PTRBC

PTRBC predicts required pre-surgery blood volume (PBV) for surgical patients using machine learning models to estimate red blood cell (RBC) transfusion needs.

Key Features:

• Data cleaning and preparation: Processes 181,027 medical documents collected over six years to derive 92,057 relevant blood transfusion records for model training.
• Machine learning integration: Employs six machine learning algorithms and selects Random Forest as the best-performing model with 72.9% accuracy, outperforming surgeon experience-based estimates by 30.4%.
• Predictive factor identification: Identifies 39 predictive factors associated with RBC transfusions to inform model inputs.
• Model validation and accuracy: Validated on 118,823 data points from patients who received allogenic RBCs or did not require transfusion, with total volumes <10 units and pre-surgery laboratory examinations within seven days, using 90% of the data for training.

Scientific Applications:

• Preoperative transfusion planning: Provides PBV estimates to support clinician decisions about pre-surgery RBC transfusion requirements.
• Blood inventory management: Informs allocation of blood resources by predicting likely transfusion volumes prior to surgery.
• Perioperative risk assessment: Uses perioperative factors and identified predictors to refine assessments of transfusion need and related patient safety considerations.

Methodology:

Extracted relevant factors from medical records including surgeon experience volumes and actual transfused RBC volumes; cleaned 181,027 documents to obtain 92,057 transfusion records; identified 39 predictive factors; trained six machine learning algorithms and selected Random Forest (72.9% accuracy); used 90% of data for training and validated on 118,823 data points meeting specified clinical criteria (allogenic RBCs or no transfusion, total volumes <10 units, pre-surgery labs within seven days).