Related Experiment Video
Updated: Oct 6, 2025
07:42
A Data-Driven Approach to Quantifying Immune States in Sepsis
Published on: February 7, 2025
330
Evaluating machine learning models for sepsis prediction: A systematic review of methodologies
Hong-Fei Deng1,2, Ming-Wei Sun3, Yu Wang1,2,3,4
1Institute for Emergency and Disaster Medicine, Sichuan Academy of Medical Sciences, Sichuan Provincial People's Hospital, School of Medicine, University of Electronic Science and Technology of China, Chengdu, Sichuan 610072, China.
Iscience
|January 14, 2022
Summary
Machine learning models for sepsis prediction show promise, but inconsistent definitions and methods hinder progress. New standards and criteria are proposed to improve sepsis prediction model development and clinical application.
Area of Science:
- Medical Science
- Artificial Intelligence in Healthcare
- Clinical Informatics
Background:
- Machine learning (ML) is increasingly applied to sepsis prediction.
- Existing studies exhibit significant heterogeneity in methodology and evaluation.
- A need exists for standardized evaluation criteria in ML-based sepsis prediction.
Purpose of the Study:
- To propose novel evaluation criteria and reporting standards for ML models in sepsis prediction.
- To critically assess 21 ML models for sepsis prediction using PRISMA guidelines.
- To identify key factors influencing the performance of ML models in sepsis prediction.
Main Methods:
- Systematic review of 21 ML models for sepsis prediction based on PRISMA.
- Development and application of new evaluation criteria and reporting standards.
- Analysis of data sources, preprocessing, model types, feature engineering, and inclusion criteria.
Main Results:
- Inconsistent sepsis definitions and diverse methodologies across studies were observed.
- Model performance, measured by AUROC, improved closer to sepsis onset.
- ML's role in feature engineering significantly enhanced AUROC.
- Deep neural networks with Sepsis-3 criteria performed well on time-series patient data.
Conclusions:
- Standardized evaluation criteria are crucial for advancing ML in sepsis prediction.
- Addressing methodological inconsistencies will improve model reliability and clinical utility.
- Future research should focus on deep learning and Sepsis-3 criteria for time-series sepsis data.
