Evolving consolidation patterns and outcomes for a large cohort of patients with primary CNS lymphoma

Primary central nervous system (CNS) lymphoma (PCNSL) is an aggressive non-Hodgkin lymphoma originating in the CNS, with diffuse large B-cell lymphoma comprising the most common subtype (>95% of cases). Although PCNSL is rare, the incidence has been increasing.^1,2 The standard treatment approach involves a high-dose methotrexate–containing induction regimen followed by consolidation therapy. There are several guideline-supported consolidation strategies for PCNSL including whole-brain radiotherapy (≤24 Gray [Gy] reduced-dose [RD-WBRT]; or >24 Gy standard-dose [SD-WBRT]) with cytarabine, nonmyeloablative chemotherapy (NMC) such as cytarabine alone, or autologous hematopoietic stem cell transplantation (AHCT).^3,4 WBRT has a long history of efficacy in this disease; however, neurotoxicity is a major concern with historically used doses of 45 Gy, particularly in patients aged >60 years.⁵ Reduced doses of WBRT have shown encouraging results^6,7; however, use remains low⁸ likely because of lingering toxicity concerns. Meanwhile, significant enthusiasm for AHCT as a consolidation strategy has grown, given high complete response rates and encouraging early progression-free survival (PFS).^9-14 Prospective trials comparing AHCT and RD-WBRT have not been performed.

In the modern era, there is a range of consolidation options and patient selection is typically guided by comorbidities, functional status, disease characteristics, and treatment toxicity profile. With multiple consolidation strategies and the evolution of clinical trial data, current understanding of practice patterns and outcomes by treatment is limited. The aims of this study were to (1) analyze longitudinal patterns of PCNSL consolidation use in a large cohort, and (2) assess clinical outcomes of the different consolidation strategies in a homogeneously treated population.

Immunocompetent patients with PCNSL (histologically confirmed diffuse large B-cell lymphoma of the brain, cerebrospinal fluid [CSF], or eyes/ocular compartments) diagnosed from 1983 to 2020 were identified from an institutional database (supplemental Figure 1). Patients were excluded if they had primary low-grade or T-cell lymphoma, evidence of systemic disease outside the CNS at initial presentation (ie, secondary CNSL), or lacked at least 1 radiographic (computed tomography or magnetic resonance imaging [MRI]) assessment after therapeutic initiation. Patient demographics and Memorial Sloan Kettering Cancer Center (MSKCC) recursive partitioning analysis (RPA) classification were evaluated.¹⁵ This retrospective study was approved by the institutional review board with waived patient consent.

Baseline CSF status was recorded when available. Induction and consolidation regimens were collected for first-line therapy. Specifically, first-line consolidation regimens were categorized as AHCT, WBRT with cytarabine, NMC alone, or other. WBRT was further classified as standard-dose (SD-WBRT) if >24 Gy or reduced-dose (RD-WBRT) if ≤24 Gy. Of note, NMC at our institution is typically cytarabine dosed at 3 g/m² on 2 consecutive days for 2 28-day cycles. For AHCT, most (n = 74, 93%) patients received thiotepa-based conditioning (n = 73 thiotepa/busulfan/cyclophosphamide, n = 1 thiotepa/carmustine) and few received carmustine/etoposide/cytarabine/melphalan conditioning (n = 6, 7%). Common consolidation regimens categorized under “other” included maintenance methotrexate, maintenance ibrutinib, and focal radiotherapy. Patients with radiographic or clinical progression of disease (POD) during induction or early relapse within 3 months of induction were classified as primary refractory. For patients who progressed before consolidation, each subsequent therapy received was classified as salvage (not consolidative) intent. Maintenance therapy without intercurrent POD was considered consolidation. Surgical debulking beyond the extent of a diagnostic biopsy was classified as resection.

Neuroaxis imaging (MRI and/or computed tomography) was analyzed at initial presentation and then longitudinally until first relapse or final available scan (for censoring). If the original images were not available for direct interpretation, the formal radiology report text was analyzed. Disease response was classified per International PCNSL Collaborative Group guidelines on standardized evaluation of patients with PCNSL: complete or unconfirmed (CR/CRu), partial (PR), stable, or POD.¹⁶ 

Median follow-up was calculated by reverse Kaplan-Meier method. Multinomial logistic regression models were performed to evaluate factors associated with receipt of specific consolidation regimens. Variables significant on univariable logistic regression at level P = .05 were included in the multivariable model. Time was categorized by decade of diagnosis (eg, 1980s, 1990s, 2000s, and 2010-2020). PFS and overall survival (OS) were estimated using the Kaplan-Meier method and calculated from the date of diagnosis or consolidation start. The reference group for consolidation strategies was NMC, given this was the largest group. Multivariable Cox proportional hazards models were performed incorporating all clinical variables (except for “receipt of rituximab” given “induction regimen” was included instead) given clinical relevance for both PFS and OS. A landmark analysis at 6 months after diagnosis was performed to compare PFS and OS between receipt of any consolidation vs none. Patients who died or had last follow-up before the 6-month landmark time were excluded from the analysis. Longitudinal consolidation use patterns were analyzed using Fisher exact test. Consolidation strategies and outcomes were stratified by decade and MSKCC RPA class 1 (age of <50 years), 2 (age of ≥50 years, Karnofsky performance status (KPS) score of ≥70), or 3 (age of ≥50 years, KPS score of <70). Clinicodemographic associations with consolidation strategy were analyzed using multinomial logistic regression.

Of 645 evaluated patients, 559 were eligible and 385 (69%) were consolidated (Table 1). Median follow-up from diagnosis was 7.4 years (95% confidence interval [CI], 6.4-8.6), and median OS was 5.7 years (95% CI, 4.9-7.4). Median age was 63 years (interquartile range, 54-72). Most (n = 321, 57%) patients were RPA class 2. RPA class was significantly associated with both PFS and OS (Figure 1). Most patients (n = 532, 95%) received a methotrexate–containing induction regimen, mostly R-MPV (rituximab, methotrexate, procarbazine, and vincristine; n = 257, 46%) followed by MPV (n = 144, 26%). Induction regimens categorized under “other” included methotrexate monotherapy (n = 54, 10%), other combination methotrexate–containing combination therapy (n = 77, 14%), and nonmethotrexate-containing therapy (n = 27, 5%).

Among the 385 who received consolidation, most received R-MPV for induction (n = 211, 55%) and most received NMC for consolidation (n = 167, 43%) followed by AHCT (n = 79, 21%). Most patients presented with bilateral (n = 157, 44%), multifocal (n = 188, 52%), and/or supratentorial disease (n = 299, 83%); and 117 patients (20.9%) presented with leptomeningeal disease defined by MRI or CSF. Of the 351 who achieved a CR/CRu to consolidation (91% of 385 consolidated), median OS and PFS from the start of consolidation were 9.2 years (95% CI, 7.4-12.0) and 4.6 years (95% CI, 3.4-5.8), respectively (Table 2). In those who achieved CR/CRu, 1-, 2-, and 5-year OS rates from the start of consolidation were 93% (95% CI, 90-96), 85% (95% CI, 81-89), and 65% (95% CI, 59-71), respectively; 1-, 2-, and 5-year PFS rates were 80% (95% CI, 76-84), 67% (95% CI, 62-72), and 46% (95% CI, 41-53), respectively. For this cohort with CR/CRu to consolidation, those who received R-MPV had the most favorable outcomes (median OS and PFS of 12 years and 7 years, respectively) compared with those who received MPV (median OS and PFS of 6.2 years and 3.0 years, respectively) or other regimens (median OS and PFS of 4.8 years and 3.0 years, respectively).

Over the study period, there was significant change in consolidation with declining use of WBRT (61% in 1990s vs 12% in 2010s) and rising use of AHCT (4% in 1990s vs 32% in 2010s) and NMC (27% in 1990s vs 52% in 2010s; Figure 2A, P < .001). This temporal evolution was seen across RPA classes, with greater NMC usage for higher RPA in recent time (Figure 2B-D).

Controlling for decade of diagnosis and response to induction, compared with RPA class 1, RPA class 2 was significantly less likely to receive AHCT (odds ratio [OR], 0.23; P = .001), RD-WBRT (OR, 0.31; P = .02), or SD-WBRT (OR, 0.08; P < .001) vs NMC (Table 3). Similarly, compared with RPA 1, RPA class 3 was less likely to receive AHCT (OR, 0.08; P < .001) or SD-WBRT (OR, 0.05; P < .001) vs NMC. Those with PR to induction were also less likely to receive AHCT (OR, 0.36; P = .02) vs NMC.

Among the 351 with CR to consolidation, on univariable analysis, AHCT (hazard ratio [HR], 0.37 [95% CI, 0.23-0.60], P < .001) and RD-WBRT (HR, 0.52 [95% CI, 0.35-0.79], P = .002) were both significantly associated with improved PFS compared with NMC (Table 4). However, on multivariable analysis accounting for decade of diagnosis, induction regimen, and the interaction between consolidation strategy and RPA, only receipt of R-MPV induction was associated with improved PFS (HR, 0.52; P = .006). On sensitivity analysis excluding the 6 patients who had received a nonmethotrexate–containing induction regimen, receipt of R-MPV was still associated with improved PFS (HR, 0.50; P = .004). There was no significant PFS association with any consolidation strategy. However, on 6-month landmark analysis, compared with no consolidation, receipt of any consolidation was associated with improved PFS (P < .001; Figure 3). For RPA class 1, median PFS for AHCT and RD-WBRT were not reached, and median PFS for SD-WBRT and NMC were 7.6 years (95% CI, 3.2-17.0) and 2.5 years (95% CI, 1.1 to not reached [NR]), respectively (Figure 4). For RPA class 2, median PFS was numerically highest in AHCT and RD-WBRT (9.4 years [95% CI, 9.4-NR] and 6.7 years [95% CI, 4.1-NR], respectively), followed by SD-WBRT and NMC (3.5 years [95% CI, 2.3-7.4] and 3.0 years [95% CI, 2.1-6.0], respectively). For RPA class 3, the median PFS was numerically greatest among RD-WBRT (4.6 years [95% CI, 1.4-NR]) followed by NMC (1.7 years [95% CI, 1.0-4.6]). On further stratified pairwise analysis, RD-WBRT had significantly improved PFS compared with NMC for patients with RPA 1 and RPA 2 (supplemental Figure 2). Similarly, on stratified, unadjusted analysis PFS outcomes were significantly improved in patients receiving AHCT vs NMC for RPA 1 and 2 (supplemental Figure 3). On pairwise analysis comparing RD-WBRT and AHCT, although there was a trend toward improve outcomes with RD-WBRT, the difference was not statistically significant among RPA 3. Of note, only 5 patients received AHCT in this subset (supplemental Figure 4).

On 6-month landmark analysis, compared with no consolidation, receipt of any consolidation was associated with improved OS (P < .001; Figure 3). On multivariable analysis, patients who received a consolidation strategy other than AHCT, WBRT, or NMC had worse OS (HR, 2.95; 95% CI, 1.36-6.39; P = .006). By multivariable analysis, the only factor identified associated with OS was RPA class 3 (vs class 2; HR, 1.8; P = .034; Figure 2). For RPA class 1 and 2, among all consolidation strategies, those treated with AHCT and RD-WBRT had similar, numerically best OS outcomes (Figure 3). For RPA class 1, median OS was not reached for AHCT or RD-WBRT and was 13 years after NMC (95% CI, 13-NR). For RPA class 2, median OS was 13 years after RD-WBRT and cytarabine (95% CI, 7.9-NR), 9.4 years after AHCT (95% CI, 9.4-NR), and 7.7 years after NMC (95% CI, 6.0-13.0). On stratified, unadjusted pairwise analyses, there was no statistically significant difference between either AHCT or RD-WBRT vs NMC for RPA class 2 (supplemental Figures 2 and 3).

We present the largest single-institutional analysis of patients with PCNSL, summarizing changing practice patterns and clinical outcomes with regards to consolidation strategy. This analysis is critical to evaluate the past and current landscape for PCNSL consolidation to evaluate whether the direction of practice is consistent with optimizing clinical outcomes. Here, we show upon adjusting for clinicodemographic variables, PFS and OS outcomes were comparable across consolidation strategies. Importantly, receipt of any consolidation was associated with both improved PFS and OS compared with none. Interestingly, we show increased use of AHCT, especially among the fittest patients, which is supported by the numerically favorable PFS outcomes compared with NMC. However, we do show increased use of NMC, especially in less fit patients, with decreased use of RD-WBRT despite numerically improved PFS outcomes with RD-WBRT vs NMC. This incongruent practice pattern with clinical outcomes has important implications for how we may reconsider consolidation in AHCT-ineligible patients.

In evaluating consolidation use patterns over nearly 40 years at a high-volume PCNSL center, we revealed significant decline in the use of WBRT and a rise in both AHCT and NMC. Patient age and performance status (ie, RPA) were factors contributing to consolidation selection, and this shift away from WBRT consolidation was most prominent in the RPA 2 subgroup (age ≥50 years, KPS score ≥70). In 2010, prospective data from G-PCNSL-SG-1 revealed significant neurotoxicity with doses of 45 Gy, therefore this shift away from SD-WBRT likely reflects appropriate concerns regarding neurotoxicity.⁵ Moreover, preliminary randomized data after a median follow-up of 55 months from NRG-RTOG 1114 comparing R-MPV+ cytarabine with or without RD-WBRT showed median intent-to-treat PFS was significantly improved in the arm receiving RD-WBRT consolidation vs NMC alone (HR, 0.51; 95% CI, 0.27-0.95; P = .015).⁶ Importantly, there were no significant differences in terms of neurotoxicity as reported by the investigators; however, completing neuroimaging and neurocognitive assessments were not provided in their preliminary report. Although OS results are maturing, these data support previous prospective data published in 2007 showing RD-WBRT consolidation after R-MPV was not associated with neurocognitive decline and was associated with excellent disease control (overall response rate of 93%).¹⁷ Given the potential for neurologic compromise with PCNSL relapse and potential morbidity of multiple lines of CNS-directed therapy, durable remission as estimated by PFS is a meaningful end point. Considering these data, the shift away from all forms of WBRT (including RD-WBRT) toward NMC alone identified in our cohort may not be the optimal trend in certain cases. In our series, the stratified analyses showed significantly improved PFS outcomes with RD-WBRT vs NMC for RPA 1 and 2, in addition to numerically improved outcomes for RPA 3. Although we acknowledge that the multivariable analyses show comparable outcomes across consolidation strategies, this may be related to inadequate statistical power, and the visible differences are worth noting. RD-WBRT also had improved PFS compared with non-AHCT consolidation approaches and no consolidation receipt in the 6-month landmark analysis, and a significant difference was seen in receipt of any consolidation compared with none. Although selection bias is likely at play given the rationale for withholding consolidation (ie, POD, death), this finding reemphasizes the importance of consolidation in PCNSL.

Fitter patients and those with CR to induction were more likely to receive AHCT in our analysis. This selection of favorable patients likely reflects weighing potential treatment-associated toxicities with superior outcomes achieved with AHCT vs NMC or SD-WBRT.^9,18 Updated results from the randomized phase 3 MATRix/IELSG43 trial showed that consolidative AHCT is superior to NMC-immunotherapy in patients aged ≤70 years.¹⁸ Specifically, PFS and OS were significantly higher after AHCT compared with R-DeVIC (rituximab, dexamethasone, etoposide, ifosfamide, and carboplatin) despite similar remission rates after consolidation. There were no differences in neurocognitive outcomes between the groups. Given the emphasis on achieving CR before AHCT, perhaps additional emphasis should be placed on optimizing induction and potentially bridging regimens, such as those used in extracranial lymphomas, to achieve adequate treatment sensitive disease before AHCT in eligible patients.¹⁹ Recently, the long-term update of the French PRECIS study that compared 40 Gy WBRT vs thiotepa-busulfan-cyclophosphamide AHCT showed improved event-free survival and neurocognitive outcomes for AHCT in younger patients.⁹ To date, there have been no randomized trials comparing AHCT and RD-WBRT, and, as stated, the neurocognitive changes are dose dependent. In our study, consolidation strategy was not independently significantly associated with OS or PFS; however, AHCT and RD-WBRT numerically had the most favorable outcomes and on 6-month landmark analysis, the 2 strategies appeared to have similar OS. For patients who are frail (ie, RPA class 3) the median PFS appeared highest among RD-WBRT (4.6 years [95% CI, 1.4-NR]). These findings suggest that in patients deemed ineligible for AHCT, RD-WBRT can be more strongly considered as an alternative to NMC.

Future directions in this disease are focused on reducing toxicity and improving outcomes via novel therapies. There are efforts to improve the safety and tolerability of the conditioning regimen before AHCT, particularly in patients aged ≥70 years.²⁰ Although the study presented here focused on consolidation approaches, efforts are also being made to optimize induction regimens, such as shortening therapy to reduce toxicity as is currently being investigated in the randomized OptiMATe trial.²¹ Additionally, novel agents like Bruton's tyrosine kinase inhibitors are currently tested in the first-line setting to further increase complete response rates. Chimeric antigen receptor (CAR) T-cell therapy has been studied primarily in the relapsed/refractory PCNSL setting and initial studies have been promising, with early data demonstrating a CR rate of 60% and improved relapse-free survival in those who achieve CR vs non-CR to CAR T-cells (response rate of 85% at 30 months after infusion).²² This study also showed improved outcomes in those who achieved CR/PR before CAR T-cells (response rate of 77% in CR/PR vs 22% in stable disease/POD), highlighting the potential role for effective bridging therapy to improve post–CAR T-cell outcomes in these patients. In the future, consideration of multimodal consolidation regimens personalized to an individual’s age, performance status, comorbidities, and response to induction therapy may be necessary to optimize outcomes. For instance, incorporating involved-site radiotherapy as opposed to WBRT to reduce normal tissue toxicity in patients with localized refractory disease may render patients in remission to increase eligibility for novel and effective therapies such as AHCT or CAR T-cell therapy.^23,24 

This study has several potential limitations. Because of the retrospective study design over the course of decades, it is difficult to account for all patient characteristics and nuances of treatments. For instance, consolidation regimens here were defined as the first-line therapy after induction and therefore do not include a broader base of patients who received any of these regimens as later lines of therapy or after initial progression. In addition, intention of treatment (ie, palliative vs curative) is difficult to ascertain retrospectively; however, we attempted to account for this via sensitivity analyses excluding patients who received nonmethotrexate–containing induction regimens. This series may have inherent selection bias in that it is a large, single institutional series of MSKCC patients, and therefore the outcomes may not be generalizable to other institutions or practices. For example, our patient cohort is slightly younger than previously reported registries of patients with PCNSL, possibly because older patients not suitable for treatment may not be referred to our center. Patients were not limited to receiving treatment at a single institution, so it is possible that follow-up points do not include some missing records. In addition, most patients (92%) received thiotepa/busulfan/cyclophosphamide conditioning, which is the most intensive AHCT conditioning regimen, and may possibly explain why fitter patients received AHCT. In more recent years, there has been a shift toward using more tolerable conditioning regimens (ie, thio-carmustine) to treat older patients, and therefore patient selection may continue to evolve with ever-changing practices.

From the 1980s to the modern era, AHCT has been increasingly used for PCNSL consolidation, especially for fitter patients. Although consolidation strategies were statistically comparable on multivariable analysis of PFS and OS, AHCT did have favorable outcomes statistically on pairwise PFS comparisons vs NMC. NMC appeared not as effective for most patients, but it was increasingly used, in lieu of RD-WBRT for patients with poor performance status. Importantly, we show encouraging PFS outcomes with RD-WBRT that support preliminary data from the ongoing NRG-RTOG 1114 trial. Our data suggest RD-WBRT can be considered as an effective alternate strategy for AHCT-ineligible patients (or those who decline AHCT) who would otherwise receive NMC.

The study was supported by a research grant from the National Institutes of Health for National Cancer Institute Cancer Center (P30 CA008748) as well as a research grant from the Lacher Lymphoma Foundation and the Connecticut Community Foundation.

Contribution: K.R.T. is responsible for concept development, data collection and organization, data analysis, data interpretation, manuscript writing, table and figure generation, and manuscript editing; M.S. is responsible for concept development, data collection, data interpretation, manuscript writing, and manuscript editing; J.Y. is responsible for data collection, data interpretation, and manuscript editing; C.W. is responsible for data analysis, data interpretation, manuscript writing, table and figure generation, and manuscript editing; Z.Z. is responsible for data analysis, data interpretation, manuscript writing, table and figure generation, and manuscript editing; J.S., G.C., and L.R.S. are responsible for data interpretation and manuscript editing; L.D. is responsible for data collection and organization and manuscript editing; B.S.I. is responsible for is responsible for concept development, data interpretation, manuscript writing, and manuscript editing; and C.G. is responsible for concept development, data collection and organization, data interpretation, manuscript writing, and manuscript editing.

Conflict-of-interest disclosure: M.S. served as a paid consultant for McKinsey & Company, Angiocrine Bioscience Inc, and Omeros Corporation; received research funding from Angiocrine Bioscience Inc, Omeros Corporation, and Amgen Inc; served on ad hoc advisory boards for Kite (a Gilead company); received honoraria from i3Health, Medscape, and CancerNetwork for CME-related activity; and received honoraria from IDEOlogy. L.R.S. reports scientific advisory board participation for BTG, plc; reports research funding from BTG, plc, Merck, and DebioPharm; reports consultant fees from ONO; and has a patent pending related to the use of low-dose glucarpidase. B.S.I. reports professional services related to GT Medical Technologies Inc, Ono Pharma, and Telix Pharmaceuticals Limited (uncompensated); and receives research funding outside this work (to the institution) from AstraZeneca, Novartis, Bayer, and Kazia Therapeutics. C.G. reports consulting services from BTG International, Curis, Ono Pharma, and Roche; and reports speakers bureau services for Ono Pharma and Curis. The remaining authors declare no competing financial interests.

Correspondence: Christian Grommes, Department of Neurology/Neuro-oncology, Memorial Sloan Kettering Cancer Center, 1275 York Ave, New York, NY 10065; email: grommesc@mskcc.org.