Biotech Values

[DewDiligence]

>GNVC – Can the endpoint "overall survival with a P<0.245" be met when there is no difference in median survival?<

In theory, yes. Median survival is simply the intersection of the K-M curve with a horizontal line drawn at a height of 50% on the y-axis. There is no magical power to this 50% number — it’s merely a convention to report data that way.

The observed hazard ratio is—to a first-order approximation that’s accurate enough for this discussion—the ratio of the area under the K-M curve for the control arm to the area under the K-M curve for the treatment arm.

In this case, where HR=0.75 at the interim look, there is roughly 75% as much area under the K-M curve for the control arm as there is under the K-M curve for the treatment arm. However, the difference in the areas under the respective K-M curves comes almost entirely from the interval between 12 and 18 months on the x-axis. This is a bizarre finding that suggests the HR=0.75 result is a fluke and it will not hold up at the final analysis.

>If so, how strong does hazard ratio have to be [to achieve statistical significance]?<

The HR per se does not have to improve to achieve a statsig outcome; however, the 95% confidence interval on the HR needs to shrink considerably. GNVC reported that the 95% CI at the just-completed interim look was [0.49-1.15]; a statsig outcome at the final analysis will require that the upper bound of this interval be <1.0. (Actually, it will have to be slightly lower than that because a small amount of alpha [the “penalty” you referred to] gets expended on the interim looks; however, this is a minor consideration in the overall scheme of things.)

All told, I don’t think this trial is even remotely promising and it certainly would not induce me to seek a license for the drug if I were in the business development department of a Big Pharma. Regards, Dew