← Back
NeuroTrials.ai
Neurology Clinical Trial Database

MATRix/IELSG43

High-dose chemotherapy followed by autologous stem-cell transplantation versus non-myeloablative consolidation in primary CNS lymphoma (MATRix/IELSG43): a randomised phase 3 trial

Year of Publication: 2026

Authors: Illerhaus G, Ferreri AJM, Wendler J, et al.

Journal: The Lancet

Citation: Lancet 2026;408(10554):532-544

Link: https://doi.org/10.1016/S0140-6736(26)00917-7

Bottom Line

In fit patients with newly diagnosed PCNSL who respond to MATRix induction, thiotepa-based HCT-ASCT consolidation nearly halved the risk of progression or death and yielded superior 3-year PFS (78% vs 51%) and OS (86% vs 71%) compared with non-myeloablative R-DeVIC, at the cost of substantially higher toxicity — establishing HCT-ASCT as the preferred consolidation strategy.

Major Points

  • Largest randomised phase 3 trial ever conducted in untreated PCNSL (229 randomised of 346 who started induction, from 368 enrolled) across 56 European centres.
  • Uniform MATRix induction (rituximab, methotrexate 3.5 g/m2, cytarabine 2 g/m2 BID, thiotepa 30 mg/m2 x 4 cycles) was used in all patients, allowing isolation of the consolidation effect.
  • Primary endpoint met: 3-year PFS 78% (HCT-ASCT) vs 51% (R-DeVIC); HR 0.43 (95% CI 0.27-0.68), p=0.0003.
  • Overall survival was also significantly prolonged with HCT-ASCT (3-year OS 86% vs 71%; HR 0.46, 95% CI 0.26-0.81, p=0.0075).
  • 3-year cumulative relapse incidence was more than halved: 19% (HCT-ASCT) vs 45% (R-DeVIC); SHR 0.40 (95% CI 0.24-0.66).
  • Toxicity of HCT-ASCT was substantially greater (mean 14.6 vs 9.3 AEs/patient; febrile neutropenia 61% vs 14%; stomatitis 79% vs 9%; diarrhoea 59% vs 7%).
  • Treatment-related mortality during consolidation: 5 deaths in the HCT-ASCT arm (4 infections, 1 pulmonary embolism) vs 2 in the R-DeVIC arm (both secondary AML).
  • One-third of enrolled patients (116/346) did not reach randomisation, mainly due to induction toxicity or progression — underscoring the need to optimise induction.
  • Quality-of-life (EORTC QLQ-C30) trends favoured HCT-ASCT over time; detailed neurocognitive analysis will be reported separately.

Design

Study Type: Randomized Controlled Trial (Phase 3)

Randomization: 1

Blinding: Open-label (central neuroradiological response review was blinded to treatment allocation)

Enrollment Period: July 16, 2014 - Aug 31, 2019

Follow-up Duration: Median 45.3 months (IQR 31.7-58.3)

Centers: 56

Countries: Germany, Italy, Denmark, Norway, Switzerland

Sample Size: 229

Analysis: Intention-to-treat (full analysis set, excluding patients with major violations of entry criteria or who did not reach at least partial response after induction). Safety analysis set: 219 randomised patients who initiated consolidation therapy.

Trial Registration: ClinicalTrials.gov NCT02531841; EudraCT 2012-000620-17

Randomisation Timing: 1:1 after induction, in patients reaching at least partial response with adequate stem-cell harvest (>=3 x 10^6 CD34+ cells/kg); stratified by response status (CR vs PR)


Inclusion Criteria

  • Untreated, immunocompetent patients with B-cell primary CNS lymphoma (PCNSL)
  • Age 18-65 years regardless of ECOG performance status
  • Age 66-70 years with ECOG performance status 0-2
  • B-cell lymphoma confirmed on biopsy, CSF cytology, or vitrectomy (with central neuropathological review)
  • Pretreatment corticosteroids permitted
  • Adequate cardiac, pulmonary, hepatic, and renal function
  • For randomisation: at least partial response after 4 cycles of MATRix induction plus stem-cell harvest >=3 x 10^6 CD34+ cells/kg
  • Written informed consent

Exclusion Criteria

  • Lymphoma manifestation outside the CNS
  • Congenital or acquired immunodeficiency, including HIV seropositivity
  • Renal impairment (creatinine clearance <60 mL/min by MDRD)
  • Inadequate cardiac, pulmonary, or hepatic function at investigator discretion
  • Active hepatitis B or C virus disease
  • Isolated ocular lymphoma without brain parenchymal or spinal cord involvement
  • Symptomatic coronary artery disease, uncontrolled cardiac arrhythmias, or myocardial infarction within the prior 6 months

Baseline Characteristics

CharacteristicControlActive
GroupR-DeVIC (n=115)HCT-ASCT (n=114)
Median Age (years)60 (IQR 54-64)58 (IQR 51-64)
Age >=65 years24% (28/115)20% (23/114)
Sex - Male54% (62/115)57% (65/114)
Sex - Female46% (53/115)43% (49/114)
ECOG 016% (18/115)22% (25/114)
ECOG 161% (70/115)49% (56/114)
ECOG 214% (16/115)25% (29/114)
ECOG 36% (7/115)3% (3/114)
ECOG 43% (3/115)0
Increased serum LDH63% (72/115)69% (79/114)
Neurological symptoms at diagnosis92% (106/115)92% (105/114)
DLBCL histology98% (111/113)97% (111/114)
CD20 positive99% (113/114)100% (114/114)
Multiple brain lesions64% (72/113)57% (65/114)
Supratentorial only74% (83/113)72% (81/114)
Infratentorial only7% (8/113)4% (5/114)
Both supra- and infratentorial19% (21/113)24% (27/114)
Meningeal involvement4% (5/115)4% (5/114)
Possible ocular infiltration (slit lamp)3% (4/115)10% (11/114)
Positive CSF cytology10% (11/115)4% (5/114)
Increased CSF protein35% (41/115)38% (43/114)
Cumulative pre-treatment dexamethasone (mg), median (IQR)152.0 (84.0-240.0)156.0 (72.0-272.0)

Arms

FieldControlHCT-ASCT (myeloablative consolidation)
InterventionAfter 4 cycles of MATRix induction (rituximab 375 mg/m2 days 0 and 5; methotrexate 3.5 g/m2 day 1; cytarabine 2 g/m2 BID days 2-3; thiotepa 30 mg/m2 day 4; every 21 days) - 2 cycles of R-DeVIC: rituximab 375 mg/m2 day 0, dexamethasone 40 mg days 1-3, etoposide 100 mg/m2 days 1-3, ifosfamide 1500 mg/m2 days 1-3, carboplatin 300 mg/m2 day 1, every 21 daysAfter 4 cycles of MATRix induction - carmustine 400 mg/m2 day -6 (or busulfan 3.2 mg/kg days -8 and -7 if carmustine unavailable) plus thiotepa 5 mg/kg BID days -5 and -4, followed by reinfusion of autologous stem cells on day 0
DurationTwo 21-day consolidation cycles after 4 induction cyclesSingle high-dose conditioning + stem-cell reinfusion after 4 induction cycles

Outcomes

OutcomeTypeControlInterventionHR / OR / RRP-value
Progression-free survival (time from randomisation to progression, relapse, or death from any cause), full analysis setPrimary3-year PFS 51% (95% CI 41-60); median 39 months (95% CI 26-not reached); 56 events3-year PFS 78% (95% CI 69-85); median not reached; 27 events0.430.0003
Overall survival (3-year)Secondary71% (95% CI 61-78); median 54 months86% (95% CI 78-91); median not reached0.460.0075
5-year overall survivalSecondary63% (95% CI 52-73)84% (95% CI 76-90)not reported for 5-year timepoint
Complete response rate on day 60 after randomisationSecondary65% (75/115; 95% CI 56-74)69% (79/114; 95% CI 60-78)1.22ns
3-year cumulative incidence of relapseSecondary45% (95% CI 35-54)19% (95% CI 12-26)0.4significant per SHR CI
EORTC QLQ-C30 median change (baseline to 1-year follow-up)Secondary+8.3 points (IQR -8.3 to 25.0)+16.7 points (IQR 0 to 25.0)descriptive only; higher scores = better global health
Mean AEs per patient (consolidation)AdverseR-DeVIC 9.3 (SD 4.4) vs HCT-ASCT 14.6 (SD 5.8)
Any AE (consolidation)AdverseR-DeVIC 100% (108/108) vs HCT-ASCT 100% (111/111)
Grade >=3 AE (consolidation)AdverseR-DeVIC 94% (102/108) vs HCT-ASCT 100% (111/111)
Grade 5 (fatal) AEs during consolidationAdverseR-DeVIC 2/108 (2%) - both secondary acute myeloid leukaemia; HCT-ASCT 5/111 (5%) - 4 infections, 1 pulmonary embolism (all but the PE judged possibly treatment-related)
Febrile neutropeniaAdverseR-DeVIC 14% vs HCT-ASCT 61%
Thrombocytopenia (any grade)AdverseR-DeVIC 94% vs HCT-ASCT 97%
Neutropenia (any grade)AdverseR-DeVIC 69% vs HCT-ASCT 77%
Infections (any)AdverseR-DeVIC 29% vs HCT-ASCT 68%
Grade >=3 infectionsAdverseR-DeVIC 14% vs HCT-ASCT 53%
Stomatitis (any grade)AdverseR-DeVIC 9% vs HCT-ASCT 79% (grade >=3: 0 vs 54%)
NauseaAdverseR-DeVIC 21% vs HCT-ASCT 57%
DiarrhoeaAdverseR-DeVIC 7% vs HCT-ASCT 59%
VomitingAdverseR-DeVIC 8% vs HCT-ASCT 40%
AlopeciaAdverseR-DeVIC 35% vs HCT-ASCT 44%
FatigueAdverseR-DeVIC 23% vs HCT-ASCT 33%
PyrexiaAdverseR-DeVIC 9% vs HCT-ASCT 45%
Secondary AML (during consolidation)AdverseR-DeVIC 2 cases vs HCT-ASCT 0 cases
Induction-phase SAEsAdverse58% of 346 (200 patients) with at least one SAE; 18 patients (5%) fatal SAEs (13 infections, 3 respiratory disorders); 11/18 (61%) considered possibly related to induction

Subgroup Analysis

Results consistent across prespecified subgroups; PFS benefit maintained in per-protocol population (appendix). Findings apply only to the 66% of enrolled patients who reached randomisation with at least partial response after induction.


Criticisms

  • Only 229/346 patients (66%) who started induction reached randomisation - one-third dropped out due to induction toxicity or early progression, so the survival benefit does not apply to the full enrolled cohort.
  • Open-label design: treatment allocation was not masked to patients, investigators, or local outcome assessors (central neuroradiological review was blinded).
  • Investigator discretion in judging fitness for HCT-ASCT (cardiac, pulmonary, hepatic) introduces potential selection bias.
  • Population restricted to Western European centres with established transplantation expertise; race/ethnicity data were not collected; generalisability to other populations and to less resourced settings is uncertain.
  • R-DeVIC arm delivered only 2 consolidation cycles - some argue longer or more intensive non-myeloablative regimens (e.g., CALGB50202 EA) might narrow the gap.
  • Detailed neurocognitive and quality-of-life outcomes reported only in abbreviated form; full analysis pending in a separate publication.
  • 26/115 (23%) patients in the R-DeVIC arm went on to receive off-protocol HCT-ASCT after progression, which may have attenuated the OS difference.

Funding

German Federal Ministry of Research, Technology and Space (grant 01KG1207); Swiss Cancer Research Foundation; Riemser Pharma. Funders had no role in study design, conduct, analysis, or reporting.

Based on: MATRix/IELSG43 (The Lancet, 2026)

Authors: Illerhaus G, Ferreri AJM, Wendler J, et al.

Citation: Lancet 2026;408(10554):532-544

Content summarized and formatted by NeuroTrials.ai.