BackgroundOutpatient antibiotic misuse is common, yet it is difficult to identify and prevent. Novel methods are needed to better identify unnecessary antibiotic use in the outpatient setting.MethodsThe Twitter developer platform was accessed to identify Tweets describing outpatient antibiotic use in the United States between November 2018 and March 2019. Unique English-language Tweets reporting recent antibiotic use were aggregated, reviewed, and labeled as describing possible misuse or not describing misuse. Possible misuse was defined as antibiotic use for a diagnosis or symptoms for which antibiotics are not indicated based on national guidelines, or the use of antibiotics without evaluation by a healthcare provider (Figure 1). Tweets were randomly divided into training and testing sets consisting of 80% and 20% of the data, respectively. Training set Tweets were preprocessed via a natural language processing pipeline, converted into numerical vectors, and used to generate a logistic regression algorithm to predict misuse in the testing set. Analyses were performed in Python using the scikit-learn and nltk libraries.Results4000 Tweets were included, of which 1028 were labeled as describing possible outpatient antibiotic misuse. The algorithm correctly identified Tweets describing possible antibiotic misuse in the testing set with specificity = 94%, sensitivity = 55%, PPV = 75%, NPV = 87%, and area under the ROC curve = 0.91 (Figure 2).ConclusionA machine learning algorithm using Twitter data identified episodes of self-reported antibiotic misuse with good test performance, as defined by the area under the ROC curve. Analysis of Twitter data captured some episodes of antibiotic misuses, such as the use of non-prescribed antibiotics, that are not easily identified by other methods. This approach could be used to generate novel insights into the causes and extent of antibiotic misuse in the United States, and to monitor antibiotic misuse in real time. Disclosures All authors: No reported disclosures.