Assessing contribution of treatment phases through tipping point analyses using rank preserving structural failure time models

TL;DR

This paper introduces a novel method combining tipping point analysis and RPSFT modeling to evaluate the contribution of specific treatment phases in multi-phase clinical trials, demonstrated with cancer trial data.

Contribution

It develops a new approach for assessing individual treatment phase contributions in multi-phase trials using combined tipping point and RPSFT methods.

Findings

Effective in identifying the impact of treatment phases

Applicable to real-world clinical trial data

Provides tailored thresholds for different inferential goals

Abstract

In clinical trials, an experimental treatment is sometimes added on to a standard of care or control therapy in multiple treatment phases (e.g., concomitant and maintenance phases) to improve patient outcomes. When the new regimen provides meaningful benefit over the control therapy in such cases, it proves difficult to separately assess the contribution of each phase to the overall effect observed. This article provides an approach for assessing the importance of a specific treatment phase in such a situation through tipping point analyses of a time-to-event endpoint using rank-preserving-structural-failure-time (RPSFT) modeling. A tipping-point analysis is commonly used in situations where it is suspected that a statistically significant difference between treatment arms could be a result of missing or unobserved data instead of a real treatment effect. Rank-preserving-structural-failure-time modeling is an approach for causal inference that is typically used to adjust for treatment switching in clinical trials with time to event endpoints. The methodology proposed in this article is an amalgamation of these two ideas to investigate the contribution of a treatment phase of interest to the effect of a regimen comprising multiple treatment phases. We provide two different variants of the method corresponding to two different effects of interest. We provide two different tipping point thresholds depending on inferential goals. The proposed approaches are motivated and illustrated with data from a recently concluded, real-life phase 3 cancer clinical trial. We then conclude with several considerations and recommendations.