Sepsis Subphenotyping Based on Organ Dysfunction Trajectory

Abstract Purpose Although organ dysfunction is a defining element of sepsis, its trajectory is not well studied. We sought to identify whether there are distinct Sequential Organ Failure Assessment (SOFA) score trajectory-based subphenotypes in sepsis. Methods We created 72-hour SOFA score trajectories in patients with sepsis from two diverse intensive care unit (ICU) cohorts. We then used Dynamic Time Warping (DTW) to compute patient similarities to capture evolving heterogeneous sequences and establish similarities between groups with distinct trajectories. Hierarchical agglomerative clustering (HAC) was utilized to identify subphenotypes based on SOFA trajectory similarities. Patient characteristics were compared between subphenotypes and a random forest model was developed to predict subphenotype membership, within 6 hours of ICU arrival. The model was then tested on the validation cohort. Results A total of 4,678 and 3,665 unique sepsis patients were included in development and validation cohorts. In the development cohort, four subphenotypes of organ dysfunction were identified: Rapidly Worsening (n=612, 13.08%), Delayed Worsening (n=960, 20.52%), Rapidly Improving (n=1,932, 41.3%) and Delayed Improving (n=1174, 25.1%). In-hospital mortality for patients within different subphenotypes demonstrated distinct patterns over time. Similar subphenotypes and their associated outcome patterns were replicated in the multicenter validation cohort. Conclusion Four novel, clinically-defined, trajectory-based sepsis subphenotypes were identified and validated. Trajectory based subphenotyping is useful for describing the natural history of sepsis in the ICU. Understanding the pathophysiology of these differential trajectories may reveal unanticipated therapeutic targets for patients with sepsis and identify more precise populations and endpoints for the predictive enrichment of clinical trials.