Data-Driven Automated Classification Algorithms for Acute Health Conditions: Applying PheNorm to Anaphylaxis

Accurate identification of anaphylaxis using observational data is important for medical product safety surveillance, but difficult to diagnose clinically or recognize algorithmically. Traditional phenotyping methods rely on expensive gold standard training data and manual feature engineering. The Sentinel Innovation Center instead applied an automated approach, PheNorm, to create a computable phenotype for identifying patients with anaphylaxis using Natural Language Processing (NLP), machine learning, and low-cost silver-standard training labels. Performance was comparable to a recently published, higher-cost manual phenotyping effort. It was presented at the American Medical Informatics Association (AMIA) 2024 Annual Symposium on November 13, 2024.