Robust Meta-Model for Predicting the Likelihood of Receiving Blood Transfusion in Non-traumatic Intensive Care Unit...

Area of Science:

• Critical Care Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Blood transfusions are vital in intensive care units (ICUs) for anemia and coagulopathy.
• Current decision support systems have limitations in patient diversity and transfusion type focus.
• Accurate transfusion prediction is essential for resource management and patient risk stratification.

Purpose of the Study:

• To develop and evaluate an advanced machine learning model for predicting blood transfusion necessity in non-traumatic ICU patients.
• To assess the model's performance across a diverse patient cohort over a 24-hour period.
• To identify key biomarkers influencing transfusion decisions.

Main Methods:

• Retrospective cohort study of 72,072 non-traumatic adult ICU patients (2016-2020).
• Development of a meta-learner and various machine learning predictors.
• Annual model training and 5-year iterative evaluation on unseen data.

Main Results:

• The meta-model demonstrated superior performance compared to other models.
• Achieved an area under the receiver operating characteristic curve of 0.97.
• Reported an accuracy rate of 0.93 and an F1 score of 0.89 in the optimal scenario.

Conclusions:

• This study introduces a pioneering machine learning approach for predicting blood transfusions in critically ill patients.
• The model effectively predicts transfusion likelihood and identifies significant biomarkers.
• Findings support improved transfusion decision-making and resource allocation in ICUs.