Marjac, HK, BB, LB and all who contributed to the Rule 24/60 filing – Thank you for your diligence and resolve to develop this truly outstanding response. Your concise and compelling argument should force Judge Du and AMRN management to take a stand and address this error.

Moving forward, I am very anxious to see how AMRN responds to the filing. While AMRN was clearly focusing its legal effort on the Supreme Court, ignoring the 24/60 filing turns their back on AMRN shareholders and Dr. Bhatt. Furthermore, it seems likely that the Supreme Court will not review the case until Rule 24/60 is resolved by the lower courts. Hopefully AMRN will quickly realize that supporting the Rule 24/60 is in their customers’, shareholders’, and management’s best interest.

Kiwi - Sorry I inadvertently mixed up significance in my response. My point was that a differential analysis between the EPA and DHA arms is only important if Judge Du neglected the fact that LDL increased in both arms and instead focused on the EPA arm not increasing a significant amount.

Raf - True but should also add #3:

3. Judge must disregard the fact that that LDL was raised in both the EPA and DHA groups and fixate on the fact that DHA increased a somewhat significant amount (compared to DHA Pre-study arm) while EPA did not.

I see where you are going but still think that the specific wording of Du's decision highlights the need for the interaction analysis. As such, I don't see any potential negligence claim against Covington going anywhere.

I don't know how Singer is proposing to get the paper into the court record but I would not be surprised to see it show up as part of his response brief.

Raf - Good question. With hindsight, it is easy to say that Covington should have hired a biostats expert to pick apart the existing studies. But we need to remember that this would only be necessary if the judge:

1. Determined that this small 60-person exploratory study was remotely relevant; and
2. Totally disregarded the 2-population argument that TG>500 would react differently to TG<500.

Given limited time and need to summarize key arguments for the judges comprehension, I can see how it may have made sense to focus efforts on the two aspects above.

There appears to be much confusion and debate regarding the following statement from the Curfman, Bhatt and Pencina paper:

Furthermore, our detection of an error in the prior art does not constitute new evidence in the case that could not be considered by the Court of Appeals for the Federal Circuit. Rather, our observation constitutes a re-interpretation of evidence that was already cited multiple times in the judge’s opinion.

The “error in the prior art” is Du’s deduction that Mori’s study would have provoked a POSA to ask if DHA or EPA (if not both) was responsible for the increase in LDL cholesterol. The paper indicates that this question would have been inconceivable by a POSA because Mori did not address the question of differential effects of DHA and EPA on LDL cholesterol levels and if they had, the difference would have been determined to be statistically insignificant. Therefore, Du’s extrapolation of the Mori results conflict with the basic principles of statistical analysis. Mori should not be faulted for not performing this analysis because the purpose of the small study was to “generate hypothesis” and not to provide definitive results. The “error” is solely Du’s and was only brought to light in her post-trial decision. Since this error appears to be a primary driver or her opinion on obviousness and was not available prior to her decision, it is directly relevant and should be allowed in the appeal.

I would also expect that the Curfman, Bhatt and Pencina paper was well planned, coordinated and reviewed by the Singer legal team to maximize the impact on the appeal. Neither the above excerpt nor the timing of the paper’s release was an accident. Singer focused on the procedural errors in his brief and is now tapping into the scientific experts to demonstrate that Du’s deduction is not supported by sound statistical analysis.

Raf - I'm an engineer who initially developed this spreadsheet a couple of years ago. I've been long AMRN since 2012 and frequently review this message board. I haven't previously posted and don't plan to post unless I can something unique to offer.

HDG - Sorry for the confusion. I should have rephrased only as "Lost to Followup". I understand that dropouts will be included in the study if follow-up information can be obtained.

Please note that once the model was together, I backed into the 3% value for the LtFu rate because it nicely matched the 5.2% rate presented by AMRN at ADCOM.

Here's my model results that appears to be in line with HDG's current thoughts and AMRN "expectations".

Goal: Model Reduce-It primary events to determine potential event rates and RRR that meet that actual interim and end dates. The model considers the following factors:
1. Increasing event rate with age
2. Dropouts who discontinue study and are lost to follow-up
3. Varying event rates vary based on enrollment date (before & after May 2013)


Assumptions:
1. Event rate increases exponetially at 2%/year extrapolated from KPNW study for high risk patients age 60-70
2. Discontinue study & lost to followup rate (LtFu) of 3%/year (approximately 970 over the trial)
3. A range of initial placebo rates both before and after AMRN increased Trig enrollment requirements in May 2013
4. Constant RRR over time. JELIS data did not result in a conclusive pattern
5. Enrollment data per ADCOM and AMRN PRs

Results:


Conclusions:
1. Model results with 5.2% composite rate closely mimics linear "Expected" results shown by AMRN at ADCOM.
This provides confidence in the model and indicates that the effects of increasing event rates with age and reduction of the study group size (from LtFu and events) roughly cancel out.
2. The "band" of potential event rate combinations that meet all the actual interim and end dates is narrow and can be approximated with a composite average rate of 4.95%.
3. With initial placebo rates of 5% and 5.5% and secondary placebo rates (after May 2013) of 6%, the corresponding RRR is 18% and 32% respectively. My guess is illustrated by the orange line (P1i = 5.2%, P2i = 6.0%, RRR = 24%)
4. PY > 35,000 represents the total participant years of everyone remaining in the study, including those who have already had a primary event. With the above assumptions, the model results in PY = 34957 (close enough for me) through March 2018. The subset of PY elgible for primary events is approximately 32650. With no death or LtFu, PY would be approximately 38,800.
5. The event rate should increase approximately 0.5% over the course of the 5-year study resulting in in an average event rate approximately 0.25% than at the start of the study.
6. Given the duration of the study, the "bimodal" effect of enrolling lower trig participants before May 2013 and higher trig participants after May 2013 is not significant and may be closely approximately by a single average rate (the orange line is very close to the dashed black line).
7. LtFu rate assumption of 3% annually results mostly closely matches ADCOM "Expected" results shown by AMRN. If a smaller LtFu rate is assumed, the 5.2% composite rate results in 1612 events prior to the end of 2017. With a larger LtFu rate, 5.2% composite rate stretches the trial into 2018.
8. If the LtFu rate is increased to 4% or more annually, there is no straight line average event rate that meets all actual dates; either event 1290 occurs prior to March 2017 or event 1612 occurs after March 2018.