Development and validation of a novel pathologically-based prognostic model for primary central nervous system lymphoma

Abstract Background: Although the IELSG and MSKCC prognostic scoring systems for PCNSL are widely used, their clinical applicability remains limited. The IELSG scoring systems is hampered by the variable clinical availability of LDH and CSF protein assessments. The MSKCC scoring system demonstrates limited prognostic comprehensiveness due to its exclusive incorporation of age and Karnofsky performance score. This study aims to develop and validate a novel prognostic model based on real-world clinical data to improve prediction accuracy and clinical utility. Methods: We retrospectively analyzed the clinical data of 180 newly diagnosed PCNSL patients aged ≥18 years between January 2012 and January 2025 in our center. The primary endpoint was overall survival (OS), and the secondary endpoint was progression-free survival (PFS). Patients were randomly assigned to the training (70%) and validation (30%) cohorts. We performed univariable and multivariable cox regression analyses to identify factors associated with OS. Based on the multivariable analysis results, we developed a novel prognostic model, which subsequently underwent rigorous evaluation for calibration, discrimination, and clinical utility in both training and validation cohorts. Results: The study enrolled 180 patients with PCNSL. Their clinical and treatment characteristics are as follows. The median age at diagnosis of the patients was 60 years (range: 18-82 years), and 59% (101/180) were male. The pathological type was mainly diffuse large B-cell lymphoma (DLBCL) (96%, 172/180). Regarding treatment: 134 cases (74.4%) underwent subtotal/total tumor resection, 46 cases (25.6%) only underwent biopsy, 78 cases (43.3%) received high-dose methotrexate (HD-MTX) chemotherapy, 30 cases (16.7%) received combined treatment (including radiotherapy or hematopoietic cell transplantation). Multivariate cox regression analyses showed that age >60 years (HR: 2.5, P <0.05), Eastern Cooperative Oncology Group (ECOG)≥2 (HR: 4.5, P <0.05), and positive MYC expression by immunohistochemistry (HR: 2.0, P <0.05) were independent predictors of OS. Based on key risk factors, we developed a novel predictive model with the following validation outcomes: 1)excellent discriminative ability, demonstrated by area under the ROC curve (AUC) values of 0.74 in the training set and 0.88 in the validation set; 2)satisfactory calibration performance, showing close agreement between predicted probabilities and actual observations; 3)significant clinical utility, with decision curve analysis confirming consistent net clinical benefit across a wide range of risk thresholds. Conclusions: This study established a novel prognostic model incorporating age, ECOG, and high-risk immunohistochemical markers to predict OS in newly diagnosed PCNSL patients. Considering the correlation between high-risk clinicopathologic features and malignancy, this system holds the promise in contribute to individualized treatment decisions.