Replies to post #147318 on Biotech Values

[NP1986]

Robert, I don't think anyone is unduly downplaying the results of the trial. You have to remember that the full results have not yet been presented at a scientific meeting, and there are still many unanswered questions.

Firstly, as others have already pointed out, overall survival was not a primary endpoint of the trial. Thus, self-acclaimed "PPHMites" who describe it as having a "gold standard design" are mistaken. If measuring overall survival was the main objective of the trial, unblinding would not have occured until a pre-specified number of deaths had occured. Measuring OS after unblinding introduces potential bias.

Secondly, there are still a lot of unanswered questions about the trial. For instance, were baseline factors such as performance status, EGFR status, etc balanced between all three treatment arms? Furthermore, were there differences in treatment that patients received post-progression in each arm? Are the OS results consistent across all geographic regions?

If the treatment arms were well-balanced and the results are consistent across different subgroups, then it certainly warrants further exploration in a phase III study that is actually designed to assess OS.

Finally, don't forget that this is still a relatively small trial (only ~40 patients in the control arm). There have been a number of instances where promising results in randomized phase II studies weren't replicated in a larger phase III study.

By the way, has PPHM offered hints as to whether there is a difference in OS in the 1st line NSCLC trial?

[iwfal]

For example, if a HR = 0.75 corresponds to an increase in the 1- and 2-year OS rate of 10% and 20%, respectively, between the treatment arms in an advanced non-small cell lung cancer trial, these might be deemed clinically meaningful improvements. If the median survival difference between arms is next considered, then an improvement of 50 days may also be regarded as clinically significant, whereas an improvement of approximately 10 days may not. Only if the descriptive absolute measures reveal a fairly consistent pattern of clinically meaningful improvements should a statistically significant HR be hailed as a clinical advancement.

This is misleading. The author is implying that medians can be trumped by medians (either good or bad). But good luck finding a trial approved by the FDA for having a 'good' median separation without a good log rank p value (log rank p value is, effectively, the p value associated with the HR.). Doesn't happen.

That said I don't discount difference of medians as much as Dew. As a VERY rough rule of thumb I treat medians as having the same noise as would exist in HR for a trial of <1/2 the size.