Impact of Dose-Difference in Comparator Arms Among Clinical Trials on Treatment Effect Size in a Network Meta-Analysis: A Case of First-Line Therapies in Chronic Lymphocytic Leukemia (CLL)

Introduction:

Network meta-analyses (NMA) are used commonly to allow formal comparisons between drugs that were not compared head-to-head in a clinical trial through a common comparator. Two main assumptions are made to allow such methodology: a) the study samples are drawn from similar populations and b) the common comparator in the trials represents the same intervention. Clinical trials for first-line treatment of chronic lymphocytic leukemia (CLL) offer an opportunity to perform NMA through a common comparator chlorambucil (CHL). In cases where the common comparators differ, the results of the NMA may be impacted; for example, the dose for CHL for most trials was 0.5 to 0.8 mg/kg of body weight given orally on days 1 and 15 of each 28-day cycle (total dose = 70 to 112 mg; denoted as CHL-low arm) in CLL-11 for obinutuzumab (OBI)-CHL and rituximab (RTX)-CHL, in RESONATE for ibrutinib (IBR), and in Knauf 2012 for bendamustine (BEN). The dose for CHL for COMPLEMENT-1 trial for ofatumumab (OFA)-CHL was 10 mg/m² per day given orally on days 1-7 of each 28 day cycle (total dose = 132 mg; denoted as CHL-high arm). We sought to evaluate the impact of the dose difference, CHL0.5 vs. CHL10 on efficacy outcomes in CLL patients receiving these treatments.

Methods:

A systematic literature review was conducted in PubMed, Embase, and Cochrane library for the time period of 2010-2016 and in conference proceedings of American Society of Hematology (ASH), the American Society of Clinical Oncology (ASCO), and the European Hematology Association (EHA) for 2014-2016. Search was limited to clinical trials conducted on humans and published in English language. Randomized clinical trials (RCTs) evaluating first-line CLL therapies were included. Bayesian NMA of efficacy outcomes, including progression-free survival (PFS), complete response (CR), and overall response rate (ORR) were conducted without accounting for CHL dose-difference (base-case analyses) and with accounting for CHL dose-difference (secondary analyses) using SAS® (v9.3). For the base case analyses, the CHL-high and CHL-low comparator arms were assumed to have similar effects and treated as the same intervention. For secondary analyses, the two doses of CHL were evaluated as different interventions and trials where fludarabine treatment arms were a common comparator was included.

Results:

Of the 77 studies screened, four RCTs (n = 1,624) were included in the NMA. In the base-case NMA where CHL dose-difference is not accounted, hazard ratios (HRs) for PFS with OFA-CHL in comparison to OBI-CHL and RTX-CHL were both non-significant (95% CI: 0.01, >100), the odds ratios (95% CI) for ORR with OFA-CHL in comparison to OBI-CHL and RTX-CHL were both non-significant, and the odds ratios for CR with OFA-CHL in comparison to OBI-CHL and RTX-CHL were 0.07 (0.01-0.90) and 0.29 (0.01-6.03) respectively. In the secondary NMA where CHL dose difference is taken into account, HRs for PFS and odds ratios for ORR with OFA-CHL in comparison to OBI-CHL and RTX-CHL were both also non-significant, and the odds ratios for CR were 0.03 (0.01- >100) and 0.01 (0.01- >100) respectively.

Conclusion:

The differences in estimations and their respective confidence intervals demonstrate the importance of employing appropriate methodology and clinical considerations in NMA. In this case, conclusions arising from the results of two scenarios differed based on accounting for any clinically significant dose differences in "a potential common comparator." When CHL dose was not accounted for, OBI-CHL appeared to be superior to OFA-CHL in CR. After accounting for the dose-difference, OBI-CHL, OFA-CHL and RTX-CHL were similar to each other. Accurate estimation of treatment effects is important as these results are used to inform health policies and treatment guidelines through regulatory or HTA bodies and professional clinical bodies respectively. These data are also used to inform health economic models and budget impact models. Inaccurate results are propagated to future evidence synthesis efforts that further misinform policies and guidelines.