Comparison between machine learning methods for mortality prediction for sepsis patients with different social...

Area of Science:

• Critical care medicine
• Health informatics
• Machine learning in healthcare

Background:

• Sepsis is a life-threatening condition in critically ill patients, with diagnosis challenges due to evolving criteria.
• Social determinants of health can significantly impact the performance of machine learning models for sepsis risk prediction.

Purpose of the Study:

• To investigate the impact of social determinants on machine learning-based sepsis mortality prediction.
• To identify performance disparities in sepsis diagnosis across different patient subgroups.

Main Methods:

• Analysis of a critical care patient cohort from the Medical Information Mart for Intensive Care (MIMIC)-III database.
• Training 16 machine learning classifiers to predict in-hospital mortality for sepsis patients (identified by Sepsis-3 criteria).
• Evaluating model performance across subpopulations stratified by race, sex, marital status, insurance type, and language.

Main Results:

• Significant differences in social determinants were observed among patients identified by various sepsis criteria.
• Machine learning model performance for mortality prediction decreased significantly for Asian, Hispanic, and Spanish-speaking sepsis patients.
• Performance discrepancies were noted between Asian and White patients, and between English-speaking and Spanish-speaking patients.

Conclusions:

• Disparities in patient identification by sepsis criteria exist across social determinant groups.
• Universal machine learning models may compromise mortality prediction accuracy for specific subpopulations.
• A versatile sepsis diagnostic system is required to mitigate the impact of social determinants on patient outcomes.