasmus MC). In the UW model, men and women acquire HIV through sexual partnerships. Although HIV is modeled independently from HPV, the risk of acquiring HIV is elevated if persons have HPV. Once infected with HIV, individuals transition from acute infection through declining CD4 counts and increasing viral load based on empiric data. Persons can start antiretroviral therapy (ART) and drop out of treatment. HPV does not impact HIV progression. However, HIV/HPV co-infected women experience higher rates of HPV persistence and progression to CIN. These rates are based on CD4 count, viral load, and ART status.
The STDSIM model is calibrated to KwaZulu-Natal, South Africa, based on data on demographics, sexual behavior, and sexually transmitted infection (STI) prevalence from the Africa Center. STI transmission is modeled through a dynamic (hetero-)sexual network. Individuals infected with HIV progress from an acute stage to asymptomatic, symptomatic and AIDS stages. The facilitating effects of HSV-2, syphilis, gonorrhea, and chlamydia on HIV acquisition are taken into account by assigning cofactors. HPV is included in the model, with a cofactor to reproduce the observed doubled risk of HIV acquisition. In addition, HIV-infected individuals have a higher risk of acquiring HPV, and experience more persistent HPV infections. Interventions such as ART and HPV vaccination are included in the model.
Outcomes: Each of the models have the capability to produce a common set of analytic outcomes, including epidemiologic outcomes (e.g., age- and type-specific HPV prevalence, CIN prevalence, cancer incidence, cancer mortality), screening and diagnostic outcomes (e.g., number needed to screen per death averted, number of false positive screens, colposcopies, precancer treatments), clinical outcomes (e.g., number of diagnosed cancers and cancer-related deaths, cancer by stage of detection, life expectancy, quality-adjusted life expectancy), and economic outcomes (e.g., short-term or lifetime costs).
Public Health Impact
The Cervical CISNET group will utilize these models to evaluate the health benefits, harms and costs of cervical cancer prevention strategies in the US, including HPV vaccination and screening. They will also identify the most efficient and cost-effective cancer control strategies considering new technologies such as the 9v vaccine and emerging screening techniques. The models will also be used to identify the most cost-effective cancer control strategies for HIV-positive women in the US and three low- and middle-income countries (Kenya, Uganda, and South Africa).