Machine Learning-Based Prediction of Postoperative Complications Using Laboratory and Clinical Parameters in Total Hip and Total Knee Arthroplasty: A Scoping Review

Background: Total joint arthroplasties (TJAs) are well-established orthopedic procedures that provide effective pain relief and restore functional mobility. Although these procedures are generally highly successful, postoperative complications may still occur. This review examined the literature on how machine learning (ML) can predict postoperative complications based on laboratory and clinical parameters for total knee (TKA) and total hip arthroplasty (THA). Methods: A scoping review was conducted and reported in accordance with the PRISMA criteria for scoping reviews (PRISMA-ScR). We searched PubMed/MEDLINE, Scopus, and EMBASE for studies evaluating the use of ML to predict adverse events following TKA and/or THA, using routinely available laboratory values combined with demographics and clinical parameters. Results: The initial search yielded 144 studies, of which 18 were finally deemed eligible for inclusion. Key predictors of complications in the examined studies included hemoglobin, hematocrit, white blood cell count, platelet count, and renal markers. Need for transfusion was the most frequently studied outcome, followed by severe adverse events and postoperative delirium. The ML models, especially gradient boosting machines (GBMs), random forests (RFs), and neural networks (NNs), demonstrated promising discriminative performance in predicting adverse events. Conclusions: Combining laboratory data with demographics and clinical information may improve the accuracy of complication prediction. This review emphasizes the potential of ML analysis of laboratory parameters to create risk assessment tools, which could detect complication risk and potentially improve outcomes for patients undergoing TKA or THA.