I am an advanced fellow in health services research, working with the Nudge Unit and Prof. Mitesh Patel at the University of Pennsylvania and the Center for Health Equity Research and Promotion (CHERP) at the Philadelphia VA Medical center. I completed my doctoral work in the NIH Oxford Cambridge Scholars Program, where I was advised by Dr. Laura Koehly at the Social and Behavioral Research Branch (SBRB) of the National Human Genome Research Institute (NHGRI), Prof. Felix Reed-Tsochas at the Said Business School at Oxford University, and Dr. Chris Marcum, also at the SBRB. You can view my dissertation here.
My research currently focuses on leveraging social data to understand and improve health. One avenue I am particularly involved in is how we can glean socail data from Electronic Medical Records (EMRs) and Hospital Administrative Data (HAD): databases very much not designed to capture social information. Using this approach, I have shown that who one is in the same hospital ward as (which I term "co-presence") can predict cancer mortality and infection. I have also applied this approach to secondary analyses of behavioral interventions and simulation studies. Through all of these studies I have shown that who we are around is important to our health, and health interventions should both be aware of this and leverage it if possible.
In addition to empirical work, I have developed a number of methodological improvements when I have found the current methodology limiting. This has made my methodological contributions directly linked to my ongoing empirical work, ensureing immediate implementation of the given methods. For instance, in order to detect the effect of co-presence on chemotherapy patients, I had to control for hospital standard operating procedure. Due to the complexity of the healthcare system, I developed a method to determine when two patients were in a ward more than expected by chance, net of underlying scheduling protocol at the hospital. This was immediately put into use in my empirical work. As another example, I developed an algorithm to efficiently caluclate the colored triad census with the immediate aim of using the statistics generated by the algorithm to understand infection networks and identify asymptomatic infections (work ongoing). In approaching methodological contributions this way, I can immeduately show a use case that will increase uptake of a given algorithm.
I have recently begun incorporating behavioral phenotyping into my work as a way to identify subsets of patients most at-risk of certain outcomes, or who would benefit most from certain interventions. As part of the former, I used co-presence networks built from electronic medical records along with biomarker data to infer a class of patients with nosocomial infection below the threshold of microbiological tests. These inferred infections are "subclinical infections" (Chapter 4), and may have many adverse effects on both individuals and the disease dynamics of nosocomial infection spread. I am also working on a project analyzing the results of the Lose It trial to examine latent groups in the participants, and determine if they responded differentially to the trial. Future work will include using this type of latent class analysis to optimally allocate participants to different arms of a trial.