Diffuse large B cell lymphoma (DLBCL), an aggressive subtype of non-Hodgkin lymphoma, is a curable disease, with long term remissions seen in 60-70% of patients treated with the standard first-line rituximab-based chemoimmunotherapy (CIT) (1). For patients with refractory and/or relapsed disease, the standard of care (SOC) consists of high dose chemotherapy followed by autologous stem cell transplantation (ASCT) in patients with chemosensitive disease. Nearly half of these patients can achieve long term remission with ASCT (2). In contrast, patients who are refractory to first line treatment or relapse shortly after, have dismal outcomes with median overall survival of 6 months (1). In current practice, patients who do not respond to salvage chemotherapy (therefore unable to proceed to ASCT) or relapse after ASCT can be treated with an approved anti CD-19 chimeric antigen receptor T cell (CAR-T) therapy such as axicabtagene ciloleucel (axi-cel), lisocabtangene maraleucel (liso-cel), and tisagenlecleucel (tisa-cel) in third line setting or later (3–5).

Recently, three randomized controlled trials (RCTs) TRANSFORM (6), ZUMA-7 (7), and BELINDA (8) were conducted in hopes to establish CAR-T as the second line of treatment in DLBCL. These trials compared the outcomes of anti-CD19 CAR-T cell therapy against SOC in patients with either primary refractory DLBCL or relapsing within 12 months after first line CIT. TRANSFORM (6) and ZUMA-7 (7) reported positive outcomes of CAR-T cell therapy with respect to event free survival compared to SOC while in BELINDA trial (8), CAR-T cell therapy failed to improve event free survival compared to SOC. Thus, we performed a meta-analysis to quantify the relative and absolute benefit of CAR-T cell therapy compared to SOC as second line treatment for R/R DLBCL.

This systematic review and meta-analysis is reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) statement (9) ( Figure 1 and eMethods 1efd ).

Of 1803 studies initially identified ( Figure 1 ), three trials (6–8) with a total of 865 patients and assessing axi-cel, tisa-cel, liso-cel were included in this systematic review and meta-analysis ( Table 1 ). The overall risk of bias for all studies was low ( eFigure 1 in the Supplementary ).

The results of this systematic review and meta-analysis showed that treatment of DLBCL with CAR-T cell therapy in second line setting achieves significantly higher response rates, longer duration of remission, and delayed disease progression with no statistically significant increase in overall toxicity as compared to SOC. Although, there appears to be a signal of overall survival benefit with CAR-T cell therapy, the difference was not statistically significant at current follow up ( Table 2 ). Treatment related mortality was comparable with 4% (18/434) in the CAR-T cell therapy arm versus 3.9% (17/431) in the SOC arm. The incidence rates for all-grade and grade ≥3 CRS were approximately 70%, and 4%, respectively in patients who received CAR-T cell therapy. All-grade and grade ≥3 NT were observed in 25% and 7.5% of the patients who received CAR-T cell therapy. Among different CAR-T cell products, delayed disease progression, and increased objective response were observed with axi-cel and liso-cel as compared to tisa-cel. In terms of OS improvement, no difference was observed among different CAR-T cell products at current follow-up. However, the relative safety of different CAR-T cell products was different, with tisa-cel being the safest among others.

The results of this study suggests statistically significant improvement in EFS with liso-cel (6) and axi-cel (7) as compared to tisa-cel (8). However, it is important to view these results in the context of the current evidence and the differences across the included trials. First, there was variability in EFS definitions; more specifically, the BELINDA trial defined EFS as the time from randomization to either progressive or stable disease at or after 12 weeks in addition to death at any time. This suggests that the events occurring prior to week 12 would be counted as events in the ZUMA-7 (7) and TRANSFORM (6) trials but not in the BELINDA trial (8) which could explain this underlying heterogeneity and discrepant EFS observed across these trials ( eTable 1 in the Supplementary ). Considering these differences in EFS across trials, PFS might have been a more optimal endpoint, however, PFS was not reported in the BELINDA trial (8). Second, longer median time from leukapheresis to CAR-T cell infusion (52 days in BELINDA (8)) compared to a median time from randomization to infusion of 29- and 34-days in the ZUMA-7 (7), and TRANSFORM (6) trials, respectively might explain the disparate outcomes observed in these trials. Third, mixed treatment comparisons showed that the response rates were significantly lower with tisa-cel as compared to axi-cel and liso-cel. This pattern is consistent with the findings from studies assessing CAR-T cell therapy in the third line setting (58% CR with axi-cel from the ZUMA-1 trial (3), 53% CR with liso-cel from the TRANSCEND trial (5); and 40% CR with tisa-cel from the JULIET trial (4)). Interestingly, the complete response to tisa-cel in the second line setting was observed in only 28.3% of the patients in the BELINDA trial which is even lower than the 40% observed in the JUILET trial (4) in the third line setting. This lower complete response rate with tisa-cel in the second line setting may potentially explain the negative EFS outcome. Fourth, it is also important to highlight that trials were different in their use of bridging therapies prior to CAR-T cell therapy. ZUMA-7 (7) only allowed the use of glucocorticoids and did not allow the use of bridging chemotherapy prior to CAR-T cell therapy which by design excludes patients with highly aggressive and advanced disease; conversely, approximately 60% of the patients in TRANSFORM trial (6) and 83% in BELINDA trial (8) received bridging chemotherapy. However, patients in the trial TRANSFORM trial were only allowed one cycle of bridging chemotherapy while 47% patients in the BELINDA trial received (3) 2 cycles. Patients who receive bridging chemotherapy are known to harbor worse prognosis than those who do not, and these patients are more reflective of the real-world clinical setting (13). Fifth, while BELINDA (8) and TRANSFORM (6) trials allowed the receipt of two lines of salvage CIT in the control arm, ZUMA-7 trial (7) only allowed one line of salvage CIT. Sixth, only 66.4% of the patients in the BELINDA trial (8) had primary refractory disease as compared to 73.5% and 73.3% in ZUMA-7 (7) and TRANSFORM (6) trials, respectively. Our subgroup analysis based on limited data showed no potential effect modification by prior response status, i.e., between patients who had primary refractory disease and those who relapsed. ( eFigure 19 in the Supplementary ). Finally, the variable proportions of activated B-cell like (ABC) phenotype and germinal center B-cell (GCB) like lymphoma may have impacted the outcomes in these trials. Patients with ABC phenotype are known to harbor worse prognosis than those with GCB like lymphoma (14). Proportion of patients with ABC phenotype was greater in the tisa-cel arm compared to the SOC arm in the BELINDA trial (8). In contrast, approximately 23% and 32% had ABC phenotype in liso-cel and SOC arms, respectively in the TRANSFORM trial (6) and less than 10% of the patients exhibited the ABC phenotype in the ZUMA-7 trial (7). However, we did not find statistically significant effect modification by the cell of origin. A subgroup analysis of EFS by age group demonstrated similar efficacy among patients aged over 65 years ( eFigure 16 in the Supplementary ). These findings are supported by several encouraging reports on third-line CAR-T therapy among elderly patients with lymphoma, including pivotal studies and real-world data, which describe comparable outcomes for CAR-T therapy in older patients (15, 16).

Current evidence suggests a signal of potential OS benefit with CAR-T cell therapy compared to SOC. However, the difference was not statistically significant. Notably, a significant proportion of patients (52%) randomized to SOC arm were able to cross over to CAR-T cell therapy in the TRANSFORM (6) and BELINDA (8) trials or received commercial CAR-T cell therapy off study in the ZUMA-7 (7) trial. Similarly, no statistically significant difference was observed among axi-cel, liso-cel and tisa-cel though the direction of effect indicated potential superiority of axi-cel and liso-cel compared to tisa-cel. The lack of statistical significance in these comparisons could plausibly be explained by the fact that the OS analyses in the ZUMA-7 (7) and TRANSFORM (6) trials were interim as per protocol and hence, updated analyses based on mature OS data would provide more meaningful estimates. Toxicity profile of CAR-T cell therapy was consistent with the studies from third line setting. CAR-T cell therapy was not associated with an increased risk of all grade and grade ≥3 any AEs when compared to SOC, however, the type of adverse events were different in both arms (17, 18). The pooled incidences of grade ≥ 3 CRS and NT were 4.19% and 7.57% with CAR-T cell therapy which are consistent with the results of other studies. A detailed breakdown of adverse events observed with axi-cel, liso-cel, and tisa-cel is provided as eTable 2 in the Supplementary . Quality of life (QoL) is also an important endpoint to consider when opting CAR-T cell therapy. QoL report from the ZUMA-7 trial showed statistically significant improvement in QoL with axi-cel when compared to SOC at day 100. The results were consistent at day 150 (19). QoL data for CAR-T cell therapy in second line setting is still emerging and results from the TRANSFORM and BELINDA trials will inform further.

Moreover, a closer examination of the included trials revealed that majority of the patients were not able to proceed to ASCT in the SOC arm either because of suboptimal response to salvage chemotherapy or from progression of disease and only around 38.5% (166/431) of the patients underwent ASCT, whereas approximately 95.6% (415/434) of the patients, assigned to CAR-T cell therapy, were able to receive the CAR-T cell infusion (7, 8, 20). Acknowledging the limitations of our posthoc exploratory analysis, it is still an important finding that when directly compared, ASCT may be associated with an increased incidence of CR compared to CAR-T cell therapy. Consistent superiority of ASCT in patients who achieve at least a PR after salvage chemotherapy has been observed in a comparative analysis of Center for International Blood and Marrow Transplant Research (CIBMTR) registry by Shadman et al (21) which showed that ASCT was associated with a lower risk of relapse and an improved survival when compared to CAR-T cell therapy in R/R DLBCL patients. However, it should be noted that the analysis by Shadman et al (21) was retrospective in nature, and despite adequate adjustment for variable disease burden, different number of prior lines of therapy there could be potential confounding relationships. Also, it included patients who had exhibited chemosensitivity after salvage therapy compared to trials included in our analysis which included patients at the highest risk of chemo-refractory disease. Nevertheless, taken together these findings suggest that ASCT may still be preferrable for a subset of patients who exhibit sensitivity to salvage chemotherapy.

There are several noteworthy strengths of this study. First, we used a systematic approach to investigate the efficacy and safety of CAR-T cell therapy compared to SOC using totality of available evidence. Second, we performed a detailed and thorough review of relevant trials and provided a summary of baseline trial and population characteristics along with the limitations in each trial. Third, we conducted comprehensive pairwise analysis and used the GRADE approach to assess certainty of evidence for each patient important outcome and translated relative effects to absolute effect estimates. Lastly, we also conducted a network meta-analysis to assess comparative effectiveness of different CAR-T cell products using mixed treatment comparisons. However, this study is limited by a small number of included trials. Mixed treatment comparisons were based on an open network with sparse direct evidence which precluded the formal assessment of publication bias and incoherence. Median follow up durations varied across trials, and OS analyses in the ZUMA-7 (7) and TRANSFORM (6) trials were interim. Hence, mature OS data at longer follow up might offer different insights. We believe this study is timely as it examines data in totality, provides precise estimates for treatment effects of CAR-T cell therapy compared to SOC across different outcomes and presents a comprehensive assessment of comparative effectiveness of different CAR-T cell products.

In summary, patients with R/R DLBCL harbor considerable disease heterogeneity and we need to tailor the choice of therapy carefully based on individual patient and associated disease factors. CAR-T cell therapy can be a potential second line treatment option for patients with primary refractory DLBCL or patients relapsing within 12 months of their first line chemoimmunotherapy. While patients relapsing more than 12 months after their first line treatment or those with chemosensitive disease may still benefit from ASCT.

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding authors.

KA: Writing – original draft, Writing – review & editing. MZ: Writing – original draft, Writing – review & editing. EH: Writing – review & editing. AA: Writing – review & editing, Writing – original draft. SA: Writing – original draft, Writing – review & editing. NA: Writing – original draft, Writing – review & editing. AS: Writing – review & editing. JD: Writing – review & editing. JT: Writing – review & editing. BD: Writing – review & editing. AT: Writing – original draft, Writing – review & editing. MH: Writing – original draft, Writing – review & editing.