Résumé
Background: Relapsed/refractory primary mediastinal large B-cell lymphoma (R/R PMBL) remains a therapeutic challenge, with limited responses to conventional salvage therapies. CAR-T cell therapy has emerged as a potential second-line (2L) option, but access may be restricted. Anti-PD1 checkpoint inhibitors (CPIs) have shown promising efficacy in clinical trials (KEYNOTE-170, CHECKMATE-436), but they lack European regulatory approval and real-world data in PMBL. We conducted a multicenter retrospective study to evaluate outcomes in R/R PMBL patients treated with CPIs.
Methods: This study included R/R PMBL patients treated with CPIs, either alone or combined with other agents, across 20 LYSA-affiliated centers in France from 2014 to 2025. Tumor responses were assessed using PET or CT imaging, according to the 2014 Lugano criteria. The best overall response rate (bORR), defined as the proportion of patients achieving complete (CR) or partial (PR) response, was reported. Progression-free survival (PFS) and overall survival (OS) were estimated using the Kaplan-Meier (KM) method. Median follow-up was calculated using the reverse KM approach.
Results: A total of 100 patients were included (median age 31 years, range 18–82; 51% female), with 98% presenting with a mediastinal mass, 58% with localized disease (stage I–II), 86% with ECOG PS 0–1, 10% with B symptoms, and 9% with a CNS-IPI score ≥4. Seven patients (7%) had central nervous system (CNS) involvement at inclusion. Most patients (95%) were refractory to their last line, with a median of 2 prior regimens (range 1–5), including R-CHOP14 (34%), R-CHOP21 (15%), DA-EPOCH-R (16%), R-ACVBP (25%), and others (10%) as first line. In 2L, 84% received platinum-based chemotherapy; prior autologous SCT and CAR-T were given in 3 and 7 patients, respectively.
CPIs included pembrolizumab (n=66), nivolumab (n=33), and atezolizumab (n=1), administered as monotherapy (n=39), in combination with brentuximab vedotin (Bv; n=52), or with other agents (n=9). The median number of CPI cycles was 6 (IQR 3–15; range 1–52), and of Bv cycles was 4 (IQR 2–8; range 1–36). Consolidation therapy (57%) included radiotherapy (RT, n=20), autologous SCT (n=9), allogeneic SCT (n=3), and anti CD19 CAR-T cells (n=25).
Treatment was generally well tolerated: 72% of patients experienced no adverse events, and most toxicities were low-grade, immune-related or infectious, consistent with previous reports.
The bORR was 84%, including 48% CR and 36% PR. bORR by treatment subgroup was 89% with CPI monotherapy (CR 47%, PR 42%), 78% with CPI+Bv (CR 56%, PR 22%). The median time to best response was 2.9 months for monotherapy, 2.2 months for CPI+Bv.
After a median follow-up of 31.8 months (IQR 11.9–51.8), median PFS and OS were not reached. Estimated 2-year PFS and OS rates were 56.6% [IC95: 46.9-68.2] and 75.5% [IC95: 66.5-85.7], respectively. A total of 27 deaths were reported; main cause of death was lymphoma (92.3%). Among the 48 patients who achieved CR, only 3 relapsed (median time to relapse: 12.3 months). Additionally, 54% of patients initially in PR converted to CR over time.
In univariate Cox analysis, poor OS was associated with ECOG ≥2 (HR 3.88 [95% CI 1.53–9.84], advanced stage (HR 2.73 [1.25–5.97], CNS-IPI ≥4 (HR 6.66 [2.26–19.64], B symptoms (HR 4.18 [1.38–12.65] and LDH>ULN (HR 4.08 [1.21–13.76]. Hb ≥10.5 g/dL (HR 0.30 [0.14–0.67], platelet count ≥100 G/L (HR 0.20 [0.08–0.51], and female sex (HR 0.41, [0.18–0.94] were associated with better OS. In multivariate analysis, CNS-IPI was independently associated with shorter OS (HR 4.99 [2–3] and 4.48 [4–5]), while female sex (HR 0.19) and Hb ≥10.5 g/dL (HR 0.17) were linked to longer OS.
Predictors of shorter PFS in univariate analysis included ECOG ≥2 (HR 3.98 [1.92–8.25]), advanced stage (HR 2.50 [1.30–4.80]), CNS-IPI ≥4 (HR 5.33 [2.19–12.98]), B symptoms (HR 4.41 [1.90–10.26]), and LDH>ULN (HR 3.11 [1.21–8.03]). Conversely, Hb ≥10.5 g/dL (HR 0.38 [0.20–0.75]) and platelets ≥100 G/L (HR 0.31 [0.13–0.70]) were associated with longer PFS. In multivariate analysis, only CNS-IPI remained independently associated with shorter PFS (HR 4.59 [2.09–10.08] for score 2–3; HR 4.78 [1.34–17.05]) for score ≥4.
Conclusion: This large real-world cohort confirms the high efficacy and sustained responses of CPIs in R/R PMBL. These results support broader and earlier integration of CPIs in the treatment strategy for this high-risk population.