Work
Projects on drug-safety data. Each one ends in a finding and a decision it should change.
Every project leads with what surprised me. The method and the code are one click further in.
80→54% 01 · PV Intake Assistant A change to my own system raised its coverage from 52% to 80%. I threw it away. Python · NLP · SNOMED CT · Docker Read 46% 02 · Causal safety Anti-inflammatory painkillers looked barely riskier for your kidneys than paracetamol. They were twice as risky. Python · Causal inference · SCCS Read 0.83 vs 0.71 03 · ADR Reader Four systems read the same medical notes looking for drug side effects. The most advanced one finished last. Python · NLP · PubMedBERT Read HR 2.4 04 · EHR Survival Analysis One risk factor looked like it did nothing at all. It was doing plenty — just not at a constant rate. R · Cox regression Read 20M+ 05 · FAERS Explorer Reports of harm for one heartburn drug exploded in 2020. Nothing about the drug had changed. Python · openFDA · Streamlit Read PRR ≈ 5 06 · FAERS Signal Detection How do you tell a real drug safety signal from a coincidence, in a database of twenty million reports? Python · Disproportionality Read 3 answers 07 · FAERS in SQL The same database, the same question, three ways of collecting the rows — and three different answers about whether drug harm is rising. SQL · Python Read 08 · Synthea EHR I pulled every patient with high blood pressure and asked what they were being treated with. The top answer was a painkiller. Python · SQL · Synthea Read 09 · EHR ML Risk Model A model that flags every patient as high risk catches every case. It is also useless. Python · scikit-learn Read