Identification of Precision Sepsis Subphenotypes Using Vital Sign Trajectories

Project Abstract The scientific goal of this K23 is to apply cutting-edge data science approaches to identify novel subphenotypes within the heterogeneous sepsis syndrome. This K23 application proposes a 5-year training program to propel Dr. Sivasubramanium Bhavani towards his career as an independent physician-scientist. Dr. Bhavani’s career

goal is to be an expert in developing computer-aided diagnostic tools to map the extensive clinical and biological data in the electronic health record (EHR) to personalized treatment plans for critically ill patients. Dr. Bhavani will accomplish this career goal by completing 3 short-term goals: 1) Gain expertise in unsupervised machine

learning, 2) Gain expertise in deep learning neural networks, and 3) Gain expertise in clinical informatics principles for model application to real-world data. Dr. Bhavani has outlined an integrated program of didactics, seminars, conferences, and consistent communication with expert mentors to provide the necessary career

development. Dr. Bhavani’s mentors are Dr. Craig Coopersmith, a past president of the Society of Critical Care Medicine with a long career of NIH-funded sepsis research, and Dr. May Wang, a renowned expert in machine learning. In addition, Dr. Bhavani’s advisors are Drs. John Hanfelt, Annette Esper, Matthew Semler and Matthew

Churpek, with collective expertise in longitudinal clustering, sepsis biomarkers, and bioinformatics. With the support of the K23, Dr. Bhavani will contribute to the development of precision medicine approaches to sepsis. Sepsis is a severe and heterogeneous syndrome characterized by a dysregulated host response to infection that

results in over 270,000 deaths in the U.S. annually. Decades of clinical trials have failed to identify therapies that consistently benefit patients with sepsis. The one-size-fits-all treatment approach has not worked, and there is a need to identify sepsis subphenotypes that may have different responses to treatment. To date, most studies

have identified sepsis subphenotypes using static measurements of labs and vital signs. However, sepsis is a dynamic process with biological and physiological responses that evolve over minutes to hours. The objective of this proposal is to identify novel sepsis subphenotypes using dynamic data, specifically longitudinal vital signs.

In Aim 1, Dr. Bhavani will apply cutting-edge machine learning algorithms to longitudinal vital signs to develop and validate novel vitals trajectory subphenotypes. In Aim 2, Dr. Bhavani will investigate the immune signatures of these subphenotypes. In Aim 3, Dr. Bhavani will study the responses of the subphenotypes to one of the most

common interventions in sepsis – intravenous fluids. Identification of subphenotypes with responses to different fluids could shift sepsis management from a one-size-fits-all approach to a precision medicine approach – the ultimate objective of sepsis subphenotypes. Through the training in this K23, Dr. Bhavani will be prepared for

R01-level work in a) refining subphenotypes by combining dynamic clinical and immunological data and b) studying the responses of subphenotypes to additional treatments by using data from other RCTs.