This systematic review and meta-analysis were conducted according to the reporting guidelines in Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Meta-Analysis of Observational Studies in Epidemiology [18]. The authors performed a comprehensive search of the literature in MEDLINE, Embase, and the Cochrane Central Register of Controlled Trial databases with language restriction in English from inception to 30 June 2022. The main keywords for the literature search were b-cell lymphoma and prospective phase II or III study, and the details of the search strategy are described in the supplementary Methods. Additionally, the meeting abstracts from the American Society of Clinical Oncology, American Society of Hematology, European Society for Medical Oncology, and European Hematology Association and the references of pertinent articles were manually scanned to identify additional relevant studies.

Prospective phase II or III trials evaluating the efficacy of systemic treatment for R/R DLBCL were selected for meta-analysis. Additional inclusion criteria were full-text articles and report of the primary outcome. Studies with the following characteristics were excluded: (1) no specific subtype outcome or characteristics of DLBCL among several types of B-cell lymphoma; (2) conducted before the first-line rituximab era; (3) mainly comprised patients with central nervous system lymphoma, b-cell cutaneous lymphoma, human immunodeficiency virus-related lymphoma, or primary mediastinal lymphoma; (4) studied the efficacy of radiotherapy only; (5) studied the efficacy of ASCT only; (6) had no information on previous treatments; and (7) studied a pediatric population.

Two authors independently extracted the following records from both ASCT-eligible and ineligible studies (J.K. and J.C.): trial identifier, published journal, publication year, name of the first author, trial phase, country where trial was conducted, treatment regimens, treatment category, and median follow-up duration. Clinical information of median age (range); median number of prior therapies (range); complete remission (CR) or partial remission (PR) rate; total number of patients; and proportions of patients with age ≥ 65 years, male gender, de novo DLBCL, germinal center B-cell like subtype, stage III/IV, Eastern Cooperative Oncology Group ≥ 2, and previous anti-CD20 treatment were also collected. In the ASCT-eligible studies, the proportion of patients with CR duration from front-line treatment (< or ≥ 12 months), who proceeded to primary refractoriness, high lactate dehydrogenase, International Prognostic Index (IPI) 4/5, second-line age-adjusted IPI 2/3, and ASCT were additionally collected. In the ASCT-ineligible studies, the percentage of prior ASCT, refractory to last treatment, and IPI 0–1/2–3/4–5 were obtained. The other two authors (S.E.Y. and S.J.K.) resolved discrepancies in the extracted data.

Based on current guidelines, classification of the chemotherapeutic agents for R/R DLBCL was performed [13]. In the group intending to proceed to ASCT, the regimen containing cytotoxic chemotherapy as the main backbone and attempting ASCT in responding patients was classified as chemotherapy, while CAR-T cell therapy was separately grouped. The regimens not recommended in the current guidelines were categorized as “others.” In the non-candidates for ASCT group, treatment consisting primarily of chemotherapy was classified as chemotherapy. Several targeted agents were classified into each category itself if specified in the guidelines. Accordingly, three treatment categories were generated in the ASCT-eligible group and nine in the ASCT-ineligible group. The regimens used in the ASCT-eligible studies were (1) CAR T-cell therapy, including axicabtagene ciloleucel [6], lisocabtagene maraleucel [7], and tisagenlecleucel [11]; (2) Chemotherapy, including ICD (irinotecan, cisplatin, and dexamethasone) [19], R (rituximab)-ICE (ifosfamide, carboplatin, and etoposide) [4,20], R-DHAP (cytarabine, cisplatin, and dexamethasone) [4,21,22], R-NIMP (vinorelbine, ifosfamide, mitoxantrone, and prednisolone) [23], ofatumumab plus DHAP [22,24], ofatumumab plus ICE [24], R-GDP (gemcitabine, dexamethasone, and cisplatin) [21,25], dacetuzumab plus R-ICE [20], R-OAD (oxaliplatin, cytarabine, and dexamethasone) [26], ofatumumab plus BCE (bendamustine, carboplatin, and etoposide) [27], temsirolimus plus R-DHAP [28], and platinum-based chemoimmunotherapy [6,7,11] used as a control in recent randomized trials that evaluated the efficacy of CAR T-cell therapy; and (3) others, including R-lenalidomide [29], inotuzumab ozogamicin [30], blinatumomab [31], and R-gemcitabine plus lenalidomide [25].

Treatments in the ASCT-ineligible studies were grouped as follows: (1) CAR T-cell therapy [5,32], including axicabtagene ciloleucel [8], lisocabtagene maraleucel [10], and tisagenlecleucel [9]; (2) chemotherapy, including R-Gemox (gemcitabine and oxaliplatin) [33,34], R-B (bendamustine) [14,35–38], pixantrone [39], one of the various chemotherapies [39] (one of vinorelbine, oxaliplatin, ifosfamide, etoposide, mitoxantrone, gemcitabine, or rituximab), R-ifosfamide+etoposide [40], R-pixantrone [41], R-gemcitabine [41], R-PECC (prednisolone, etoposide, chlorambucil, and lomustine) [42], decitabine plus DHAP [43], ibrutinib plus lenalidomide plus R-EPOCH (etoposide, prednisolone, vincristine, cyclophosphamide, and doxorubicin) [44], pixantrone plus obinutuzumab [45], and lenalidomide plus R-GDP [46]; (3) ibrutinib-based therapy, including ibrutinib plus durvalumab [47], ibrutinib plus nivolumab [48], and ibrutinib monotherapy [49]; (4) lenalidomide-based therapy, including R-lenalidomide [50], lenalidomide monotherapy [34], obinutuzumab plus lenalidomide [51], and temsirolimus plus lenalidomide [52]; (5) loncastuximab tesirine [17]; (6) polatuzumab plus bendamustine and rituximab [14,53,54] and R-polatuzumab [55]; (7) selinexor [15]; (8) tafasitamab, including tafasitamab monotherapy [56] and tafasitamab plus lenalidomide [16]; (9) others, including ibritumomab [57,58], temsirolimus [59], ofatumumab [60], R-everolimus [61], dacetuzumab [62], sepantronium [63], belinostat [64], R-coltuximab [65], R-panobinostat [66], mocetinostat [67], buparlisib [68], abexinostat [69], coltuximab [70], entospletinib [71], panobinostat [72,73], R-inotuzumab [37], selumetinib [74], nivolumab [75], R-pinatuzumab [55], and copanlisib [76].

The primary outcome was 1-year progression-free survival (PFS) rate, and the secondary outcome was CR rate according to treatment category. There are two main reasons why 1-year PFS rate was used as a primary endpoint; First, disease progression takes place mostly within the first year and thereafter tends to form a plateau with most treatments, including CAR T-cell therapy, in most included trials. Second, most of the included studies reported 1-year PFS rate, event-free survival rate, or related Kaplan-Meier survival graphs. When the 1-year PFS rate was only demonstrated as a Kaplan-Meier survival graph, the software GetData Graph Digitizer 2.26 (http://getdata-graph-digitizer.com/) was used for digitizing and extracting data. The pooled primary and secondary outcomes and corresponding 95% confidence intervals (CIs) and p-values were calculated using random-effects models. Our model was a single-arm proportion meta-analysis, and the calculations were performed using the inverse variance method. To assess 1-year PFS and the CR rate according to treatment category, subgroup analyses using a Q test were performed. In addition, a meta-regression was conducted to evaluate potential moderators associated with the primary outcome. The patient proportions of stage III/IV, duration of complete response greater or less than 12 months from front-line treatment, and primary refractoriness to previous therapy were included in the ASCT-eligible studies. The proportions of stage III/IV, prior ASCT, refractoriness to the last treatment, and median number of prior lines of treatment were included in the ASCT-ineligible studies. The variables with p-value of < 0.1 in univariate meta-regression or predefined variables with clinical significance, especially the percentage of primary refractoriness to previous therapy in the ASCT-eligible group, were incorporated in the multivariate meta-regression analysis. Meta-regression was executed using the metareg function with its default parameters and a mixed-effects model. Heterogeneity was assessed using the τ² and I² statistics. All statistical tests were two-sided, and p-values ≤ 0.05 were regarded as statistically significant. R studio software (ver. 1.4.1743, R Foundation for Statistical Computing, Vienna, Austria) was utilized for quantitative pooled analysis and meta-regression modeling.

A total of 9,670 records was identified during the initial database search. After removing duplicates and screening the titles and abstracts of the studies, 811 potentially relevant studies remained for detailed review. As described above, additional exclusion criteria were applied to derive the final 17 studies [4,6,7,11,19–31] comprising 26 cohorts and 2,924 ASCT-eligible patients and 53 studies [5,8,9,10,14–17,32–76] of 59 cohorts and 3,617 ASCT-ineligible patients included in the meta-analysis (Fig. 1).

The baseline characteristics of the 26 ASCT-eligible cohorts are outlined in Table 1 and S1 Table, and the baseline characteristics of the 59 ASCT-ineligible cohorts are outlined in Table 2 and S2 Table. In the ASCT-eligible studies, six studies including 12 cohorts were phase 3 randomized trials and were published in 2008–2022. The categories of regimens were classified as follows: 19 cohorts from 14 studies [4,6,7,11,19–28] with chemotherapy; three cohorts from three studies [6,7,11] with CAR T-cell therapy; and four cohorts from four studies [25,29–31] with others. Only one cohort [29] reported a median number of prior systemic therapies of 3, while all others reported a median number of one prior systemic therapy. Studies comparing CAR T-cell therapy and chemotherapy [6,7,11] in randomized controlled trials had no patients who maintained CR for more than 12 months from front-line treatment. All but two cohorts [24] and cohorts treated with CAR T-cell reported various ranges of proportions of patients to proceed to ASCT. In ASCT-ineligible studies, 10 cohorts from five studies [34,37,39,41,55] were randomized trials and were published in 2007–2022. The following nine categories were used to classify regimens: five cohorts from five studies [5,8-10,32] with CAR T-cell therapy; 17 cohorts from 16 studies [14,33–46,61] with chemotherapy; three cohorts from three studies [47–49] with ibrutinib-based therapy; four cohorts from four studies [34,50–52] with lenalidomide-based therapy; one cohort from one study [17] with loncastuximab; four cohorts from four studies [14,53–55] with polatuzumab plus bendamustine and rituximab; one cohort from one study [15] with selinexor; two cohorts from two studies [16,56] with tafasitamab; and 22 cohorts from 21 [37,55,57–60,62–76] studies with others. The median number of previous treatments ranged from 1 to 5, and all CAR T-cell studies reported a median of three previous treatments.

The overall pooled proportion for 1-year PFS in ASCT-eligible studies was 0.33 (95% CI, 0.28 to 0.38). There were no statistically significant differences in 1-year PFS by category (p=0.30); 0.40 (95% CI, 0.15 to 0.65) in the CAR T-cell therapy group, 0.34 (95% CI, 0.30 to 0.37) in the chemotherapy group, and 0.21 (95% CI, 0.05 to 0.38) in others (Fig. 2). Meta-regression analysis with a mixed-effects model using predefined and relevant study-level moderators was conducted to evaluate whether several factors influenced the outcome (Table 3). Univariate meta-regression revealed that proportion of stage III/IV patients, duration of complete response greater or less than 12 months from front-line treatment, and primary refractoriness to previous therapy were not significantly associated with 1-year PFS rate. The only factor associated with the primary outcome was the CR rate for testing the performance of the regression analysis. According to the predefined analysis principle, primary refractoriness to prior treatment was included in the multivariable meta-regression model, and there was still no significant difference between the Chemotherapy and CAR T-cell groups. The pooled CR rate was 0.31 (95% CI, 0.25 to 0.36), and significantly different CR rates were observed according to regimen (p=0.01) (S3 Fig.). CR rate was 0.53 (95% CI, 0.29 to 0.78) for the CAR T-cell therapy group, 0.28 (95% CI, 0.23 to 0.33) for the Chemotherapy group, and 0.20 (95% CI, 0.13 to 0.27) for the others group. The pooled CR rate of CAR T-cell therapy was significantly higher than those of the Chemotherapy and Others groups.

The overall pooled 1-year PFS rate in ASCT-ineligible trials was 0.24 (95% CI, 0.20 to 0.27). The significant differences in 1-year PFS according to category (p < 0.01) were 0.40 (95% CI, 0.35 to 0.46) for CAR T-cell therapy, 0.23 (95% CI, 0.17 to 0.30) for chemotherapy, 0.24 (95% CI, 0.14 to 0.34) for ibrutinib-based therapy, 0.26 (95% CI, 0.18 to 0.35) for lenalidomide-based therapy, 0.26 (95% CI, 0.19 to 0.3,3) for loncastuximab, 0.35 (95% CI, 0.27 to 0.42) for polatuzumab plus bendamustine and rituximab, 0.21 (95% CI, 0.15 to 0.29) for selinexor, 0.47 (95% CI, 0.37 to 0.57) for tafasitamab, and 0.16 (95% CI, 0.11 to 0.21) for others (Fig. 3). Meta-regression analysis demonstrated that patient proportion of stage III/IV, prior ASCT, and refractoriness to the last treatment had no association with the outcome in univariate meta-regression. The median number of prior lines of treatment showed a weak correlation with 1-year PFS rate and was included in the multivariable meta-regression model with regimen. This model revealed that CAR T-cell therapy was significantly better than chemotherapy, ibrutinib-based therapy, lenalidomide-based therapy, and selinexor but showed no difference in efficacy compared to loncastuximab, polatuzumab plus bendamustine and rituximab, and tafasitamab, with adjustment for median number of previous lines of treatment (Table 4). The pooled proportion of CR was 0.20 (95% CI, 0.16 to 0.24) (S4 Fig.), and CAR T-cell therapy (0.45; 95% CI, 0.37 to 0.54) demonstrated a better CR rate than all other groups except loncastuximab and polatuzumab plus bendamustine and rituximab (data not shown).