Ability of a Machine Learning Algorithm to Predict the Need for Perioperative Red Blood Cells Transfusion in Pelvic Fracture Patients: A Multicenter Cohort Study in China

• DOI: 10.3389/fmed.2021.694733
• Publication Date: 2021-08-16T06:24:05Z
• Authors: Xueyuan Huang; Yongjun Wang; Bingyu Chen; Yuanshuai Huang; Xinhua Wang; Linfeng Chen; Rong Gui; Xianjun Ma
• Keywords: Gradient boosting

Background: Predicting the perioperative requirement for red blood cells (RBCs) transfusion in patients with pelvic fracture may be challenging. In this study, we constructed a RBCs predictive model (ternary classifications) based on machine learning algorithm. Materials and Methods: This study included adult trauma hospitalized across six Chinese centers between September 2012 June 2019. An extreme gradient boosting (XGBoost) algorithm was used to predict need transfusion, data being split into training test (80%), which subjected 5-fold cross-validation, set (20%). The ability of compared preparation surgeons' experience other models, including random forest, decision tree, K-nearest neighbor, logistic regression, Gaussian naïve Bayes classifier models. Data 33 from one hospitals were prospectively collected validation. Results: Among 510 patients, 192 (37.65%) have not received any 127 (24.90%) less-transfusion (RBCs < 4U), 191 (37.45%) more-transfusion ≥ 4U). Machine learning-based produced best performance accuracy 83.34%, Kappa coefficient 0.7967 methods (blood 65.94%, 0.5704; forest method an 82.35%, 0.7858; tree 79.41%, 0.7742; neighbor 53.92%, 0.3341). prospective dataset, it also had food 81.82%. Conclusion: multicenter retrospective cohort described construction accurate that could fractures.