Modeling of Infectious Disease Informed by Investigative Focus and Data Characteristics, From the Microscopic to County Community Level

Abstract

Infectious disease modeling is a tool to understand disease characteristics of interest, which is often used to make predictions and guide public health policy. These models rely heavily on observational data, which greatly influence what models can be built and their resulting quality. Some model outputs may not have well developed analyses available, obscuring the general approach to answering specific research questions. In this dissertation, I explore the general process of obtaining, handling, and analyzing data, and apply these concepts to several different infectious disease modeling problems. The problems vary in both the type of disease explored and the scale they inhabit, ranging from molecular scale investigations within the host to population level questions. In chapter 2, I employ host-level modeling to investigate why T cell activation is suppressed in Mycobacterium tuberculosis (Mtb) infections. Working at the within-host level, I modify a model of the immune system response to Mtb using an existing agent-based model (ABM). This methodology capitalizes on how ABMs facilitate capturing emergent behavior, in this case the immune system modulated formation of a granuloma to contain the Mtb infection. Our approach allowed us to view interactions of immune cells during granuloma formation and visualize how these interactions affect the ability of T cells to become activated over the span of an infection. I found recruitment of non-specific T cells and granuloma spatial characteristics contributed to crowding out of the few Mtb specific T cells within the granuloma cell, reducing the chances they could interact with and be activated by infected macrophages.

In chapter 3, I explore how factors in hospital active surveillance for vancomycin resistant Enterococcus (VRE), such as non-compliance, affect our ability to estimate endemic rates within hospitals. Using electronic health records at an individual based level, I simulated patient infections using an ABM approach to establish a baseline for hospital infection. I modeled different compliance rates and testing strategies using cost-effective analysis to judge which type of surveillance strategy is most effective in identifying cases. Our analysis revealed that increasing the compliance rate of screening under any current active surveillance strategy maximized efficacy of identifying VRE cases. In chapter 4, I explore risk factors that affected the COVID-19 case fatality rate (CFR) during the first wave of the pandemic. Having only access to county-level data, this population level study used a case-lag adjust, count-based regression approach to explore the relationship between CFR and other county level indicators such as comorbidity rates, healthcare infrastructure capacity, and non-pharmaceutical interventions. As the new disease was not yet endemic at the time of data collection, the case lag adjustment allowed consideration of the case deaths lag to properly estimate the case fatality rate. This study agreed with previous findings, including relationships between increased asthma occurrences and rates of CFR, and contributed new findings on risk factors, including reduced CFR with bans on religious gatherings. Each infectious disease problem utilized the properties of the data and best-practice methodologies in order to best answer the research question, whether the problems involved exploring disease modeling at a county, institutional, or cell-level scale. These approaches taken to answer scientific questions should be at the forefront of all projects and will be necessary to build large, effective knowledge based from these efforts.