Machine learning-based prediction of transfusion.

Area of Science:

• Medical Informatics
• Health Services Research
• Machine Learning in Healthcare

Background:

• Predicting blood transfusions can optimize blood supply management and enhance patient safety.
• Accurate prediction ensures sufficient red blood cell (RBC) availability for patients.

Purpose of the Study:

• To evaluate the predictive accuracy of four machine learning algorithms for hospital transfusions.
• To assess the models' ability to predict transfusion, massive transfusion, and RBC unit quantity.

Main Methods:

• Retrospective observational study (2008-2017) across three Australian hospitals.
• Compared four machine learning models: Neural Networks (NNs), Logistic Regression (LR), Random Forests (RFs), and Gradient Boosting (GB) trees.
• Evaluated performance using ROC AUC, F1 score, and average precision.

Main Results:

• Models accurately predicted the need for at least one RBC unit (high sensitivity and specificity).
• Prediction of massive transfusions was less successful, with varying sensitivity across models.
• Forecasting the total number of RBC units transfused proved inaccurate.

Conclusions:

• Machine learning models can reliably forecast the necessity for intrahospital transfusions.
• Predicting the precise quantity of RBC units required during a hospital stay is more challenging.