This is the partial deposition of Hal Morgenstern, Ph.D. In it, he discusses the relative risks of developing tardive dyskinesia after using Reglan or Metoclopramide.
Q. Would you calculate that for me on this piece of paper?
A. I remember that I had to solve for something in this, let's see, okay. There
were 51 metoclopramide and 51 others. Okay. I took -- multiply that times that
divided by that times that.
Q. Would you write that down for me? So your odds ratio is the product of 15 times
42 divided by the product of 36 times 9?
A. As I recall.
Q. And you say o-r with a --
A. Little hat above it.
Q. Thank you. I thought there was a technical term for it.
A. There probably is. Most people just say hat. It means estimate. I'm just used
to doing it.
Q. And why do you say this is an odds ratio estimate?
A. Because I estimated it from the data.
Q. And what I'm getting at is, do you ever calculate odds ratios that are not
estimates?
A. Not from data.
Q. And this gave you 1.94, correct?
A. Uh-huh.
Q. And the confidence interval was .76 to 4.97?
A. Yes, if you're reading from my -- yes.
Q. And that confidence interval includes one, correct?
A. Yes.
Q. And what is the significance of the fact that the confidence interval includes
one?
A. Just means that the 95 percent confidence interval includes the null value.
Q. You chose a 95 percent confidence interval; why?
A. Just standard, I could have chosen anything, most common, some people use 99,
some people use 90.
Q. But 95 percent is the most standard confidence interval used in epidemiology?
MR. PITTLE: Object to the form.
A. I think so.
Q. Now look at the paper on page 1472. Under tardive dyskinesia, it says 29
percent (n=15) of --
A. I'm sorry, where are you?
Q. 29 percent (n=15) of metoclopramide users met the case definition of tardive
dyskinesia compared with 17.6 percent (n=9) of non-users, correct?
United States District Court, N.D. Texas,
Dallas Division.
Sue C. MCNEIL, Plaintiff,
v.
WYETH f/k/a American Home Products Corporation d/b/a A.H. Robins Company,
Defendant.
No. 3:02-CV-2072-L.
Q. Do you agree that point prevalence provides no information on the risk of
developing the disease?
A. I'm sorry, I was distracted by the phone. It'll take it, but, I'm sorry, say it
again, ask me the question again.
Q. Sure. Do you agree that point prevalence does not provide information about the
risk of developing the disease?
A. Yes.
Q. And the reason --
A. Well, that's not totally true. The prevalence and incidence are related, so
it's not as if it doesn't tell you anything about it. Prevalence depends on
incidence.
Q. And how would you describe that incidence -- or relationship?
A. Oh, well, that's complicated, but if you make certain assumptions, you can
express it mathematically. If you assume steady state conditions of a population,
and you assume that there's no net migration of cases into or out of the
population, then the prevalence is equal to the incidence rate times the mean
duration of the disease divided by that quantity plus one.
[Note: Page 50 missing in original document]
A. You're going to misunderstand if I say yes to that, I'm sure, because if you
want to measure -- if you're interested in measuring the prevalence of disease in
a -- I'm sorry, the incidence of disease in a population, the prevalence would not
necessarily be equivalent to that, that's true, but it doesn't follow that you
don't get any information about incidence from a cross-sectional study.
Q. Do you agree that case-control studies cannot determine the incidence rate of a
disease?
A. Disagree with that.
Q. Why?
A. Because you can do a population based case-control study; you get new cases of
disease that occur over time. Those are the cases. You know the size of the
population that gave rise to those cases. You can estimate the incidence rate or
the risk.
Excuse me a second.
Q. Sure.
(Brief Pause.)
Q. Do you agree that case-control studies cannot determine the -- I'm sorry, do
you agree that prevalence studies cannot determine the incidence rate of disease?
A. Yes. Well, no, that's not true. It would be difficult, but one can use
prevalence information to get incidence.
Q. And how would you do so?
A. It's tricky. You'd have to know, for all the prevalent cases in your
population, you'd have to know when the disease started. You'd have to know what
was the onset of symptoms probably.
Q. So in order for a prevalence study to determine the incidence rate of a
disease, you would need to know the onset of the disease, correct?
A. Right.
Q. Do you agree that prevalence studies are more susceptible to selection bias
than incidence studies?
A. Yes, well, than a cohort study.
Q. Than a cohort study?
A. Yes. You see, a case-control study can involve incidence. Case-control studies
can be done cross-sectionally or longitudinally.
Q. What is relative risk?
A. It's the risk of developing a disease in one group divided by the risk of
developing the disease in another group.
Q. What is a risk ratio?
A. Same thing.
Q. Is it the number of people who are exposed who develop the disease divided by
the number of people who
[Note: Pages 53-62 missing in original document]
A. Not by me.
Q. I'm trying to understand how you use -- how do you use relative risk?
A. The ratio of two risks, risk in one group divided by risk of the same outcome
in another group.
Q. And relative risk is only obtained through prevalence -- I'm sorry, incidence
information, correct?
A. No. You can estimate it in case-control, cross-sectionally as well.
Q. And the way you estimate it is through an odds ratio?
A. If you want to -- depends what you want to estimate. If you want to estimate
the incidence rate ratio, yes.
Q. Do you agree that because a prevalence study begins with a group that has
developed the disease, it is not possible to determine the rate at which the
disease developed?
A. Didn't we just talk about this? You're asking me if you can determine the
incidence rate from prevalence data, and my answer was yes, but it's kind of
difficult. You have to have additional information to do that and it's not done
very often.
Q. And relative risk cannot be computed from the data generated by a prevalence
study?
[Note: Pages 64-77 missing in original document]
A. I published literature on neuroleptics, yes.
Q. And in those publications, you define neuroleptics as antipsychotics, correct?
A. Oh, no, course not.
Q. Do you refer to neuroleptics as antipsychotics?
A. I was studying schizophrenics, for the most part, schizophrenics who were being
treated in an outpatient population in New Haven, Connecticut. The reason we were
studying them is because they were on antipsychotics. I did not define
neuroleptics as antipsychotics.
Q. Did you consider -- what other medications are neuroleptics?
A. I'm not sure.
Q. So you believe antipsychotics are neuroleptics, correct?
A. Yes, they are.
Q. And you believe metoclopramide is an antipsychotic?
A. No, I don't believe metoclopramide --
Q. I'm sorry, metoclopramide is a neuroleptic?
A. Yes, I mean that's my understanding from reading the literature. I'm not a
pharmacologist.
Q. And you're not familiar with the binding affinities, correct?
A. Well, I've read a little bit about them, but, no, I'm not very familiar.
Q. You're not going to talk about the pharmacology of metoclopramide, correct?
A. No, I'm not.
Q. And, Doctor, as I understand it, the first time you formed the opinion that
metoclopramide was in fact a neuroleptic was after you spoke with Mr. Pittle,
correct?
MR. PITTLE: Object to the form of that question.
A. It's true, but it's not necessarily because Mr. Pittle told me that.
Q. I'm just talking about timing.
A. Yes, I believe it was.
Q. Prior to your involvement with this litigation, you knew about metoclopramide,
correct?
A. I had heard of metoclopramide.
Q. And you had read about metoclopramide?
A. I'm not sure I read about it. I think I probably just heard about it in the
context of collaborating with a number of different kinds of physicians at Yale
Medical School, possibly before then.
Q. And you knew that metoclopramide could cause
[Note: Pages 80-84 missing in original document]
And to do a meta-analysis might be useful for this case, but it wouldn't, I don't
think, be necessarily original science that anyone would be willing to publish in
a journal. Why? Because I'm not showing anything more than what those two authors
already claimed. It literally took me, I don't know, maybe an hour at the most to
do it.
Q. So your analysis of these two studies or combination of these studies took you
only an hour?
A. Something like that.
Q. And as I understand it, you have no plans to publish, correct?
A. That's right.
Q. The combination of these studies was done solely in the context of this
litigation?
A. Yes, it was.
Q. And you don't believe it would be publishable material?
A. Because I don't think it's original enough, not because I don't think it's
valid.
Q. And you don't believe it's original enough because it simply restates what is
already in the medical literature?
A. It's not a -- well, it doesn't -- no, it's not restating what's in the medical
literature because I
[Note: Pages 86-87 missing in original document]
Most journals are not terribly interested in them except if it's a very elaborate,
major issue, you know, and it's very controversial and people really think that
the meta-analysis is going to contribute beyond what everyone knows from the
existing literature.
But combining two studies, it's like I'm almost stating the obvious from the two.
You can look at the two studies and see that they give similar results. No one's
going to require quantitative analysis to see that. I just did it formally because
of the context of this case.
Q. But my question is, if you were to present this for publication, would you
affirmatively describe your combination of Sewell and Ganzini as a meta-analysis?
A. I could.
Q. Would you?
A. It would be --
Q. Would you as an --
A. I don't know.
Q. -- epidemiologist?
A. I was just telling you, I have no intent -- it never even occurred to me that I
would publish this. So what I call it, what I call it, I think people would laugh
at me if I called it a meta-analysis, frankly, nowadays.
Q. And the reason for that is it's simply two studies?
A. It's only two studies, yes.
Q. And you've done meta-analysis in the past, correct?
A. Yeah, I've done meta-analysis.
Q. In fact, one of your meta-analysis was published back in 1987?
A. On tardive dyskinesia.
Q. Yes.
A. Yeah.
Q. And have you reviewed meta-analysis for publication?
A. You mean as a reviewer for a journal?
Q. Yes, yes.
A. I think I got one once. It had nothing to do with tardive dyskinesia.
Q. And as a reviewer of epidemiological papers, would you accept your combination
as, quote/unquote, a meta-analysis?
A. What I just did?
Q. Yes.
A. Yes.
Q. Meaning you would accept it if the author
[Note: Pages 90-93 missing in original document]
and that's actually stated in here. They're the only two papers I know that
examine the association between metoclopramide use and tardive dyskinesia, so I
could have been fancier than that, but these were the only two studies.
Q. The first study is the Ganzini study, paper?
A. Yes.
Q. What summary measures of association did you take from the Ganzini study?
A. Odds ratio, prevalence odds ratio.
Q. Do you have that paper in front of you?
A. Oh, the paper?
Q. Yes.
And what was that number?
A. What was what number?
Q. The prevalence odds ratio?
A. Well, I have to look in what I wrote. It's not in the paper. I had to calculate
it; 1.94.
Q. Where is that in the paper?
A. It's not in the paper. The authors didn't do that. I'll have to find what they
did do.
Now I see where you got the word controls from.
Yeah. I got it from Table 1. The information that I needed to do the calculations
were from Table 1.
Q. And how did you do the calculation?
A. I estimated the odds ratio.
Q. How did you estimate it?
A. How do you estimate an odds ratio?
Q. Yes.
A. I'm not sure how to say that to you verbally. I can define the odds ratio. I
can tell you how to calculate it, if you want me to show you.
Q. Let's do it very easily. I'm going to hand you a piece of paper. We'll mark
that as Exhibit 3.
You've told me that the prevalence odds ratio that you used or calculated was
1.94?
A. Yeah.
(Deposition Exhibit No. 3 marked for identification.)
Q. Would you calculate that for me on this piece of paper?
A. I remember that I had to solve for something in this, let's see, okay. There
were 51 metoclopramide and 51 others. Okay. I took -- multiply that times that
divided by that times that.
Q. Would you write that down for me? So your odds ratio is the product of 15 times
42 divided by the product of 36 times 9?
A. As I recall.
Q. And you say o-r with a --
A. Little hat above it.
Q. Thank you. I thought there was a technical term for it.
A. There probably is. Most people just say hat. It means estimate. I'm just used
to doing it.
Q. And why do you say this is an odds ratio estimate?
A. Because I estimated it from the data.
Q. And what I'm getting at is, do you ever calculate odds ratios that are not
estimates?
A. Not from data.
Q. And this gave you 1.94, correct?
A. Uh-huh.
Q. And the confidence interval was .76 to 4.97?
A. Yes, if you're reading from my -- yes.
Q. And that confidence interval includes one, correct?
A. Yes.
Q. And what is the significance of the fact that the confidence interval includes
one?
A. Just means that the 95 percent confidence interval includes the null value.
Q. You chose a 95 percent confidence interval; why?
A. Just standard, I could have chosen anything, most common, some people use 99,
some people use 90.
Q. But 95 percent is the most standard confidence interval used in epidemiology?
MR. PITTLE: Object to the form.
A. I think so.
Q. Now look at the paper on page 1472. Under tardive dyskinesia, it says 29
percent (n=15) of --
A. I'm sorry, where are you?
Q. 29 percent (n=15) of metoclopramide users met the case definition of tardive
dyskinesia compared with 17.6 percent (n=9) of non-users, correct?
A. Uh-huh.
Q. You have to say yes.
A. Yes, yes.
Q. The relative risk of developing tardive dyskinesia was 1.67, 95 percent
confidence interval, 0.93 to 2.97?
A. Yes.
Q. Do you agree that that's an incorrect use of the term relative risk?
A. Yes.
Q. And why is that?
A. They were actually comparing two prevalences. I would have called it the
prevalence ratio or relative prevalence.
Q. So the authors should have used the term prevalence odds ratio?
A. No. That's not an odds ratio. It's a prevalence ratio or relative prevalence,
people could call it. Instead of being a relative risk, it's a relative
prevalence.
Q. And would you agree that the relative risk, as they use the term, of 1.67, was
not statistically significant?
A. No, I wouldn't agree with that.
Q. Why?
A. Because I don't believe that what you call statistical significance has any
scientific meaning.
Q. In general?
A. In general.
Q. So to say that something is not statistically significant has no meaning in
terms of epidemiological conclusions?
A. That's right. It can be misleading, as a matter of fact. It often is.
Q. So my question again is, since the confidence interval includes one, do you
agree that this is not a statistically significant conclusion?
A. No, I don't agree.
Q. And why was it that you calculated -- would you explain why you did the
calculations the way you did them?
A. To guesstimate the odds ratio.
Q. I know, but why did you multiply 15 times 42?
A. Why did I multiply 15 by 42?
Q. Sure, sure.
A. Well, because the odds ratio is the ratio of two odds.
Q. Go ahead and write it out.
A. Say the odds in the exposed divided by the odds in the unexposed, okay. The
odds of being a case in the exposed group is 15 over 51 divided by 1 minus 15 over
51. That's an odds, probability of being a case divided by 1 minus the probability
of being a case, or probability of being a non-case; do the same thing in the
denominator, 9 over 51 divided by 1 minus 9 over 51. If you solve for that, it'll
come out to 15 times 42 divided by 36 times 9.
Q. Have you seen the meta-analysis done by Dr. Strom?
A. Meaning the -- I think I did a long time ago.
Q. Have you seen the combination of the Sewell and Ganzini studies by Dr. Strom?
[Note: Pages 100-111 missing in original document]
A. No.
Q. Is there anything else that it's based on other than the confidence interval
being one?
A. That's what I based it on.
Q. And you note, it should be noted, however, that one cannot conclude there is no
effect or association simply because the P value is greater than 0.05 in each
study; why do you say that?
A. Because apparently I was predicting what you were going to ask me today. You
think it is because you've read it everywhere, I'm sure that you've seen it in
print, and people often assume it, but it's fallacious.
Q. Would you agree that it is widely published that there is no effect or
association because the P value is greater than .05?
A. There are a lot of incorrect things that are widely published and, yes, it is
widely -- well, it's widely published, yes, and widely believed.
Q. And why do you believe that it is incorrect to conclude that there is no effect
or association simply because the P value is greater than .05?
A. Well, fundamentally because the statistical information never provides an all
or none answer to whether there's an effect or an association or not, so
[Note: Pages 113-122 missing in original document]
et al and Sewell, et al should be replicated in other populations, I have no
reason to expect the results to be substantially different. That's how I would
answer the question.
Q. Yes or no, are the results of your combining Sewell and Ganzini generalizable
to the overall population of metoclopramide users?
MR. PITTLE: Object to the form.
A. I believe they probably are.
Q. And what is that based on?
A. What I just told you, it's based primarily on my knowledge of tardive
dyskinesia and studying the effect of antipsychotics, which appears to have
similar activities, similar causal action as the other neuroleptics.
Q. Any other basis?
A. No, I guess that's the primary reason. Well, yeah, one other thing, these two
studies are in different populations, they're both VA populations, but they were
done different places, different states, different VA systems, and they get
similar results.
So when one gets similar results from different studies conducted in very
different populations, that kind of replication enhances your ability to
generalize. Admittedly, there's only two
[Note: Page 124 missing in original document]
I'm just going to help you out. I'm in this paragraph that begins, there are
several case-series investigations, but the results of these studies cannot be
used to estimate the effect of metoclopramide on TD because investigators do not
have information on the population from which the TD cases were identified,
correct?
A. Yes.
Q. And, therefore, they cannot measure the statistical association between
metoclopramide and TD status, correct?
A. Yes.
Q. In other words, case-series or case reports cannot be used to measure the
statistical association between metoclopramide and tardive dyskinesia, correct?
A. Right.
Q. Would you also agree that case reports or case-series cannot be used to
determine the amount, or the duration, rather, of prescription use?
A. Well, that depends on how the cases were assembled. If the cases are regarded
as a complete number of cases or a random sample of cases from some well-defined
population, and one were able to get accurate medical records for everyone to know
when medication was used, then at least for that group you'd
[Note: Pages 126-141 missing in original document]
A. Yes.
Q. From an epidemiological standpoint, would it be correct to conclude, based on
the Sewell -- I'm sorry. Let me start over.
From an epidemiological standpoint, is it correct to conclude that 29 percent of
people who take metoclopramide will develop tardive dyskinesia?
A. No, that's not correct.
Q. So if someone testified that the Ganzini paper stands for the proposition that
29 percent of people who take metoclopramide will develop tardive dyskinesia,
that's an incorrect statement?
MR. PITTLE: Object to the form.
A. It's an incorrect statement.
Q. Why?
A. Well, the way you expressed the statement, you expressed it in terms of
incidence, but you didn't say incidence, and you're basing it on a prevalence
study.
Q. You don't have any incidence information about tardive dyskinesia association
with metoclopramide, correct?
A. Correct.
Q. Now if you look at the Sewell study, those authors concluded that 27 percent of
those who took metoclopramide had tardive dyskinesia at that point in time,
correct?
A. Yes, I believe so. I'll have to look at the numbers again. It's down here, 27
percent, yeah.
Q. Am I correct?
A. Yes.
Q. From an epidemiological standpoint, would it be correct to state, based on the
Sewell study, that 27 percent of people who take metoclopramide will develop
tardive dyskinesia?
MR. PITTLE: Object to the form.
A. No.
Q. That would be an incorrect statement?
A. Yes.
Q. For the same reasons as you explained with respect to the Ganzini study?
A. Yes. By the way, it could be much more than that.
Q. We just don't know?
A. Right.
Q. At the break, I asked Dr. -- let me back up.
You pulled two files that are sitting on your desk. They're about an inch thick.
Are those the files that --
A. No, they're not an inch thick.
Q. Well, I'm sorry, about a half inch thick?
