Machine Learning for Revision Joint Arthroplasty: A Systematic Review of Distinct Challenges, Current Performance, and Clinical Readiness

Background: Given the high complication rates and economic burden of revision total joint arthroplasty, machine learning (ML) models may offer a tool to improve outcomes.This study aims to review the application of ML for predicting outcomes following revision total hip arthroplasty and total knee arthroplasty.Methods: A comprehensive literature search using Ovid MEDLINE, Embase, and Web of Science was conducted in May 2024.Studies were screened in Covidence.Primary exclusion criteria included (1) systematic reviews; (2) not utilizing ML; and (3) not involving revision arthroplasty.Thirteen studies met the inclusion criteria.Results: Studies focused on patient complications (n = 7), inpatient status and length of stay (n = 3), and readmissions and discharge dispositions (n = 2).One study looked at all 3 outcomes.Studies involved revision total knee arthroplasty (n = 6), revision total hip arthroplasty (n = 4), or a combination (n = 3).Eleven studies reported predictive success using the area under the receiver-operating characteristic curve.6.7% were in the poor range, 36.7% in the acceptable range, 48.3% in the excellent range, and 8.3% in the outstanding range.Ten studies were internally validated, and 3 were externally validated.Conclusions: ML algorithms used in revision total joint arthroplasty are complicated by greater heterogeneity, data sparsity, and outcome imbalance.They have utility in clinical settings for risk stratification, predicting complications, length of stay, and discharge disposition.However, these algorithms should be critically assessed before clinical adoption.