Research

Updated 12 Sep, 2024

Helping patients and surgical excellence are the foundations of my professional mission. My research is broadly focused on applying emerging technologies to solve clinical problems.

I am currently exploring novel methods that leverage computation in order to anticipate and mitigate surgical risk. Our ability to do so is severely inadequate, owing to several factors:

• Evidence-based medicine is reliant upon the literature, despite consensus that most published research findings are false, flawed, biased, or irrelevant to the specific patient population in question.

  • The modern era of generative AI has exacerbated the fact that more than 5,000 biomedical papers are published every day, producing a volume of research that surpasses any clinician's ability for review.

• Most clinicians are unable to mitigate their own implicit biases, much less additionally remain aware of and mitigate the biases already present in the literature.

• There is a general lack of insistence on necessary and sufficient evidence when discussing surgical complications.

• The Bitter Lesson: methods that leverage computation are ultimately the most effective, but most researchers seek to leverage their human knowledge of the domain to achieve incremental progress by building upon the literature. Most available surgical risk prediction models attempt to assess individualized risk with statistical methods, but rely on sanitized, structured EHR data. This data is low validity (eg: ICD codes are often inaccurate) and doesn't include important features that inform actual surgical decision making (eg: what do the last operative note and discharge summary say? what do the CT and echocardiogram look like?).

Integrated multimodal AI systems demonstrate emergent capabilities exceeding those of ensemble systems. The development of surgical risk prediction systems that are 1) optimized for auditability, 2) capable of direct interpretation of both structured and unstructured health information, and 3) able to interface directly with the surgeon (and later, the patient) via audio, visual, and text media — may dramatically improve our ability to assess and mitigate surgical risk.

I was previously the Clinical Research Fellow at the Penn Center for Human Appearance, where my research included methods to augment surgical risk modeling with unstructured data, innovations in surgical technique for the management of ventral hernia, and applications of 3-D computer vision technologies for pre- and post-operative care. See selected publications.

During medical school, I participated in research at the Eskandar Lab at MGH. Activities included the use of rapid-sampling EEG to investigate neural activity during decision making and overhauling the design of the OpBox, an open-source operant conditioning and experimental environment for small rodents in use at 5+ MGH labs and internationally; I prototyped its design and built 12 functional units, reducing per-unit cost by 35% to <10% cost as compared to commercial solutions. I additionally developed diagnostic software to debug EEG signal anomalies and performed EEG signal analysis.