Project Abstract Patients with cancer face unacceptable morbidity and mortality from sepsis, a life-threatening dysregulated response to infection. Oncologic sepsis contributes to > 15% of cancer hospitalizations and 10% of cancer deaths in the US, with far greater morbidity and mortality than noncancer sepsis. Timely evidence-based sepsis care bundles improve outcomes, but are frequently initiated too late or not at all in cancer, suggesting that earlier accurate recognition may improve care and outcomes. Current approaches to detecting and treating oncologic sepsis suffer from interrelated limitations including poor accuracy of sepsis prediction tools in patients with cancer as well as general and oncology-specific barriers to their effective implementation. This Career Development Award will support Patrick G Lyons, MD, MS, in addressing this challenge while completing his development into an independent physician-investigator with the training and experience necessary to improve cancer care delivery in the hospital. The overall goal of Sepsis Prediction in Oncology Through Implementation Science and Technology (SPOT-IT) is to use EHR data to develop an oncology- specific sepsis prediction model using machine learning and to use human centered design methods to design and evaluate the usability of a stakeholder-informed implementation strategy for this model. These themes fit with the NCI’s goal of “rapid development, testing, and refinement of innovative approaches to implement... evidence-based cancer control interventions” and the DCCPS’s priority areas in healthcare delivery research and implementation science and are reflected in the Aims: 1) develop an oncology-specific sepsis prediction model using machine learning on EHR data; 2) design and refine implementation strategies to improve oncologic sepsis management; and 3) conduct a pilot trial to determine the early implementation and process outcomes of SPOT-IT. These Aims link to Dr. Lyons’s career development objectives: 1) develop core cancer care delivery knowledge, (2) enrich his knowledge in sepsis epidemiology and outcomes, (3) advance his skills in machine learning and informatics, (4) gain advanced skills in implementation science and human centered design, and (5) enhance his scientific leadership skills. Dr. Lyons will achieve these goals via a 5-year career development plan incorporating didactics, fieldwork and experiential research, and intensive mentoring by Terri Hough, MD (an international leader in sepsis epidemiology and pragmatic implementation research), Brandon Hayes-Lattin, MD (an oncologist specializing in stem-cell transplantation and clinical trials), and Matthew Churpek, MD, PhD (a critical care physician and informaticist with expertise in machine learning using EHR data). Dr. Lyons’s experienced multidisciplinary team of mentors and advisors, combined with the exceptional research environment at Oregon Health & Science University, will provide the support and training necessary to achieve his long-term goal of becoming a leading cancer care delivery scientist.

Project Narrative Because hospitalized patients with cancer face unacceptable morbidity and mortality from sepsis (a life- threatening dysregulated response to infection), there is urgent need for interventions to identify oncologic sepsis early and enable timelier appropriate treatments. However, incumbent electronic health record-based sepsis prediction tools – linked to improved mortality in other populations – have failed to deliver on their promise in oncology because of design and implementation limitations. This project aims to advance the simultaneous and iterative development of a machine learning tool to predict oncologic sepsis alongside the user-centered design of implementation strategies to improve sepsis care and outcomes for hospitalized patients with cancer.