Biotech Values

[DewDiligence]

Re: Median values not indicative of hazard ratio

>…is it safe to assume that these median overall survival results are good indicators for the hazard ratio?<

No—it is not. The median value of a survival curve is simply a linear projection of the curve at the 50% point on the y-axis. It is no more indicative of the entire survival curve than any other single point on the curve.

The hazard ratio takes into account the entire span of the survival curves for the active arm and the control arm, respectively, and it is thus much more representative of the relative efficacy of two treatment than a comparison of point values. (You can get an “eyeball” gauge of the hazard ratio from the relative amount of area under the two survival curves.)

In spite of the above, median values of survival curves are generally reported prominently in press releases and conference calls. However, this is due to convention—and the implicit assumption that investors need to have statistics dumbed down—rather than any inherent importance of the median values.

An example that shows how the comparison of median values of the survival curves is a poor indicator of the hazard ratio is DNDN’s 9902a trial:



To get the median values of the Provenge-arm and control-arm survival curves, draw a horizontal line through them at the 50% value on the y-axis. As you can see, the curves are widely separated in this vicinity, but the curves are hardly separated at all at values a little higher or a little lower on the y-axis.

As it turned out, the ratio of the Provenge median survival to the placebo median survival in the 9902a trial is almost the same as the ratio of these values in the 9901 trial, but the hazard ratio in 9902a is very much worse than the hazard ratio in 9901.