Predicting postoperative transfusion in elective total HIP and knee arthroplasty: Comparison of different machine learning models of a case-control study

Abstract Background Total joint arthroplasty (TJA) is a very successful orthopedics procedure but associates with a significantly high transfusion rate. Objective In this study, we aimed to determine predictors of postoperative blood transfusion in patients undergoing elective hip and knee TJA patients and compare the accuracy of machine learning (ML) algorithms in predicting transfusion risk. Methods We utilized data from 12,642 patients undergoing primary unilateral TJA. Risk factors and demographic information were extracted, and predictive models were developed using seven ML algorithms. The area under the receiver operating characteristic curve was used to measure the predictive accuracy of each algorithm. Results The overall transfusion rate was 18.7%. Patient-related risk factors for transfusion included age 65–85 (Odds Ratio (OR): 1.175–1.222), female (OR: 1.246), American Society of Anesthesiologists grade II or greater (OR: 1.264–2.758). Surgical factors included operation time (OR: 1.736), drain use (OR: 2.202) as well as intraoperative blood loss (OR: 7.895). Elevated preoperative Hb (OR: 0.615), Hct (OR: 0.800), BMI (≥24 kg/m 2 ) (OR: 0.613–0.731) and tranexamic acid use (OR: 0.375) were associated with decreased risk of postoperative transfusion. The long short-term memory networks (LSTM) and random forest (RF) models achieved the highest predictive accuracy ( p < 0.001). Conclusion The risk factors identified in the current study can provide specific, personalized postoperative transfusion risk assessment for a patient considering lower limb TJA. Furthermore, the predictive accuracies of LSTM and RF algorithms were significantly higher than the others, making them potential tools for future personalized preoperative prediction of risk for postoperative transfusion. Highlights