Network meta-analysis of targeted therapies for diffuse large B cell lymphoma

This meta-analysis was performed according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology [12]. Throughout December 2019, we systematically searched the PubMed, EmBase, and Cochrane Central Register of Controlled trials databases with the following keywords: (“Diffuse large B cell lymphoma” OR “DLBCL”) AND (“random” or “blind”). We also conducted manual searches of reference lists from all the relevant original and review articles to identify additional eligible studies.

The literature search and study selection were independently carried out by two authors, and any disagreement was resolved by a third author. Studies were included if the following inclusion criteria were met: (1) Patients: all patients were diagnosed with DLBCL; (2) Intervention: rituximab-, I-tositumomab-, bevacizumab-, bortezomib-, dacetuzumab-, ibrutinib-, ofatumumab-, obinutuzumab-, or lenalidomide-based treatment regimens were used; (3) Control: chemotherapy or rituximab-based chemotherapy was used as a control; (4) Outcomes: the primary outcomes were overall survival (OS), events-free survival (EFS), and overall response rate (ORR), while the secondary outcomes included any potential adverse events; and (5) Study design: all included studies had to have a randomized controlled trial (RCT) design. Exclusion criteria included basic studies and genotype-related studies. Further, reviews, editorials, letters, and conference papers without sufficient data were excluded.

The following items were extracted from each study: first authors’ surname, publication year, country, sample size, mean age, number of men and women, disease status, stage, intervention, chemotherapy regimen, and reported outcomes. The methodological quality of the included studies was assessed using the JADAD scale, which is based on the following five items: randomization, concealment of the treatment allocation, blinding, completeness of follow-up, and the use of intention-to-treat analysis [13]. Data extraction and quality assessment were conducted independently by two authors. Information was examined and adjudicated independently by an additional author referring to the original studies.

A network meta-analysis was conducted for indirect and mixed comparisons of various agents [14]. The loop-specific approach, which assesses the difference between direct and indirect estimates for a specific comparison in the loop, was employed to check for the presence of inconsistency [15]. A design-by treatment interaction inconsistency model was used to check the assumption of consistency across the entire network [14]. After this, an inconsistent model was employed due to the potential heterogeneity among included patients. The surface under the cumulative ranking (SUCRA) probabilities were calculated to rank the treatments for each outcome [16]. Publication biases for primary outcomes were calculated using comparison-adjusted funnel plots [17]. Moreover, the pooled results for potential grade 3 or greater adverse events were calculated using the relative risk (RR) with corresponding 95% confidence intervals (CI) using a random-effects model [18, 19]. The potential impacts of disease status on the prognosis of DLBCL were also illustrated by subgroup analysis. Heterogeneity across included trials was calculated using the I² and Q statistics, and P < 0.10 was considered as significant heterogeneity [20, 21]. A two-side p value < 0.05 was considered statistically significant for all analysis. All statistical analyses were performed using STATA (Version 10.0, Stata Corp., College Station, TX, USA).

In our initial searches, 326 articles were identified from electronic databases, and 201 articles remained after duplicates were removed. One hundred and forty-seven studies were excluded due to irrelevance after checking titles and abstracts. The remaining 54 studies were retrieved for full-text evaluation, and 36 studies were excluded due to the following reasons: affiliate study (n = 19), insufficient data (n = 9), and no appropriate control (n = 8). Reviewing the reference lists of relevant studies did not yield any new eligible studies. Eventually, 18 RCTs assessing 8207 DLBCL patients were collected in our study [22‐39] (Fig. 1).

The baseline characteristics of the included studies are presented in Table 1. To summarize, the studies were published from 2002 to 2019, and 102–1418 patients were included in individual trials. Ten of the included studies were conducted in multiple countries, and the remaining eight studies were conducted in a single country. Mean age of included patients ranged from 47.0–69.5 years, and the disease status ranged from low to high risk. The quality of included studies was evaluated using the JADAD scale, and studies with a score of 4 or 5 were regarded as high quality. Overall, six studies scored 4, six studies scored 3, four studies scored 2, and the remaining two studies scored 1.

The eligible comparisons of OS in the network plot including I-tositumomab, bevacizumab plus rituximab, bortezomib plus rituximab, dacetuzumab plus rituximab, ibrutinib plus rituximab, lenalidomide, obinutuzumab, ofatumumab, and rituximab treatments are presented in Supplemental 1. The nodes and the edges are weighted based on the number of studies in each treatment and on the precision of the direct pair-wise comparison, respectively. The SUCRA probabilities (%) were ranked to obtain comparative effects of these agents on OS, and the results indicated that the addition of dacetuzumab (74.8%) to rituximab-based regimens and lenalidomide (77.1%) was the most likely to improve OS (Fig. 2). The results of pair-wise comparisons agents are presented in Supplementals 2 and 3. Finally, no significant publication bias was detected through reviewing the funnel plot (Supplemental 4).

The eligible comparisons of EFS in the network plot including I-tositumomab, bevacizumab plus rituximab, bortezomib plus rituximab, dacetuzumab plus rituximab, ibrutinib plus rituximab, lenalidomide, obinutuzumab, ofatumumab, and rituximab are presented in Supplemental 1. The results of the SUCRA probabilities (%) indicated that dacetuzumab (80.4%) or bortezomib (70.8%) added to rituximab was the treatment most likely to improve EFS (Fig. 3). Supplementals 2 and 3 shows the details regarding the therapeutic effects of pair-wise comparisons agents on EFS. No significant publication bias was observed upon reviewing the funnel plot (Supplemental 4).

The network meta-analysis comparing the effects of various agents including I-tositumomab, bevacizumab plus rituximab, bortezomib plus rituximab, dacetuzumab plus rituximab, ibrutinib plus rituximab, lenalidomide, obinutuzumab, ofatumumab, and rituximab on ORR is presented in Supplemental 1. The SUCRA probabilities (%) indicated that lenalidomide (93.8%) produced the best therapeutic effect on ORR, and that I-tositumomab (77.2%) had a relatively good effect on ORR (Fig. 4). The results of pair-wise comparisons agents on ORR are listed in Supplementals 2 and 3. The funnel plot showed that there was no publication bias (Supplemental 4).

We firstly noted targeted therapies are significantly associated with improved OS, irrespective for previous untreated patients (HR: 0.82; 95%CI: 0.71–0.95; P = 0.008) or patients with relapsed or refractory DLBCL (HR: 0.85; 95%CI: 0.75–0.97; P = 0.016). Moreover, targeted therapies could significantly improved EFS for previous untreated patients (HR: 0.76; 95%CI: 0.63–0.92; P = 0.005), while it did not yield significant improvement in EFS for patients with relapsed or refractory DLBCL (HR: 0.77; 95%CI: 0.58–1.04; P = 0.084). Finally, targeted therapies have no significant effects on ORR, irrespective for previous untreated patients (RR: 1.00; 95%CI: 0.96–1.05; P = 0.869) or patients with relapsed or refractory DLBCL (RR: 1.11; 95%CI: 0.91–1.34; P = 0.302) (Table 2).

The pooled results for the targeted therapies on the risk of grade 3 or greater adverse events are summarized in Table 3. Overall, we noted that targeted therapies were associated with an increased risk of diarrhea (RR: 2.63; 95%CI: 1.18–5.86; P = 0.019), and thrombocytopenia (RR: 1.48; 95%CI: 1.08–2.02; P = 0.015), whereas no significant differences were observed among groups for the risks of fever (P = 0.470), infection (P = 0.267), mucositis (P = 0.615), liver toxicity (P = 0.307), cardiac toxicity (P = 0.197), neurologic toxicity (P = 0.393), renal toxicity (P = 0.136), lung toxicity (P = 0.539), nausea or vomiting (P = 0.232), constipation (P = 0.560), neutropenia (P = 0.363), anemia (P = 0.096), leucocytopenia (P = 0.342), or granulocytopenia (P = 0.549).