It is time to rethink randomized clinical trial approaches

This letter to the editor comments on the recently published editorial by Fojo, LaRose, and Bates, and agrees that changes are needed in clinical research approaches.

It makes little sense for randomized controlled trials (RCTs) to compare drugs with differing mechanisms of action. 2 Therapies may preferentially benefit different subpopulations.The rational approach is to offer one then the other, not one instead of the other.One might reasonably assess optimal drug sequence (A->B vs B->A), but it is unreasonable to discard one based on A-vs-B RCTs if each may benefit patients refractory to the other. 2n CodeBreak200, 3 median progression-free survival (PFS) was 1.1 months longer with sotorasib vs docetaxel.However, as frequently happens, 4 medians were distorted by Kaplan-Meier midpoint deviations.Half-life assessments mitigate this.Curve exponential decay nonlinear regression analysis using published methods 4 indicated PFS half-lives of 6.1-vs-4.3months with sotorasib-vs-docetaxel.
For most therapies, if overall survival (OS) is undistorted by crossover, PFS half-life gains ≥ 1.5 months indicate a 90% probability of OS gains ≥ 2 months. 5Crossover ≥ 20% substantially reduces probability that OS differences will achieve significance. 4While OS is important, it can be an unreliable RCT endpoint due to crossover 4 and long post-progression survival. 6We agree with Fojo et al, 1 we must rethink OS as a primary RCT endpoint. 5e also agree that further trials should primarily assess who is most likely to benefit.This is most efficiently done using response as the outcome measure. 2 This can subsequently be confirmed with non-randomized PFS/OS assessments.Subsequent assessment of real-world patients managed using different approaches can elucidate whether a factor is prognostic rather than predictive.We do not need RCTs for this.
Fojo et al are correct that targeted therapies can be toxic. 1However, the VP4-2023 trial demonstrated that dose reductions reduced sotorasib toxicity much more than efficacy. 7any therapies have relatively flat dose-response curves.
With respect to setting the bar higher, 1 it depends.Using RCTs to assess efficacy requires high accrual to show small gains. 8This slows progress by deviating resources from assessment of other options.Far fewer patients are needed to demonstrate high gains. 2A higher bar reduces trial costs.
However, only 20-30 patients are needed in phases I-II trials if single agent response rate is used to assess efficacy.In 25 trials of placebo/best supportive care, median objective response rate was 1%. 2 The highest response rate was 4%.Hence, if even a minority of patients (≥10%?) experience durable responses, this may indicate a drug worth trying in patients with no good alternatives if toxicity and price are reasonable.RCTs are needed to confirm modest contributions of a drug to a combination, 2 but are unnecessary with single agents.
Overall, we should rethink RCT approaches in assessing new therapies. 2

Conflicts of interest
None directly related to this work.Unrelated: honoraria or advisory board or consulting fees from Merck, AstraZeneca, Abbvie, Pfizer and Amgen; owns a 3% interest in US Patent no.9.675.663(a test to predict response to TUSC2/FUS1 gene therapy).The Ottawa Hospital receives research support from a broad range of pharmaceutical companies.Royalties from book "A Short Primer on Why Cancer Still Sucks".