Re maturity of OS data - IWFAL
Agree with your observation. If the median OS is not too long, and if patent life and competion is not a concern, most smaller companies prefer to run small trials and wait to extract the most milieage.
It's also true that to date there has yet to surface a silver bullet against cancer, an effect size of 33% prolongation in the median seems to be the upper limit. Even the AVASTIN trial, when paired with an inferior chemo regimen at most reduced the risk by 30%, lasted till 70% to 80% patients had an event.
What's more problematic is in the advanced disease setting, more drugs got approved in later disease stage and proceed to earlier setting, or similar drug got approved in different indications can be used in off-label or on-label after a patient fails a study drug, therefore causing cross-over even if there is no built in cross-over in the trial, which in essence minimized the treatment difference.
In this case, the longer the follow-up, the worse the separation of the curves. It's in the sponsor's interest to either put in clever mechanism to minimize cross-over or delay the time frame a patient can receive later line effective therapy (such as the hurwitz trial which continue to give avastin after subject fails first-line or discontinue chemo), or to front-load accrual and conduct mega trials.
Under the latter senario, the event rate will be low or immature, which means it becomes an art of coaxing the FDA to think that the OS data is mature. Knowing that the longer goes and the earlier the lines of therapy is, the less the separation will there be, the more difficult it is to show survival benefit. I believe they are aware of the fact, but I am curious to hear what you think on how to strike such a balance. WIll it be sufficient to show survival difference, and let the trial run slightly after 50% event to provide a 95% confidence interval, and provide long term OS udpate?
To me this is less evil compared to a large trial with a high event rate powered to detect infinitesimallly small difference and should be allowed.
On the other hand, provenge seemed to have bucked the trend with the whopping treatment effect of 50% prolongation of OS. It also more or less shown there is a greater divergence between the KM curves over long-term followup - so higher event rate, even if it means one needs to wait longer, is probably useful.
