ANALYSIS AND POLICY
Tackling Viruses of Global Health Importance
Human papillomavirus (HPV), human immunodeficiency virus (HIV), and hepatitis C virus (HCV) can result in serious sequelae years after initial infection. Clinical and public health decisions are challenging because of heterogeneity of disease progression and uncertainties about the optimal timing, intensity, and technology for prevention and treatment. Decision analytic methods, such as disease modeling, empiric calibration and cost-effectiveness analysis, can inform decisions about research priorities, clinical guidelines, new technologies, and vaccine policy.
HPV-Related Cervical Cancer: Why Use Models?
For viruses like HPV, the complete disease course cannot be directly observed and clinical trials that compare all strategies in every population are not feasible. Mathematical models can simulate the natural history of disease, extend empiric information by extrapolating outcomes beyond the time horizon of a single study, synthesize data sources in an internally consistent and epidemiologically plausible way, and contextualize strategies for different settings. Read an early paper about decision analytic models. Scan abstracts about modeling cervical cancer prevention and evaluating vaccination and screening.
Country-Specific Analyses of HPV Vaccination
A major applied area of work addresses cervical cancer prevention policy and practice through HPV vaccination of pre-adolescent girls and screening of adult women. Individual-based microsimulation models capture risk stratification by HPV type, the role of HPV persistence and age in disease progression, and detailed prevention and treatment strategies with different epidemiological profiles. Empirical calibration of models to more than a dozen countries allowed for the assessment of both vaccination and screening. For example, read about vaccinating girls in India, in Peru, and about the incremental value of vaccinating boys in Brazil.
Projecting Benefits, Costs, and Value by Region
For countries with limited data, we used population-based models with a simplified set of inputs (e.g., population demographics, cancer risk and mortality, HPV attributable fraction) to project benefits, cost-effectiveness, and affordability over time and with alternative patterns of introduction and scale-up. These analyses informed vaccine financing for Gavi, the Vaccine Alliance, for 72 of the poorest countries (see abstract), as well as 33 Latin American countries (see abstract). Read a paper summarizing these analyses and insights. Learn about more recent analyses assessing a portfolio of vaccines for low- and middle-income countries.
Prototypical Public Health Problem
HPV infection and HPV-associated cervical cancer, represent a prototypical public health problem given the communicable and non-communicable nature of disease, opportunities for intervention along the entire disease spectrum (e.g., primary and secondary prevention, diagnosis, treatment), the varied ages at which interventions are targeted (e.g., adolescence, adulthood), the sharing of a single etiological agent across multiple health conditions, the differential burden of HPV across populations, and the health disparities worldwide. No clinical trial or single longitudinal cohort study would be able to consider all of these factors; in fact, just predicting the population-level impact of cervical cancer prevention is challenging as the time course from infection to disease spans several decades and most data are based on intermediate endpoints. As a result, mathematical models have played a critical role in synthesizing evidence and data, predicting the expected values of competing health decisions or strategies, and exploring the uncertainty that is inherent in every decision.
Looking Under the Hood: Mathematical Models
Our models are biologically-based models which simulate the underlying disease process, while remaining consistent with observed epidemiological data. Evaluating the descriptive epidemiology of complicated diseases for which limited empirical data exist is challenging. In many cases, observational studies provide only partial information, e.g. on cross-sectional prevalence at a given time. Through empiric calibration to multiple data sources and leverage of region-specific data on important cofactors, models can be used to impute unobserved natural history parameters. This process not only allows us to simulate the chronic course of disease in the absence of complete data, it enhances our understanding of the descriptive epidemiology, and often leads to new hypotheses about the mechanism of underlying disease. Read about one of our earliest models, and an improved empirically calibrated model.