CNS Relapse in T-Cell Lymphoma Index: A Risk Score to Predict Central Nervous System Relapse in Patients with T-Cell Lymphomas

Background: Central nervous system (CNS) relapse (CNSr) in patients with aggressive non-Hodgkin lymphoma is uncommon but carries a high morbidity and mortality. Data in T- and NK-cell lymphomas (TCL) are sparse, though the few small case series describing CNSr in TCL report an incidence up to 9% with median overall survival (mOS) less than 3 months (mo). We previously reported a large (n=75) case series of CNSr in TCL (Abstract #1382, ASH 2021). Herein, we developed a prediction score for CNSr in newly diagnosed patients with TCL.

Methods: We retrospectively collected clinicopathologic and treatment data from 19 US academic centers from patients diagnosed with TCL between 1/1/09-1/1/19 who experienced CNSr (CNSr cohort). We combined these data with that from patients diagnosed with TCL without CNSr from a single institution (Penn cohort) to create a training set (T-set) enriched for patients with CNSr. Patients with leukemic or cutaneous TCL subtypes were excluded. We applied a LASSO Cox proportional hazards model to the T-set to create a predictive score for factors associated with CNSr and validated this score in an independent population-based validation set (V-set) from the Swedish Lymphoma Registry, which was not enriched for CNSr.

Results: We evaluated data from 91 and 135 patients in the CNSr and Penn cohorts, respectively, which were combined to form the T-set and compared them to the V-set (n=745). Median age in the T-set was lower than in the V-set (60 vs 67; p<0.001). The most common TCL subtypes in the T-set and V-set were peripheral TCL, not otherwise specified (PTCL, NOS; 23% vs 34%, respectively; p=0.002), TCL with T-follicular helper phenotype (23% vs 14%, respectively; p=0.001), and ALK-negative anaplastic large cell lymphoma (18% vs 15%, respectively; p=0.352). CNS prophylaxis, defined as intrathecal therapy or high-dose methotrexate, was more common in the V-set than in the T-set (15% vs 9%, respectively; p=0.012). Rates of stem cell transplantation (SCT) as consolidation in first remission (CR1) did not differ significantly between the T-set and V-set (p=0.140). The median follow-up time was 37.6 mo in the T-set and 95.9 mo in the V-set. The median progression free survival (mPFS) and mOS after diagnosis were 9.8 mo and 33.2 mo in the T-set, respectively, and 8.5 mo and 17.9 mo in the V set, respectively. Within the T-set, mPFS and mOS were shorter for patients in the CNSr cohort vs the Penn cohort (6.0 mo vs 37.1 mo [p<0.001] and 17.3 mo vs 78.1 mo [p<0.001], respectively). In univariate analysis of the CNSr-enriched T-set, CNS prophylaxis, SCT in CR1, or frontline etoposide did not significantly decrease the risk of CNSr.

Using several clinicopathologic characteristics, we fitted a LASSO Cox proportional hazards model to the T-set, which selected histology, LDH, stage, B-symptoms, and ≥2 sites of extranodal involvement (ENI) for a weighted risk score of CNSr (Fig. 1A). Variables most strongly associated with CNSr were PTCL, NOS (HR 2.64), enteropathy-associated TCL (HR 2.92), and ≥2 sites of ENI (HR 2.66). We stratified the T-set into tertiles based on weighted scores (x), which was low-risk if x≤0.14, intermediate-risk if 0.14<x ≤1.11, or high-risk if x>1.11. Cumulative incidence of CNSr in each respective group was 19.6%, 55.1%, and 83.7%. When the same cutoffs were applied to weighted scores in the V-set, we confirmed separation into distinct low-, intermediate-, and high-risk groups with cumulative incidences of CNSr of 1.9%, 4.9%, and 7.3%, respectively (Fig. 1B). As the V-set population was not enriched for CNSr, the relative incidence is expectedly lower compared to the T-set. Compared to the low-risk group in the V-set, risk of CNSr was significantly increased in the high-risk group (HR 5.72, p=0.002) and increased with a trend toward significance in the intermediate-risk group (HR 3.35, p=0.057).

Conclusions: In a cohort of patients with TCL enriched for CNSr, we were able to characterize several risk factors associated with CNSr and developed a predictive CNSr in T-cell lymphoma Index (CITI) that could stratify patients into 3 distinct risk groups. PTCL, NOS, EATL, and ≥2 sites of ENI had the highest impact on risk of CNSr. The CITI score was validated in an independent population-based cohort. Although CNSr was uncommon even in high-risk patients (7.3%), our data will inform clinical decision making and may allow for identification of patients with TCL at high-risk of CNSr in future studies.