Prognostic model for primary CNS lymphoma (PCNSL): Recursive partitioning analysis (RPA) of the MSKCC PCNSL population
1531 Background: Increasingly there is a need to develop a simple prognostic score that can be used in the analysis and design of PCNSL studies as well as for clinical management. Recently the IELSG published a 3 group prognostic model incorporating patient age, performance status, serum LDH, location of brain lesions and CSF total protein; however, only 105 of their 378 patients had all of the variables available to develop this score. Methods: We analyzed 338 patients (median age 60; median KPS 70) seen and treated for PCNSL at MSKCC between 1983 and 2003. The median survival was 37 months and median follow up of surviving patients is 35 months. Univariate analysis of potential prognostic factors was performed using the Kaplan Meier product limit method. Significant univariate variables were included in a multivariate analysis using the Cox proportional hazards regression model. Patients were separately analyzed using the IELSG prognostic score. Finally, RPA was employed as an independent method of developing specific prognostic categories. Results: In the univariate analysis, age, hemiparesis, mental status changes, creatinine clearance and KPS were significant predictors of overall survival; in the multivariate model only age and KPS remained as significant predictors. 113 patients had adequate information (all 5 variables) to be analyzed using the IELSG prognostic score; while this correlated significantly with overall survival, the comparison between groups 2 and 3 was not statistically significant (p = 0.10). RPA of all 338 patients identified 3 subgroups: age ≤ 50 (median OS 9.2 y), age > 50 and KPS ≥ 70 (median OS 3.2 y) and age > 50 and KPS < 70 (median OS 1 y) that significantly separated our entire PCNSL population (p < 0.001). Conclusions: The use of RPA allows for easy discrimination of 3 prognostic groups of patients with PCNSL. In contrast to the IELSG score the MSK RPA classification includes information that is readily available on all patients and can be easily incorporated into the analysis or design of clinical research. No significant financial relationships to disclose.