BackgroundCNS dissemination is an uncommon but lethal event in non-Hodgkin lymphomas. Early detection of CNS disease and a timely and effective CNS prophylaxis are the main strategies to reduce related mortality. However, both the criteria for recognition of lymphoma patients (pts) with increased risk of CNS involvement and the most effective prophylaxis modality remain important, unmet clinical needs. Some international guidelines recommend intrathecal chemotherapy by lumbar injection as exclusive prophylaxis; however, this strategy results in erratic, short-lived drug bioavailability and does not prevent brain parenchymal relapses. Herein, we report a retrospective analysis of the value of clinical variables and immunohistochemical ontogenic stratification in predicting CNS dissemination and of risk-tailored CNS prophylaxis in a mono-institutional series of 194 pts with DLBCL treated in the rituximab era. MethodsConsecutive HIV- adults with DLBCL without CNS involvement at diagnosis treated with first-line rituximab-CHOP or similar ± radiotherapy were considered. Primary CNS, mediastinal and cutaneous leg-type lymphomas were excluded. ‘High risk' of CNS relapse was defined by the involvement of the testis, spine, skull, orbit, nasopharynx, kidney, and/or breast or by IPI ≥2 (including two among extranodal sites ≥2, advanced stage and high serum LDH). DLBCLs were ontogenically subclassified in ‘germinal-centre B-cell-like' (GCB) and ‘non-germinal-centre B-cell-like' (non-GC) by immunohistochemistry following the Hans algorithm. Results194 patients were analyzed (median age 65, range 18-89; M:F ratio 1.1). Risk of CNS relapse was low in 90 pts and high in 104. Low-risk pt did not receive CNS prophylaxis, while 40/104 (38%) high-risk pts received 3-4 courses of methotrexate 3 g/m2 ± intrathecal (IT) liposomal cytarabine (n=30), cytarabine 16 g/m2 in 4 days (n=2) or IT chemotherapy (n=8). In the high-risk group, IPI ≥2 was more common among pts who did not receive prophylaxis (89% vs. 68%; p=0.006), while “high-risk” extranodal lymphomas were more common among pts who did (88% vs. 33%; p= 0.0001). One hundred and forty-one cases were assessable for Hans algorithm: 74 (52%) were GCB and 67 (48%) were non-GCB DLBCL. GCB DLBCLs were significantly associated with low CNS risk (55% vs. 31%; p= 0.004), and normal LDH levels (57% vs. 36%; p= 0.02); ontogenic stratification was not associated with high-risk extranodal sites, IPI ≥2, bone marrow infiltration, stage and systemic symptoms.After first-line treatment, 160 pts achieved a CR (82%; 95%CI= 77-87%), 34 pts had PD. At a median follow-up of 60 months (13-156), a single low-risk pt and 9 high-risk pts (1% vs. 9%; p= 0.016) experienced CNS relapse (exclusive site in all cases; brain in 5 pts, meninges in 5), with a median TTP of 12 months (7-55). CNS relapses occurred in 3 pts with IPI ≥2, in 1 pt with extranodal disease (testis) and in 5 pts with both features (kidney 3; testis, orbit). Ontogenic stratification was not associated with CNS recurrence, which was 5% for GCB and 6% for non-GCB; these figures were confirmed when analysis was limited to high-risk pts managed without prophylaxis.In the high-risk group, CNS relapses occurred in 7/64 (11%) pts who did not receive prophylaxis, in 2/8 (25%) pts who received only IT chemotherapy, whereas no CNS relapses were detected in the 32 pts treated with intravenous (IV) prophylaxis. CNS relapse rate was 13% for pts treated with “inadequate” prophylaxis (none or IT only) and 0% (p= 0.03) for pts managed with IV prophylaxis.Eight pts with CNS relapses died of lymphoma after 7-37 months (median 12), which represented 28% of all lymphoma-related deaths (n=29) in the high-risk group. Pts treated with IV prophylaxis had a significantly better OS than the other high-risk pts (5-yr: 94 ± 7% vs. 49 ± 6%; p= 0.001). ConclusionsStratification by specific extranodal sites and IPI is superior to ontogenic stratification to recognize CNS risk groups in DLBCL. However, the low sensitivity of predictive clinical variables suggests that molecular studies focused on the predictive and pathogenic role of molecules involved in CNS tropism will contribute to a more accurate definition of lymphoma candidates for CNS-directed strategies. In this context, IV high-dose methotrexate-based prophylaxis may significantly reduce CNS failures in high-risk pts. Disclosures:No relevant conflicts of interest to declare.