Unveiling Sub-Populations in Critical Care Settings: A Real-World Data Approach in COVID-19

Wesley Anderson^1*Ruth Gould²Namrata Patil³Nicholas Mohr⁴Kenneth Dodd⁵Danielle Boyce^6,7Pam Dasher¹Philippe Guerin⁸Reham Khan⁹Sreekanth Cheruku¹⁰Vishakha Kumar⁹Ewy Mathe¹¹Aneesh Mehta¹²Andrew Michelson¹³Andrew Williams⁶Smith F Heavner^1,14,15Jagdeep Podichetty¹

• ¹Critical Path Institute, Tucson, United States
• ²Centers for Disease Control and Prevention (Georgia), Atlanta, Georgia
• ³Brigham and Women's Hospital, Harvard Medical School, Boston, Massachusetts, United States
• ⁴The University of Iowa, Iowa City, Iowa, United States
• ⁵Advocate Aurora Health, Milwaukee, Wisconsin, United States
• ⁶Tufts University, Medford, Massachusetts, United States
• ⁷School of Medicine, Johns Hopkins University, Baltimore, Maryland, United States
• ⁸Infectious Diseases Data Observatory, University of Oxford, Oxford, England, United Kingdom
• ⁹Society of Critical Care Medicine, Mount Prospect, Illinois, United States
• ¹⁰University of Texas Southwestern Medical Center, Dallas, Texas, United States
• ¹¹National Center for Advancing Translational Sciences (NIH), Bethesda, Maryland, United States
• ¹²Emory University, Atlanta, Georgia, United States
• ¹³School of Medicine, Washington University in St. Louis, St. Louis, Missouri, United States
• ¹⁴Department of Public Health Sciences, College of Behavioral, Social and Health Sciences, Clemson University, Clemson, South Carolina, United States
• ¹⁵Department of Biomedical Sciences, School of Medicine Greenville, University of South Carolina, Greenville, South Carolina, United States

Background: Disease presentation and progression can vary greatly in heterogeneous diseases, such as COVID-19, with variability in patient outcomes, even within the hospital setting. This variability underscores the need for tailored treatment approaches based on distinct clinical subgroups.Objectives: This study aimed to identify COVID-19 patient subgroups with unique clinical characteristics using real-world data (RWD) from electronic health records (EHRs) to inform individualized treatment plans.Materials and Methods: A Factor Analysis of Mixed Data (FAMD)-based agglomerative hierarchical clustering approach was employed to analyze the real-world data, enabling the identification of distinct patient subgroups. Statistical tests evaluated cluster differences, and machine learning models classified the identified subgroups.Results: Three clusters of COVID-19 inpatients with unique clinical characteristics were identified. The analysis revealed significant differences in hospital stay durations and survival rates among the clusters, with more severe clinical features correlating with worse prognoses and machine learning classifiers achieving high accuracy in subgroup identification.Conclusions: By leveraging RWD and advanced clustering techniques, the study provides insights into the heterogeneity of COVID-19 presentations. The findings support the development of classification models that can inform more individualized and effective treatment plans, improving patient outcomes in the future.