Development and Evaluation of Electronic Health Record (EHR) Information Extraction Pipeline and Tree-Based Scan Statistic (TBSS) Methods for EHR-Based Signal Detection (DA2)

Tree-based scan statistics (TBSS) are data mining methods used to prioritize statistical alerts when screening for potential adverse events. TBSS uses a hierarchical tree to group and map relationships between thousands of correlated outcomes and account for multiple hypotheses being evaluated. Historically, TBSS has used insurance claims data - captured with International Classification of Diseases, Ninth Edition, Clinical Modification (ICD-9-CM) and (ICD-10-CM) codes - to screen for potential adverse events for associations with a drug of interest. While a valuable drug safety surveillance method for insurance claims data, TBSS methods, to date, have not been developed to leverage the rich clinical information, including laboratory test results, vital signs, and clinical notes, available in electronic health records (EHR). This project will develop approaches for abstracting and combining structured and unstructured EHR data. This project will also expand TBSS methods to allow identification of statistical alerts for outcomes available only through EHR data (e.g., natural language processing, laboratory values). Led by partner teams at Vanderbilt University Medical Center (VUMC) and Brigham and Women’s Hospital (BWH), this project will ultimately inform the Sentinel Innovation Center’s (IC) detection analytics roadmap for the integration and development of methods of signal identification from EHR data.