What is the statistical rationale for a phase 2 clinical trial?

Question

There is a lot of contradictory information about the purpose of phase 2 clinical trials.

Many sources claim that the aim is to test whether a treatment works, and sample size calculators exist to compare a response rate in a treatment group (as a binary statistic) to a historical control.

But it's clear that a phase 2 trial is almost by definition underpowered to detect minimum clinically important differences, the use of a binary endpoint would reduce the power further, and if not randomised then the lack of concurrent control means that any positive detected effect is very likely to arise from some bias or other.

So, if we can't trust positive or negative results from these trials (because they are underpowered or biased) then what, from a statistical point of view, can we hope to learn about efficacy?

What research objective is actually being met by a typical phase 2 design?

Answer 1

What area are we talking about? Oncology or non-Oncology?

• non-Oncology: There's often two different types:
  • Phase 2a: Often "proof of concept" type of trials that try to give enough efficacy evidence to make the decision to continue further development sensible. If one simulates these things out with realistic assumptions (e.g. many drugs do not work all that well, realistic costs, discounted value of potential future drug far in the future) and assumes that one also has other promising options to invest money/effort and allocate patients to trials, it often turns out that makes sense to have such an early hurdle, in part of efforts involved in subsequent stages. The powering of such studies is usually a decision theoretic trade-off between their cost vs. the value for (potentially) avoiding future cost or making future investment more certain.
  • Phase 2b: Dose finding (often parallel group or sometimes cross-over studies) to identify doses that provide the right trade-off between efficacy and safety. Often involving extensive modeling (e.g. using MCP-Mod or similar approaches). Again, there's huge costs in only determining your final dose in Phase 3 (including longer-term studies in Ph3, required size of the safety database for the dose to be approved, if not already done developing a final market image of a drug etc.) so that it makes sense to find a single dose that is promising to take forward. These types of studies are usually pretty well powered (at least for an analysis that uses modeling of dose-response and potentially tries to exploit things like patterns over time etc., and possibly uses shorter-term or different outcomes than Phase 3).
• Oncology:
  • Traditionally dose finding has not been as much of a focus in Oncology, but that may be changing due to regulatory changes.
  • You seem to be referencing a particular type of (often) single arm nature, where one looks at something like an outcome such as "objective response" (=tumor shrinking sufficiently based on specific criteria). There are several complications here. * Firstly, is this a surrogate outcome for what we really care about (in Oncology usually people surviving longer), which may not be the same between different cancer types. * Secondly, are we sure there's a causal effect of the drug? That is not always clear if there's no control group. Formally, one could do a Bayesian comparison with historical controls (e.g. using robust meta-analytic predictive priors or any of the many alternatives). Often it's rather looked at in terms of threshold crossing, which in the end may be justified on a Bayesian basis or in other ways. Lots of issues play into this like sampling variation, changes in medical practice, other differences between current patients and historical controls etc. On a whole, the ICH E10 guidance is rather sensible on this topic, which is rather thoroughly debated with lots of references to whether medical interventions are parachutes or not. * If the assumptions are done sensibly, such studies can be well powered. And, of course, their interpretation may be affected by the knowledge that underpowered studies are more likely to produce false positive "significant" results. * Depending on how regulators judge the previous points in combination with the unmet need, the results of such trials might lead to accelerated/conditional approval (or not), but then there's usually still a Phase 3 trial that looks at progression free survival/overall survival.

In the end, a lot may just come down to eNPV optimization.

Answer 2

I would just like to add to Björn's excellent answer but to focus on the pharmacokinetic (PK) and pharmacodynamic (PD) work that is usually carried out in Phase II as part of what Björn talks about. These trials, especially the 2b phase, are important in refining our understanding of a drug's behaviour in the target population.

In Phase II, particularly in non-oncology settings as Björn mentioned, the main goal is often to identify the optimal dosage and understand the drug's kinetics and dynamics. This intensifies the focus on PK/PD aspects and involves fairly complex mathematical and statistical modelling, extending the initial models developed in Phase I (which are usually based on data from healthy humans). In Phase II, these models are adapted and expanded to the target patient population, aiming to determine key factors such as:

Efficacy and Safety Windows: This involves identifying the range of doses at which the drug is both effective and safe for the target population.

Minimum/Maximum Effective Dose: Determining the smallest dose that produces the desired therapeutic effect and the maximum dose before adverse effects outweigh benefits.

Maximum Tolerated Dose: Establishing the highest dose that the target population can tolerate without significant side effects.

Timing of Efficacy and Tolerability Effects: Understanding how long it takes for the drug to exhibit therapeutic effects and at what point side effects might become intolerable.

Titration Steps and Dosing Intervals: Developing guidelines for safely increasing the drug dose and determining the frequency at which the drug should be administered.

Identification of Potential Subgroups for Dose Adjustment in Phase III: This involves recognising specific patient subgroups that might need different dosing strategies in the subsequent phase of the trial.