Adjusting For Treatment Switching in Rcts – Identifying, Analysing and Justifying Appropriate Methods: a Case Study in Metastatic Melanoma

Abstract

Participants in oncology randomised controlled trials (RCT) are often permitted to switch from their randomised treatments onto alternative treatments. Intention-to-treat (ITT) assessments are prone to bias in the presence of switching and consequently the overall survival (OS) benefit and cost-effectiveness of the novel treatment may be underestimated. Regardless, decision-makers frequently reject statistical analyses which adjust for treatment switching in health technology assessment, often due to poor justification of methodological assumptions. This study applies adjustment methods to an RCT comparing dabrafenib to dacarbazine in patients with BRAF V600E/K mutation-positive metastatic melanoma, and investigates which adjustment method best fits this specific case study. The adjustment methods applied included the Rank Preserving Structural Failure Time Model (RPSFTM), Inverse Probability of Censoring Weights (IPCW), and two-stage adjustment. The suitability of each method was assessed by investigating their assumptions and trial characteristics. 37/63 (58.7%) dacarbazine patients switched to dabrafenib (direct switching). Also, 16 (25.4%) dacarbazine patients and 27 (14.4%) of 187 dabrafenib patients received other small molecule targeted treatments post-study (indirect switching). The ITT hazard ratio (HR) for OS was 0.81 (95% confidence intervals (CI) 0.56 - 1.16,) favouring dabrafenib. An RPSFTM analysis to adjust for direct switching, combined with a two-stage analysis to adjust for indirect switching, appeared most appropriate, producing an adjusted HR of 0.68 (95% CI 0.33 - 1.63). It was not possible to robustly adjust for direct and indirect switching simultaneously using standalone IPCW or two-stage methods due to small patient and event numbers. Whilst it is not possible to perfectly test the common treatment effect assumption required by the RPSFTM, our investigations did not find strong evidence against this. Adjusting for switching showed an increased OS effect for dabrafenib. Methodological assumptions must be rigorously investigated to demonstrate whether and which adjustment methods are justified.