Adjusting for the Confounding Effects of Treatment Switching-The BREAK-3 Trial: Dabrafenib Versus Dacarbazine

Abstract

Background: Patients with previously untreated BRAF V600E mutation-positive melanoma in BREAK-3 showed a median overall survival (OS) of 18.2 months for dabrafenib versus 15.6 months for dacarbazine (hazard ratio [HR], 0.76; 95% confidence interval, 0.48-1.21). Because patients receiving dacarbazine were allowed to switch to dabrafenib at disease progression, we attempted to adjust for the confounding effects on OS.

Materials and methods: Rank preserving structural failure time models (RPSFTMs) and the iterative parameter estimation (IPE) algorithm were used. Two analyses, "treatment group" (assumes treatment effect could continue until death) and "on-treatment observed" (assumes treatment effect disappears with discontinuation), were used to test the assumptions around the durability of the treatment effect.

Results: A total of 36 of 63 patients (57%) receiving dacarbazine switched to dabrafenib. The adjusted OS HRs ranged from 0.50 to 0.55, depending on the analysis. The RPSFTM and IPE "treatment group" and "on-treatment observed" analyses performed similarly well.

Conclusion: RPSFTM and IPE analyses resulted in point estimates for the OS HR that indicate a substantial increase in the treatment effect compared with the unadjusted OS HR of 0.76. The results are uncertain because of the assumptions associated with the adjustment methods. The confidence intervals continued to cross 1.00; thus, the adjusted estimates did not provide statistically significant evidence of a treatment benefit on survival. However, it is clear that a standard intention-to-treat analysis will be confounded in the presence of treatment switching-a reliance on unadjusted analyses could lead to inappropriate practice. Adjustment analyses provide useful additional information on the estimated treatment effects to inform decision making.

Implications for practice: Treatment switching is common in oncology trials, and the implications of this for the interpretation of the clinical effectiveness and cost-effectiveness of the novel treatment are important to consider. If patients who switch treatments benefit from the experimental treatment and a standard intention-to-treat analysis is conducted, the overall survival advantage associated with the new treatment could be underestimated. The present study applied established statistical methods to adjust for treatment switching in a trial that compared dabrafenib and dacarbazine for metastatic melanoma. The results showed that this led to a substantially increased estimate of the overall survival treatment effect associated with dabrafenib.

Trial registration: ClinicalTrials.gov NCT01227889.

Keywords: BRAF V600E mutation-positive melanoma; Dabrafenib; Dacarbazine; Overall survival; Survival analysis.

Conflict of interest statement

Disclosures of potential conflicts of interest may be found at the end of this article.

©AlphaMed Press.

Figures

Figure 1.

The potential impact of treatment switching. The first 2 rows illustrate a “perfect” RCT in which no treatment switching occurs and the new intervention extends progression-free survival (PFS) and postprogression survival (PPS). From such a study, the true overall survival (OS) difference can be estimated and the intent-to-treat (ITT) analysis will be a robust estimate of the treatment effect (ignoring any need for extrapolation). The third row illustrates the potential impact on survival in the control group, if switching of treatment is permitted after disease progression. In such a study, PFS is unaffected, but, under the assumption that control group patients who switch treatment will benefit from the new intervention after disease progression, PPS is longer if patients switch treatment than if patients did not switch treatment. Therefore the ITT analysis (RCT OS difference) can underestimate the true effect of treatment on OS. From [9], with permission from SAGE Publications, Inc. Abbreviations: OS, overall survival; PFS, progression-free survival; PPS, postprogression survival; RCT, randomized controlled trial.

Figure 2.

Study schema for the BREAK-3 trial (December 2012) [7].

Figure 3.

Unadjusted overall survival by randomized treatment for patients receiving dabrafenib or dacarbazine in the BREAK-3 trial (December 2012) [7]. Abbreviations: CI, confidence interval; HR, hazard ratio; NR, not reported.

Figure 4.

Observed and counterfactual Kaplan-Meier curves for overall survival from rank preserving structural failure time models “treatment group” (A) and “on-treatment observed” (B) analyses. Abbreviation: ITT, intent-to-treat.