Probability of success of a clinical trial: strategic context
Strategic context
Implementing a DDCP computation based on prior data and getting the relevant numbers, changes in DDCP after not stopping at interim analyses etc. out is certainly an important aspect. However, in a drug development context this is only the starting point. Strategic considerations are important. Summarizing lots of discussions that are available in the various PTS handbook and adding our own experience from projects we invite consideration of the following points.
Definition of “success”
Definition of what “success” of the trial is you compute DDCP for is key. Is it to beat target TPP? Minimal TPP (which is typically equivalent to be statistically significant)? Make sure you and the team understand and are crystal clear on what you are computing DDCP for.
Derivation of the prior
Another important aspect is derivation of the prior for computation of DDCP.
- Ideally, the prior should summarize and synthesize all prior knowledge you have about the treatment effect of interest. An easy example would be availability of a randomized Phase 2 trial. Then a prior, normal or pessimistic as introduced in Rufibach et al. (2016), centered at this estimate could be a good starting point.
- Since effect estimates of Phase 2 are typically optimistic it might make sense to attenuate the observed estimate.
- The variance of the prior could be (inversely) proportional to number of observations (or events).
- If endpoints (or estimands in general, e.g. also handling of intercurrent events) are different between early and pivotal trial an option is to model the association between effect estimates. An example is provided in the Gazyva case study.
- However, do not over-use DDCP computations. If you do not have data you can trust stick to the generic framework. See also the PTS handbooks.
- Sometimes, there is a wish to “back-engineer” a prior from a clinical PTS that the teams want to achieve, e.g. to find a matching prior to a generic PTS. How to do that and what to watch out for is discussed here.
- Collecting and synthesizing all the available evidence on a treatment effect to define a prior is a task for the entire project team: the clinicians and RWD scientists need to know and scan the literature for the data and work with the statistician to define the prior. Do not let the team push that activity to the statistician!