Sepsis is defined as a severe infection with dangerous physiologic changes, organ dysfunction or death, which hospitalizes over 1.6 million people in the U.S. annually. Sepsis is a priority for the Center for Medicaid and Medicare Services due to its healthcare impact, incidence and staggering annual cost, which exceed $20 billion and 5% of all U.S. hospital costs. All sepsis patients, even those with mild sepsis are at risk for in- hospital complications and death, but have improved outcomes if identified early. Sepsis recognition, however, is challenging due to the heterogeneity of patients who may manifest a wide array of clinical presentations. Electronic medical record (EMR) linked computer programs, known as clinical decision support (CDS) tools, have become ubiquitous to assist providers, including identifying sepsis patients. Unfortunately, all CDS tools in the literature miss 20-30% of sepsis patients and frequently misidentify non-sepsis patients as sepsis patients. Inaccurate CDS tools generate far too many false positive alerts, creating the dangerous condition of ?alert fatigue? in which providers become habituated to all alerts, threatening patient safety and even leading to fatal consequences. The PI and Co-I of this proposal collaboratively developed a CDS software called Sepsis-Alert for adult emergency department (ED) patients. It was fully implemented into Detroit Medical Center's (DMC) EMR live environment and has now been continually operational to provide real-time ongoing monitoring of all ED patients at Sinai Grace Hospital of DMC since October 2014. Our analysis of 25,000 ED visits reveals that while Sepsis-Alert's performance exceeds any reported performance, it still remains unacceptably inaccurate. All the CDS tools, including ours, have two limitations: (1) they lack a mechanism to learn from their past erroneous decisions and consequently repeat the same mistakes again and again, and (2) their decision- making process is fixed and treats all patients in the same way even in face of high heterogeneity of patients, The main thrust of this research project is to develop an innovative prototype CDS software that functions like Sepsis-Alert but without the two limitations for the same ED sepsis screening purpose. We will develop the software system by utilizing data extracted from the EMR and will test and fine tune the system in over 35,000 retrospective and prospective patients at Sinai Grace Hospital. The proposed prototype, Intelligent Sepsis Alert, will have the cutting edge capabilities of recognizing the subtleties of sepsis, categorizing patients and learning from its own mistakes to avoid repeat them. CDS tools of the future can and must be better. Machine learning is the solution to optimizing patient care without creating a harmful environment. The final deliverable of this project will be a highly accurate and advanced program readily adoptable by any health system or hospital to improve sepsis care and create a safer healthcare environment. 1
Sepsis is a toxic response to a severe infection and represents a healthcare epidemic, which afflicts millions of people and accounts for 5% of all hospital costs and over $20 billion in the U.S. annually. This project's objective is to develop a cutting edge computer based tool, called Intelligent Sepsis Alert, with the power of machine intelligence (a form of artificial intelligence) and the ability to learn that will accurately identify sepsis patients for healthcare providers extremely early in their hospital course to ensure that patients receive all of the necessary life-saving interventions they need. Intelligent Sepsis Alert will be readily adoptable by other hospitals and health systems and by providing the critical real-time, bedside support for early sepsis identification will translate into multitudes of deaths prevented, abundant intensive care unit admissions avoided, hundreds of thousands of dollars saved and thousands of wasted nursing and physician man-hours eliminated.
