Electronic Health Records + Natural Language Processing + Machine Learning = Improved Sentinel Outcome Detection Algorithms

This presentation describes the work of a multi-site team to improve automated algorithms for identifying health outcomes of interest to the FDA Sentinel Initiative by applying data-driven machine learning and natural language processing (NLP) methods to rich information from electronic health records. We will describe the motivation, approach, results and implications of this work developing and validating an algorithm to identify anaphylaxis, a rare but potentially life-threatening adverse event associated with medical products use that cannot be accurately measured using traditional, administrative claims data alone.

Target Audience:
Medical informaticists, medical product safety and real-world evidence researchers and regulators