I have been asking myself and been asked by other people at what date for the 360th event do I start to feel very secure in my belief that we will have a successful outcome. We all know that with biotech you can never say never. But one of the tests that I have done is a basic smell test. It goes like this: assume for a moment that there is no drug advantage. What date would you have to get to to have a 7 month median survival for the whole trial -- for the aggregate of both the control and the drug group assuming no advantage? When I model with the following assumptions: 1% die in month 1, 4% in month 2, 6% in month 3, 8% in month 4, 9% in month 5, 10% in month 6, 12% in month 7, 9% in month 8, 7% in month 9, 5% in month 10, 5% in month 11, 4% in month 12, 3 % in month 13, 2% in months 14,15 and 16, 1% in each of months 17 through 22 and 5% live for years, I get a 7 month median survival with the 360th even occurring in the first week of March. Now I know these numbers are not exact... but the key is, they are reasonable. They try to solve with a projected median rather than solve for the 360th event and determine the median ex post. The assumptions provide for a long survival tail >20% beyond 1 year,10% still surviving at month 18 and 5% superheros who live for years. And they provide for a smooth survival curve that is at least broadly consistent with the survival curves that have been observed in earlier trials. A couple of interesting things: the model yields a death count of 163 in the first week in August, consistent with what the company reported. (When I tried to solve for a 6 month aggregate median survival I was getting 179 events in the first week in August -- too big a difference from the 163 reported number. The 6 month aggregate median predicts a 360th event in mid Feb, which I don't think is going to happen, but even if it does there is a good chance that we will be able to achieve success-depending on the actual median of the control group)
Now comes the smell test part of my 7 month model. I know with a high degree of certainty that the control group is NOT going to achieve median survival of 7 months. That long a period would be hugely inconsistent with anything that has been observed for late stage mcrc patients in any reasonably sized trial that has ever been run. With that high degree of certainty I then do some analysis. If the control group comes in with a median survival of 5 months then by definition the drug group must have come in with some number closer to 9 months (I will leave it to the mathematicians to figure out the exact number) for the whole trial to have achieved a 7 month aggregate median. Maybe 5 months is too much to hope for for the control. But if the median survival was even an unusually long 6 months for the control group (far longer than I am expecting), then you would need a median survival of over 8 months for the drug group to get the aggregate of 7 months. What I am essentially saying is that though I believe we can achieve success if the 360th event occurs as early as next week, if this thing gets into early March,which I believe that it will, we are in very very good shape. That is just my view anyway.
August 3rd. 2011, when the interim numbers determined that there was no way that perifosine wasn't working in a signifcant manner.
For the Keryx management this may of been January 1st 2012. Since they have access to the enrollment data and possibly other information. They had always expected Q4 11 completion with a positive outcome (before the DSMB data), even without the over-enrollment. And I presume this wasn't Dec 31st either!!
<August 3rd. 2011, when the interim numbers determined that there was no way that perifosine wasn't working in a signifcant manner.>
I think this is a continuing misconception. The interim was not an efficacy look, it was futility and safety. With an event fraction <50% it was very immature at the time and the medians would not have been reached in one or both arms. The futility boundary was likely near HR of 1. It is essentially non-informative for efficacy. I think it's importance is in calibrating the blended survival function. But with it being at such a relatively low event count the tails of the curve can still fluctuate. Another event count anchor point would be useful.
Ocker - I have ran my own model as of late and using a 5.5 month median for Cap and 9.5 median for Pcap I get on "avg" the correct number of events in Aug and and on "avg" the 350th event in mid March... BUT I ran my model ~100 times using the same curves using a random number generator to decide which person started on each date (meaning when did each 38 month survivor start and when did each 3 month survivor start) and got ~ 1 month of variance in the data points for the 163rd event and ~1.5 months of variance for the 350th event (normal distribution).
Noid - I would like to see the shape of your Cap and Pcap curves since you have spent more time looking at different studies, if they are different from mine I would like to run them in my model and see what they yield.
Although I've compared my curves to other studies, I generated my curves by simply doing a time scaled average of the two curves from the P2 trial (with some smoothing, but they are still not as smooth as a curve derived from a mathematical formula, due to the low number of participants)
Anyway, if it helps, here is what I get for 5.5 month median (each line is a month)
And here is 9.5:
(note, I forced 0 deaths in month 1 to make my model more conservative, hence at the end of month 1 in both curves we are still at 1, which means a lower number of deaths at the august peek for the same median, which should yield slightly more conservative results, but not much, since most patients were beyond a month by then anyway)
Ock: Your model, however close to actually being correct, is still based on assumptions... as a long time investor in science the day I start to feel comfortable is the day positive results are announced... obviously, one can feel hopeful under certain circumstances but you can never be comfortable when the wheel is still spinning...
I am long and will play devil's advocate.
The 360th event could have already occured and they could be scrubbing data. There is no exact time on how long they will take to announce the 360th event after it happens. Isnt that correct?