Predictors of Failure After Two-Stage Revision for Periprosthetic Joint Infections of the Knee.

BACKGROUND: Prosthetic joint infection (PJI) following total knee arthroplasty remains a serious complication. Although two-stage revision is considered the gold standard, predictors of failure remain unclear. This study aimed to identify factors associated with failure after two-stage revision for knee PJI using conventional statistics with exploratory machine-learning validation. METHODS: A retrospective review was conducted on 154 patients who underwent two-stage revision for knee PJI between 2018 and 2022 at a high-volume arthroplasty center. All patients completed a minimum 24-month follow-up (mean 36.3 ± 20.3 months). Treatment success and failure were defined using MSIS Tier-1 criteria. Independent predictors of failure were identified using multivariate logistic regression and Random Forest classification model. RESULTS: Most infections were late in onset (91.6%), and 68.2% were culture-positive. Difficult-to-treat (DTT) organisms were identified in 22.1% of cases and were associated with significantly worse postoperative WOMAC, KSS, and OKS outcomes (all P < .001). Multivariate analysis identified DTT organisms (odds ratio 0.035, P < .001) and shorter time from index surgery to infection (odds ratio 1.033, P = .016) as independent predictors of failure. The random Forest model demonstrated an overall accuracy of 84.2% and identified DTT organisms, pre-revision C-reactive protein levels, and time from index surgery as the most influential variables contributing to classification performance. Most failures occurred within 30 months of reimplantation. CONCLUSIONS: DTT organisms and shorter time from index surgery to infection were independent predictors of failure following two-stage revision for knee PJI. Exploratory machine-learning analysis identified similar predictor patterns and supports the importance of microbiological factors in determining outcomes.