Research Interests

My work is primarily in two domains: genetics and digital phenotyping.

My genetics side is focused on developing methodology for testing related sets of single nucleotide polymorphism (SNP-sets, or genes) in genetic association studies. My doctoral thesis generalized the higher criticism test to fit into this context by allowing for correlation (SNPs in a SNP-set are frequently in linkage disequilibrium) and by not relying on asymptotics (SNP-sets can sometimes contain few SNPs).

Digital phenotyping is based on the fundamental idea that a person's phone can inform their behavior. Smartphones offer an array of sensors such as GPS, accelerometers, WiFi, bluetooth, etc.. which can be used to see how people sleep, move, interact with one another, and function in day-to-day life at an incredible level of detail. I am currently a postdoc in the Onnela lab in the Biostatistics department at the Harvard School of Public Health, and our lab has developed a smartphone application, Beiwe, which can be used to collect both surveys designed by researchers as well as passive data. My research is focused on the development of software that practitoners can use to make sense of the abundance of data that Beiwe produces.

A few example collaborations I am involved in are:
1) Post-surgery patient progress (collaborator: Timothy Smith, MD, PhD, MPH)
2) Stress/anxiety (collaborator: Randy Buckner, PhD)
3) Schizophrenia (collaborator: John Torous, MD)

While Beiwe surveys are low-stress for the user, my primary interest lies in making behavioral and outcome inferences based on the passive data collection. The idea that we would be able to predict the onset of a PTSD episode, or be able to detect a problem during the recovery of a patient after back surgery by simply having the patient install an app that runs non-invasively in the background, is absolutely fascinating to me. Compared to many antiquated practices and standards of care (can't we do better than simply asking a patient if they are feeling better, the same, or worse after a procedure?), smartphones have the capability of both improving the quality of these health outcomes while also reducing the burdens (such as travel to hospital for the followup visits) of the patient.

In this new area of digital phenotyping, while the data is bountiful and full of possibilities it also is noisy and frequently has missingness. For example, GPS cannot constantly be active or else a phone will lose battery too quickly. When GPS is inactive, this translates to a missing data problem. This is just one example of the types of problems that require statistics, and so this area is where I have focused my efforts.