Diagnostic and prognostic capabilities of a biomarker and EMR-based machine learning algorithm for sepsis

Area of Science:

• Biomedical Informatics
• Clinical Medicine
• Machine Learning

Background:

• Sepsis is a leading cause of hospital mortality globally.
• Early antibiotic administration improves sepsis outcomes.
• Timely sepsis recognition remains a significant clinical challenge.

Purpose of the Study:

• To evaluate the diagnostic and prognostic capabilities of a machine-learning (ML) algorithm for sepsis detection.
• To assess the ML model's performance using electronic medical record (EMR) data and uncommon biomarkers.
• To determine the algorithm's ability to predict patient outcomes like length of stay and mortality.

Main Methods:

• A two-center cohort study involving 1400 adult emergency department patients with suspected sepsis.
• Prospective sample collection and analysis.
• Development and validation of an ML model utilizing clinical data and three non-routine biomarkers.

Main Results:

• The ML model achieved an Area Under the Receiver Operating Curve (AUROC) of 0.83 for sepsis diagnosis.
• Higher predicted risk scores correlated with septic shock, organ dysfunction, and adverse outcomes.
• Risk stratification into low, medium, and high-risk groups showed significant differences in length of stay, 30-day mortality, and readmission rates (all p < 0.0001).

Conclusions:

• The ML algorithm demonstrates strong diagnostic accuracy for sepsis.
• The model possesses significant prognostic power, predicting hospital length of stay, 30-day mortality, and readmission.
• This ML approach, integrating EMR data and novel biomarkers, shows promise for early sepsis management at initial blood culture.