Phenotyping COVID-19 Patients by Ventilation Therapy: Data Quality Challenges and Cohort Characterization

Abstract

The COVID-19 pandemic introduced unique challenges for treating acute respiratory failure patients and highlighted the need for reliable phenotyping of patients using retrospective electronic health record data. In this study, we applied a rule-based phenotyping algorithm to classify COVID-19 patients requiring ventilatory support. We analyzed patient outcomes of the different phenotypes based on type and sequence of ventilation therapy. Invasive mechanical ventilation, noninvasive positive pressure ventilation, and high flow nasal insufflation were three therapies used to phenotype patients leading to a total of seven subgroups; patients treated with a single therapy (3), patients treated with either form of noninvasive ventilation and subsequently requiring intubation (2), and patients initially intubated and then weaned onto a noninvasive therapy (2). In addition to summary statistics for each phenotype, we highlight data quality challenges and importance of mapping to standard terminologies. This work illustrates potential impact of accurate phenotyping on patient-level and system-level outcomes including appropriate resource allocation under resource constrained circumstances.