[00:00:00] Speaker 00:
Case number 23, that 1045 et al, Huntsman Petrochemical LLC Petitioner versus Environmental Protection Agency.

[00:00:09] Speaker 00:
Mr. Lazerati for the petitioners, Ms.

[00:00:11] Speaker 00:
Chen for the respondent.

[00:00:14] Speaker 03:
All right, Mr. Lazerati, good morning.

[00:00:17] Speaker 01:
Good morning.

[00:00:25] Speaker 01:
May it please the court.

[00:00:26] Speaker 01:
My name is John Lazzarelli.

[00:00:28] Speaker 01:
I'm here today representing Huntsman Petrochemical, American Chemistry Council, and Louisiana Chemical Association.

[00:00:34] Speaker 01:
I'd like to request four minutes for rebuttal.

[00:00:37] Speaker 01:
Thank you.

[00:00:38] Speaker 01:
Your honors, we are here today challenging EPA's ethylene oxide risk value for the miscellaneous organics NESHAP, or Mon Rule.

[00:00:46] Speaker 01:
The reason parties put this issue first is that it is dispositive.

[00:00:50] Speaker 01:
If the risk value is wrong, the Mon Rule is wrong.

[00:00:53] Speaker 01:
And here the risk value is wrong for several reasons.

[00:00:56] Speaker 01:
EPA did not follow the Clean Air Act.

[00:00:58] Speaker 01:
It did not follow the National Academy of Sciences recommendations.

[00:01:01] Speaker 01:
And it did not follow its own guidance.

[00:01:04] Speaker 01:
EPA ignored significant flaws raised in public comments.

[00:01:07] Speaker 01:
Rather than base its risk value on sound science, it boiled 17,000 data points into five categories, graphed them, and eyeballed the graph.

[00:01:16] Speaker 01:
That is not what the Clean Air Act requires and resulted in an arbitrary and capricious rule.

[00:01:22] Speaker 01:
I'd like to start by addressing two statutory violations arising from the Monroe.

[00:01:26] Speaker 01:
First, EPA can impermissible preference to its IRIS value or the integrated risk information system value.

[00:01:33] Speaker 01:
And second, it ignored the National Academy of Sciences recommendation.

[00:01:39] Speaker 01:
On the first issue, the Clean Air Act requires EPA to consider all credible and relevant toxicity information and setting risk values.

[00:01:46] Speaker 01:
It cannot just use the latest entry from its IRIS database.

[00:01:50] Speaker 01:
This is what Congress adopted when it incorporated the benzene Nishap into the Clean Air Act in 7412 F2, and it is what EPA told Congress it would do in conducting residual risk assessments in its report to Congress in 1999.

[00:02:05] Speaker 01:
EPA does not argue otherwise in its briefs.

[00:02:08] Speaker 01:
EPA deviated from this required approach in the Mon Rule.

[00:02:11] Speaker 01:
The final risk value relies exclusively on the IRS value.

[00:02:15] Speaker 01:
EPA put Iris on a pedestal and made it the default value in the rule itself.

[00:02:19] Speaker 02:
So I agree it would be a problem if there were studies you had pointed to that EPA disregarded entirely, which seems to be the type of argument you're making.

[00:02:28] Speaker 02:
But it seems to me that for each of the issues you raised, EPA at least discussed it in some way.

[00:02:36] Speaker 02:
And so what we're really doing is fighting over the adequacy of EPA's explanation for why it didn't modify the Iris value.

[00:02:45] Speaker 02:
based on any of these studies is that is that fair you might still be that they didn't give adequate explanations but is that what's going on.

[00:02:55] Speaker 01:
There's a statutory issue that EPA had a preference for iris.

[00:02:59] Speaker 01:
And through that, did not view other evidence in the right light.

[00:03:03] Speaker 01:
Under 7412, that would be a statutory violation separate of whether they even addressed the issues.

[00:03:08] Speaker 01:
They've also identified a number of issues that they didn't address.

[00:03:11] Speaker 01:
And I do want to distinguish, because you use the word studies.

[00:03:14] Speaker 01:
And certainly, the agency has pointed to topics that are discussed that are similar to the issues.

[00:03:19] Speaker 01:
So if we raise a concern about smoker data, the fact that the iris value predicts

[00:03:25] Speaker 01:
a linkage between smoking and lymphoid cancer that simply doesn't exist.

[00:03:31] Speaker 01:
They've addressed another topic on smoking.

[00:03:33] Speaker 01:
So they may have addressed the study, or they mentioned the study, but they haven't addressed the issues that we were raising.

[00:03:38] Speaker 01:
And so that disconnect between the reality checks we've identified and the response that they've made is problematic, even if they had the correct standard under 7412.

[00:03:48] Speaker 02:
Well, I think maybe the smoker studies are a useful example, right?

[00:03:51] Speaker 02:
Because you pointed to specific, my understandings, you pointed to specific studies and EPA discusses those studies on a specific page of the response at the reconsideration stage.

[00:04:08] Speaker 02:
And it explains, maybe not in the most articulate way, that there were two problems with those studies.

[00:04:13] Speaker 02:
They ignored confounding factors, and they used this hemoglobin marker that EPA doesn't view as validated.

[00:04:22] Speaker 02:
Don't you need to persuade us that those explanations are irrational or arbitrary, or a different point you're making?

[00:04:29] Speaker 01:
Irrational or non-responsive, Your Honor.

[00:04:32] Speaker 01:
So for example, on the first one, our point is that there is ethylene oxide in cigarette smoke.

[00:04:37] Speaker 01:
There's a good deal of it.

[00:04:38] Speaker 01:
There's so much that if the iris value were correct, there would be a roughly 1 in 10 risk of smokers developing lymphoid cancer.

[00:04:45] Speaker 01:
So there would be a strong signal of a risk.

[00:04:48] Speaker 01:
Lymphoid cancer would be linked.

[00:04:50] Speaker 01:
We have decades of research from the CDC that

[00:04:53] Speaker 01:
smoking in lymphoid cancer.

[00:04:55] Speaker 01:
Now, their response is, well, you haven't validated the data, so we don't know exactly what the amount of lymphoid cancer should be when you smoke.

[00:05:07] Speaker 01:
But the point is it doesn't matter.

[00:05:08] Speaker 01:
If it's one or two, if it's supposed to be 10 and we're not seeing that, that's the reality check that they were missing with the iris value.

[00:05:19] Speaker 02:
Well, I don't want to spend all day on these studies, but assume, you know, I'm just a judge.

[00:05:25] Speaker 02:
Assume the hemoglobin marker, you know, the wrong baseline to measure whether ethylene oxide is causing cancer.

[00:05:35] Speaker 02:
If they're right about that, what value do these studies finding no correlation, aren't they basically irrelevant?

[00:05:44] Speaker 01:
Your honor, they're not saying that those are, in fact, they use those adducts as well for determining a correlation.

[00:05:49] Speaker 01:
So what they're not saying is that it's an unreliable adduct in general.

[00:05:53] Speaker 01:
They're only saying that that relationship, while well-established at higher concentrations, hasn't been scientifically validated for lower concentrations.

[00:06:00] Speaker 01:
Now, that's a little untrue as well.

[00:06:02] Speaker 01:
The data was available for EPA to do that validation.

[00:06:07] Speaker 01:
They simply didn't do it.

[00:06:09] Speaker 01:
But even if the agency wants to raise, perhaps the number would be slightly different.

[00:06:14] Speaker 01:
What we're talking about here is orders of magnitude different.

[00:06:17] Speaker 01:
So it's simply addressing the issue with respect to whether the IRIS value was reasonable to use for setting a risk value for the model.

[00:06:26] Speaker 01:
And again, that's getting to a technical issue.

[00:06:30] Speaker 01:
First, EPA has to show that it is authorized to use the Irish value in the way that it has in this rule.

[00:06:35] Speaker 01:
And it hasn't by giving it a preference over other values and over other evidence.

[00:06:41] Speaker 01:
Your Honor, another statutory issue is the failure to address the National Academy of Sciences recommendations.

[00:06:48] Speaker 01:
Section 7607D3 requires EPA to consider, or even proposing a rule, the pertinent findings and recommendations of the National Academy of Sciences.

[00:06:57] Speaker 01:
Here, there are almost 200 pages of recommendations going to the reliability and accuracy of virus assessments that the NAS published in 2011 and 2014.

[00:07:05] Speaker 01:
The EPA made no mention of them in the mod proposal.

[00:07:10] Speaker 01:
Both the Supreme Court and this Court have noted that 7607 imposes substantive obligations on EPA to consider the NAS's recommendations.

[00:07:19] Speaker 01:
Had EPA complied with the statutory requirement, it would have been forced to address many of the same flaws that we're raising today.

[00:07:24] Speaker 01:
In fact, EPA has since overhauled the IRIS program in response to those recommendations.

[00:07:30] Speaker 01:
But it has not gone back to address the same flaws in the development of the IRIS value for ethylene oxide that it's used in the Monro.

[00:07:37] Speaker 01:
This was contrary to law and renders the Monro arbitrary and nutritious and in violation of the statute.

[00:07:46] Speaker 01:
Now, Your Honor, we did address on the factual question of whether the iris value is arbitrary and capricious, already the smoker issued, or a few other examples that I would like to address today.

[00:07:55] Speaker 01:
Another issue is the exposure assessments that were used to support the iris value.

[00:07:59] Speaker 01:
The iris value is a dose-response relationship, so relevant to that study is not only the response, how many people are getting lymphoid cancer, the dose, how much ethylene oxide we're exposed to.

[00:08:10] Speaker 01:
EPA used the NIOSH study, which did not have exposure data for the vast majority of the time in that study from 1938 to 19, roughly, 76 or 78.

[00:08:20] Speaker 01:
So it modeled exposure during that time period.

[00:08:23] Speaker 01:
To do this, it assumed there was no change in work practices to reduce ethylene oxide exposure before 1978.

[00:08:29] Speaker 01:
That's simply not supported by the record.

[00:08:32] Speaker 01:
Petitioners submitted several studies identifying specific work practices that reduced exposure to ethylene oxide over that time.

[00:08:40] Speaker 01:
And Your Honor, again, when you talk about studies that were addressed or not addressed, one study they address is Bogan, where they say that one portion of that study was not adequately documented.

[00:08:49] Speaker 01:
It does not address the issue of whether there were work practices involved.

[00:08:52] Speaker 01:
And in fact, for that, Bogan cites eight other published studies to identify those work practices.

[00:08:58] Speaker 01:
as well as FDA data demonstrating that medical devices that were subject to sterilization had decreasing levels of ethylene oxide residue on them over time, demonstrating again that those work practices were effective in reducing ethylene oxide exposure.

[00:09:16] Speaker 02:
The result is that... Could I ask you to address the figures over which there was a lot of back and forth in the briefing?

[00:09:27] Speaker 02:
And in particular, I think it would make sense to start with figure four, three, and maybe what is your position about what EPA did with that figure that it shouldn't have done?

[00:09:39] Speaker 02:
Because it seemed to me that whenever you said it did something it shouldn't have done, EPA said we didn't do that.

[00:09:46] Speaker 01:
Yes, Your Honor.

[00:09:47] Speaker 01:
And from our perspective, our legal concern is that they did not follow their own guidance with respect to selecting the model to use for the risk value.

[00:09:56] Speaker 01:
So once you have the NIOSH data or the study data, the question is, what's the dose-response relationship of finding in that data?

[00:10:03] Speaker 01:
Their guidance makes two things clear.

[00:10:05] Speaker 01:
First, you're supposed to use biology and statistics to identify a model.

[00:10:11] Speaker 01:
EPA did not have a biological reason for its model selection.

[00:10:15] Speaker 01:
It asserts that ethylene oxide damages DNA.

[00:10:18] Speaker 01:
EPA was recognized in the record that this biologically means that the dose response curve should be a line, linear, or curving upward.

[00:10:28] Speaker 01:
The model they chose is the exact opposite.

[00:10:30] Speaker 01:
It starts, it skyrockets, and then plateaus.

[00:10:33] Speaker 01:
So this gets to what exactly they mean by linear.

[00:10:37] Speaker 02:
Yes, your honor.

[00:10:37] Speaker 02:
It seems, well, their position stated explicitly at the reconsideration stage is that a spline, two lines, is what they call a linear model, including their guidelines.

[00:10:49] Speaker 02:
And I'm not sure what basis you had to disagree with that.

[00:10:53] Speaker 01:
In the Carson education guidelines themselves, there is the footnote, though, that says that a spline is a nonlinear.

[00:10:59] Speaker 02:
Where does it say a spline is nonlinear?

[00:11:02] Speaker 02:
It says nonlinear includes, for example, a quadratic model or a threshold model.

[00:11:09] Speaker 02:
And in the definition of a low dose linear model, this case is really fun, acknowledges that it will not be linear at all values.

[00:11:26] Speaker 02:
Seems like a it might not be a lay person's definition of linear, but they have defined linear in a way that would encompass the spline.

[00:11:36] Speaker 01:
Yes, your honor.

[00:11:36] Speaker 01:
And I believe in that footnote three that you're citing to it does reference splines as an as another example of a nonlinear model.

[00:11:45] Speaker 01:
Even if it doesn't, the ultimate issue is that they didn't use this biology as a basis.

[00:11:52] Speaker 01:
So biology is off the table.

[00:11:53] Speaker 01:
What they've said is they don't have, they're not aware of a mechanistic explanation for the shape of the exposure response curve.

[00:11:59] Speaker 01:
So they've abandoned biology.

[00:12:01] Speaker 01:
They thought they had statistics to support their model.

[00:12:04] Speaker 01:
But there was a problem with their calculation of those statistics, the degrees of freedom.

[00:12:09] Speaker 01:
The result is what they thought was a statistically significant model is not statistically significant.

[00:12:14] Speaker 01:
In other words, it doesn't predict the data any better than the assumption that there is no relationship between ethylene oxide and lymphoid cancer.

[00:12:21] Speaker 02:
Can I ask you about the biological plausibility?

[00:12:24] Speaker 02:
Yes.

[00:12:25] Speaker 02:
Because it seemed to me repeatedly they said that other carcinogens do have similar spline-like

[00:12:34] Speaker 02:
dose response models on J2393, 1950 other places too.

[00:12:42] Speaker 02:
Isn't that the basic type of biological

[00:12:48] Speaker 01:
If they were talking about the right carcinogen, it might be, but they're not.

[00:12:54] Speaker 01:
And the reason is the mechanism by which something causes cancer.

[00:12:57] Speaker 01:
And that's why the issue here is that they've identified ethylene oxide as a direct acting mutagen, which means that it directly mutates DNA.

[00:13:06] Speaker 01:
That's not something that would, so the biological process would either be linear when it contacts DNA, it damages it, or it would be upward curving.

[00:13:15] Speaker 01:
There might be some biological defense mechanism, DNA repair, that gets overwhelmed at some point to allow the curve to move upward.

[00:13:22] Speaker 01:
The other type of

[00:13:24] Speaker 01:
chemical that they've identified, the plateauing effect, is when your body needs to first metabolize the chemical to make it a carcinogen.

[00:13:31] Speaker 01:
So in other words, you need a biological mechanism to make it carcinogenic.

[00:13:35] Speaker 01:
That can be overwhelmed.

[00:13:37] Speaker 01:
And that's how you would get a plateauing effect.

[00:13:39] Speaker 01:
But there's nothing in the record and no allegation that ethylene oxide works that way.

[00:13:44] Speaker 02:
Is there anything in the record that says for carcinogens with this direct effect on DNA

[00:13:55] Speaker 02:
they cannot have a spline dose response model, which is, I think, the point you're making.

[00:14:03] Speaker 01:
Yes, sir.

[00:14:03] Speaker 01:
In the IRIS assessment itself, it identifies that for the direct acting muted, the expectation would be linear or upward curving.

[00:14:11] Speaker 01:
And because of that reason.

[00:14:12] Speaker 02:
Doesn't that go down to the dispute over what they mean by linear?

[00:14:18] Speaker 01:
Not in this context, Your Honor.

[00:14:20] Speaker 02:
This is an important point.

[00:14:21] Speaker 02:
So if you have a site for that point that you can give us on rebuttal, that would be very helpful.

[00:14:31] Speaker 01:
And then, Your Honor, well, I see him running out of time.

[00:14:36] Speaker 01:
But if you'd like to address the interpretation then with that in mind of the graph, I'm happy to do that as well.

[00:14:46] Speaker 01:
I think it would be helpful to at least hear the argument.

[00:14:50] Speaker 01:
OK.

[00:14:50] Speaker 01:
So what they did.

[00:14:53] Speaker 01:
They did not rely.

[00:14:54] Speaker 01:
And the record is clear.

[00:14:54] Speaker 01:
They did not rely on biology or statistics to support it.

[00:14:57] Speaker 01:
What they did was apply what they call visual fit, which is essentially attempting to eyeball this graph that we're looking at on page 45 of Petitioner's Brief, which are referred to as figure 4-3.

[00:15:07] Speaker 01:
They tried to say, what looks the most like of these straight lines or solid lines?

[00:15:14] Speaker 01:
What looks the most like these purple dots?

[00:15:17] Speaker 01:
And that has two problems.

[00:15:19] Speaker 01:
First, the purple dots are not the data.

[00:15:21] Speaker 01:
These purple dots were a categorical model that EPA ran.

[00:15:26] Speaker 01:
So in other words, they took the 17,000 data points.

[00:15:29] Speaker 01:
They identified the 45 or so leukemia mortalities that they wanted to plot.

[00:15:35] Speaker 01:
They grouped those into these five dots.

[00:15:39] Speaker 01:
And then they ran it through, or they grouped them into five categories.

[00:15:42] Speaker 01:
And then they ran it through a model, a categorical model, to say, what's the excess risk?

[00:15:47] Speaker 01:
When they send that to peer review, their science advisory board said, don't do that.

[00:15:51] Speaker 01:
Don't use a categorical model to attempt to detect a pattern in the data.

[00:15:58] Speaker 01:
When EPA came back, what they did was they used the model, but then used that as their standard for determining what other model, continuous model, would look like the data.

[00:16:07] Speaker 01:
And so that's when we say these dots are not the data.

[00:16:10] Speaker 01:
These are just an earlier model EPA ran that was rejected that they're now trying to find a line that's most similar to.

[00:16:16] Speaker 01:
The other problem is.

[00:16:18] Speaker 03:
And so what's wrong with that?

[00:16:21] Speaker 01:
What's wrong with that is that this categorical model is just a model.

[00:16:24] Speaker 01:
It doesn't actually reflect the pattern in the data.

[00:16:27] Speaker 03:
Of course, it's just the beginning.

[00:16:29] Speaker 03:
I mean, you act as though everything is the end.

[00:16:34] Speaker 03:
And there's only one way to do things.

[00:16:37] Speaker 03:
And that's not correct.

[00:16:38] Speaker 03:
And you know that.

[00:16:41] Speaker 01:
There are certainly many ways to identify an appropriate model, but what EPA's guidance says is that you're supposed to use the biology and statistics and only deviate from a linear model if biology and statistics support it.

[00:16:56] Speaker 01:
And to start with a linear model and only deviate from that if you have a reason to do so.

[00:17:02] Speaker 01:
What the Science Advisory Board specifically said with respect to this is it is dangerous to use a categorical model to try to find the pattern.

[00:17:11] Speaker 01:
You need to use the actual data to identify a pattern.

[00:17:15] Speaker 01:
And so the starting point needs to be the data itself.

[00:17:18] Speaker 01:
If you run through a model.

[00:17:21] Speaker 01:
I'm sorry, Your Honor?

[00:17:27] Speaker 03:
You say you have to use the actual data yourself.

[00:17:31] Speaker 03:
Is that correct?

[00:17:33] Speaker 01:
You have to start with the data.

[00:17:34] Speaker 01:
Otherwise, what you're finding is not a pattern in the data.

[00:17:36] Speaker 03:
And why didn't they start with the data?

[00:17:40] Speaker 03:
You don't like where they ended up, but that's a different issue.

[00:17:45] Speaker 01:
Your Honor, we've objected to them not starting with the data, but they have not provided an answer as to why they haven't.

[00:17:52] Speaker 01:
I think the answer is inertia.

[00:17:54] Speaker 01:
They had a model and they wanted to use it.

[00:17:57] Speaker 01:
They were told not by the SAB, rather than go back to the actual data, they simply tried to find a continuous model that would look like the model they'd already run.

[00:18:05] Speaker 01:
The problem is the model they originally ran doesn't accurately reflect the data.

[00:18:11] Speaker 02:
So I have a question about the EPA says in their brief that after the SAB gave this feedback, they actually did go back.

[00:18:18] Speaker 02:
and run this model using the individual data points.

[00:18:22] Speaker 02:
And of course, the categorical dots are based on those individual data points.

[00:18:26] Speaker 02:
But they incorporated that feedback from the SAB and addressed it before finalizing in 2016.

[00:18:35] Speaker 02:
Is that just false?

[00:18:38] Speaker 01:
Well, they may have used the individual data to generate these lines, but to select which line they wanted to use, they went back to their categorical model.

[00:18:47] Speaker 01:
And that's because you can generate a whole lot of lines using the individual data they've got.

[00:18:53] Speaker 03:
But their own answer has acknowledged that they went back to the original data.

[00:19:01] Speaker 01:
To generate models, but not to select a model.

[00:19:04] Speaker 03:
I understand, but you know,

[00:19:06] Speaker 03:
You have to be careful what you say to the court, because as Judge Garcia said, we're just judges.

[00:19:13] Speaker 03:
So you have to be careful what you tell us.

[00:19:15] Speaker 03:
And you told us they didn't go back to the original data.

[00:19:19] Speaker 03:
Well, they did.

[00:19:19] Speaker 03:
And then they went somewhere you don't like, but that's a different issue.

[00:19:30] Speaker 01:
We have raised concerns with the underlying data, with respect to the exposure assessment data, with respect to the model.

[00:19:37] Speaker 03:
I understand that, but the question is, to what extent do those objections stand up?

[00:19:44] Speaker 03:
And maybe they do, and maybe they don't.

[00:19:47] Speaker 03:
But we have to understand what the agency did and not be misled.

[00:19:53] Speaker 03:
That's all I'm getting at.

[00:19:55] Speaker 01:
Yes, Your Honor.

[00:19:56] Speaker 01:
And this court has on several occasions reviewed models.

[00:20:00] Speaker 01:
And the question is, is the model supported by the record evidence?

[00:20:04] Speaker 01:
Are the assumptions he pays making supported by reality?

[00:20:07] Speaker 03:
I understand that.

[00:20:08] Speaker 03:
And you disagree with where they come out.

[00:20:12] Speaker 03:
But don't mislead us as to how they got there.

[00:20:14] Speaker 03:
That's all I'm getting at.

[00:20:16] Speaker 03:
We don't need to belabor this.

[00:20:21] Speaker 01:
Yes, Your Honor.

[00:20:21] Speaker 01:
And perhaps just to clarify, the reason that the iris value is so off is the selection of the model.

[00:20:28] Speaker 01:
So choosing the 1600 knotted spline, that's what drives a lot of the problem with the end result in the iris value.

[00:20:37] Speaker 01:
So it was, when you're looking at all these various models, which one do we use?

[00:20:41] Speaker 01:
That question was a significant error, selecting a model that isn't supported by a biology or statistics.

[00:20:48] Speaker 01:
When they tried to support it with visual fit, our first point is visual fit is not an appropriate way to select a model, period.

[00:20:56] Speaker 01:
But if you are going to use visual fit, this particular graph is not one you can use for those two reasons.

[00:21:01] Speaker 01:
First, they're trying to fit it to just another model.

[00:21:04] Speaker 01:
They're not actually trying to fit it to data.

[00:21:06] Speaker 01:
And the second is the figure itself says,

[00:21:09] Speaker 01:
don't use me to try to determine which is closer to another, because this is relative risk.

[00:21:17] Speaker 01:
This is not the actual risk from the data set.

[00:21:26] Speaker 03:
We'll give you some minutes and replant.

[00:21:29] Speaker 03:
Ms.

[00:21:30] Speaker 03:
Chen?

[00:21:48] Speaker 04:
Good morning, and may it please the court, Sue Chen for the United States.

[00:21:52] Speaker 04:
The court should uphold EPA's use of its cancer risk estimate for ethylene oxide because that estimate is scientifically sound.

[00:22:00] Speaker 04:
Now, before I get into my arguments, I'd like to just take a moment to clarify some of the confusion I heard this morning about different parts of EPA's quantification of risk.

[00:22:11] Speaker 04:
And just bear in mind, I'm going to throw out a lot of nuances and maybe even oversimplify for the sake of clarity.

[00:22:19] Speaker 04:
So once EPA has concluded, based on the weight of the evidence, that ethylene oxide is a carcinogen, the next task was to quantify that risk.

[00:22:28] Speaker 04:
And for that, it used the biggest and most comprehensive human study it had available, and the Science Advisory Board agreed with the choice of that data set.

[00:22:37] Speaker 04:
So step one is that EPA took all the 17,000 plus individual data points and used them to create different dose response models.

[00:22:48] Speaker 04:
And those are the models you see at JA 2400.

[00:22:51] Speaker 04:
It is true that earlier on, EPA had tried to create a dose response model using the categorical data.

[00:22:58] Speaker 04:
And that is shown at JA 2401.

[00:23:01] Speaker 04:
But the Science Advisory Board said, no, that's not really a good idea.

[00:23:04] Speaker 04:
And so EPA went back to the drawing board and looked at the models created with individual data points.

[00:23:11] Speaker 04:
So step one was developing different dose response models.

[00:23:16] Speaker 04:
Step two is choosing one of those models as the dose response model to use.

[00:23:21] Speaker 04:
And this is a step where issues like model fit, categorical breakouts, shape of the dose response relationship come into play.

[00:23:33] Speaker 04:
Step three is extrapolation, which petitioners reference in their reply brief.

[00:23:40] Speaker 04:
And at this step, EPA takes its dose response model and uses it to calculate a value called the point of departure, which

[00:23:49] Speaker 04:
then basically became an input into the extrapolation calculation.

[00:23:55] Speaker 04:
And you can see those calculations at JA 2692.

[00:23:57] Speaker 04:
And petitioners here don't dispute the methodology for extrapolation.

[00:24:03] Speaker 04:
They're really disputing one of the inputs into the extrapolation calculation.

[00:24:07] Speaker 04:
Step four, EPA takes the risk calculated from step three and does more statistical analysis and comes up with the cancer risk estimate that we're talking about here.

[00:24:22] Speaker 04:
So that at the 50,000 foot level is the different steps EPA took in quantifying risk.

[00:24:28] Speaker 04:
Now I'm gonna start by addressing the choice of the dose response model and a few things here.

[00:24:33] Speaker 04:
The kind of default linear model that petitioners are talking about, that applies to the extrapolation.

[00:24:42] Speaker 04:
So that's at step three.

[00:24:43] Speaker 04:
And you can see this discussed in the cancer risk guidelines at JA212930.

[00:24:49] Speaker 04:
and 22, 24.

[00:24:51] Speaker 04:
That is for the extrapolation step.

[00:24:53] Speaker 04:
When it comes to finding the dose response model for the observed data, which is what we're talking about at step two, and which is what petitioners are really trying to get at, there is no default linear anything.

[00:25:05] Speaker 04:
The guidelines say that a linear model is appropriate unless the fit is poor.

[00:25:18] Speaker 04:
But here, and I'm sorry, let me just give you the site for that.

[00:25:26] Speaker 04:
This is at JA 2212.

[00:25:31] Speaker 04:
Nowhere in the guidelines does it say that for step two, there is a linear model as a default.

[00:25:37] Speaker 04:
So I just want to clear the air on that issue.

[00:25:41] Speaker 04:
On petitioners claim that all EPA did is eyeball five dots

[00:25:47] Speaker 04:
to come up with the model, I think that portrays a fundamental misunderstanding of statistical analysis.

[00:25:54] Speaker 04:
Because categorical breakouts are the common form in statistical analysis to make sense of a big data set.

[00:26:03] Speaker 04:
Because all you're seeing there is really just a cloud of points.

[00:26:07] Speaker 04:
And categorical breakouts allow you to see where the weight of those individual data points fall

[00:26:14] Speaker 04:
within an interval and also across different intervals.

[00:26:18] Speaker 04:
And that's important because there can be a lot of variation between individuals in how they respond to exposures or other stimuli.

[00:26:28] Speaker 04:
And EP is interested in the average person's response.

[00:26:31] Speaker 04:
It's not making decisions based on the atypical response.

[00:26:35] Speaker 04:
And the average response is what the categorical breakout allows you to see.

[00:26:40] Speaker 04:
I also want to.

[00:26:41] Speaker 02:
Is that breakout drawn from the underlying the same 17,000 data points?

[00:26:45] Speaker 04:
Yes, it is.

[00:26:47] Speaker 04:
So if you look at JA2680 table D54, it shows you the interval breakouts.

[00:26:56] Speaker 04:
And there are about three to four thousand data points within each interval.

[00:27:01] Speaker 04:
I also want to push back against petitioners' assertion that somehow the only respectable way to do statistics is to look at numbers and equations.

[00:27:14] Speaker 04:
That's just not right.

[00:27:16] Speaker 04:
Graphs and other visuals are a critical tool in statistical analysis.

[00:27:21] Speaker 04:
I mean, you flip open any statistics textbook and their graphs all over the place because that kind of analysis shows you things that you can't see from just numbers and equations.

[00:27:36] Speaker 04:
On whether the models are comparable to each other in terms of relative risk, in figure 4-3, the y-axis tells you that it measures relative risk.

[00:27:47] Speaker 04:
And so if you look at different models, you can see that at a given exposure, one model might tell you that the risk you face is one and a half times relative to your baseline risk, while another model might tell you it's two and a half times.

[00:28:02] Speaker 04:
So Model 2 predicts higher relative risk than Model 1 does, and you can absolutely compare the models in terms of relative risk.

[00:28:13] Speaker 04:
What they don't tell you is the baseline risk that you face.

[00:28:16] Speaker 04:
That depends on who we're talking about in a given analysis.

[00:28:24] Speaker 04:
On biological plausibility, I think there's ample support in the record that EPA's analysis is supported by biological realities.

[00:28:35] Speaker 04:
First of all, the analysis underwent two rounds of external peer review by experts in statistics and epidemiology and toxicology and experts who are well-versed in the relevant literature.

[00:28:49] Speaker 04:
And they agreed with EPA's

[00:28:53] Speaker 04:
choice of the data set, and they also express their preference for the use of a spline model as its dose response model.

[00:29:05] Speaker 04:
And this is at JA 3320-27.

[00:29:11] Speaker 04:
EPA also subjected its analysis to extensive public comments during which commoners submit a lot of studies, including studies that post-state the cancer risk estimate.

[00:29:24] Speaker 02:
Can I stop you on this, the biological plausibility point?

[00:29:30] Speaker 02:
Yes.

[00:29:30] Speaker 02:
Petitioners say that

[00:29:33] Speaker 02:
Generally, a spline like model is not that EPA's own other statements say a spline like dose response model is not common for this type of carcinogen.

[00:29:46] Speaker 02:
What's the response to that?

[00:29:47] Speaker 04:
Well, first of all, they have no citations to any authority for that statement.

[00:29:52] Speaker 04:
which is a big problem.

[00:29:53] Speaker 04:
The other response is, you know, it is true that for mutagenic substances, an increase in exposure means an increase in risk, and that is reflected even in a spline model.

[00:30:04] Speaker 04:
When you increase the exposure, the risk increases.

[00:30:08] Speaker 04:
Now, there are other factors at play that affect our risk.

[00:30:13] Speaker 04:
Some of that might not be completely understood, but we're seeing it in the data.

[00:30:17] Speaker 04:
I can give you a couple of theories of why we're seeing this plateauing effect.

[00:30:23] Speaker 04:
One is

[00:30:25] Speaker 04:
No, people who are just more susceptible to a chemical have already gotten sick or died at lower exposures.

[00:30:35] Speaker 04:
So the people who stayed on long enough to have these very high cumulative exposures could just be naturally less susceptible to the chemical in the first place.

[00:30:45] Speaker 04:
Another theory is that our bodies activate certain defense mechanisms only at high exposures.

[00:30:54] Speaker 04:
So we don't know exactly why, but that is what we're seeing in the data for not only ethylene oxide, but other occupational carcinogens.

[00:31:03] Speaker 02:
The other question I just wanted to go back to was to, I want to make sure I'm understanding the dispute over whether this is linear and whether that matters.

[00:31:10] Speaker 02:
Yeah.

[00:31:10] Speaker 02:
Because in the reconsideration decision at 4,345 to 4,6, this is where EPA is explaining why it's not going to use the Texas model.

[00:31:20] Speaker 02:
And it says the two-piece linear spline model is a low-dose linear model as defined by the guidelines.

[00:31:28] Speaker 02:
And that's why I was asking them questions about the fact that it seems like EPA is saying, this is a linear model that aligns with our guidelines.

[00:31:37] Speaker 02:
But what you said made me worried I'm in the wrong ballpark.

[00:31:42] Speaker 04:
The low dose linear, first of all, the low dose linear is defined as something with a positive slope, which if you look at EPA's model, it has a positive slope and that the low dose linear discussion applies to extrapolation.

[00:31:57] Speaker 04:
So step three, what we're talking about here with choice of the dose response model is step two.

[00:32:04] Speaker 04:
So we are really in a different ballpark when it comes to what the guideline is saying about low dose linear.

[00:32:13] Speaker 04:
Unless there are other questions about the dose response model, I'd like to move to some of the studies that petitioners.

[00:32:20] Speaker 03:
Could I just ask counsel all of that is relevant then at step two?

[00:32:29] Speaker 03:
All of what you were just talking about with Judge Garcia, your low dose linear

[00:32:37] Speaker 04:
No, the low-cost linear as discussed in the guidelines is talking about that as a default for the extrapolation, which is step three.

[00:32:47] Speaker 04:
My point is simply that at step two, there is no default linear or anything.

[00:32:52] Speaker 04:
Thank you.

[00:32:58] Speaker 04:
I just want to say a couple things about the pre-1978 exposure and the smoking studies.

[00:33:04] Speaker 04:
On the exposure, the Bogan studies looking at work practices, if you look at the record at JA 4818, which is the Bogan studies, it's talking about increased wash cycle since the 1980s, which is after the period we're talking about.

[00:33:22] Speaker 04:
So I don't see how that's even relevant here.

[00:33:27] Speaker 04:
On the smoker studies, I mean, the problem with the smoker studies is really that absence of evidence is not the same as evidence of absence.

[00:33:39] Speaker 04:
All we have in the smoker studies is really the absence of evidence because they didn't reliably measure ethyl oxides exposure and they didn't quantify the cancer risk.

[00:33:49] Speaker 04:
So they don't say anything one way or the other about ethyl oxides cancer risk, which is what EPA is interested in here.

[00:33:59] Speaker 02:
But I take part of the point to be that the effects of smoking are one of the most studied impacts there is.

[00:34:06] Speaker 02:
And if EPA were right about the impact of ethylene oxide, we'd at least at a general level expect to see higher cancer rates of this type of cancer in that population if ethylene oxide is causing cancer.

[00:34:22] Speaker 04:
Well, we still have an absence of evidence, because those studies are pretty vague and noncommittal about lymphoid cancer deaths in smokers.

[00:34:34] Speaker 04:
It seems like they're saying there are not enough lymphoid cancer deaths out there in smokers.

[00:34:40] Speaker 04:
Well, how many are there?

[00:34:42] Speaker 04:
And how does that compare with non-smoker's death rates?

[00:34:46] Speaker 04:
We have no idea.

[00:34:47] Speaker 04:
There's an absence of information, because the studies didn't look at this particular question.

[00:34:59] Speaker 04:
I'm happy to address any more questions on the merits.

[00:35:04] Speaker 04:
Otherwise, I can turn to the procedural issues.

[00:35:12] Speaker 03:
Go ahead.

[00:35:13] Speaker 04:
Okay.

[00:35:14] Speaker 04:
Just a few things on procedural issues.

[00:35:19] Speaker 04:
You know, our brief shows that EPA's use and calculation of the cancer risk assessment is procedurally sound, and I want to just spend a couple minutes on some new arguments raised in the reply about specific National Academy recommendations, which

[00:35:35] Speaker 04:
Go to things like greater clarity and explaining methods that are tables to use in comparing different studies and recommendations to have to create some general protocol and and guys.

[00:35:49] Speaker 04:
And although we think the National Academy's recommendations don't even apply here, even if they did, they don't set

[00:35:57] Speaker 04:
the Standard for Arbitrary Precious Review.

[00:36:00] Speaker 04:
Those recommendations are aspirational goals that are aiming for something higher than just not arbitrary and so fair to some aspirational goals does not translate into arbitrary action as a legal matter.

[00:36:15] Speaker 04:
For that you need to look at what happened here

[00:36:18] Speaker 04:
And what happened here is that EPA reviewed the comments that were submitted.

[00:36:23] Speaker 04:
And yes, it disagreed with some of petitioners' comments on the merits.

[00:36:28] Speaker 04:
But that's not a reason to find procedural defect.

[00:36:31] Speaker 02:
So do you agree if the NAS had said something like, it's not specific to ethylene oxide, but there are such fundamental problems with this iris system that EPA needs to stop until it makes XYZ changes?

[00:36:46] Speaker 04:
I think that might change the calculus, but the reality is that EPA explained in its response to comments at JA 1884, the National Academies recognize a lot of improvements that EPA had made, but also that a lot of its recommendations are for the long term, and that there was no need for EPA to just stop all its risk assessments in the meanwhile.

[00:37:06] Speaker 02:
I think it even says, affirmatively, it shouldn't delay EPA's risk assessments.

[00:37:11] Speaker 04:
Right, so the National Academies, the way these recommendations are forced, they don't really say anything about whether what EPA did was arbitrary or capricious.

[00:37:23] Speaker 04:
I'm happy to answer any other questions the report may have.

[00:37:27] Speaker 02:
One last follow-up.

[00:37:28] Speaker 02:
You were talking about biological explanations for a plateauing effect.

[00:37:34] Speaker 02:
I wondered if you had JA citations for those points.

[00:37:38] Speaker 02:
I understand that you might not.

[00:37:40] Speaker 04:
Yes, it's at J.A.

[00:37:42] Speaker 04:
2617.

[00:37:42] Speaker 04:
Thank you.

[00:37:48] Speaker 04:
Thank you very much.

[00:37:51] Speaker 03:
I'm going to take a minute.

[00:38:17] Speaker 01:
Your Honor, Judge Garcia, one of your questions was for the site in the record to the biological mechanism, and that's at JA 2476.

[00:38:24] Speaker 01:
That's in the IRIS assessment.

[00:38:29] Speaker 01:
With respect to the NAS recommendation, we've heard some discussion about them, long-term recommendations.

[00:38:35] Speaker 01:
There are a number of recommendations in there that would have addressed this issue, including the requirement that EPA better consider biology, better elucidate the selection of studies, better rely on statistics.

[00:38:45] Speaker 01:
The fact that they may be long-term and not addressed in every IRS value is not a reason not to comply with 76073, which required EPA to identify and consider those recommendations.

[00:38:57] Speaker 01:
They don't have to comply with all of them before proceeding with rulemaking.

[00:39:00] Speaker 01:
But they have to show that they have considered them and their relevance to the rulemaking before they proceed.

[00:39:04] Speaker 01:
That's the violation we've identified here.

[00:39:07] Speaker 01:
We did not hear any response on the preference for Iris.

[00:39:11] Speaker 01:
The fact is that that poisoned the entire rulemaking.

[00:39:14] Speaker 01:
The EPA was not going to consider everything on the way to the evidence, like they told Congress they would do, and what Congress incorporated through the Benzie-Nishap.

[00:39:20] Speaker 01:
They have not conducted a rulemaking as they were required to do by law.

[00:39:25] Speaker 01:
We heard a reference that we do not dispute the method of extrapolation.

[00:39:30] Speaker 01:
We absolutely do.

[00:39:31] Speaker 01:
That is their model selection.

[00:39:33] Speaker 01:
There is no data at the concentrations EPA is attempting to regulate.

[00:39:37] Speaker 01:
All of this is an exercise of extrapolation.

[00:39:40] Speaker 01:
And the way they're extrapolating is by trying to rely on a model that does not have a statistical basis.

[00:39:45] Speaker 01:
They've described the graphs as a physical tool or as a critical tool, but they cite nothing for that.

[00:39:51] Speaker 01:
There is no guidance that says visual fit should be a primary basis for selecting a model.

[00:39:56] Speaker 01:
The most that's in their guidance is that it can be a confirmatory basis of a model driven by biology and statistics.

[00:40:04] Speaker 01:
We continue to have a dispute over the y-axis.

[00:40:07] Speaker 01:
Most of their brief and an argument heard that you can compare them to see how close they are.

[00:40:11] Speaker 01:
The footnote speaks for itself, however.

[00:40:16] Speaker 01:
It says that the graphs are not strictly comparable across the y-axis.

[00:40:21] Speaker 01:
And that's because what they're showing here is rates of increase, not absolute increases.

[00:40:30] Speaker 01:
We've heard today that they emphasize that this selection has undergone peer review.

[00:40:35] Speaker 01:
It's worth clarifying that this model selection did not go through peer review.

[00:40:39] Speaker 01:
This model selection was done after the last peer review by the SAB.

[00:40:47] Speaker 01:
In addition, on the issue of smokers, we heard that there is perhaps not enough smokers to elucidate the data factor identifying.

[00:40:56] Speaker 01:
That's unsupported by the record.

[00:40:57] Speaker 01:
There are more than enough smokers in the country to identify, if the iris value were correct, that there would be a link to lymphoid cancer.

[00:41:11] Speaker 01:
Your honors, because EPA did not follow its own guidance, disregarding the National Academy of Sciences recommendations, replaced the required statistical and biological analysis with eyeballing a chart, this was not a rational basis for rules with major real-world ramifications.

[00:41:27] Speaker 01:
Particularly since the results failed multiple reality checks, we requested the rule be vacated and the case remanded to EPA for a new risk assessment.

[00:41:35] Speaker 03:
I believe Judge Garcia wanted to cite.

[00:41:39] Speaker 01:
Thank you.

[00:41:41] Speaker 03:
Thank you.