Sex-stratified statin toxicity prediction
A model that estimates the risk of adverse statin reactions in female patients, a group that cardiovascular modeling often under-represents. It started at In Transcription @ UCLA's 2026 BioHackathon.
The problem
Statins are prescribed widely, but adverse reactions don't fall evenly across patients. The goal here was a sex-stratified view of risk, looking specifically at outcomes like myopathy, glucose intolerance, and liver toxicity in female patients.
What I built
The model is a stacked ensemble: XGBoost, LightGBM, a random forest, and logistic regression as base learners, combined through a meta-learner. It trains on federal adverse-event reports (FAERS) from 2020 through 2024, filtered to female patients on statins, with imputation and encoding steps to handle the messy parts of that data.
I wrapped the model in a FastAPI backend and a React interface so the predictions can be explored rather than read out of a notebook.
This is a research prototype built on reported adverse-event data. It is not a clinical recommendation tool and shouldn't be used to make treatment decisions.