Multi-Institutional Machine-Learning Predictor of Gross-Total Resection in Skull-Base Chondrosarcoma

Objective: Develop and validate an anatomy-driven model that predicts gross-total resection (GTR) for skull-base chondrosarcomas (SCBs). Methods: We analyzed 179 consecutive SBCs resected at three academic centers. Thirteen preoperative variables were abstracted: tumor location, internal carotid artery (ICA) encasement, compartmental extensions, cranial-nerve involvement, prior radiotherapy, and approach. Data were split 75/25 into training ( n = 135) and validation ( n = 44). Five algorithms were tuned for validation. For interpretability, multivariable logistic GLM and nomogram were refitted. Performance was summarized with AUC, accuracy, sensitivity/specificity, and Brier score. Results: Anatomic burden varied by zone, cavernous-sinus invasion clustered in peri-lacerum (66.7%) and petroclival (57.0%); jugular-foramen extension mainly petroclival (37.2%); sinonasal/orbital spread characterized midline lesions. Approach selection mirrored: EEA predominated in midline (80%) and common in petroclival (62%), whereas lateral tumors were mostly open (84.6%). GTR was 56% (101/179) overall. In petroclival disease, corridor choice was decisive ( p < 0.001): ETPA 70.3% GTR vs open 25.7% and midline EEA 36.4%. Operative Approach, EOR, and Residual Disease Petroclival ( N = 121) Peri-lacerum ( N = 30) Lateral ( N = 13) Midline ( N = 15) p-value EOR GTR 61 (50%) 20 (67%) 10 (77%) 10 (67%) 0.116 STR 60 (49%) 10 (35%) 3.0 (23%) 5 (33%) GTR% x approach Open 9/35 (26%) 12/19 (63%) 9/11 (82%) 2/3 (67%)