I'm a PhD candidate in Epidemiology at Vanderbilt, studying how to construct valid inference from data. My interests are medication safety and large-language-model capability. Currently, I'm funded by NASA to study the association between low-dose radiation exposure and Parkinson's disease to prepare for a potential Mars mission.
Which drug-safety signals in large healthcare databases are real, and which are artifacts of study design. Recent work: the depression warning on 5-alpha reductase inhibitors, and their metabolic effects across national databases.
A system that reasons from a patient's symptoms toward a ranked rare-disease diagnosis — disease and causal gene — the way a geneticist would, judged against curated ground truth and blinded physician adjudication.
// manuscript in preparationThe method that connects the two: treat a benchmark like an experiment — the model is a population, the prompt a treatment you assign, the score an estimator with error, not a fact. Is it powered? Representative? Measured against a reliable standard?
Four independent lines of evidence — meta-analysis, Mendelian randomisation, and two pharmacovigilance databases — converge on the same conclusion: the finasteride-depression signal is an artefact of study design and stimulated reporting, not pharmacology. The warning itself may be causing more harm than the risk it describes.
15 min readA deep dive into my meta-analysis on 5-alpha reductase inhibitors and depression risk. Why some studies overestimated the risk by 200% and what this means for patients taking finasteride or dutasteride.
12 min readNew evidence from South Korea's national health database sheds light on whether finasteride and dutasteride affect diabetes risk. Using robust active-comparator methods, we found minimal to no increased risk.
10 min readDoes finasteride affect blood sugar control in men with type 2 diabetes? Using New Zealand's national pharmaceutical data, we compared long-term outcomes between finasteride and tamsulosin users.
8 min read