Sepsis, defined as life-threatening organ dysfunction in response to infection, is a devastating condition that contributes to up to half of hospital deaths and over $24 billion in healthcare costs in the U.S. annually. Over 750,000 patients develop sepsis in the U.S. each year, and survivors suffer long-term cognitive impairment and physical disability. Historically, sepsis research has focused on patients who are already critically ill. However, up to 50% of patients with sepsis receive their care on the inpatient wards, and only 10% of patients with sepsis are initially diagnosed in the intensive care unit (ICU). Because early intervention improves outcomes in sepsis, it is important to optimize the detection and treatment of sepsis outside the ICU. The current sepsis paradigm has several problems. The first problem is that early identification of infection relies on clinician intuition, and caregivers often disagree regarding which patients are infected. This leads to delays in therapy and increased mortality in some patients and unnecessary therapies and adverse medication side effects in others. A second problem is that there is a lack of accurate tools to risk stratify infected patients outside the ICU after they are identified. Some patients with infection are treated outside the ICU and are later discharged home, while others develop life-threatening complications and die in the hospital. Accurate risk stratification of infected patients would bring additional critical care resources to the bedside of the high-risk patients that need them most. A third problem with the current sepsis paradigm is that it is often treated as a one-size-fits-all syndrome. However, patients with sepsis have a wide range of clinical presentations and outcomes due to the complex interactions between patient risk factors, the infectious organism, and the host immune response. These data suggest that the impact of timely and more aggressive interventions on outcomes may differ based on a patient's clinical phenotype. Identifying important subphenotypes of infected patients is critical to delivering more personalized care at the bedside. The purpose of this project is to use data from the electronic health record and statistical modeling techniques to identify high-risk infected patients and important new subphenotypes of this syndrome.
In Aim 1, we will develop a novel tool for identifying infected patients outside the ICU using modern machine learning techniques.
In Aim 2, we will develop a tool for risk stratifying infected patients outside the ICU using machine learning methods. Finally, in Aim 3 we will use cluster analysis techniques to determine whether the benefit of early and more aggressive interventions varies based on clinical phenotype. Our project will provide clinicians with powerful new tools to identify high-risk infected patients and important new subphenotypes of this common and deadly syndrome. This work will help to deliver early, life-saving care to the bedside of septic patients and lead to future interventional trials aimed at decreasing preventable death.

Public Health Relevance

Sepsis, defined as life-threatening organ dysfunction in response to infection, is a devastating condition that contributes to up to half of hospital deaths and over $24 billion in costs in the US. The purpose of this project is to use data from the electronic health record and statistical modeling techniques to identify high-risk infected patients and determine which subgroups of patients benefit most from early and more aggressive interventions. This work will result in novel algorithms to identify high-risk infected patients that can be implemented in the electronic health record to decrease preventable death.

Agency
National Institute of Health (NIH)
Institute
National Institute of General Medical Sciences (NIGMS)
Type
Research Project (R01)
Project #
7R01GM123193-04
Application #
10056599
Study Section
Biomedical Computing and Health Informatics Study Section (BCHI)
Program Officer
Brazhnik, Paul
Project Start
2017-05-01
Project End
2022-03-31
Budget Start
2019-08-01
Budget End
2020-03-31
Support Year
4
Fiscal Year
2019
Total Cost
Indirect Cost
Name
University of Wisconsin Madison
Department
Internal Medicine/Medicine
Type
Schools of Medicine
DUNS #
161202122
City
Madison
State
WI
Country
United States
Zip Code
53715
Lyons, Patrick G; Edelson, Dana P; Churpek, Matthew M (2018) Rapid response systems. Resuscitation 128:191-197
Rojas, Juan C; Carey, Kyle A; Edelson, Dana P et al. (2018) Predicting Intensive Care Unit Readmission with Machine Learning Using Electronic Health Record Data. Ann Am Thorac Soc 15:846-853
Blagev, Denitza P; Collingridge, Dave S; Rea, Susan et al. (2018) The Laboratory-Based Intermountain Validated Exacerbation (LIVE) Score Identifies Chronic Obstructive Pulmonary Disease Patients at High Mortality Risk. Front Med (Lausanne) 5:173
Sanchez-Pinto, L Nelson; Luo, Yuan; Churpek, Matthew M (2018) Big Data and Data Science in Critical Care. Chest 154:1239-1248
Parker, William F; Anderson, Allen S; Hedeker, Donald et al. (2018) Geographic Variation in the Treatment of U.S. Adult Heart Transplant Candidates. J Am Coll Cardiol 71:1715-1725
Churpek, Matthew M; Snyder, Ashley; Sokol, Sarah et al. (2017) Investigating the Impact of Different Suspicion of Infection Criteria on the Accuracy of Quick Sepsis-Related Organ Failure Assessment, Systemic Inflammatory Response Syndrome, and Early Warning Scores. Crit Care Med 45:1805-1812