318 Optimizing the Identification and Prediction of Statin Intolerance to Improve Statin Adherence Using Natural Language Processing and Machine Learning

Abstract

OBJECTIVES/GOALS: Aim1: To develop a natural language processing (NLP) algorithm to effectively identify statin associated muscle symptoms (SAMS) in patients’electronic health records (EHRs). Aim2: To develop a machine learning model based on clinical features within EHRs that predict the likelihood of SAMS occurrences. METHODS/STUDY POPULATION: A retrospective cohort of adult patients initiated on statins within the Minnesota Fairview Healthcare System EHRs from 2010 to 2020 will be analyzed. NLP-PIER (Patient Information Extraction for Research) platform will be used to search and identify patients who developed SAMS after statin initiation. Manual annotation of clinical notes will be completed to validate the accuracy of identified SAMS cases. Then, a selection of clinical features within the EHRs will be input as predictors for machine learning algorithms development. Select machine learning classifiers will be deployed to generate models for the prediction of SAMS and the best-performing model will be selected based on model performance. RESULTS/ANTICIPATED RESULTS: The expected outcomes include generation of a fine-tuned NLP algorithm that can rigorously identify SAMS occurrences within EHRs. Further, we anticipate having a practical risk model that accurately predicts patients’risks of developing SAMS when taking statins. DISCUSSION/SIGNIFICANCE: The positive and translational impact of our research will be to equip healthcare providers with such informatics tools to improve statin adherence, ultimately promoting patient optimal health and outcomes by maximizing the tolerance and thus realizing the therapeutic benefits of statins.