Learning a Severity Score for Sepsis: A Novel Approach based on Clinical Comparisons

Area of Science:

• Critical Care Medicine
• Biomedical Data Science
• Machine Learning in Healthcare

Background:

• Sepsis is a major cause of mortality in the United States, necessitating accurate severity assessment.
• Current scoring systems like APACHE II and SOFA have limited sensitivity in differentiating sepsis stages.
• Early treatment is crucial for reducing sepsis-related mortality and morbidity.

Purpose of the Study:

• To evaluate the feasibility of the Disease Severity Score Learning (DSSL) framework for developing a sepsis severity score.
• To compare the performance of the DSSL-derived sepsis score against established scoring systems (APACHE II, SOFA).

Main Methods:

• The Disease Severity Score Learning (DSSL) framework was utilized to automatically derive a sepsis severity score from clinical data.
• The DSSL framework learns by analyzing pairs of disease states ordered by severity.
• Performance was assessed by comparing the DSSL score's sensitivity in distinguishing sepsis stages against APACHE II and SOFA.

Main Results:

• The DSSL-derived sepsis severity score significantly outperformed both APACHE II and SOFA in distinguishing between sepsis stages.
• The learned score demonstrated sensitivity to changes in patient severity leading up to septic shock.
• The score was also sensitive to changes in severity following treatment administration.

Conclusions:

• The DSSL framework provides a feasible and effective method for developing a sepsis severity score.
• The DSSL-based sepsis score offers superior performance compared to APACHE II and SOFA for sepsis staging.
• This novel scoring system has potential applications in monitoring disease progression and treatment response in sepsis patients.