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

Abstract

INTRODUCTION: Acute blood loss anemia requiring allogeneic blood transfusion is still a postoperative complication of total knee arthroplasty (TKA). This study aimed to use machine learning models for the prediction of blood transfusion after primary TKA and to identify contributing factors. METHODS: A total of 2,093 patients who underwent primary TKA at our institution were evaluated using data extracted from the hospital quality improvement database to identify patient demographics and surgical variables that may be associated with blood transfusion. A multilayer perceptron neural network (MPNN) machine learning algorithm was used to predict risk factors for blood transfusion and factors associated with increased length of stay. Statistical analyses including bivariate correlate analysis, Chi-Square test, and Student t-test were performed for demographic analysis and to determine the correlation between blood transfusion and other variables. RESULTS: The results demonstrated important factors associated with transfusion rates, including preoperative hemoglobin level, preoperative creatinine level, length of surgery, simultaneous bilateral surgeries, tranexamic acid usage, American Society of Anesthesiologists Physical Status score, preoperative albumin level, ethanol usage, preoperative anticoagulation medications, age, and TKA type (conventional versus robotic-assisted). Patients who underwent a blood transfusion had a markedly greater length of stay than those who did not. The MPNN machine learning model achieved excellent performance across discrimination (AUC = 0.894). DISCUSSION: The MPNN machine learning model showed its power as a statistical analysis tool to predict the ranking of factors for blood transfusion. Traditional statistics are unable to differentiate importance or predict in the same manner as a machine learning model. CONCLUSION: This study demonstrated that MPNN for the prediction of patient-specific blood transfusion rates after TKA represented a novel application of machine learning with the potential to improve preoperative planning for treatment outcomes.