Dissertation Defense: Elizabeth Lee
Candidate Name: Elizabeth Lee
Major: Infectious Diseases
Advisor: Shweta Bansal, Ph.D.
Title: Epidemiological Inference and Surveillance with High Resolution Medical Claims Data: The Case of Influenza
Disease surveillance, the activity of monitoring disease burden and its associated risk factors, is critical for situational awareness, disease control, and public health coordination and planning. Due to the opportunistic recruitment of sentinel health care providers for routine, passive surveillance, it is challenging to make informed decisions about the best uses of available surveillance data and the design of future surveillance system data collection. Medical billing claims are an alternative potential surveillance source with features advantageous to public health monitoring, such as near real-time reporting, high population coverage, fine spatial resolution, and symptom diagnoses from health care providers.
In this dissertation, I approach these challenges by answering the following questions: 1) Can medical claims compare to and expand upon the information gathered from traditional surveillance systems? 2) Can medical claims be leveraged to improve the design of surveillance system data collection? and 3) Can medical claims be used to understand biases among existing surveillance system data? I investigate these questions by examining a high resolution medical claims dataset for influenza-like illness in the United States and assessing the value of these data with statistical tools.
First, I develop two indexes of population-level influenza intensity that enable a comparison of medical claims relative to traditional systems. While both indexes accurately capture end-of-season influenza disease burden, the claims-based index has good performance for early warning and relies on a sole data stream for information, thus enhancing its applicability in operational contexts. Next, I mimic multiple sentinel surveillance system designs with statistical models of the medical claims, and find that across-year consistency in sentinel locations and the data reporting-related factors improve system performance. Finally, I explore how the spatial aggregation of surveillance data affects epidemiological inference. I find that errors are most prevalent early in the flu season, and environmental factors may help to infer finer scale information from aggregated data with accuracy.
The results of this dissertation offer a framework to optimize the use and data collection design of surveillance systems, while also demonstrating the potential for medical claims to contribute new epidemiological insights.
Friday, December 8, 2017 at 1:00pm to 3:00pm
Reiss Science Building, 112
37th and O St., N.W., Washington