Replies to post #5200 on Dendreon Corp fka DNDNQ

[steveporsche]

<With a given number of events at the interim and an allocated p value, is the Hazard Ratio determined automatically or does the shape of the curve or some other factor(s) make it a variable independent of events and p value?>

I am not the person to answer this, but I think the answer is both. It can be both instantaneous at all points along the curve, and overall at the end. The instantaneous HRs along the curve builds to overall HR. What worries me about the interim is the potential low HRs for patients enrolled less than 24 mos. (over 40% of the pop. at 4Q08 ) squeezing the hell out of the overall HR and p-value to that point.

% alive
9901 6 months 12 months 18 months 24 months 30 months 36 months

Provenge 91.5 76.8 67.1 52.4 43.9 34.1
Placebo 93.3 66.7 55.6 40.0 24.4 10.7
dif -1.8 10.1 11.5 12.4 19.5 23.4
hr 0.98 1.15 1.21 1.31 1.80 3.19

% alive
9902a 6 months 12 months 18 months 24 months 30 months 36 months

Provenge 92.3 72.3 52.3 44.6 41.5 31.6
Placebo 90.9 69.7 45.5 39.4 27.3 21.2
dif 1.4 2.6 6.8 5.2 14.2 10.4
hr 1.02 1.04 1.15 1.13 1.52 1.49

This table will probably be garbled, but if others agree with its validity I can PM it to you.

[iwfal]

While both DNDN and CEGE have used the terms "allocated alpha" or alpha spend" both the CEGE CEO and DNDN's Schiffman seem to concentrate on the importance of the Hazard Ratio at the interim rather than the p value.

This is just the way that drug companies talk. As far as the FDA is concerned it is only p value vs alpha that matters.

Slide 9 states that statiscians like Hazard Ratios because all data is used and there is smoothing.

There is just as much 'smoothing' with LogRank in comparison to other p values you could calculate such as Chi Squate of % alive at 36 months.

With a given number of events at the interim and an allocated p value, is the Hazard Ratio determined automatically or does the shape of the curve or some other factor(s) make it a variable independent of events and p value?

Given a number of events and an HR the p value is very constrained. Given HR and # of events the p value can move a little based upon KM curve shape - but only a little.