Validation of the CNS-IPI-C Prognostic Model in Patients with Systemic DLBCL in the Real World

Introduction CNS relapse in diffuse large B-cell lymphoma (DLBCL) is a rare, but mostly fatal event. The CNS International Prognostic Index (CNS-IPI) represents widely adopted prognostic model. As previously shown, there is a different CNS relapse rate in different subgroups defined by gene expression profiling (GEP; GCB vs ABC; Klanova, 2019), as well as higher frequency of distinct molecular subtypes (MCD) among patients (pts) who developed CNS relapses (Ollila, 2021). Integration of COO by GEP into the CNS-IPI led to a modified prognostic CNS-IPI-C model (one point for high CNS-IPI, and one for ABC/unclass. COO), which identifies pts at high risk of CNS relapse more accurately (Klanova). However, COO assessed by GEP and mutational profile-defined DLBCL subtypes are still not available in daily practice. IHC-assessed COO has been tested as biomarker for CNS relapse as well (Savage, 2016). We aimed to verify if the GEP-defined COO in CNS-IPI-C could be replaced by IHC-defined COO in an CNS-IPI-IHC index in a real-world cohort of DLBCL pts. Methods Using the prospective NiHiL project (NCT03199066), we identified 1201 pts with histologically confirmed DLBCL, HG B-NHL, or THRBCL diagnosed between 2010-2021 who received R-CHOP as frontline regimen at two academic centers in the Czech Republic. Pts with CNS involvement at diagnosis were excluded. Out of 1201 pts, 932 (78%) pts had available COO of DLBCL assessed by IHC-based Hans' algorithm. The COO results were reviewed by two hematopathologists; subsequently 11 pts with unknown COO were excluded resulting in 921 (77%) pts entering the analysis. CNS-IPI-C was approximated using CNS-IPI (1 point for CNS-IPI 4-6) index and COO by IHC-based Hans' algorithm (1 point for non-GCB), resulting in CNS-IPI-IHC model (low-, LR = 0; intermediate-, IR = 1; high-risk, HiR = 2 points). Univariate analysis (UVA) was performed by log-rank test, cumulative incidence of CNS relapse by the Kaplan-Meier method, and multivariate analysis (MVA) by the Cox regression model. Results Out of 921 cases, 486 (53%) were classified as GCB and 435 (47%) as non-GCB DLBCL. According to CNS-IPI, 265 (29%), 398 (43%), and 258 (28%) pts were categorized as LR, IR, and HiR for developing CNS relapse, resp. According to CNS-IPI-IHC, 354 (38%), 441 (48%), and 126 (14%) were categorized as being at LR, IR, and HiR, resp. With a median follow-up time of 5.4 years (IQR 2.6-8.9 years), 32 (3.5%) pts developed CNS relapse (26 isolated CNS, 6 systemic and CNS relapses). Among these pts, 5 (16%) cases were categorized as LR for both CNS-IPI and CNS-IPI-C, 8 (25%) and 15 (47%) as IR, and 19 (59%) and 12 (38%) as HiR, resp. The CNS relapse rate was 3.0% at 2 years, and 3.7% at 5 years, with a median time to CNS relapse of 1.2 years (range 0.4-9.7 years). The 2-year PFS and OS of these 32 pts was 12.5% and 16.1%. The 5-year CNS relapse rates in pts with LR, IR, and HiR CNS-IPI vs CNS-IPI-IHC were 1.2% vs 1.0%, 2.2% vs 3.5%, and 9.8% vs 14.7%, resp. In UVA, there was trend for non-GCB DLBCL association with a HiR CNS relapse risk compared to GCB (5.5% vs 2.2% at 5 years, HR 1.96, 95% CI0.98 - 3.93, P = 0.059). HiR subgroups of both indexes were associated with higher CNS relapse risk in comparison to LR and IR subgroups (P < 0.001), but statistical significance between IR and LR was retained only in the CNS-IPI-IHC (P = 0.048). In MVA of factors associated with CNS relapse, HiR CNS-IPI was significantly associated with CNS relapse (HR 5.82, 95% CI 2.16-15.7, P < 0.001), while association of non-GCB DLBCL with CNS relapse did not reach statistical significance (HR 1.98, 95% CI 0.96-4.05, P = 0.063). Conclusion High CNS-IPI was associated with significantly higher risk of CNS relapse in this study. There was a trend towards non-GCB DLBCL to be associated with high CNS relapse risk. The CNS-IPI-IHC model (in which COO was assessed by Hans' algorithm) in comparison to CNS-IPI identifies a smaller subgroup of HiR pts (14% vs 28% of all DLBCL pts) with cumulative incidence of CNS relapse at 5 years 14.7% vs 9.8%. The size reduction of HiR group led to an incorrect classification of 7 (22%) pts who ultimately developed a CNS relapse to an IR CNS-IPI-IHC group. The implementation of IHC-based COO to a CNS-IPI prognostic model is limited in a daily practice; thus, using GEP or mutational analyses will be needed in the future to better stratify pts at different CNS relapse risk. First two authors contributed equally. Supported by CU Hem-Onco Cooperatio Program and grant NU21-03-00127.