PREDICTIVE MODEL FOR INCISIONAL COMPLICATIONS FOLLOWING REVISION ARTHROPLASTY

Background and Aims Revision total joint arthroplasty (rTJA) carries a high risk of postoperative incisional complications (PIC), with an incidence rate ranging from 10% to 20%. These complications are critical precursors to periprosthetic joint infections and surgical failure. Existing risk models often overlook dynamic host-status indicators, particularly systemic immune competence and nutritional reserves. Therefore, this study aimed to identify independent risk factors for PIC in rTJA patients. Furthermore, we sought to construct and validate a robust machine learning-based clinical predictive model to facilitate personalized risk stratification and early prophylactic intervention. Methods A total of 100 rTJA patients were retrospectively enrolled. Perioperative clinical and laboratory data were comprehensively collected. A predictive model was constructed using the XGBoost algorithm and logistic regression, culminating in the development of a nomogram. Model performance was evaluated via Receiver Operating Characteristic (ROC) curves, calibration curves, Decision Curve Analysis (DCA), and SHAP (Shapley Additive exPlanations) analysis. Results This study included 100 patients undergoing rTJA (70 hip, 30 knee revisions). During follow-up, 25 patients developed postoperative incision complications (PIC), indicating a 25% incidence rate. Multivariate Logistic regression identified preoperative total lymphocyte count (TLC) (P < 0.01) and serum albumin levels (P < 0.01) as independent risk factors for PIC. A machine learning-based predictive model incorporating these two markers demonstrated excellent discrimination, with an AUC of 0.987 (95% CI: 0.966-1). The Hosmer-Lemeshow test (P = 0.950) and calibration curves confirmed optimal goodness-of-fit. SHAP analysis verified the dominant predictive roles of TLC and albumin. Furthermore, survival analysis indicated that concurrent low albumin and TLC significantly increase the risk of long-term complications. Conclusion Preoperative albumin (< 33.6 g/L) and TLC (< 1.4 × 109/L) are critical predictive factors for PIC in rTJA patients. The nomogram model based on these factors exhibits superior predictive efficacy