I have been curious for my own understanding about the differences between the test and control groups that would be required to generate a RRR of 30% on 3 point MACE and what level of statistical significance would be generated by the sample sizes in this study. Perhaps this is all contained in the trial filing…I haven’t checked.

Keep in mind I am not a bio-statistician and I am very rusty on my stats.

I have used the information available and that is 8 MACE events per 100 patient years as confirmed by Claydon through BDAZ’s request. I used a simple binomial model based on equal sample sizes of 1200 patients in each of the test and control groups.

Caveat – I used the information available but there is far more statistical information in the raw data.

This is a time to first event trial.

Therefore, I believe the basic dependent variable that will be used for statistical analysis will be the time of the first event minus the time when the patient started the trial. If they started at week 12 of the trial and had a MACE at week 36 their time to event will be 24 weeks for that individual.

However, we also know that patients had to have had a MACE prior to entering the trial. Perhaps this could also be used as a base for “time to first event” as well if it is validated?

Obviously these are individual scores and thus would apply to all sub analysis.

This measure allows for analysis of variance (ANOVA) across all subgroups. ANOVA is fundamentally different than the simple binomial model I used (because the data is not available).

The time to first event data would apply to all of the other dependent variables such as 5 point MACE and various biochemical measures.

Some Background

Being confused by all of the 3600 patients years posts I reviewed them again.

1) BDAZ Feb 20, 4:34 p.m.

I heard back from Clayton about this today. He responded:

"To answer your questions, yes the actual events we are observing in the blinded analysis from BETonMACE is close to 8 per 100 patient years as we had planned. As you know this trial remains blinded and therefore, we do not have any access to who is on placebo or who is on drug, so these numbers are derived from all patients."

2) BDAZ Feb 16th, 5 p.m.

Second, my point of making this 3600 patient year post was to focus on the predicted patient years and get away from event rates. Yes, the two are related in the sense that a lower overall event rate will require a larger number of total patient years to achieve a given number of MACE events. Conversely, a higher overall event rate will require less patient years to achieve the same number of MACE events. There are still a lot of unknowns with the event rates. However, the tracking of patient years over time will help give us a sense of where we stand relative to the original goal by using the number of patients enrolled and how long each wave of patients has been enrolled.

3) BDAZ

posted on Feb 27, 2018 11:59 a.m.

Koo,

Keep in mind that an event rate is only useful if there is a unit of time. It's actually not very informative to express as events/patient (288 events/2400 patients) without a unit of time. Some patients might just start in trial today. Others have been in trial for 2 years. That's why patient years is a good normaliser. It takes into account not just patient number but duration in trial. Comparing an event rates from one trial at 6 months to another trial at 12 months to another trial at 18 months is an unfair comparison unless these individual event rates are normalized to a time component.

BearDownAZ.

The Binomial Model

Keeping in mind the caveat I mentioned I used the 8 MACE events per 100 patients years as the dependent variable and assumed 1200 patients in the test group and 1200 in the control (SOC) group.

MACE Rate                       Absolute                                        Statistical

Test            Control           Difference              RRR               Significance

7.5%              8.5%            1.0% point             11.8%            less than 90%

7.0%              9.0%            2.0% points           22.2%            90% confidence level

6.9%              9.1%            2.2% points           24.2%            95% confidence level

6.5%              9.5%            3.0% points           31.6%            99% confidence level

So based on this simple binomial model given that the current MACE event rate in the trial of 8 out of 100 person years then this hypothetical model shows that statistical significance (as defined by the standard 95% confidence level) is not achieved unless the MACE rate is 6.9% in the apabetalone test group and 9.1% in the SOC control group. The relative risk reduction in this scenario would be 24.2%, which is below the 30% RRR target.

At this minimal level of significance (95% confidence) I would think statistical significance at the subgroup levels might be a challenge.

The RRR target of 30% is exceeded if the MACE rate in the test group is 6.5% and in the control it is 9.5% and the statistical significance is at or above the 99% confidence level.

So as I understand it the committee would be looking for evidence of a 3% point difference test vs. control.

As I pointed out previously the actual dependent variable in the study is time to first MACE event from when a patient entered the trial…not the variable I used in this analysis.

I did this for my own perspective. I welcome your thoughts.

DYODD.

Toinv