Pharmacogenomics · Claims analytics · Explainable ML
R. Jerome Dixon
PhD student in Integrative Life Sciences at Virginia Commonwealth University and part-time data scientist at UVA’s School of Data Science. I build partition-first pipelines and consensus-based attribution for opioid and polypharmacy risk in large-scale claims data.
About
I build decision-ready analytics and production ML systems where the stakes are clinical or operational — not experimental. My work sits at the intersection of personalized medicine and supply chain analytics, two domains where the wrong prediction has real consequences and where rigorous, explainable models aren’t optional.
At CANA LLC, I’ve spent 9+ years as a Senior Operations Research designing end-to-end analytics solutions for logistics and supply chain clients. That means translating complex operational data into actionable insights through advanced analytics, data visualization, and cloud data engineering on AWS infrastructure.
My PhD research at Virginia Commonwealth University takes the same rigor into clinical AI. I’m building adverse drug event risk models from Virginia’s All Payers Claims Database — applying XGBoost, CatBoost, survival analysis, SHAP, and Formal Feature Attribution (FFA) to map drug interaction networks and quantify individualized risk.
At the UVA School of Data Science (2022–2026), I conducted ML research on pediatric heart transplant outcomes — one of the highest-stakes prediction problems in clinical medicine. At SurgicalEd VR, I design within a serverless architecture construct — building event-driven AWS data lakes, automated ETL pipelines, and predictive models that evaluate surgical skill in VR environments without the overhead of persistent infrastructure. My stack spans Python, R, SQL, AWS (Lambda, S3, EMR, DynamoDB, EC2), XGBoost, CatBoost, SHAP, FFA, and serverless streaming architectures.
The throughline across supply chain, surgical performance, and clinical pharmacology is the same: building systems that produce decisions you can defend — to a regulator, a clinician, or an operations team. That’s the standard I hold my work to, and it shapes everything from model architecture to how I communicate and visualize results.
Live projects
PGx Risk Calculator
VCU APCD opioid & polypharmacy ED risk — ensemble models, Consensus Filter, serverless dashboard.
PHTS Risk Calculator
UVA pediatric heart transplant graft-loss risk models with SHAP/FFA scenario tools.
Research artifacts
Network visualizations and association-rule exports by age band and year.
Publications
Dissertation manuscripts in the journal pipeline for Clinical and Translational Science and npj Digital Medicine (with Elvin T. Price). Status as of July 2026; DOIs will be added when available.
-
Accepted
Temporal Drivers of Opioid-Related ED Visits: An Ensemble Machine Learning Study with Consensus-Based Feature Attribution
-
Accepted
A Serverless Pharmacogenomic Risk Dashboard: Translating Ensemble Models and Model-Based Scenario Rules to Clinical Decision Support
-
Approved MS received
A SHAP-Informed Formal Feature Attribution Framework for Drug–Drug Interaction Risk in Large-Scale Claims Data
-
Under review
Bridging Explainable Artificial Intelligence and Pharmacogenomics for Opioid and Polypharmacy Risk Prediction: A Systematic Quantitative Literature Review
-
Under consideration
Low cost big data framework for scalable claims analytics
