Prediction Models for Sepsis-Associated Thrombocytopenia Risk in Intensive Care Units Based on a Machine Learning...

Area of Science:

• Critical Care Medicine
• Computational Biology
• Hematology

Background:

• Sepsis-associated thrombocytopenia (SAT) is a frequent and serious complication in intensive care units (ICUs).
• SAT significantly elevates mortality rates and worsens patient prognosis.
• Machine learning (ML) offers potential for predicting critical conditions in ICU patients.

Purpose of the Study:

• To develop and compare ML models for predicting platelet decrease and severe platelet decrease in ICU sepsis patients.
• To identify the most effective ML algorithm for SAT prediction.
• To enable early identification of high-risk patients for tailored treatment.

Main Methods:

• Utilized data from 1,455 ICU sepsis patients (January 2015 - October 2019).
• Developed prediction models using Random Forest, Neural Network, Gradient Boosting Machine, and Bayesian algorithms.
• Validated models on the MIMIC-III database.

Main Results:

• Thrombocytopenia occurred in 49.7% of patients, associated with longer ICU stays and higher mortality.
• Model performance varied: AUCs for thrombocytopenia ranged from 0.54-0.72.
• AUCs for severe thrombocytopenia ranged from 0.70-0.77, with Bayesian models showing the best predictive power.

Conclusions:

• Neural Network and Gradient Boosting Machine models effectively predict SAT occurrence.
• Bayesian models demonstrate superior performance in predicting severe SAT.
• These ML models can aid in early risk stratification and personalized treatment strategies for SAT patients.