Computational Epidemiology: Methods and Applications for Global Health
Dissertation ZW - 20210504.pdf (35.32Mb)
Access StatusFull text of the requested work is not available in DASH at this time ("dark deposit"). For more information on dark deposits, see our FAQ.
Ward, Zachary J
MetadataShow full item record
CitationWard, Zachary J. 2021. Computational Epidemiology: Methods and Applications for Global Health. Doctoral dissertation, Harvard University Graduate School of Arts and Sciences.
AbstractIn this dissertation I discuss the development and application of simulation-based approaches to major challenges in global health. I refer to this general methodological approach as “computational epidemiology”, which I broadly define as the use of advanced computing capabilities to understand and solve complex epidemiologic problems.
Chapter 1 examines the obesity epidemic in the United States, and describes the development of new modelling methods to analyze individual-level and population-level trends in body-mass index (BMI). These analyses find that on current trends, obesity will continue to rise to high levels in the US, with nearly 60% of today’s children projected to develop obesity by the age of 35, and severe obesity projected to become the most common category of BMI for adults in many states and demographic groups by 2030. These studies provide projections of the future of the obesity epidemic in the US, highlighting both the magnitude of the problem and substantial disparities by race/ethnicity, sex, and income.
Chapter 2 presents analyses of global childhood cancer incidence and survival, modelling the care cascade of health systems from incident cancer through to five-year survival outcomes. To address the fact that many countries do not have cancer registries, and those that do often cover only a small proportion of the population, I developed a model to simulate the total incidence and survival of 48 childhood cancer diagnoses in 200 countries/territories. We find that current estimates of childhood cancer incidence based on registry data suffer from large underreporting bias, with an estimated 43% of childhood cancer cases going undiagnosed globally, ranging from 3% in western Europe and North America to 57% in western Africa. I also estimate childhood cancer survival among all diagnosed cases, accounting for availability of treatment modalities, treatment abandonment, and quality of care. Global five-year net childhood cancer survival is estimated to be 37%, with large variation by region, ranging from 8% in eastern Africa to 83% in North America. Evaluating the comparative effectiveness of various policy interventions, we find that although expanding access to treatment (chemotherapy, radiation, and surgery) and addressing financial toxicity are essential, investments to improve quality of care, at both the health-system and facility-level, are needed to improve childhood cancer outcomes globally.
Chapter 3 presents the development and calibration of a microsimulation model of global maternal health. The model simulates the reproductive lifecycles of individual women in 200 countries/territories, using Bayesian hierarchical models to synthesize the best available data on demographic, epidemiologic, clinical, and health system factors, including underreporting of maternal deaths. I discuss the benefits of using structural models over aggregate regression-based approaches, and compare estimated maternal deaths and maternal mortality ratios to models by the Global Burden of Disease and the United Nations. I find that these models may underestimate the magnitude and the uncertainty of total levels of maternal mortality. I also find that although maternal mortality is improving, on current trends the projected declines will not be sufficient to achieve the Sustainable Development Goal of a global maternal mortality ratio (MMR) below 70 by 2030, with many countries projected to have an MMR above 140 in 2030, mostly in sub-Saharan Africa and South Asia.
Chapter 4 describes the main features of Amua, a new open source graphical modeling framework and probabilistic programming language, intended to provide a user-friendly platform to help expand decision science computational methods to a wider audience of students and researchers. Models can be run in Amua or exported to other programming languages (C++, Java, Python, and R), allowing conceptual models to be separated from any particular implementation. As a probabilistic programming language, Amua allows model parameters to be treated as random variables and fit to data using various model calibration techniques.
Citable link to this pagehttps://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37368237
- FAS Theses and Dissertations