Pilot Prognostic Model for Survival in r/r DLBCL Patients Receiving Palliative Radiation Therapy

Abstract

Patients with relapsed/refractory (r/r) diffuse large B-cell lymphoma (DLBCL) often require palliative radiotherapy (pRT) for symptomatic relief, bridging therapy, or salvage treatment (i.e., irradiation of only site of disease). Prognostication remains a challenge, especially with the advent of novel therapies such as chimeric antigen receptor (CAR)-T cells. We aimed to construct a predictive model for survival in r/r DLBCL treated with pRT.Patients with r/r DLBCL who received their first course of pRT between 2009 and 2020 were included. OS was defined as time from pRT consult to death. A multivariable prognostic model was developed using Cox proportional hazard regression with LASSO (Least Absolute Shrinkage and Selection Operator)-regularization, which optimizes variable inclusion based on the model's overall prediction accuracy. Model discrimination performance was summarized using the concordance index after optimism-adjustment with the bootstrap (internal validation).82 patients met the inclusion criteria. Mean age at pRT was 62 years, with the majority of patients being male (59%) and Caucasian (80%). Most patients had an ECOG 0-1 (78%), despite being heavily pretreated (50% ≥3 systemic therapies, 22% transplant, 20% CAR-T cell therapy). pRT was most commonly delivered for symptoms (54%), followed by salvage (34%) and bridging (12%) to another systemic therapy. With a median follow-up of 6.3 months (range, 0.1- 91.2), 57 deaths (70%) were observed with median survival (MS) of 7.8 months. Treatment intent was associated with differences in MS: 4.1 mo (symptomatic), 10.4 mo (bridging), and 11.9 mo (salvage; P = 0.026). The LASSO method selected 5 co-variates for the multivariable OS model (Table 1), which was associated with an optimism-adjusted concordance index of 0.70 (95% CI, 0.61-0.80). Conversely, age, tachycardia, double/triple hit lymphoma, prior transplant or CAR-T cell therapy, and time from diagnosis were not selected by the LASSO for the model.Our pilot prognostic model for survival of patients with r/r DLBCL treated with pRT has good predictive performance (concordance index 0.7), with performance status, treatment intent, and number of prior lines of therapy having the most impact on predictions. Interestingly, types of prior therapies (e.g., CAR-T cell) and pathologic features at diagnosis were not selected for the model. Future work will focus on validating these findings in other cohorts, where larger numbers may permit development and validation of a stronger prognostic model.