Comparative Analysis of Three Machine-Learning Techniques and Conventional Techniques for Predicting Sepsis-Induced...

Area of Science:

• Medical research
• Computational biology
• Critical care medicine

Background:

• Sepsis-induced coagulopathy has a poor prognosis.
• There is a lack of established tools for predicting its progression.
• Early prediction is vital for improving patient outcomes.

Purpose of the Study:

• To develop predictive models for coagulopathy progression in sepsis using machine learning.
• To compare the predictive accuracy of machine learning techniques against conventional methods.
• To evaluate the efficacy of the ΔDIC score for predicting disseminated intravascular coagulation (DIC) progression.

Main Methods:

• A post-hoc analysis of the Japan Septic Disseminated Intravascular Coagulation retrospective study.
• Calculation of the ΔDIC score (Day 3 DIC score - Day 1 DIC score).
• Application of machine learning models (Random Forests, Support Vector Machines, Neural Networks) and multiple linear regression for prediction.

Main Results:

• Random Forests (RF) demonstrated the highest predictive accuracy (67.0%) for DIC progression.
• The difference between predicted and real ΔDIC was lowest with RF (1.54).
• Machine learning models generally outperformed conventional methods in predicting coagulopathy progression.

Conclusions:

• Random Forests show promise as a tool for predicting sepsis-induced coagulopathy progression.
• Machine learning techniques offer improved accuracy over conventional methods for forecasting DIC.
• Further validation of these models is warranted for clinical application.