Prognosticating salvage stereotactic radiosurgery outcomes in relapsed primary central nervous system lymphoma: A machine learning-driven decision tree analysis.

Purpose: To identify key clinical risk factors affecting therapeutic outcomes in relapsed primary central nervous system lymphoma (r-PCNSL) patients undergoing stereotactic radiosurgery salvage therapy (SRS-ST) and develop a decision tree-based predictive model.

Patients And Methods: A retrospective analysis was performed on r-PCNSL patients undergoing SRS-ST at The Second Affiliated Hospital of Xuzhou Medical University between January 2012 and November 2021. The cohort was randomly divided into training and validation sets (7:3 ratio). The C5.0 algorithm was employed to develop a decision tree model for predicting treatment response. Model performance was evaluated using diagnostic metrics including accuracy (ACC), sensitivity, and specificity.

Results: A cohort of 209 patients meeting inclusion/exclusion criteria were enrolled. Survival analysis revealed a mean progression-free survival (PFS) of 7.5 ± 2.6 months and overall survival (OS) of 13.8 ± 4.1 months. Using multivariate analysis, a decision tree model was developed incorporating three critical prognostic parameters: Karnofsky Performance Status (KPS); deep brain structure involvement; and International Extranodal Lymphoma Study Group (IELSG) score. The model demonstrated robust predictive accuracy, with sensitivities of 0.880-1.000 in the training set versus 0.667-0.880 in the validation set, and corresponding specificities of 0.926-1.000 and 0.854-0.984, respectively.

Conclusions: Our analysis identified critical determinants of therapeutic response in r-PCNSL patients receiving SRS-ST, developing a clinically applicable decision tree model to guide hematologists and neuro-oncologists in personalizing treatment approaches.