Predicting Factors for Blood Transfusion in Primary Total Knee Arthroplasty Using a Machine Learning Method

Area of Science:

• Orthopedic Surgery
• Data Science
• Anesthesiology

Background:

• Acute blood loss anemia is a significant postoperative complication following total knee arthroplasty (TKA).
• Allogeneic blood transfusions are frequently required to manage this anemia.
• Predicting transfusion risk is crucial for optimizing patient outcomes.

Purpose of the Study:

• To develop and validate machine learning models for predicting blood transfusion requirements after primary TKA.
• To identify demographic and surgical factors contributing to transfusion necessity.
• To assess the potential of machine learning in improving preoperative planning for TKA.

Main Methods:

• A cohort of 2,093 patients undergoing primary TKA was analyzed.
• Data from a hospital quality improvement database were used to extract patient demographics and surgical variables.
• A multilayer perceptron neural network (MPNN) machine learning algorithm was employed for prediction, alongside traditional statistical analyses.

Main Results:

• Key predictors for transfusion included preoperative hemoglobin and creatinine levels, surgery duration, simultaneous bilateral procedures, tranexamic acid use, ASA score, albumin levels, ethanol use, anticoagulation status, age, and TKA type.
• Patients receiving blood transfusions experienced a significantly longer hospital stay.
• The MPNN model demonstrated high predictive performance with an AUC of 0.894.

Conclusions:

• The MPNN model effectively ranks factors influencing blood transfusion risk in TKA patients.
• Machine learning offers superior predictive capabilities compared to traditional statistical methods for TKA transfusion prediction.
• This novel application of MPNN in TKA has the potential to enhance preoperative planning and patient management.