Meta-analysis of survival by phased-variant ctDNA and PET response in large B-cell lymphoma

Visual Abstract

TO THE EDITOR:

Large B-cell lymphoma (LBCL), the most common B-cell malignancy in adults, is cured in 60% to 70% of patients with frontline chemoimmunotherapy.¹ End of treatment (EOT) response is assessed using positron emission tomography (PET) imaging, and complete metabolic response (CMR) is defined as a Deauville score of 1 to 3.² CMR is associated with improved survival, but 20% to 30% will relapse.^3,4 Measurable residual disease (MRD) assays identify cancer that may not be detected by conventional imaging. MRD testing is part of the standard assessment for other B-cell malignancies, including acute lymphoblastic leukemia and multiple myeloma, but is not incorporated into response assessment in LBCL.^5,6 In LBCL, circulating tumor DNA identified by phased variants is currently the most sensitive and specific method of detecting MRD and is associated with survival.^7,8 Phased-variant assays use the identification of multiple mutations in a single circulating tumor DNA strand to reduce background noise and improve the accuracy of MRD detection. The aim of this meta-analysis is to more precisely define the clinical impact of phased-variant MRD testing, particularly in the context of PET scan response, after frontline curative-intent chemoimmunotherapy in patients with LBCL.

A literature search was performed using PubMed with the cutoff date of 23 February 2026. Primary search terms were “lymphoma” and “phased-variant.” Studies were included if prospectively conducted in newly diagnosed LBCL, treated for curative intent with anthracycline-based regimens, and reported progression-free survival (PFS) in patients in whom both a phased-variant method of MRD detection and PET response were reported at EOT. The following data elements were independently extracted by 2 authors (M.J.G. and G.O.): age, treatment regimen, LBCL subtype, International Prognostic Index score, median follow-up, and PFS. The primary outcome of the study was PFS. The secondary outcome was estimated 2-year PFS rate.

The meta-analysis was conducted using a random-effects model based on restricted maximum likelihood estimation with the Hartung-Knapp method.^9,10 Hazard ratios (HR) for PFS were calculated for EOT MRD status (undetectable MRD [uMRD] vs detectable MRD [dMRD]) and PET scan response (CMR vs no CMR), and 2-year PFS was estimated after stratification by EOT MRD and PET imaging status. In the pooled 2-year PFS analysis, proportions were analyzed using a random-effects model with logit transformation for normality assumptions. To account for boundary constraints in studies with 100% PFS, a continuity correction was applied by adjusting the upper confidence limit based on sample size (ie, $\left(n-0.5\right)/n$). Subgroup differences were assessed by χ². P value of <.05 was considered significant. All analyses were conducted using R (version 4.4.2).

Out of 19 studies that matched the initial search terms, 3 met criteria for inclusion in the meta-analysis (supplemental Figure 1).11, 12, 13 In 2 studies, MRD testing was performed by Foresight Diagnostics, and the third study was conducted by the Cancer Molecular Diagnostic Laboratory, University of Cambridge.11, 12, 13 Clinical characteristics are shown in supplemental Table 1.

Among 367 patients, 68% had uMRD and PET-CMR at EOT (n = 250), 21% had dMRD (n = 77), and 22% had no CMR (n = 81). Among patients with dMRD, 53% also had no CMR (n = 41), and in patients with no CMR, 51% had dMRD (n = 41).

We evaluated the association of PFS with EOT MRD or PET result (Figure 1A). The HR for PFS was 14.02 (95% confidence interval [CI], 3.49-56.35) when comparing uMRD with dMRD, and 5.09 (95% CI, 2.51 – 10.34) in PET when comparing CMR with no CMR. dMRD was associated with HR for PFS that was >2.75 than no CMR at EOT (P = .005).

Figure 1.

Progression fress survival stratified by MRD or PET. (A) Forest plots of the HR for uMRD vs dMRD and CMR vs no CMR by PET scan response for PFS at EOT are shown separately. (B) Forest plots of the HR for PFS comparing uMRD vs dMRD by PET scan response.

We therefore next asked if MRD status affected PFS specifically within patients with CMR and no CMR at EOT (Figure 1B). In the CMR group, dMRD was associated with an HR of 10.84 (95% CI, 1.17-100.66), and in the no CMR group, dMRD was associated with an HR of 8.75 (95% CI, 2.25-33.98). In dMRD, there was no significant difference in PFS between the CMR and no CMR groups (P = .72).

The estimated PFS rate at 2 years in all patients was 73.1% (95% CI, 44.4-90.3). Two-year PFS was higher in uMRD regardless of EOT PET status (P < .0001; Figure 2). Among patients with CMR/uMRD (n = 250), 2-year PFS was 95.7% (95% CI, 75.7-99.4). No patients with CMR/uMRD (n = 40) had a similar 2-year PFS of 91.3% (95% CI, 76.8-97.1). In contrast, 2-year PFS was 40.2% (95% CI, 14.5-72.7) in CMR/dMRD (n = 36) and 23.4% (95% CI, 9.8-46.2) in no CMR/dMRD (n = 41).

Figure 2.

Forest plots of 2-year PFS proportion by combined MRD and PET scan status.

Phased-variant dMRD at EOT is a robust predictor of survival after curative-intent frontline therapy in LBCL. This study also demonstrates similar results using 2 distinct assays. Compared with the “gold standard,” EOT response assessment by PET imaging, phased-variant dMRD was associated with a nearly threefold increase in risk for PFS events when compared with no CMR. Although these results should be interpreted as descriptive, they do support further investigation of phased-variant MRD detection in prospective clinical trials.

In this study, patients with CMR and uMRD had <5% rate of PFS events at 2 years, a validated surrogate end point in LBCL.14, 15, 16 In addition, patients who did not have CMR but had uMRD had a 2-year risk of PFS events <10%, suggesting that outcomes are favorable for patients with uMRD regardless of PET response.

In contrast, patients who were in CMR but had dMRD had an estimated 2-year PFS rate of only 40%. Patients with no CMR and dMRD had 2-year PFS rates of 23%. Although PFS rates are substantially lower in dMRD, a significant proportion did not have events by 2 years in this study. Ongoing trials are evaluating consolidation with allogeneic or autologous CD19 chimeric antigen receptor T-cell therapy in dMRD at EOT (ClinicalTrials.gov identifiers: NCT06500273; NCT07316010) and will hopefully better define the role of MRD-directed consolidation in LBCL.

This meta-analysis is based on 3 relatively small nonrandomized studies, and the results should be interpreted as exploratory. In addition, individual patient data were not available for the meta-analysis, which could contribute to unmeasured bias. Larger prospective and ideally randomized studies are needed to formally compare MRD and PET response assessment and their optimal use in the clinical care of patients with lymphoma. This study also only includes patients who completed frontline therapy, excluding treatment-refractory patients, who would be expected to have particularly poor outcomes. We also note that the median time from EOT to clinical progression is <6 months.¹¹ Therefore, to be clinically meaningful, an MRD test needs to have a quick turnaround time.

Phased-variant MRD testing is a promising new technology that appears to enhance imaging-based response assessment at EOT. Future studies focused on specific clinical scenarios may improve the definition of the populations in which phased-variant MRD testing may be most impactful. The strength of the current data suggests phased-variant assays should be incorporated into response assessment at EOT in LBCL as soon as clinically available. Because of the significant risk of overtreatment, we would not recommend initiating salvage therapy based only on dMRD or lack of CMR at EOT.

Conflict-of-interest disclosure: The authors declare no competing financial interests.