Mortality and Antibiotic Timing in Deep Learning-Derived Surviving Sepsis Campaign Risk Groups: A Multicenter Study

Area of Science:

• Critical Care Medicine
• Infectious Diseases
• Artificial Intelligence in Healthcare

Background:

• Current Surviving Sepsis Campaign (SSC) guidelines recommend antibiotic timing based on sepsis probability and shock presence.
• Objective patient stratification and outcome analysis based on these risk groups are lacking.
• Need for data-driven approaches to refine sepsis management protocols.

Purpose of the Study:

• To objectively stratify patients with suspected sepsis using deep learning (DL) models.
• To analyze patient outcomes, specifically short-term mortality, based on DL-derived risk groups and antibiotic timing.
• To evaluate the impact of antibiotic administration timing on mortality in different sepsis risk strata.

Main Methods:

• Observational cohort study utilizing prospectively collected data from two health systems (2016-2024).
• Two DL models used to classify patients into four risk groups based on sepsis probability and shock likelihood.
• Primary outcome: short-term mortality (in-hospital mortality or transition to hospice).

Main Results:

• 34,163 adult patients with potential sepsis were analyzed.
• Mortality rates varied significantly across risk groups, with higher mortality in groups with high shock likelihood.
• Patients with probable sepsis unlikely to develop shock had similar mortality rates whether antibiotics were given within 1 or 3 hours.

Conclusions:

• Objective risk stratification using DL models can inform sepsis management.
• For low-risk patients (unlikely to develop shock), a more lenient antibiotic administration time (e.g., 3 hours) may not increase mortality.
• Findings suggest potential for more judicious antibiotic use in certain sepsis patient groups, warranting further prospective validation.