Validation of Anaphylaxis Using Machine Learning

The goal of this pilot project is to develop a draft framework to use machine learning and natural language processing (NLP) techniques to improve health outcome of interest (HOI) detection algorithms that may later be used in the larger Sentinel Distributed Database. This project has 4 aims: 

• Aim 1: Conduct medical record chart validation of an adverse event outcome algorithm for anaphylaxis using machine-readable electronic medical records (i.e., not paper charts). 
• Aim 2: Conduct “deep annotation” of charts to identify features consistent with the case definition that is being applied and features that the clinical experts use in their determination that aren’t captured by the case definition.  
• Aim 3: Given the establishment of a “ground truth” of validated anaphylaxis cases using expert medical chart review, use machine learning and NLP techniques to develop risk prediction models that are designed to improve the accuracy with which anaphylaxis can be identified using structured and unstructured electronic data.
• Aim 4: Leverage the work conducted in Aims 1-3 to draft a general methodological framework for developing improved HOI identification algorithms in the Sentinel Distributed Data Network using machine learning and NLP techniques.