Abstract Background Characterizing, diagnosing, and caring for “long COVID” patients has proven to be challenging due to heterogenous symptoms and broad definitions of these post-acute sequelae. Here, we take a machine learning approach to identify discrete clusters of long COVID symptoms which may define specific long COVID phenotypes. Figure 1: (A) Principal component analysis followed by K-means clustering identified three groups of participants. (B) Heatmap depicting three distinct clusters (high values are in red and low value are in blue); Cluster 1 exhibits sensory symptoms (e.g., loss of smell and/or taste), Cluster 2 exhibits fatigue and difficulty thinking (e.g., changes in ability to think) symptoms, and Cluster 3 exhibits difficulty breathing and exercise intolerance symptoms. (C) Clinical and demographic characteristics of 97 military health system beneficiaries by identified clusters Methods The Epidemiology, Immunology, and Clinical Characteristics of Emerging Infectious Diseases with Pandemic Potential (EPICC) study is a longitudinal COVID-19 cohort study with data and biospecimens collected from 10 military treatment facilities and online recruitment. Demographic and clinical characteristics were collected using case report forms and surveys completed at enrollment and at 1, 3, 6, 9, and 12 months. For this analysis, we identified those who reported any moderate to severe persistent symptoms on surveys collected 6-months post-COVID-19 symptom onset. Using the survey responses, we applied principal component analysis (PCA) followed by unsupervised machine learning clustering algorithm K-means to identify groups with distinct clusters of symptoms. Results Of 1299 subjects with 6-month survey responses, 97 (7.47%) reported moderate to severe persistent symptoms. Among these subjects, three clusters were identified using PCA (Figure 1A). Cluster 1 is characterized by sensory symptoms (loss of taste and/or smell), Cluster 2 by fatigue and difficulty thinking, and Cluster 3 by difficulty breathing and exercise intolerance (Figure 1B). More than half of these subjects (57%) were female, 64% were 18-44 years old, and 64% had no comorbidities at enrollment (Figure 1C). Those in the sensory symptom cluster were all outpatients at the time of initial COVID-19 presentation (p < 0.01). The difficulty breathing and exercise intolerance symptom-clusters had a higher proportion of older participants (Age group ≥ 45-64) with more comorbidities (CCI ≥ 1-2). Conclusion We identified three distinct ‘long COVID’ phenotypes among those with moderate to severe COVID-19 symptoms at 6-months post-symptom onset. With further validation and characterization, this framework may allow more precise classification of long COVID cases, and potentially improve the diagnosis, prognosis, and treatment of post- infectious sequelae. Disclosures Ryan C. Maves, MD, AiCuris: Grant/Research Support|Sound Pharmaceuticals: Grant/Research Support|Trauma Insights, LLC: Advisor/Consultant Julia S. Rozman, n/a, Astra Zeneca: The HJF, in support of the USU IDCRP, was funded to conduct or augment unrelated Phase III Mab and vaccine trials as part of US Govt. COVID19 response Mark P. Simons, PhD, AstraZeneca: The HJF, in support of the USU IDCRP, was funded to conduct or augment unrelated Phase III Mab and vaccine trials as part of US Govt. COVID19 response David R. Tribble, DrPH, AstraZeneca: The HJF, in support of the USU IDCRP, was funded to conduct or augment unrelated Phase III Mab and vaccine trials as part of US Govt. COVID19 response Timothy Burgess, MD, MPH, AstraZeneca: The HJF, in support of the USU IDCRP, was funded to conduct or augment unrelated Phase III Mab and vaccine trials as part of US Govt. COVID19 response Simon Pollett, MBBS, Astra Zeneca: The HJF, in support of the USU IDCRP, was funded to conduct or augment unrelated Phase III Mab and vaccine trials as part of US Govt. COVID19 response.