Measuring and Learning from Care Variation in Sepsis

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Sepsis is a syndrome of life-threatening organ failure resulting from serious infection that contributes to at least 1.7 million U.S. hospital admissions annually. With mortality of 15-20%, sepsis kills 11 million people each year around the world. Prompt, appropriate therapy with antibiotics and IV fluids can improve sepsis outcomes, but many patients with sepsis do not receive optimal therapy. Unwarranted variation in practice between physicians is an important contributor to this problem. In some areas of treatment, however, practice variation reflects the fact that the optimal sepsis treatment strategy is not clear. In both situations, the mechanisms underlying between-physician variation in sepsis care are largely unknown. This research program will apply advanced epidemiologic, statistical, machine learning, and qualitative research methods to large, granular clinical datasets to determine how physicians make critical management decisions — including decisions related to antibiotic initiation, intravenous fluid resuscitation, and emergency department disposition — for patients with sepsis. A key focus will be understanding how different styles of decision making interact with different phenotypes of this heterogeneous syndrome to affect patient treatment and outcomes. Investigations will also harness observed variation to inform causal inference about optimal sepsis management strategies where current evidence is inadequate. In the process, this research will produce validated, scalable, and generalizable tools to support high-quality “big data” sepsis research. The proposed research, which is well matched to the NIGMS sepsis research priorities, will help health systems, researchers, and clinicians design and deliver sepsis care, provide evidence to build decision support tools customized to patient phenotypes and clinician decision styles, and lay the groundwork for future trials of sepsis care strategies and implementation methods.