Abstract
BackgroundConventional trial design and analysis strategies fail to address the typical challenge of immune-oncology (IO) studies: only a limited percentage of treated patients respond to the experimental treatment. Treating non-responders, we hypothesize, would in part drive non-proportional hazards (NPH) patterns in Kaplan-Meier curves that violates the proportional hazards (PH) assumption required by conventional strategies. Ignoring such violation incurred from treating non-responders in the design and analysis strategy may result in underpowered or even falsely negative studies. Hence, designing innovative IO trials to address such pitfall becomes essential.MethodsWe empirically tested the hypothesis that treating non-responders in studies of inadequate size is sufficient to cause NPH patterns and thereby proposed a novel strategy, p-embedded, to incorporate the dichotomized response incurred from treating non-responders, as measured by the baseline proportion of responders among treated patients p%, into the design and analysis procedures, aiming to ensure an adequate study power when the PH assumption is violated.ResultsEmpirical studies confirmed the hypothetical cause contributes to the manifestation of NPH patterns. Further evaluations revealed a significant quantitative impact of p% on study efficiency. The p-embedded strategy incorporating the properly pre-specified p% ensures an adequate study power whereas the conventional design ignoring it leads to a severe power loss.ConclusionThe p-embedded strategy allows us to quantify the impact of treating non-responders on study efficiency. Implicit in such strategy is the solution to mitigate the occurrence of NPH patterns and enhance the study efficiency for IO trials via enrolling more prospective responders.
Highlights
The unprecedented growth in immuno-oncology (IO) trials has outstripped the development of proper design and analysis strategies
If p% increased to 90%, the overwhelming majority of simulated trials displayed a clear separation of survival curves between arms right after the 3month lag and only 20% of trials demonstrated diverse non-proportional hazards (NPH) patterns
Among trials simulated under a small p% of 20%, a sample size of 2000 patients resulted in 23% of trials involving NPH patterns, whereas N of 200 led to 89% of them showing complex patterns
Summary
The unprecedented growth in immuno-oncology (IO) trials has outstripped the development of proper design and analysis strategies. A variety of complex survival patterns frequently arise in conventionally designed IO trials with time-to-event endpoints, including the delayed treatment effect pattern where Kaplan-Meier (KM) curves for the two treatment groups overlay during the early trial stage (Fig. 1a), the belly-shape diminishing effect pattern where the KM curves first separate join after sufficient follow-up (Fig. 1b), the crossing hazards pattern where the KM curve for treatment starts out being worse than that for active control but ends up being better (Fig. 1c), or the combination patterns that combine the aforementioned patterns in various fashions (Fig. 1d–f) These complex patterns reveal that the underlying hazard rate of the treatment arm is no longer proportional to that of the control arm over time, violating the proportional hazards (PH) assumption required. Treating non-responders, we hypothesize, would in part drive non-proportional hazards (NPH) patterns in KaplanMeier curves that violates the proportional hazards (PH) assumption required by conventional strategies Ignoring such violation incurred from treating non-responders in the design and analysis strategy may result in underpowered or even falsely negative studies.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have