Predictive power to assist phase 3 go/no go decision based on phase 2 data on a different endpoint.

One of the most critical decision points during drug development is to make a phase 3 go/no go decision after a phase 2 proof of concept trial is conducted. It is particularly challenging in oncology drug development where oftentimes the primary endpoint for phase 3 trial is overall survival (OS), but the phase 2 proof of concept trial is powered only for an early endpoint, typically progression-free survival (PFS), whose relationship to OS is often unclear. We propose the use of predictive power to assist the phase 3 go/no go decision by evaluating the strength of actual observed phase 2 efficacy effects in terms of how likely it will predict the chance of OS success in the subsequent phase 3 trial. The formula is provided for calculation of predictive power based on either observed PFS effect only, or observed OS effect only, or both. An example is provided to compare these three predictive powers, which shows that when there is little prior information about PFS and OS, the predictability based on the observed phase 2 PFS effect is low and not sensitive to the size of the trial and extent of the observed PFS effect, and it also has limited added value to the predictability based on the observed phase 2 OS effect. Therefore, one should be cautious of inherently high risk of making a phase 3 go/no go decision based on phase 2 PFS outcome alone and should take the phase 2 OS data into consideration whenever possible.