Statistical analysis

We used an adaptive design based on a two-stage approach, with two interim analyses to explore the safety and efficacy in the germinal centre B-cell group treated with RB-CHOP after a defined number of events. The first interim analysis was to take place once 55 patients in the germinal centre B-cell group had been randomly assigned to receive RB-CHOP. If progression-free survival at 12 months was assessed to be below 70% in this subgroup, the trial would stop recruiting into the germinal centre B-cell group. The second interim analysis was to take place when 73 patients in the germinal centre B-cell group had been randomly assigned to receive RB-CHOP and followed up for 1 year. If the progression-free survival at 12 months was assessed to be below 85% in this subgroup, the trial would stop recruiting into the germinal centre B-cell group.

The trial was powered to detect an improvement in progression-free survival at 30 months of 10% in the combined activated B-cell and germinal centre B-cell groups, from 75% in the R-CHOP group to 85% in the RB-CHOP group (corresponding to a hazard ratio [HR] of 0·56), on the basis of a log-rank test with a significance level of 5% (two-sided) and 90% power, requiring a total of 129 events. The intention-to-treat (ITT) population comprised all patients for whom gene-expression profiling was attempted (classified as activated B-cell, germinal centre B-cell, or unclassified subgroups, or for whom gene-expression profiling failed). The safety population was formed of all patients in the ITT population who received at least one dose of any study drug.

We assessed the primary outcome of 30-month progression-free survival in a modified ITT (mITT) population comprising the activated and germinal centre B-cell subgroups who were randomly assigned to receive treatment, using a Cox proportional hazards model, adjusted for cell-of-origin subtype and IPI score.

Secondary outcome analyses included repeating the primary outcome analysis in the activated B-cell ITT population alone, the germinal centre B-cell ITT population, and the unclassified ITT population, adjusting for IPI score only. We produced Kaplan-Meier curves for time-to-event data and we described follow-up maturity by the reverse Kaplan-Meier method. We used summary statistics to describe baseline characteristics for participants in the R-CHOP group, RB-CHOP group, and patients for whom gene-expression profiling had failed in the ITT population, and by cell-of-origin subgroups in the ITT population, with formal comparisons between cell-of-origin subgroups using Pearson χ² tests. Toxicity information was summarised by treatment group, and we did post-hoc analyses to compare toxicity information by treatment using Pearson χ² tests for the safety population. Further post-hoc analyses included repeating the primary outcome analysis and adjusting for time from diagnosis to the start of treatment to ascertain whether or not the interval from diagnosis to treatment affected the progression-free survival outcome. We also did post-hoc analyses to assess progression-free survival and overall survival by treatment group in the mITT population in the IPI low, intermediate, and high score groups, assessed using a Cox proportional hazards model, adjusted for cell-of-origin subtype, and also repeated the primary outcome analysis excluding patients who had a dose reduction in any treatment drug. We made no adjustment for multiple comparisons.

Post-hoc analyses to assess progression-free survival also included: comparing double-hit lymphomas to non-rearranged cases, separated by treatment group; comparing double-expressor lymphomas to all other cases, separated by treatment group; and comparing by treatment groups in subgroups with mutations in components of the NF-κB pathway.

We used Stata statistical software (version 15.1) for all analyses. This trial is registered with ClinicalTrials.gov, number {"type":"clinical-trial","attrs":{"text":"NCT01324596","term_id":"NCT01324596"}}NCT01324596.