PROJECT ABSTRACT
Sepsis is a leading cause of death. It is also a major driver of antibiotic utilization, in large part because
current guidelines and mandates compel clinicians to administer antibiotics immediately for all patients with
possible sepsis even though more than a third of patients initially suspected to have sepsis turn out to have
viral or non-infectious conditions. The Centers for Medicare & Medicaid Services (CMS) “SEP-1” measure, for
example, requires hospitals to give broad-spectrum antibiotics within 3 hours of to all patients with suspected
sepsis as well as measure lactate levels, draw blood cultures, and administer at least 30cc/kg of fluids for
hypotension. Newer guidelines are pushing for even faster antibiotic administration and bundle completion.
Aggressive, rigid, and reflexive sepsis care may therefore benefit some patients but also unnecessarily risks
promoting antibiotic resistance, drug adverse events, C.difficile infections, and fluid overload in others.
A major barrier to providing appropriate sepsis care to the right patients is our limited understanding of
which specific clinical presentations are most likely to benefit from immediate antibiotics and which can safely
be managed more conservatively. Similarly, we have imperfect understanding of which specific processes of
care are critical to optimize outcomes and which are not. To address these gaps, we need a better way to
define sepsis “time zero,” a concept that embodies both the criteria that should trigger immediate interventions
when infection is suspected and the point from which timeliness of care is measured. The current definition of
sepsis includes a heterogeneous mix of clinical signs, some of which are clearly urgent and some that may
tolerate more time for investigation before administering antibiotics. The problem is compounded by the fact
that SEP-1's time zero definition is subjective and labor-intensive to abstract, undermining its credibility and
utility for benchmarking hospitals' quality of care.
In this project, we propose to leverage detailed electronic health record data from millions of encounters
from 167 hospitals in two datasets to develop more evidence-based sepsis time zero criteria and quality
metrics. These goals are reflected our Specific Aims: 1) Develop evidence-based, objective, and electronically
computable definitions of sepsis time zero by using electronic health record data to identify the clinical signs
that are associated with higher mortality when antibiotics are delayed, 2) Systematically evaluate the
associations between each sepsis bundle component and mortality and compare the full bundle to simpler,
streamlined versions, and 3) Assess whether and how optimal treatment strategies differ for commonly
encountered and easily recognizable sepsis phenotypes based on infection site, comorbidities, and clinical
signs. This proposal directly addresses the antibiotic stewardship focus of AHRQ's HAI Prevention Portfolio
and AHRQ's mission to improve health care safety and quality by providing new insights into identifying and
treating sepsis, informing more rational antibiotic use, and facilitating better measurement of quality-of-care.
1
