Background: Revision total joint arthroplasty (TJA) becomes necessary when a primary TJA fails, leading to worse patient outcomes, higher complication rates, and increased economic burdens on the healthcare system. With the global aging population, the demand for revision TJA’s are projected to rise significantly by 2030 which highlights the need for effective machine learning (ML) models to improve outcomes in revision TJA. Although prior systematic reviews have evaluated the utility of ML in predicting outcomes for primary TJA, there is a notable gap in assessing ML’s potential in the context of revision TJA . This study aims to systematically review the application of ML for predicting outcomes following revision total hip arthroplasty (THA) and revision total knee arthroplasty (TKA).

Methods: A comprehensive literature search was conducted to May 2024 in Ovid MEDLINE, Ovid Embase, and Web of Science. Studies were imported into Covidence for screening. The exclusion criteria were: (1) other systematic reviews; (2) animal models; (3) studies not utilizing artificial intelligence; (4) studies not involving revision arthroplasty; (5) case reports; (6) technical notes; (7) editorial notes; (8) studies on joints other than the hip and knee; and (9) studies without full-text availability. The search strategy yielded 1289 articles, of which 19 were assessed as eligible for this review.

Results: Search criteria yielded 19 relevant studies. Topics of study included patient complications (n = 12), readmissions (n = 2), discharge dispositions (n = 2), patient-reported outcome measures (n = 0), inpatient status and length of stay (LOS) (n = 4), and patient function (n = 1). Studies involved revision TKA (n = 8), revision THA (n = 7), or a combination (n = 5). Studies were of solely revision arthroplasty (n = 16) or included primary and revision arthroplasty (n = 4). Of reported AUC, greater than 71% of predictive outcomes for at least one model had an area under the receiver operating characteristic curve (AUC) in the excellent or outstanding range (AUC > 0.8). Additionally, 14 studies were internally validated and only 3 studies were externally validated.

Discussion/Conclusion: Machine learning algorithms are being increasingly used to predict patient outcomes after revision THA and TKA. Machine learning has shown utility in clinical settings in the context of risk stratification and prediction of complications, patient reported outcome scores, and readmissions. While machine learning algorithms have been received with considerable interest, they should be critically assessed and externally validated prior to clinical adoption.