Page 155 - 2021_07-Haematologica-web

Meta-analysis of radiotherapy for DLBCL
(aggressive[Title] OR malignant[Title] OR advanced[Title] OR his- tiocytic[Title] OR diffuse[Title] OR undifferentiated[Title] OR mixed[Title] OR high grade[Title] OR centroblastic [Title] OR immunoblastic[Title]).
Inclusion criteria and trial selection
Three investigators independently screened the studies. The flow diagram according to the PRISMA statement8,9 depicted in Figure 1 illustrates the search and selection process. We aimed at identifying randomized trials that had enrolled at least 50 adult patients (≥18 years of age) per arm with newly diagnosed DLBCL (or aggressive lymphomas) at any stage according to the Ann Arbor classification. Patients had to be treated with a CHOP based chemotherapy (+/- rituximab), and randomized to subsequent consolidation radiotherapy or no radiotherapy. The 50-patients cut-off was chosen to exclude therapeutic explorato- ry trials. Although the cut-off is arbitrary, it is safe to assume that no confirmatory trials were excluded given the high (pro- gression-free) survival rates observed in this population. Patients with previously treated or relapsed DLBCL were excluded. The full text report of identified trials was independently checked by the three investigators. Disagreements regarding trial selection were discussed until consensus was found. Each report was scrutinized to eliminate duplicates and to ensure that it was pub- lished as an original article.
Outcome measures
Progression-free survival (PFS) and overall survival (OS) were the outcomes of interest. PFS was considered as tumor progres- sion i.e., growth of the tumor during treatment, relapse i.e., growth after previous shrinkage or stabilization, or death. For trials that did not report outcome data that fit this definition, we used data of an outcome that was as close as possible to this def- inition e.g., event-free survival.
Data extraction
Data was extracted in duplicate and disagreements were resolved by consensus. We used the Cochrane ‘Risk of Bias’ approach to assess methodological quality of trials.10 We used the data from the original publications, from intention-to-treat analyses, and from randomized patients only, and for the longest follow-up available for a particular outcome. The hazard ratio (HR) was used as effect measure for both outcomes. If HR and a measure of precision (standard error, variance, or 95% Confidence Interval [CI]) was not available, we digitized Kaplan-Meier curves, reconstructed the underlying time-to- event data, and calculated (log) HR and standard errors using a Cox regression model. Details on the outcome data of the 11 tri- als are shown in Table 2.
Statistical analysis
Outcome data were pooled with a random-effects model using restricted maximum likelihood. We also did bivariate meta-analysis considering both outcomes in one analysis. Correlation between OS and PFS was estimated from two of the identified trials.11,12 We performed random-effects meta-regres- sion for PFS over time using the mid of enrolment period as an independent covariate. Stratified analyses to explore possible reasons for heterogeneity were also done using meta-regression. Analyses were done using Stata (StataCorp. 2017. Stata Statistical Software: Release 16. College Station, TX, USA). Taking into account criticisms of meta-analysis,8,9 the Online Supplementary Appendix provides additional details on the analy- sis methods used and all outcome data. The latter, used in the meta-analysis, is provided in Table 2.
Results
After deduplication, our search strategy generated 3,181 references (Figure 1). With the aim to identify clin- ical trials that assessed the role of consolidation radiother- apy in a randomized manner as part of the first-line ther- apy, our search revealed 11 trials amenable for this meta- analysis (details are listed in Table 1). Three of the four trials published by Aviles13-16 on this topic were later retracted.14-16 As of September 2019, these retracted papers have together received a total of 39 citations. Their data are provided in the respective figures, but were excluded from the meta-analyses. One trial was stopped early when the benefit of rituximab became evident,12 or as a result of a planned interim analysis.17 Older trials included lymphomas classified by the Kiel classification18 or included DLBCL according to the Working Formulation.19 Six of the trials included patients with localized disease only, but five of the 11 trials included also advanced stages. With the exception of the GELA LNH 93-111 where doxorubicin, cyclophosphamide, vin- desine, bleomycin and prednisone (ACVBP) instead of CHOP was given in the comparator arm or SWOG,20 where the non-irradiated patients received eight cycles of CHOP (instead of three), the same chemotherapy was given to the randomized patients. The current standard R-CHOP was used in four of the 11 trials.15-17,21,22 Only the recent Lysa/GOELAMS 02 03 trial21 used PET, although not for guided treatment. Radiotherapy was given to both localized stages 1 and 2, but also advanced disease, and either to all or only patients in complete remission or bulky disease. GOELAMS 02 0321 was a non-inferiority trial whereas all other trials used for this meta-analysis used a superiority design.
Seven trials with a total of 2,488 patients contributed to the analysis of the primary endpoint PFS (Figure 2). Data were extracted from the original publications. The UNFOLDER trial was presented in part at the 12th International Congress on Malignant Lymphomas,17 and again at the American Society of Clinical Oncology (ASCO) 2018,22 albeit with different endpoints. The latter have been used for this meta-analysis. Data from Engelhard18 was not available for the PFS analysis (Table 2). For PFS, the pooled HR was 0.77 (95% CI: 0.51-1.17), and in the pooled bivariate meta-analysis HR was 0.80 (95% CI: 0.53-1.21) (Figure 2). For OS, eight trials with a total of 2,744 patients were included. The pooled HR was 0.93 (95% CI: 0.61-1.40) and 0.86 (95% CI: 0.58-1.27) in the bivariate meta-analysis (Figure 3). Between-trial het- erogeneity was high for both outcomes (PFS, tau2: 0.25, I2: 85%; OS, tau2: 0.25, I2: 74%). The total of 4,584 patients included in this meta-analysis were recruited between 1983 and 2013. However, the lack of benefit of the com- bined treatment modality remained stable over time, and time alone cannot explain the observed heterogeneity in the meta-analysis (P-value for time trend =0.95; tau2: 0.32; I2: 88%; Figure 4).
Given the significant heterogeneity (see also Table 1), we analyzed the data by using the following stratifica- tions: i) the applied chemotherapy was similar in both arms, ii) whether rituximab was used, iii) the dose of radiotherapy, and iv) whether the radiotherapy dose was given only in complete morphologic remission. In addi- tion, we stratified according to the following trial popula- tion characteristics: v) mean age of the treated patients,
haematologica | 2021; 106(7)
1925