Prognostic value of inflammation-based scores in patients with R/R LBCL treated with CD3×CD20 bispecific T-cell engagers

Key Points

• •Pretreatment inflammation-based scores (HEMATOTOX, InflaMix) independently predict poor response to CD3×CD20 bsAb in patients with R/R LCBL.
• •High baseline inflammation and early relapse after CAR T-cell therapy lead to particularly poor outcomes with CD3xCD20 bsAbs.

Visual Abstract

Abstract

T-cell–redirecting bispecific antibodies (bsAb) offer a novel therapeutic approach for relapsed/refractory large B-cell lymphoma (R/R LBCL). However, predictive biomarkers are needed to identify patients most likely to respond. As both bsAbs and chimeric antigen receptor (CAR) T cells represent T-cell–based therapies, we hypothesized that the established CAR-HEMATOTOX (HT) and InflaMix models—reflecting the degree of systemic inflammation—could be of prognostic utility for bsAb therapy. We applied both scores to a multicenter international cohort of 174 patients with R/R LBCL treated with bsAbs across 15 sites. Patients with a high HT score (≥3, 35%) displayed inferior median progression-free (PFS; 1.4 vs 7.4 months; P < .0001) and overall survival (OS; 2.0 vs 21.7 months; P < .0001) compared with patients with a low HT score. When applying the InflaMix score, 49% of the patients were assigned to the inflammatory cluster, translating into a significantly shorter median PFS (1.9 vs 17.8 months; P < .0001) and OS (4.1 vs 21.7 months; P < .0001). In a multivariable Cox regression analysis accounting for various prognostic factors, HT and InflaMix remained independent adverse risk factors for both PFS and OS. Patients presenting with both elevated HT score and the inflammatory signature showed markedly shorter OS and PFS compared with patients deemed low-risk by either one of the scores. In the CAR-pretreated subcohort, the combination of early CAR T-cell relapse (≤3 months) and elevated inflammation led to particularly detrimental outcomes. Overall, these data highlight the prognostic utility of baseline inflammatory markers in identifying patients who may benefit from combinatorial strategies alongside bsAb.

Introduction

The treatment landscape of relapsed/refractory large B-cell lymphoma (R/R LBCL) has been revolutionized by the advent of T-cell–based immunotherapies, encompassing chimeric antigen receptor (CAR) T-cell therapies and bispecific T-cell–recruiting antibodies (bsAb).1, 2, 3, 4, 5, 6 However, >60% of patients ultimately experience disease progression or relapse after CAR T-cell therapy, with dismal post-progression treatment outcomes.7, 8, 9 Currently, 3 CD20-directed bsAb—glofitamab, epcoritamab, and (by the European Medicines Agency only) odronextamab—have been approved as monotherapy for third-line treatment of R/R LBCL. Nevertheless, with an overall response rate (ORR) of ∼50% in the pivotal clinical trials and the published real-world analyses,^4,5,10, 11, 12, 13, 14, 15, 16, 17 the unmet clinical need remains high. Prognostic biomarkers are currently lacking but would help identify patients most likely to benefit from bsAb treatment vs patients who may require combinatorial or alternative treatment strategies.

In the field of CAR T-cell therapy, several scores have been designed to identify high-risk candidates for severe toxicity and dismal outcomes. The CAR-HEMATOTOX (HT) score not only predicts the length of neutropenia but, importantly, also progression-free survival (PFS) and overall survival (OS) across multiple disease entities (eg, LBCL, mantle cell lymphoma, multiple myeloma, and acute lymphoblastic leukemia).18, 19, 20, 21, 22 Determined prior to lymphodepletion, it integrates variables associated with hematopoietic reserve and inflammation (absolute neutrophil count [ANC], hemoglobin, platelet count, C-reactive protein [CRP], and ferritin). The recently developed InflaMix model²³ is based on 14 laboratory measures obtained on day 0 (preinfusion), integrating markers of organ function (albumin, hemoglobin, platelets, white blood cell count, aspartate aminotransferase, alkaline phosphatase, bilirubin, lactate dehydrogenase) and systemic inflammation (CRP, ferritin, D-dimer, interleukin-6, interleukin-10, tumor necrosis factor-alfa). Although originally derived from an extended panel, InflaMix can be calculated using as few as 6 routinely available markers. It stratifies patients into inflammatory and noninflammatory clusters, with the inflammatory assignment consistently associated with inferior outcomes following CD19-directed CAR T-cell therapy across multiple B-cell lymphoma cohorts. However, it remains unclear whether these inflammation-based scores carry prognostic value in patients with R/R LBCL treated with bsAbs.

The aim of this study was to assess the discriminatory capacity of the HT and InflaMix scores for survival outcomes in patients with R/R LBCL treated with bsAb.

Methods

Study design

We performed a retrospective, multicenter, international study examining 210 patients with R/R LBCL treated with CD3×CD20 monotherapy bsAb at 15 academic centers in Europe, the United States, and Israel. Patients were treated on a clinical trial (n = 42) or as standard-of-care (n = 168) between May 2018 and December 2024. Clinical trial participants treated with a dose that significantly differed from the authorized dose and patients receiving combination therapy were excluded from the analysis (n = 36), resulting in a final analysis cohort of 174 patients (supplemental Figure 1). Cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS) were graded according to the American Society for Transplantation and Cellular Therapy consensus guidelines.

Inflammation-based scores

For HT, a cut-off of 3 was chosen as it was shown to be the most prognostic threshold in the original publication.¹⁸ Accordingly, patients presenting with a score ≥3 were deemed “high-risk HEMATOTOX” (HR-HT), whereas those presenting with a score of ≤2 were defined as “low-risk HEMATOTOX” (LR-HT). InflaMix was applied as per Raj et al²³ using the publicly available online app (https://ssraj017.github.io/inflamix_app_prod/). All patients had at least 7 out of 14 variables available for calculation. All laboratory values were collected within the last 24 hours before the first bsAb application or prior to obinutuzumab pretreatment for patients treated with glofitamab. Because ferritin has been shown to be a stable parameter over the course of time,²⁴ a leniency period of up to 4 weeks prior to treatment start was accepted.

Statistical analysis

PFS was defined from the date of the first bsAb application until either relapse or death, censoring at time of last follow-up. OS was defined from the first bsAb application until death from any cause, censoring at last follow-up. Survival analyses were performed using Kaplan-Meier estimates and compared using log-rank test. The Benjamini-Hochberg method was used to control the false discovery rate in pairwise analyses. Univariate and multivariable analyses of PFS and OS were conducted using Cox regression models. Variables with a P value ≤.1 in the univariate analysis were included in the multivariable modeling. Categorical data were analyzed using the Fisher exact test. For continuous variables, the unpaired t test or the Mann-Whitney U test was employed depending on data distribution. Statistical analyses and data visualization were conducted using GraphPad Prism (version 10) or SPSS (version 29.0).

This study was approved by the ethics committee of the Ludwig Maximilian University of Munich, and conducted in accordance with institutional and ethical guidelines.

Results

Patient characteristics and toxicity profile of the study cohort

Of the 210 analyzed patients, 174 met the inclusion criteria. Baseline demographics and clinical characteristics are summarized in Table 1 and supplemental Tables 1 and 2. Overall, 81.0% of patients were treated with glofitamab (141/174), 17.8% with epcoritamab (31/174), and 1.1% with odronextamab (2/174). Median age was 67 years (range, 23-93). In the analyzed cohort, 41.4% of patients (72/164) were female, and 18.5% (30/162) displayed an Eastern Cooperative Oncology Group performance status ≥2. The predominant histology was diffuse large B-cell lymphoma, not otherwise specified (58.0%), followed by transformed diffuse large B-cell lymphoma (23.0%), high-grade B-cell lymphoma (17.2%), and primary mediastinal B-cell lymphoma (1.7%). Overall, 79.0% of patients (132/167) presented with an Ann Arbor stage ≥III at bsAb initiation, and 74.9% (128/171) had extranodal disease. Patients had received a median of 3 prior therapy lines (range, 1-10), 117 of 174 patients (67.2%) had undergone prior CAR T-cell therapy, and 82 of 165 patients (49.7%) were refractory to their latest line of therapy. The median time between last treatment and bsAb application was 3.7 months (range, 0.3-30.8 months).

Table 1.

Baseline demographics and clinical characteristics

Parameter	All patients (N = 174)
Demographic features
Age, median (range), y	67 (23-93)
Sex (female)	72 (41.4%)
ECOG ≥2	30/162 (18.5%)
Disease features
Histology
DLBCL-NOS	101 (58.0%)
HGBL	30 (17.2%)
PMBL	3 (1.7%)
tDLBCL	40 (23.0%)
END	128/171 (74.9%)
LDH, median (range), U/L	326 (105-3667)
Inflammation markers
CRP, median (range), mg/dL	1.36 (0.03-30.8)
Ferritin, median (range), ng/mL	270 (17-5900)
Hematological markers
Hemoglobin, median (range), g/dL	10.7 (6.3-15.8)
Platelets, median (range), × 109/L	149 (8-641)
Absolute neutrophil count, median (range), × 109/L	3.1 (0.2-22.3)
Prior therapy
Number of prior therapy lines, median (range)	3 (1-10)
Time between last therapy and bsAb infusion, median (range), mo	3.7 (0.3-30.8)
Previous CAR T-cell therapy	117 (67.2%)
Best response to previous therapy
CR	36/165 (21.8%)
PR	47/165 (28.4%)
PD/SD	82/165 (49.7%)
bsAb product
Epcoritamab	31 (17.8%)
Glofitamab	141 (81.0%)
Odronextamab	2 (1.1%)

DLBCL-NOS, diffuse large B-cell lymphoma not otherwise specified; ECOG, Eastern Cooperative Oncology Group; END, extranodal disease; HGBL, high-grade B-cell lymphoma; LDH, lactate dehydrogenase; PD, progressive disease; PMBL, primary mediastinal B-cell lymphoma; PR, partial remission; SD, stable disease; tDLBCL, transformed diffuse large B-cell lymphoma.

We observed 51 complete remissions (CR rate 30.4%) and 33 partial remissions (19.6%), resulting in an ORR of 50.0% (supplemental Figure 2A). With a median follow-up of 5.0 months, the median PFS and OS were 3.4 months and 10.4 months, respectively (supplemental Figure 2C-D). CRS occurred in 64% of patients (grade ≥3 in 8%), and ICANS was reported in 8% of patients (grade ≥3 in 1.3%; supplemental Figure 2B).

Identification of prognostic biomarkers for PFS after bsAb treatment

First, we examined the association of a range of patient-, disease-, and treatment-related variables on survival outcomes in a univariate Cox regression analysis (Table 2). We found that a shorter time between last treatment and bsAb application (hazard ratio [HR], 2.51; P < .001), Ann Arbor stage of ≥III (HR, 1.97; P = .016), and the presence of extranodal disease (HR, 1.62; P = .050) were adverse prognostic markers. On the other hand, Eastern Cooperative Oncology Group performance status, histology, number of prior lines of therapy, and prior CAR T-cell therapy were not significantly associated with PFS. Of interest, a trend toward improved PFS was noted for advanced age (HR, 0.99; P = .068) and female sex (HR, 0.71; P = .093).

Table 2.

Markers associated with PFS in univariate analysis

Parameter	n	Univariate HR (95% CI)	P value
Patient-related factors
Age (continuos variable)	174	0.99 (0.97-1.00)	.068
Sex (female vs male)	174	0.71 (0.48-1.06)	.093
ECOG (≥2 vs 0-1)	161	1.52 (0.91-2.54)	.113
Disease-related factors
Histology (HGBL vs other)	174	1.26 (0.76-2.07)	.170
Histology (tDLBCL vs other)	174	0.85 (0.53-1.34)	.475
Ann Arbor stage (≥III vs I-II)	167	1.97 (1.14-3.43)	.016
END (yes vs no)	171	1.62 (1.00-2.61)	.050
Treatment-related factors
Number of prior lines of therapy	174	1.07 (0.96-1.19)	.203
Prior CARs (yes vs no)	174	1.07 (0.71-1.62)	.742
Time between last treatment and bsAb infusion (≥3 mo vs <3 mo)	170	2.51 (1.70-3.70)	<.001
Product (glofitamab vs epcoritamab)	172	1.39 (0.79-2.45)	.252
Risk scores
InflaMix	171	2.79 (1.87-4.16)	<.001
CAR-HT	119	3.90 (2.46-6.30)	<.001

Bold values indicate statistically significant results P < .05.

Pretreatment HT and InflaMix are independent risk factors for poor PFS and OS

Due to limited data availability, HT could be calculated for 119 patients. Thirty-five percent of patients (n = 42) were classified as HR-HT, while the remaining 77 (65%) were classified as LR-HT. The HR-HT group exhibited a significantly inferior ORR (10% vs 64%; P < .0001) and CR rate (7% vs 43%; P < .0001; Figure 1A) compared with the LR group. Moreover, assignment to the HR-HT group was associated with a significantly inferior median PFS (1.4 months vs 7.4; P < .0001; Figure 1C) and OS (2.0 vs 21.7 months; P < .0001; Figure 1E) compared with the LR-HT group. In the multivariable model adjusting for age, sex, disease stage, presence of extranodal disease, and time between last treatment and bsAb administration, an elevated baseline HT score remained an independent adverse prognostic factor for both PFS (adjusted HR [aHR], 3.10 [95% confidence interval (CI), 1.81-5.30]; P < .001; Figure 2A) and OS (aHR, 3.61; 95% CI, 1.99-6.52; P < .001; Figure 2B).

Figure 1.

CAR-HT and InflaMix discriminate survival outcomes. (A-B) Best overall response after application of, respectively, CAR-HT (A) and InflaMix (B) to our cohort. Significant values were determined by Fisher exact test (∗P < .05, ∗∗P < .01, ∗∗∗P < .001, ∗∗∗∗P < .0001). (C-D) Kaplan-Meier estimates of PFS comparing patients who were HR-HT vs LR-HT (C), and patients who were inflamed vs noninflamed (D). (E-F) Kaplan-Meier estimates of OS comparing patients who were HR-HT vs LR-HT (E), and patients who were inflamed vs noninflamed (F). The number at risk at each time point is reported below the x-axis. The P value was calculated using the Mantel-Cox log-rank test. PR, partial remission.

Figure 2.

The CAR-HT and the InflaMix scores are predictors of PFS and OS in the multivariable analysis. Forest plots of the multivariable Cox regression analysis for PFS (A, C) and OS (B, D) adjusted for the baseline risk factors of age (continuous variable), Ann Arbor stage III to IV, presence of extranodal disease, time between last treatment and bsAb infusion >90 days, as well as HT risk group (A-B) or assignment to the inflammatory cluster by InflaMix (C-D). Due to missing values, only 111 of 119 and 160 of 163 patients were included in the multivariable analysis for CAR-HT and InflaMix, respectively. END, extranodal disease.

InflaMix was evaluable in 171 of 174 patients based on available baseline laboratory values; 49% of patients (n = 83) were assigned to the inflammatory cluster and 51% (n = 88) to the noninflammatory cluster. Both ORR (33% vs 66%; P < .0001) and CR rate (14% vs 45%; P < .0001; Figure 1B) significantly differed between the 2 groups. Patients in the inflammatory cluster displayed an inferior median PFS (1.9 vs 17.8 months; P < .0001; Figure 1D) and OS (4.1 vs 21.7 months; P < .0001; Figure 1F) compared with the noninflammatory cluster. In the multivariable Cox regression model adjusting for age, sex, disease stage, presence of extranodal disease, and time from last treatment, assignment to the inflammatory cluster remained a negative prognostic indicator for both PFS (aHR, 2.37; 95% CI, 1.52-3.70; P < .001; Figure 2C; supplemental Figure 3A) and OS (aHR, 2.20; 95% CI, 1.33-3.62; P = .002; Figure 2D; supplemental Figure 3B).

To analyze the discriminative capacity of the scores in different subgroups, we repeated the analysis stratifying patients by bsAb product. Interestingly, an elevated CAR-HT score remained a significant predictor for both PFS and OS in patients treated with glofitamab as well as in those treated with epcoritamab (supplemental Figure 4A-D), while InflaMix only showed significance in the glofitamab-treated cohort (supplemental Figure 4E-H).

The CAR-HT and InflaMix scores identify patients at risk for high-grade CRS and ICANS

The proportion of patients experiencing grade ≥2 CRS was significantly higher in the HR-HT compared with the LR-HT group (38.4% vs 17.8%; P = .0221; supplemental Figure 5A). Moreover, we observed a more frequent incidence of ICANS in patients presenting with HR-HT (15.4% vs 2.8%; P = .0215; supplemental Figure 5B). Similarly, assignment to the inflammatory cluster of InflaMix was associated with an elevated risk of higher grade CRS and ICANS occurrence (34.2% vs 9.4%; P < .0001; and 13.9% vs 2.4%; P = .078; supplemental Figure 5C-D).

Patients with both adverse risk HT and InflaMix exhibit dismal survival following bsAb treatment

The percentage of patients assigned to the inflammatory cluster was higher than the percentage of patients with a HR-HT score ≥3 (49% vs 35%; Figure 3A). In addition, we observed significant differences concerning the ORR between the 3 risk categories. Patients at low risk in both models displayed an ORR of 75% vs 41% for patients with discordant InflaMix and HT cluster assignment, and 10% for patients at high risk in both models (Figure 3B).

Figure 3.

Combination of CAR-HT and InflaMix in response prediction. (A) Contingency table displaying the allocation of patients to the different risk categories. (B) Best overall response in patients assigned to the high- or low-risk category by both models and patients with discordant score assignment. Significant values were determined by Fisher exact test (∗P < .05, ∗∗P < .01, ∗∗∗P < .001, ∗∗∗∗P < .0001). (C-D) Kaplan-Meier estimates of PFS and OS comparing patients at high or low risk in both models and patients with discordant InflaMix and HT cluster assignment. The number at risk at each time point is reported below the x-axis. The P value was calculated using the Mantel-Cox log-rank test. HTh, CAR-HT high; HTl, CAR-HT low; i, inflamed; ni, noninflamed; PR, partial remission.

Moreover, patients at high risk in both models showed a significantly shorter PFS and OS compared with patients with discordant InflaMix and HT cluster assignment or classified as low risk according to both scores (Figure 3C-D).

For example, the 6-month PFS and OS rate in the double high-risk patients was 3.3% and 10%, respectively. Conversely, patients deemed low-risk by both scores (n = 46) showed excellent survival outcomes with a 6-month PFS and OS rate of 64% and 82%, respectively. The patients with either HR-HT or Inflamix exhibited intermediary survival outcomes.

Figure 4 shows a heat map of the singular laboratory values contributing to the scores, providing enhanced granularity between different clusters. As expected, patients with elevated inflammatory parameters are enriched in the CAR-HT high/inflammatory cluster. As shown by the annotations, group clustering is of clinical significance as CAR-HT high/inflamed patients are characterized by detrimental survival outcomes as well as a higher incidence of nonhematological toxicities compared with the CAR-HT low/noninflamed group. When comparing the distribution of individual laboratory parameters between the different clusters, ferritin and CRP showed the strongest statistical difference between the CAR-HT high/inflamed and the CAR-HT low/inflamed + CAR-HT high/noninflamed group followed by lactate dehydrogenase, hemoglobin and platelets count, while markers of end-organ function as well as white blood cell count and ANC displayed only moderate or nonsignificant differences. These findings suggest that the inflammatory parameters mainly drive group separation.

Figure 4.

Distribution of laboratory values in the different clusters. Heat map displaying the distribution of individual score laboratory parameters (CRP, ferritin, LDH, AST, ALP, Tbil, albumin, WBC, Hgb, Plt, ANC) according to cluster assignment. Columns represent single patients, rows represent Z-score standardized laboratory parameters, with low values shown in blue and high values in red. The upper annotations display the differential score allocation, while relevant clinical parameters including response and toxicity are shown below. Missing values are represented by gray fields. ALP, alkaline phosphatase; ANC, absolute neutrophile count; AST, aspartate aminotransferase; Hgb, hemoglobin; HTh, CAR-HT high; HTl, CAR-HT low; i, inflamed; LDH, lactate dehydrogenase; ni, noninflamed; PD, progressive disease; Plt, platelets; PR, partial remission; SD, stable disease; Tbil, total bilirubin; WBC, white blood cell count.

Due to missing laboratory values (mostly owing to missing ferritin assessments), CAR-HT was only evaluable in 119 of 174 patients. We performed a sensitivity analysis showing no difference in outcome between patients with available laboratory values and patients with no available laboratory values for the calculation of the CAR-HT score (supplemental Figure 7). In the subset of patients without available CAR-HT score, InflaMix remained associated with worse PFS (supplemental Figure 8).

Prior CAR T-cell response but not exposure impacts response to bsAb treatment

Of note, no differences in survival outcomes or response rates were observed between patients who were CAR-naïve and CAR-exposed. However, patients who were CAR-naïve displayed a significantly higher baseline inflammatory state compared with patients who were CAR-exposed, presenting with elevated CRP (median 2.8 vs 0.9 mg/dL; P = .01) and ferritin levels (median 497 vs 227 ng/mL; P = .0008), as well as a higher proportion of patients assigned to the inflammatory cluster by InflaMix (62% vs 42%; P = .022). We did not observe significant differences in the proportion of patients with HR-HT and LR-HT in patients who were CAR-naïve vs CAR-exposed (supplemental Figure 9).

Patients experiencing early relapse after CAR T-cell treatment, defined as recurring within 90 days, presented with significantly shorter PFS (2.0 vs 6.8 months; P < .0001; Figure 5A) and OS (5.9 vs 21.7 months; P = .0002; Figure 5B) compared with patients with late CAR T-cell relapse (>3 months). When combining inflammation-based scores and time to CAR T-cell failure, we observed that the combination of HR-HT (Figure 5C) or inflammatory cluster (Figure 5D) with short time to CAR T-cell failure was associated with a significantly shorter PFS compared with patients presenting with only 1 of these characteristics. In the subset of patients with early relapse after CAR T-cell therapy, significantly more patients were assigned to the inflammatory cluster of InflaMix compared with patients with late relapse (55% vs 31%, P = .0134; Figure 5F), while again no significant differences were observed for CAR-HT (Figure 5E).

Figure 5.

Survival outcomes and distribution of CAR-HT and InflaMix score according to time to CAR T-cell failure. Kaplan-Meier estimates of PFS (A) and OS (B) comparing patients with early vs late relapse after CAR T-cell treatment. Kaplan-Meier estimates of PFS according to time to CAR T-cell failure and CAR-HT (C) or InflaMix (D). The number at risk at each time point is reported below the x-axis. The P value was calculated using the Mantel-Cox log-rank test. The false discovery rate was controlled using the Benjamini-Hochberg method. Distribution of CAR-HT (E) and InflaMix (F) scores in patients with early vs late relapse after CAR T-cell treatment. Significant values were determined by Fisher exact test (∗P < .05, ∗∗P < .01, ∗∗∗P < .001, ∗∗∗∗P < .0001). n.s., not significant.

Discussion

In this multicenter international study, we applied the inflammation-based CAR-HT and InflaMix models to a predominantly real-world cohort of patients with R/R LBCL treated with CD3×CD20 bsAb.

The median PFS and OS in our cohort were comparable to outcomes reported in other real-world studies in the field of bsAb,^7,11,12,14 but slightly shorter than the survival times in the pivotal clinical trials.^4,5,10 In line with previous findings, various disease- (high Ann Arbor stage, presence of extranodal disease) and treatment-related factors (time from the latest therapy line) were associated with poor outcomes.^12,25 A high CAR-HT score and assignment to the inflammatory InflaMix cluster were found to serve as predictors for both PFS and OS after adjusting for clinically relevant confounders.

Recently, time to relapse after prior CAR T-cell treatment was identified as a strong predictor for PFS and OS in patients with R/R LBCL treated with CD3×CD20 bsAb.¹⁶ After being able to reproduce these findings in our cohort, we further wanted to analyze the interplay between inflammation-based scores and time to CAR T-cell failure. We observed a higher percentage of inflamed patients in the early-relapse group. Moreover, the co-occurrence of elevated inflammation-based score and early relapse led to particularly detrimental outcomes, suggesting that the combination of these 2 variables could help identify a subset of patients at very high-risk for relapse. T-cell lymphopenia observed after CAR T-cell therapy is likely attributable to the preceding lymphodepletion with cyclophosphamide and fludarabine, and can persist for at least 180 days after infusion.²⁶ Consequently, initiating bsAb shortly after CAR T-cell therapy may be suboptimal due to insufficient T-cell numbers and compromised T-cell fitness. However, the relevance of T-cell lymphopenia and, hence, the optimal sequencing and timing of bispecific T-cell engagers and CAR T-cell therapy remains a controversial topic. Studies addressing this question are currently ongoing.

In the CAR T-cell field, it has already been shown that inflammation-based scores as well as increased inflammation markers lead to poor outcomes.^19,21,27, 28, 29, 30, 31, 32 Both the CAR-HT and the InflaMix score share the integration of inflammatory markers, underscoring the prognostic relevance of inflammation in this setting. These associations may reflect the broader impact of systemic hyperinflammation on T-cell fitness, a key determinant of response to both bsAb and CAR T-cell therapies. Furthermore, elevated inflammatory markers may serve as surrogates for an immune-hostile tumor microenvironment, characterized by the presence of bystander cells such as myeloid-derived suppressor cells and macrophages—cell types known to impair CAR T-cell expansion and function.³⁰

Interestingly, we observed a significantly higher proportion of patients assigned to the inflammatory cluster in patients who were CAR-naïve compared with CAR-exposed. The higher inflammatory state in patients who were CAR-naïve suggests that this population is characterized by a higher disease burden. This is most likely because patients requiring prompt disease control are not ideal candidates for upfront CAR T-cell therapy due to the long manufacturing process, which causes a delay in treatment initiation. In these cases, bispecific T-cell engagers are often used as a bridging therapy. Indeed, in our cohort, 17.5% (10/57) of patients who were CAR-naïve were treated with CAR T cells following T-cell engagers therapy.

With the current knowledge, we are unable to provide a comprehensive answer to the question of whether one score should be applied rather than the other. There is a certain degree of ambivalence, and the combination of both scores provided the most solid results. There may be a specific subgroup in which 1 score performs better than the other. For instance, HT was able to discriminate for survival outcomes in both patients treated with glofitamab and epcoritamab, while InflaMix only showed significance in the glofitamab cohort. However, these results should be interpreted carefully due to the limited sample size for the epcoritamab cohort. Further analyses with larger cohorts will be needed in order to address this question.

Of interest, the reported incidence of ICANS in our cohort was higher compared with the published clinical trials. Our finding might result from a divergent definition of neurotoxicity, as preexisting neurological conditions or hospital-acquired delirium in elderly patients can be misrecognized as ICANS. Therefore, despite encouraging investigators to grade neurotoxicity according to the American Society for Transplantation and Cellular Therapy recommendations, we cannot exclude overreporting of ICANS.

This retrospective study has certain limitations, primarily related to the heterogeneity of the cohort. It includes both patients who were CAR-naïve and CAR-exposed, as well as individuals treated with different bsAb constructs. Furthermore, the median follow-up duration was only 5.0 months, and longer observation will be necessary to confirm these findings. Overall, our findings require prospective validation. If confirmed, HT and InflaMix could serve as key clinical tools to identify patients who may require alternative or combinatorial treatment strategies. In particular, targeting inflammation through combination therapies33, 34, 35, 36 or the concomitant use of immunomodulatory agents³⁷ could represent a promising approach for patients who were high-risk.

In summary, our retrospective, multicenter, international analysis of 174 patients with R/R LBCL demonstrated that both the CAR-HT and the InflaMix models reliably identify patients at higher risk for poor clinical outcomes after treatment with CD3×CD20 bispecific T-cell–recruiting antibodies. These findings are in line with previous studies in the CAR T-cell context and highlight the importance of immune dysregulation for response to T-cell–based immunotherapies. These scores can be utilized to identify a high-risk population who may benefit from inclusion in clinical trials or combinatorial strategies.

Conflict-of-interest disclosure: K.R. received research funding, consultancy fees, honoraria, and travel support from Kite/Gilead; honoraria from Novartis; consultancy fees and honoraria from BMS/Celgene; travel support from Pierre-Fabre; and consultancy fees from CSL Behring. J.K.S. received travel support from BeiGene, AbbVie, Janssen, and Novartis. N.K. received research support from AstraZeneca; honoraria from AbbVie, AstraZeneca, BMS, Janssen, Kite/Gilead, and Lilly; travel grants from AbbVie, AstraZeneca, BeiGene, Janssen, and Lilly. C.S. received travel grants from Roche and AbbVie, and honoraria from Roche. U.H. received honoraria from Roche. O.B.-K. received honoraria from Kite/Gilead; consultancy fees and honoraria from Novartis; and honoraria from AbbVie. R.R. received consultancy fees and honoraria from Kite/Gilead; consultancy fees and honoraria from Novartis; and honoraria from BMS. R.S. received speaker honoraria from Incyte and Sanofi. G.L. received research grants not related to this manuscript from Agios, Aquinox, AstraZeneca, Bayer, Celgene, Gilead, Janssen, MorphoSys, Novartis, F. Hoffmann-La Roche Ltd, and Verastem; and honoraria not related to this study from ADC Therapeutics, AbbVie, Amgen, AstraZeneca, Bayer, BeiGene, BMS, Celgene, Constellation, Genase, Genmab, Gilead, Hexal/Sandoz, Immagene, Incyte, Janssen, Karyopharm, Lilly, Miltenyi, MorphoSys, Merck Sharp & Dohme, NanoString, Novartis, PentixaPharm, Pierre Fabre, F. Hoffmann-La Roche Ltd, and Sobi. V.K.J.V. received honoraria not related to this study from BeiGene, Cellgene, Gilead, Roche, Lilly Oncology, AbbVie, AstraZeneca, and Johnsson & Johnsson, all paid to her institution. P.B. has been on the advisory board and received consultancy fees from Allogene, Amgen, Autolus, BMS/Celgene, Kite/Gilead, Incyte, Miltenyi Biomedicine, Novartis, Nektar, Pfizer, and Pierre Fabre. B.C. received speaker honoraria not related to this study from AbbVie, Ars tempi, AstraZeneca, BMS, Incyte, Johnson & Johnson, Gilead, Kompetenznetz Maligne Lymphome, Roche, Sobi, and Ono; worked not related to this study as a consultant for AbbVie, BMS, Gilead, Incyte, Johnson & Johnson, Roche, and Sobi; and is not related to this study lead clinical investigator on the R-Pola-Glo trial that is supported by Frankfurt Institute for Clinical Cancer Research/Roche. U.G. received consultancy fees and honoraria from Kite/Gilead; honoraria from Novartis; and honoraria from BMS. G.I. received consultancy fees from Autolus, BMS, Kite/Gilead, Miltenyi, and Novartis; honoraria (speaker) from AbbVie, AstraZeneca, BMS, Kite/Gilead, Miltenyi, Novartis, and Lilly; and travel support from AbbVie, AstraZeneca, Kite/Gilead, and Miltenyi. V.B. received honoraria and travel support from AbbVie; honoraria, consultancy, and travel support from Amgen; research funding from BMS; research funding, consultancy, and travel support from Kite/Gilead; research support from Miltenyi Biotech; research support, consultancy fees, and travel support from Johnson & Johnson; consultancy fees and travel support from Pierre Fabre; honoraria from Pfizer; consultancy fees from Priothera; honoraria from Roche; and research support from Takeda. M. Subklewe received industry research support from Amgen, BMS/Celgene, Gilead/Kite, Johnson & Johnson, Miltenyi Biotec, Novartis, Roche, Seattle Genetics, and Takeda; serves as a consultant/advisor to AbbVie, Crossbow, Debiopharm, Gilead/Kite, Interius, Johnson & Johnson, Molecular Partners, Novartis, and Otsuka; and serves on the speakers' bureau at Amgen, BMS/Celgene, Gilead/Kite, Miltenyi Biotec, Novartis, Roche, and Takeda. The remaining authors declare no competing financial interests.