Iconoclast,

Great post. I had to re-read it several times. It really got me thinking. Possibilities vs. probabilities. I have a few comments.

"For example, I arrive at 5000 patient years by back calculating the median 27 months and extrapolating the patients known to be in the trial as given by Dr Sweeney."

This sounds reasonable. 27 month median X 2425 patients is roughly 5456 patient years. Of course, this is median not average so not an exact number. But this kind of gives a ceiling for patient years. Then one can start chipping away based on patient discontinuations/dropouts.

"Second, I am assuming a 10% end drop out rate.  Even if you do +/-5% sensitivity analysis on the 10%, it does not significantly impact end results as cumulative patient years don't radically change."

Based on other recent ACS diabetic CVOTs (EXAMINE, ELIXA, ALECARDIO), I am expecting the discontinuation/dropout rate to be between 20-30% for both placebo and apabetalone groups. I am also expecting the discontinuation/dropouts to be higher in the apabetalone compared to placebo group due to the pre-specified liver enzyme criteria.  

"Note that a smaller number of patient years means less MACE from placebo patients.  Therefore, logically the data also skews more to 5,000 patient years than to 4,000 patient years: an independent "check" if you will."

I don't quite follow you here. We know that there were ~275 adjudicated first occurence 3-point MACE events. At this time, we are blinded to apabetalone/placebo split, so it is best to think of the overall trial event rate that combines placebo and apabetalone. As one slides the total patient years to a higher number, the combined event rate must fall in order to give one a constant 275 events. Conversely, as one slides total patient years to a lower number, the combined event rate must become greater. Either you had a typo here or I am mis-understanding your point.

"I am assuming a 20 to 30% RRR.  But no matter how much I play around with the numbers, it is impossible to get a significant gap in Crestor vs Lipitor groups MACE rates.  This is purely on statistical grounds, as Jupiter explained earlier."

I still disagree with you here. There is just not enough available information right now. Apbetalone/Crestor could behave similar to Apabetalone/Lipitor, or there could be very big differences between the two. Since we are blinded to these specifics right now, many many options are still on the table.

"The only way my spreadsheets get to the 30% RRR that is bandied about on this board is a much lower MACE rate for the placebo group than the trial design of 7%.  Allow for that, and it all makes sense. Tada's explanation of changing standard of care over time, goes a long way to explaining why."

I disagree. There are tons of potential combinations depending on the actual patient years, actual dropout/discontinuation rate, and actual event rates. You paint a picture of a limited scenario. But there are too many unknowns so many other potential outcomes are possible. For any given combined event rate, one can project RRR's ranging from 0 to really good numbers (i.e. 30%, 40%). Since we are blinded to the placebo vs. apabetalone event rates and can really only think about combined event rates right now, there are many many possibilities to consider.

Just a clarification, when you say 7% event rate, do you mean 7% annual event rate (7 events per 100 patient years?). The BETonMACE design is somewhat confusing. Way back when the trial started, I recall they were expecting 250 events in the total 2400 patient population to have a combined event rate of ~10.5% at median dosing of 18 months (3600 patient years). This equates to 6.94 events per 100 patient years for the combined population. They were also assuming a 30%RRR. So if the combined event rate was 6.94 per 100 patient years, the split would be something like 5.7 events per 100 patient years in the apabetalone group and 8.1 events per 100 patient years in the placebo group to give a 30% RRR. More recently in the design posters and publications, they now express the the placebo event rate was projected to have an annual event rate of 7% (7 per 100 patient years). This is a very big inconsistency (7 per 100 vs. 8.1 per 100) and the more recent clarification of 7 per 100 patient years for the placebo is indeed a big drop from the 6.94 average/combined event rate in the original plan that would have required the assumed placebo rate to be ~8.1 per 100 patient years.

"Finally, I want to trot out my notes from talking to DM several months ago: MACE rates came in much lower than RVX itself expected.  Again, that skews realized placebo rates much lower than design."

As mentioned above, different BETonMACE design assumptions were communicated creating a bit of confusion. Further adding to the confusion is that the risk rate for experiencing a first occurence MACE event in the trial is highest in the first several months after randomization and the risk drops over time. So by amending the trial to allow dosing beyond 24 months, they increased the original planned median dosing from 18 months to 27 months. They also potentially accumulated many more patient years than originally planned (3600); Iconoclast predicts 5000. More patient years will unavoidably lower the events per 100 patient years number. So if one combines all of this information, then it is not surprising that several contributing factors could have led to a lower final events per 100 patient years.

BearDownAZ