Cervical Cancer Model, Cancer Council NSW
Model overview. The Policy1-Cervix (CCNSW) model includes a dynamic simulation of human papillomavirus (HPV) transmission and vaccination linked to a multi-cohort static simulation model of cervical carcinogenesis, cervical screening, diagnosis, pre-cancer treatment, post-treatment surveillance, and invasive cancer survival.
HPV transmission. Heterosexual behavior is modeled by stratifying the population by sex, age, and level of sexual activity (i.e., four sexual activity groups) using data from national behavioral surveys of sexual behavior. The model has been extended to include semi-assortative and age-and sex-specific mixing parameters, a revised sexual mixing matrix, the capacity to vary the annual per-partner transmission probability according to HPV type, sex and sexual activity group, and ability to capture the effects of more rapid change in behavior (by single year of age) during adolescence and early adulthood. There is capacity to simulate alternative assumptions for the duration of naturally-conferred type-specific immunity against HPV infection and its waning. A multi-type structure is used; recently, the model was extended in order to simulate the effects of next-generation nonavalent (9v) vaccines (i.e., grouped as HPV16, HPV18, other 9v-included types, and other non-9v-included high risk types).
Cervical carcinogenesis. This component takes cohort- and type-specific HPV incidence from the dynamic model as input and involves a complex multi-cohort semi-Markov implementation of the natural history of cervical pre-cancer. Progression and regression between states representing HPV infection, cervical intraepithelial neoplasia (CIN)1, 2 and 3 (due to particular HPV types or groups) are modeled, as is progression from CIN3 to invasive cervical cancer. The model accounts for age-specific hysterectomy rates (for any cause) in the population.
Vaccination. The dynamic model simulates vaccination uptake by single year of age, sex, and time. (1-3) Vaccination of older females (and males) in catch-up programs, if applicable, is modeled by single year of age, taking into account the potential for prior HPV type-specific exposure and its impact on type-specific vaccination efficacy at different ages. Male vaccination uptake is also modeled to account for incremental herd immunity effects in females. The model allows varying vaccine properties (e.g., efficacy, waning). Herd immunity is a phenomenon where vaccination of a significant proportion of the population can reduce the prevalence of the vaccine-targeted HPV types in the population, thereby providing some protection for individuals who are not vaccinated.
Screening, diagnosis, and treatment of pre-cancer. The sensitivity and specificity of cytology are setting-specific and fitted to data on the distribution of cytology test results (e.g., cytology-histology correlations) in a particular setting. (3-6) Fitted test characteristics are constrained to be consistent with findings from international meta-analyses (7-8) that report the absolute and relative sensitivity and specificity of cytology and HPV testing. Detailed analysis of registry data on the age at which young women initiate screening is performed, and for all ages rates of return to screening or follow-up management over a 10-year period is simulated, according to last screening test result, the follow-up recommendation, and 10-year age group. Following treatment for CIN, post-treatment natural history and surveillance for recurrent disease are based on meta-analysis of the literature on outcomes after pre-cancer treatment. (9) A separate model has been developed for estimating adverse reproductive outcomes in the population given alternative screening strategies and associated CIN excisional treatment rates by age.
Cancer treatment and survival. Cancer staging and progression is modeled, accounting for symptomatic detection and the possibility of downstaging at diagnosis due to screening. Predictions for age-specific cervical cancer incidence and mortality have been calibrated to observed rates in unscreened populations. The model is then additionally validated against country-specific registry data for incidence and mortality, when run with an overlay of screening according to country-specific guidelines. The stage and interval-specific cancer survival parameters are based on analysis of data from cancer registries and validated against observed data.
Calibration and validation. Unobservable parameters, such as the duration of immunity following natural infection, were calibrated using a multi-parameter Bayesian approach to the age-specific HPV prevalence prior to vaccination (2) and the proportion of women positive for HPV-16, -18 and other high-risk types. Natural history progression and regression rates for cervical carcinogenesis have been calibrated across a large number of targets from national screening programs in Australia, England, and New Zealand, after incorporating setting-specific screening behavior and clinical management algorithms. Calibration targets from each country include age-and type-specific HPV prevalence, screen-detected high-grade abnormalities, CIN 2/3 treatment ratio, numbers of screening, diagnostic number of tests, and age-and type-specific rates of cancer incidence and mortality. The model produces accurate predictions of the relative contribution of specific HPV types to HPV infections in the population, screen-detected histologically-confirmed high-grade abnormalities, and invasive cervical cancer rates and case numbers. (3, 10) A close correspondence between prior model predictions of HPV vaccine impact and the observed post-vaccination data on HPV prevalence (11) in Australia has been demonstrated post-hoc. Ongoing validation will use findings from the Compass trial, a major Australian trial assessing primary HPV testing.
Tip: Hover your cursor over the dashed attribute links below for more information. View the details of this model in a grid with other cervical models.
You may be interested in these publications by this modeling group, which were supported by a funding source other than the CISNET grant.