Lemuel A. Moye filed this expert witness report on behalf of the plaintiff in Longs v. Wyeth.
Executive Summary
The following is my review of both the scientific literature, federal Food and
Drug (FDA) documents, deposition testimony, and documents from Wyeth that are
directly relevant to the causal link between the fenfluramines (Pondimin and
Redux) and the occurrence of primary pulmonary hypertension (PPH).The intelligent use of the principles of epidemiology are essential in elucidating
the relationship between the fenfluramines and primary pulmonary hypertension and
2) the fenfluramines and valvular heart diseaseTo a reasonable degree of scientific and medical certainty, the fenflurarnines
cause primary pulmonary hypertension. Fenfluramines are not effective in producing
health improving weight loss in the obese. The overwhelming epidemiologic
evidence, using state of the art scientific methodology, concludes that the
fenfluramines cause primary pulmonary hypertension. Fenfluramines cause primary
pulmonary hypertension even for short durations of exposure (less than three
months). PPH can be induced by the fenfluramines months or years after the
fenfluramines have been discontinued. The risks of use of fenfluramines exceed the
benefits. Wyeth knew that fenfluramines caused primary pulmonary hypertension.
Wyeth did not keep the FDA, physicians, or patients apprised of the risks
associated with the fenfluramines. Wyeth mislead the FDA Advisory Committee about
the risks associated with Redux and Pondimin. The labels for Redux and Pondimin
are both defective. In fact no label could be written to describe the safe usage
of these compound.Also, There is overwhelming scientific evidence to support my opinion that
fenfluramine and dexfenfluramine cause valvulopathy. These studies were properly
designed and executed by qualified scientists, followed accepted methodology, and
relied upon objective interpretation of scientific and medical data. These
theories have been subjected to peer-review on countless occasions including but
not limited to the reference articles. Random and systematic errors have been
appropriately controlled for by generally accepted epidemiological methods. These
theories are generally accepted as valid in the scientific community and are based
upon reliable studies. These theories have been advanced in scientific journals
and seminars and not just used for litigation purposes
United States District Court, N.D. Ohio.
Ramonia LONGS, et al.,
v.
WYETH, et al.
No. 103CV02042.
March 31, 2005.
Personal Vita:
1 Introduction My name is Lemuel A. Moye. I am over twenty-one years of age, am of
sound mind, am not a party to this action, have never been convicted of a felony,
and am otherwise competent to make this affidavit. I have personal knowledge of
all factual statements contained herein, and all such factual statements are true
and correct to the best of my knowledge.
2 Degrees: I have a M.D. and a Ph.D. degree in Community Sciences - Biostatistics.
I am a licensed physician in the states of Texas and Indiana, and I have actively
practiced medicine from 1979-1992. I am a diplomat of the National Board of
Medical Examiners. My formal training has included many courses in mathematical
statistics, epidemiology, and biostatistics. I am currently a Professor of
Biostatistics at the University of Texas School of Public Health in Houston, where
I hold a full time faculty position in biostatistics.
3 Research Experience: I have carried out cardiovascular research for seventeen
years and continue to be involved in the design, execution and analysis of
clinical trials. I have been Principal Investigator on a grant from
Schering-Plough and have been Co-Principal Investigator on two grants from Bristol
Myers-Squibb. In these studies, I was responsible for the design, execution, and
analysis of these experiments. In one case, the experiment involved over 2000
patients who were followed for 3.5 years, and in the other case, 4159 patients
were followed for five years. In each of these enterprises, an important component
of my responsibility was the reporting of adverse events from physicians. In one
of these trials, I supervised the collection of adverse events that were reported
to both the Sponsor and to the Federal Food and Drug Administration (F.D.A.). Each
of these clinical trials has resulted in several articles that were published in
the Journal of the Medical Association and The New England Journal of Medicine, as
well as other journals in the peer reviewed medical literature.
4 Current Activities I am currently Principle Investigator and in charge of the
biostatistics group that is carrying out research on the treatment of strokes,
funded by the National Institute of Neurologic Disorders and Strokes.
5 Manuscripts: I have published over one hundred manuscripts in peer-reviewed
literature that reflect my experience in the design, execution and analysis of
clinical trials. Included among these manuscripts are papers that provide the
mathematical development of tools that improve both the design and the execution
of large-scale clinical trials.
6 Books: I am the sole author of the book Statistical Reasoning in Medicine - The
Intuitive P value Primer, published by Springer-Verlag in 2000. A second book,
entitled DIFFERENCE EQUATIONS WITH PUBLIC HEALTH APPLICATIONS, authored by a
colleague, and myself was published by Marcel-Dekker in October 2000. A third book
of which I am sole author, entitled Multiple Analysis in Clinical Trials:
Fundamentals for Investigators, was published by Springer-Verlag and will appear
in the summer 2003. A fourth book titled Mathematical Statistics with Applications
will appear in the winter of 2005. A fifth book entitled Finding Your Way in
Science will appear in August, 2004.
7 Presentations: I have been an Investigator on four grants from the National
Institutes of Health, involving the design, execution, and analysis of clinical
trials and epidemiologic studies. Each of these has led to publications in the
peer-reviewed literature. I have presented results of clinical research at over
twenty different professional meetings, including a presentation in 2001 to the
Drug Information Agency and a separate presentation at the International Joint
Statistical Meetings.
8 Teaching: I have consistently taught courses in biostatistics since my full-time
appointment as a faculty member in 1987 at the University of Texas School of
Public Health. In these courses, I have taught experimental design in clinical
research studies, the statistical methodology that supports these design
principles, the statistical analysis of clinical research programs, and the role
of epidemiology in interpreting the conclusions from these clinical research
programs. For 17 years, I have been a supervisory professor for students training
in epidemiology, biostatistics, and other areas of public health.
9 Invited Lectures: I have given over thirty invited guest lectures on research
methodology topics.
10 Consultations: I have served as a clinical trial consultant to Berlex, Proctor
and Gamble, Marion Merrill Dow, Pfizer, Hoerst Roussel, Aventis, Key
Pharmaceuticals, Coromed, Dupont, Bristol Myers-Squibb, Novartis, Medtronics,
Astra-Zeneca, CryoCor and Vasogen.
11 Grants: I have reviewed grants and/or data on over fifty different occasions on
behalf of government research programs.
12 Monitoring Boards: I have served on three Data, Safety, and Monitoring Boards
that oversee the conduct of clinical trials, and I am currently the chairman of a
Data, Safety, and Monitoring Committee of a study sponsored by
Bristol-Meyer-Squibb and Key Pharmaceuticals.
13 FDA Experience: I have appeared before the F.D.A. on behalf of sponsors on
several occasions.
14 In addition to serving as a consultant with the pharmaceutical industry for
seventeen years, I have also directly and overtly supported the F.D.A. I have
served as a statistician/epidemiologist for four years on the Cardiovascular and
Renal Drug Advisory Committee to the Food and Drug Administration. In this
capacity, I have reviewed the data provided by pharmaceutical companies and an
analysis of that data by the FDA to render opinions about the safety and
effectiveness of these medical interventions being considered for FDA approval. In
this capacity, I have formally reviewed over twenty clinical trial programs of
private sponsors and commented in public testimony that is part of the Federal
Registry. In some instances, a component of this review was direct evaluation,
criticism and comment on the proposed product label by the Sponsor.
15 I currently serve as a statistical consultant to the F.D.A., and most recently
have been to a two-year term on the F.D.A. Pharmacy Sciences Advisory Committee.
16 Recent Publications I have published manuscripts as the sole author in each of
the following journals: Controlled Clinical Trials, and Statistics in Medicine.
This work has promulgated the philosophy (while advancing the supporting
mathematics) that clinical research interpretation must be disciplined and must
explicitly hold the highest regard for community protection from dangerous, false,
and misleading results from clinical research. I have been featured in health care
articles in both Money magazine (December 1998) and US News and World Report
(January 11, 1999). In 2000, an article, on which I am the sole author, appeared
in Statistics in Medicine, and it has generated important published commentary in
the peer-reviewed literature.
17 Risk Benefit Experience: I have taken part in research studies that have been
both observational and experimental. In each of these studies, functioning as an
investigator, I have participated in discussions assessing the safety and
effectiveness of the medication. This assessment has not been based on complicated
economic arguments, but instead places emphasis on balancing the health advantages
the therapy offers against the disadvantages its use produces. I have published an
article that computed the effectiveness and safety of various strategies in the
treatment of essential hypertensiona.[FNa]
FNa. Moye L.A., and Roberts. S.D. 1982. "Modeling the pharmacologic
treatment of hypertension," Management Science; 28:781 -797.
18 During the thrice-yearly meetings of the Cardio-Renal Advisory Committee to the
F.D.A., I, along with the other committee members, have been asked to weigh the
advantages and disadvantages of the medications presented to us for approval. This
assessment was not economic but a collective appraisal by physicians and health
care researchers of the relative clinical strengths and clinical weaknesses of the
intervention.
19 My service on the F.D.A. Advisory Committees, as well as my service to drug
manufacturers, has given me specialized knowledge and experience concerning the
F.D.A. policies, procedures and regulations as well as the corresponding duties of
a reasonable and prudent drug manufacturer. In my service on behalf of sponsors to
the F.D.A, and my service to the F.D.A. at Advisory Committee meetings, I have
developed expertise, knowledge, education, training and experience in F.D.A.
matters.
20 Education: My education consists of the following:
Year Degree School
1987 Ph.D. Community Sciences - Biometry University of Texas
1981 M.S. Statistics Purdue University
1978 M.D. Indiana University
1974 B.A. (mathematical sciences) The Johns Hopkins University
21 I have gained scientific, technical, and specialized knowledge of
biostatistics, epidemiology, and risk/benefit assessment. The basis for my
opinions is derived from my education, training, research, experience, expertise,
review of the peer-reviewed medical literature, review of internal corporate
documents and other documents and studies, whether published or unpublished,
concerning obesity, pulmonary hypertension, the fenfluramines and review of the
deposition testimony of certain of the defendants' current and former employees,
as well as that of FDA employees.
22 Fees: My fees are $400 per hour for review of literature, documents,
depositions and trial testimonies.
23 Previous Testimony: A list of my previous testimony over the past four years is
attached as Appendix B. My curriculum vitae is attached in Exhibit M-1 of Appendix
A.
General Principles
24 What is Epidemiology? Epidemiology is the study of the cause of disease and its
distribution in human populations. It is sometimes helpful to think of
epidemiologists as "disease detectives". Traditionally, they arrive at the crime
scene (represented by the unfortunate patients with disease). They begin to make
careful observations, building a list of suspects (possible causes of the
disease). The epidemiologists then methodically examine each suspect, examining
the evidence for and against their culpability. As suggested by consideration of
the three components of this process, there is a natural progression in
epidemiological reasoning. The process begins with an observation suggesting the
possible influence of a particular factor or exposure on the occurrence of
disease. This suspicion may arise from clinical practice, examination of disease
patterns, observations from laboratory research, or even from theoretical
speculation. Regardless of the source of the observation, the observation itself
leads to the formulation of any hypothesis. This hypothesis is then put to the
test by carrying out epidemiological studies of individuals who have been both
exposed and unexposed to the putative risk factor, measuring the occurrence of
disease in both groups. This systematic collection and analysis of data engenders
a determination of whether a statistical association exists, that is, whether the
disease more commonly occurs in the presence of the risk factor than in its
absence. Validity is assessed as the epidemiologists evaluate the role of bias,
chance, or confounding as being responsible for the observed association. Finally,
a judgment is made for cause-effect relationship between exposure and health
outcome (disease).
25 Causation. A central issue in epidemiology is the ascertainment of the true
nature of a risk factor - disease relationship. Risk factors and disease can be
related in one of two ways. An association between risk factors and disease is the
observation that simultaneously occurs in a group of individuals. Causal
relationships on the other hand are much stronger. The relationship is causal if
the presence of the risk factor produces the disease in an individual; the disease
would not have been present if the risk factor was absent. This causal
relationship is very tight, and has an embedded sense of directionality (i.e. the
risk factor was present and subsequently excited the production of the disease).
The declaration that a relationship is causal has a deeper meaning than the
statement that a risk factor and disease are merely associated. This deeper
meaning and its implications for health care require that the demonstration of a
causal relationship rise to a higher standard than just their joint occurrence.
26 Avoid Artificial Rules: Just as justice is more than simply reciting from a
book of laws, the determination that a risk factor-disease relationship is causal
requires more than the blind application of artificial rules. Risk factor-disease
relationships are unique and each requires specific attention. For example, there
are circumstances in which a strong risk factor-disease relationship, first
reported as suggestive of causation, was subsequently shown to be non-causative
and coincidentalb. On the other hand, there are clear examples of risk
factor-disease relationships identified by a single examination of a series of
cases (with no quantitative measure such as an odds ratio) that are clearly
causative. Two of the clearest examples of this are 1) the link between
thalidomide and birth defects, convincingly detected by Lenz [l] and 2) the
Seldane-Ketoconazole link with deadly irregular heartbeats elucidated by
Monahan[2]. In each of these cases, one and only one study was required to
demonstrate causality. Furthermore, each was a case series. The most important
lesson one can take from these demonstrations is that the determination of
causality is a thinking person's business. There is no substitute for clear
observation, deliberative thought, and careful deductive reasoning in
consideration of the true nature of a risk factor-disease relationship.[FNb]
FNb. A fine example of this is the study that suggested a causal
relationship between coffee ingesion and pancreatic cancer. This association
that first appeared in the 1970's was subsequently debunked.
27 The determination of the true nature of an exposure-disease relationship (i.e.,
whether the relationship is merely associative, or directly causal) can be
complex. If, in a study, the play of chance, presence of bias, and confounding are
all determined to be unlikely explanations of the findings, we can then conclude
that a valid statistical association exists between exposure and disease. However,
a statistical association alone does not imply causality, and additional criteria
are required. In their work, epidemiologists often link statistical analysis with
careful research planning, observation, and deductive reasoning to allow a clear
examination of the nature of the relationship between exposure and disease.
28 Hill's Criteria. The efforts to understand and articulate the arguments that
should be used to identify or discredit a risk factor-disease relationship have
evolved over three hundred years. In 1965, Austin Bradford Hill[3] described the
nine criteria for causality arguments in health care. These nine rules or tenets
are remarkably and refreshingly devoid of complex mathematical arguments, relying
instead on natural, honest intuition and common sense for the inquiry into the
true nature of a risk factor-disease relationship. Are there many disease cases
when the risk factor is present, and fewer disease cases when the risk factor is
absent? Does a greater exposure to the risk factor produce a greater extent of
disease? Other questions asked by Dr. Hill explore the "believability" of the
relationship. Some of these are: Is there a discernible mechanism by which the
risk factor produces the disease? Have other researches also shown this
relationship? Are there other such relationships whose demonstration helps us to
understand the current risk factor- disease relationship? The nine precise
Bradford Hill criteria are: 1) Strength of Association 2) Consistency 3)
Specificity 4) Temporality 5) Biologic gradient 6) Biologic plausibility 7)
Biologic coherence 8) Experimental evidence 9) Analogy. Definitions and examples
of these nine characteristics of a relationship follow.
a. Strength of association: The relative risk (or odds ratio) is the cornerstone
for causal inferences, measuring the strength of association. The higher the
relative risk or odds ratio the stronger the association between the exposure and
the disease. In the IPPHS[4] study, it was determined that for exposure to
anorexigens, the estimated odds ratio for PPH was 6.3. When the measure of the
strength of association is high (i.e. a high relative risk, or a high odds ratio)
it s high, it is unlikely that any other factor could be the cause of this
association. The higher the relative risk is, then the stronger the association
is, and the less likely other factors can explain the association.
b. Consistency with other knowledge: Consistency requires that the findings of one
study be replicated in other studies. The persuasive argument for causality is
much more clearly built on a collection of studies involving different patients
and different protocols, each of which identify the same relationship between risk
factor exposure and effect. The Brenot[5], Douglas[6], and Abenhaim[4] studies are
thee examples of scientific evolutions using different designs and patient
populations but which nevertheless successfully identify the same hazardous
relationship between use of the fenfluramines and the occurrence of PPH.
c. Specificity: If the exposure is associated with only one disease or type of
disease, the specificity criteria are met. The presence of specificity is
considered supportive but not necessary, and epidemiologists no longer require
that the effect of exposure to an agent such as a drug be specific for a single
disease. However, unlike diseases such as heart attacks, strokes, and cancers, PPH
has had relatively fewer causes identified. It is therefore somewhat easier to
rule out alternative causes of this disease in the identification of the
relationship between it and use of the fenfluramines.
d. Temporal relationship: A temporal relationship must exist in order to
convincingly demonstrate causation. Exposure must occur before the disease
develops for it to cause that disease. Protopathic bias (drawing a conclusion
about causation when the disease process precedes the risk factor in occurrence)
can result without appropriate attention to the condition. The Douglas[6] study
clearly demonstrated that when the fenfluramines are introduced to a patient, a
subsequent result is elevated pulmonary blood pressure.
e. Biologic gradient (dose response): This assumes that the more intense the
exposure, the greater the risk of disease. However, a dose response relationship
is not necessary to infer causation.
f. Biologic plausibility: There should be some basis in the scientific theory that
supports the relationship between the supposed "cause" and the effect. However,
observations have been made in the epidemiological studies that were not
considered biologically plausible at the time but subsequently were shown to be
correct. A number of biologically plausible mechanisms have been suggested in the
relationship between PPH and anorexigens in both human and animal studies, the
results of which have been published and analyzed in authoritative medical
journals.
g. Biologic coherence: This implies that a cause-and-effect interpretation for an
association does not conflict with, what is known of the natural history and
biology of the disease. If we claim that a newly introduced exposure of high
prevalence greatly increased the incidence of PPH, there should be an increased
incidence of PPH in the population at large. If exposure to anorexigens caused the
epidemic of PPH in Germany in the 1960's[7], we would expect a decreased incidence
with withdrawal (also observed). To my knowledge, there is not any contradictory
evidence in the scientific literature to refute the association between
anorexigens and PPH.
h. Experimental evidence: This would include both laboratory experiments on
animals as well as human experiments. Experimental evidence also includes the
results of the removal of a harmful exposure. The Douglas[6] finding is a pellucid
example of this fundamental and revealing principle of Hill's.
i. Analogy: This would include a similarity to some other known cause-effect
association. The aminorex experience in Europe demonstrated that patients exposed
to an amphetamine like diet drug can produce primary pulmonary hypertension. This
recent historical example makes it all the more plausible that another amphetamine
diet drug can produce the same disease in patients who are exposed.
Research Methodology
35 Limitations imposed by sampling: Once a research team has chosen a study
design, it has to choose its patients. It is perhaps a truism that researchers are
most times unable to examine every patient in the population they wish to study.
For ethical, fiscal, logistical and administrative reasons, the researchers can
only study a tiny fraction of the targeted patients. This small fraction, termed a
sample, is studied in great detail. However, the sample is useful only to the
degree that it provides information about the larger population from which it was
selected.
36 Sample Extension. The researchers are extremely interested in extending the
results from the sample to the population. They collect a relatively small sample,
not for the sake of the sample, but to learn about the large population from which
the sample was obtained. However, this process of extension is very delicate.
Essentially, the extension requires taking the findings from the small sample that
the researchers studied, observed, and carefully documented, and then applying
these findings to thousands (and often millions) of patients who have not been
directly observed and studied. There is no doubt that extension is both a
necessary and a hazardous process. It is made more dangerous by the recognition
that several competent researchers can each independently evaluate different
samples from the same population. Since the samples are different, the sample
results they wish to extend to the population are different. As we have seen all
too often in the late 20th century media, this cacophony of conflicting scientific
results leads to chaos and policy gridlock.
37 Two guidelines to apply to results from small samples to large population: The
scientific community has developed a collection of guidelines which, when
followed, allow the safest extension of the small sample findings to the larger
population. These guidelines are simple. First, establish a procedure by which
research is peer-reviewed, i.e. carefully scrutinized and criticized by a small
number of experts. These results appear in peer- reviewed medical and scientific
journals. This is a sign that the study's methodology is consistent with the
standard research procedures accepted by the scientific community. These articles
must be given a greater priority than publications in non-peer-reviewed journals.
Peer-reviewed journals are also superior to abstracts, which are themselves only
brief, preliminary reports of non-peer-reviewed work. Like houses built on
shifting sands, contentions based on abstracts are unstable and unreliable. The
best example of being misled by an abstract is that of Weissman [8] in which the
subsequent peer-reviewed publication contradicted the first preliminary findings
announced in the abstract. Abstracts, much as the telephone call that sets up a
blind date, promise much, but are often bitterly disappointing when the full view
(of the data) finally subsequently becomes available.
38 Dealing with "the freak of chance" in research results: Secondly, the ability
to extend a research question's answer from a sample to a population depends on
whether the research was designed to answer the question. Some research findings
in the sample occur through "the freak of chance". The fisherman who returns from
a fishing trip not with fish but with boots and claims that he was "fishing for
boots all along" must be suspected of not meeting his true aim of catching fish.
So, too, is the researcher who claims he has an important answer from his sample
when he never intended to ask that research question. He just happened to find the
result in his sample. This is an important problem in research interpretation
given the tendency among many researchers and their advocates to "analyze
everything, and report what is favorable to our belief. This regrettably
unstructured tendency chokes the research information stream by mixing the few,
good, prospectively asked research questions with those, which were not asked
prospectively. Since the research team did not plan these additional analyses when
they were designing their investigation, these findings often contain many red
herrings and false leads.[9] They are instead "tacked on" during the experiment or
at the end of the experiment after everyone has seen what the data show.
39 First Tier: Prospectively asked questions: The established way to avoid this
difficult problem is to insist that research results be provided at two levels.10
The highest level is occupied by those questions asked by the investigator before
any data is obtained. These questions are often small in number, well considered,
and, more specifically, had the research design sculptured to provide answers to
them. The maxim "first say what you plan to do, then do what you said leads to the
clearest extension of the research's results from the small sample to the large
population.
40 Second Tier: Subgroup Analysis: The second tier questions are merely
exploratory questions, or hypothesis-generating questions[11]. The definitive
answers provided to this second group of questions must wait until the next
research effort because the research that spawned them was not designed to answer
them. Subgroup analysis falls into this second tier, or exploratory analysis for
very simple reasons. First, it is difficult to identify whether a patient falls
into a subgroup or not unless some prospective criteria are defined In addition,
since subgroups are only part of the sample (for example only males, or only
Hispanics), the number of patients in the subgroup can be small. The total sample
itself is typically barely large enough to justify extending results to the
population at large. Subgroup analysis reduces the number of subjects even
further. One can imagine taking a subgroup of only Hispanics, then only Hispanic
males, then only Hispanic males <45 years old, then Hispanic males < 45
years old with weight> 200 lbs. The subgroup gets smaller and smaller, with
each data "cut", making the extension of the risk factor - disease association to
the population even more hazardous. A classic example of the misdirection
subgroups provide is from the MRFTT [12]experience, in which it was found that
some men had better health outcomes with their hypertension left untreated and
uncontrolled! This spurious result of subgroup analyses was transformed[FNc] into
a stupendous "finding", sidetracking the treatment of hypertension in the 1970's.
The extensions of findings from a sample to the population are hazardous enough
without trying to extend the findings of smaller subgroups. Subgroup analyses,
like "fool's gold", should be viewed with a skeptical eye, and not be accepted
without independent confirmation from a second study[13,14].
FNc. For example does duration of exposure mean continuous exposure or
intermitent exposure -the determination is clearest if the definition is
made up front.
41 Fallacy of relying on subgroup analysis. Recognizing the difficulties imposed
by subgroup analyses, it is well recognized in leading textbooks in the field of
biostatistics that the most reliable analysis in a study is the primary analysis
of the study. In the overwhelming majority of research, this is the finding in the
overall cohort. This overall finding is the most reliable estimate of the
relationship between the exposure and the disease in any of the subgroups.
Applying this fundamental principle to the IPPHS[4] finding, we would conclude
that the odds ratio of 6.3, reflecting a 6 fold increase in the occurrence of PPH
when exposed to diet drugs applies to each of the subgroups of the IPPHS study.
Therefore, the best estimate of the odds ratio that characterizes the relationship
between fenfluramine and PPH in the subgroup that is comprised of patients of
short duration is 6.3. Similarly, the best estimate of the odds ratio of the
subgroup that comprises the relationship between the fenfluramine use and PPH in
patients who had a remote history of fenfluramine exposure is 6.3. This is
consistent with the characterization of the relationship between diet drugs and
the occurrence of PPH in the overall cohort of PPHS. This does not mean that there
is no duration-response relationship between the fenfluramine use and PPH. It
means that the Abenhaim study cannot illuminate that relationship sufficiently
because it was not designed to do so.
42 The guiding principles in drawing conclusions from research analyses are to (1)
focus on peer-reviewed published manuscripts to the exclusion of all others and
(2) in those peer- reviewed manuscripts, focus on the answers to the questions for
which the research was designed, judging everything else in an exploratory
light[15j. This principle also supports the application of the odds ratio of 6.3
in the overall cohort to each of the subgroups in IPPHS.
43 Study designs. There are a number of different types of study designs used by
an epidemiologist: non-experimental or observational studies and experimental
studies. The major observational studies are:
44 Ecological studies: Ecological studies are useful in determining relationships
between exposures and diseases in a group of people such as those living in a
country or other geopolitical units. We look at the average values of exposure in
a given country and the incidence (number of new cases) of PPH. When Aminorex was
first introduced in Europe in the 1960s, a Swiss physician, Gurtner noted a
dramatic and unexplained increase in the incidence rate of PPH - by almost 20 fold
- from the "normal" rate for this rare disease. Similar increases were soon
detected in Austria and Germany. The first case was noted 6 12 months after
Aminorex was introduced to the market. When the drug was withdrawn from the
marketplace, the rates of PPH decreased to prior levels within three years. The
medication was never reintroduced. These data were reported in a peer-reviewed
journal in 1968.[7]
45 Case Reports and Case Series: Case reports and case series are useful for
alerting clinicians to a possible health problem. In this sense, they are
hypothesis generators. Case reports are the most common types of studies published
in medical journals. These case reports can announce the occurrence of previously
unsuspected adverse drug events such as those reported to the FDA concerning
fenfluramine and dexfenfluramine. The individual case report case be expanded to a
case series that describes characteristics of a number of patients with a given
disease. Case reports and clinical series have important historical importance in
epidemiology, as they are often used as an early means to identify the beginning
or presence of an epidemic.
46 Cross-sectional studies: Cross-sectional studies look at the status of an
individual with respect to the presence or absence of both exposure and disease
assessed at the same point in time. Cross-sectional or prevalence studies are
often used as screening and classification analyses which are followed by
follow-up studies. For example, before the incidence of heart disease in a
community is studied, the population must be evaluated for current cardiac status
and sorted by characteristics such as blood pressure, smoking and cholesterol
level. For example, in a cross sectional study, the prevalence of heart valve
disease was described by five independent echo surveys reporting to the FDA of a
30% prevalence of valvulopathy in patients who had been exposed to anorexigens
compared to the usual 1.2- 3.6%.[16] Since cross-sectional surveys must consider
prevalent (already existing in the population) rather than incident (new) cases,
the data obtained will always reflect determinations of survival as well as
etiology. For instance, if all patients with valvulopathy are gathered at one
point in time and asked about their previous use of anorexigens, patients who may
have used anorexigens but have since died would not be counted and the true
prevalence of valvulopathy in diet drug users would be under-represented.
47 Case-control studies: Case-control studies select participants based on whether
they do (cases) or do not (controls) have a particular disease under study. Cases
(diseased individuals) are identified and matched to a similar but non-diseased
group. Then, previous exposure history is ascertained by chart review, personal
interview, laboratory values, or family reports. The proportions with the exposure
of interest in each group are compared. Because of this type of study design,
case-control studies offer a number of advantages for evaluating the association
between exposure and disease. The case control study is relatively inexpensive and
quick, compared to other types of analytic designs, and can examine the role of
multiple risk factors for a single disease. This study design is especially useful
in studying diseases with long latency periods including most chronic diseases. A
less well publicized but compelling point in favor of the case-control study
design is an ethical one, since clinical trial or follow-up observational studies
cannot study some exposure-disease relationships. For example, the relationship
between the fenfluramines and pulmonary hypertension cannot be studied by an
experimental study design since the anorexigens fenfluramine and dexfenfluramine
have been withdrawn from the marketplace. Additionally, observing patients who had
been exposed to these anorexigens for a prolonged period, simply to determine who
will develop primary pulmonary hypertension would be unethical.
48 Cohort Studies: A cohort study design identifies a group of individuals based
on presence or absence of exposure to a suspected risk factor and then follows the
individuals over a period of time to determine who develops the health outcome of
interest. This study design is ideal when exposure is rare or when multiple
effects of the exposure are of interest. Since disease status is measured at the
onset of the study, incident rates can be calculated (number of new cases of
disease).
49 Randomized clinical trials: Randomized clinical trials (RCT) are experimental
studies rather than observational studies, meaning that the investigator, rather
than the individual, decides who will become exposed to an intervention, such as a
drug. The participants are followed for a predetermined period to ascertain who
develops the outcome of interest, such as weight loss. Weintraub[17] and
Guy-Grand[18] conducted long-term studies on anorexigens and weight loss that
showed that participants who took diet drug medications lost weight. However, when
they stopped taking these medications, they regained weight. Clinical trials are
often considered the "gold standard". They can control all extraneous factors by
using randomization to treatment or non-treatment groups. However, they are
plagued by attrition and often do not reflect actual usage of the drug under study
or the same patient population to which the drug will be marketed. Therefore,
there can be problems in generalizing the results. As mentioned previously, a less
well publicized but compelling point in favor of the observational cohort or
case-control study designs over a clinical trial is an ethical one. Although
experimental studies are the ideal study design, they can only be employed
infrequently in studies of disease etiology for ethical reasons. Therefore, it is
usually necessary on issues such as whether diet drugs cause PPH or valvular heart
disease to draw conclusions from observational studies.
Risk Benefit
50 Definition: Medication risk/benefit is the process by which the clinical
benefits of a medication are weighted against the risks and hazards associated
with that medication's use. Risk benefit evaluations are not based on economics
and are, therefore, different from cost-effectiveness analysis. The following
comments focus on a risk-benefit analysis for fenfluramine.
51 Purpose of fenfluramines: The fenfluramines were introduced as pharmacological
measures to reduce obesity. The effects of obesity are profound; excess weight is
associated with an increased risk of hypertension, hypercholesterolemia, diabetes,
and atherosclerotic cardiovascular disease.
52 Obesity Treatment: Obesity is notoriously difficult to treat, with many
patients expressing frustration with the interventions of diet restriction and
regular exercise. The use of pharmacological compounds such, as the fenfluramines
have been associated with weight loss. However, since obesity is a chronic
disease, its treatment will be chronic as well, and, therefore, any effective
treatment must demonstrate both long term effectiveness and long term safety.
Thus, a balanced appraisal of the fenfluramines must include 1) an examination of
the pattern of weight reduction they produce and 2) an explicit consideration of
the health risks with which the fenfluramines are associated.
53 Fallacies: Although it is widely accepted that weight gain is associated with
clinical problems, it is fallacious in general, to reason that a compound
demonstrates overall benefit if it produces weight loss. The difficulty with this
simplistic, attractive logic is that the medication may have more than one effect.
If the other effects are hazardous, these hazards can offset the benefit of the
weight loss. This is the fallacy of the surrogate endpoint.
54 Surrogate Endpoint Definition: A surrogate endpoint is an intermediate endpoint
that is itself associated with more important clinical events. For example, weight
reduction in the morbidly obese is a surrogate endpoint. Although obesity is
associated with morbidity, we cannot assume that producing weight reduction is
sufficient to avoid long-term morbidity. It is possible that a medication could
reduce weight, while simultaneously producing other problems that may worsen the
clinical disease it was meant to diminish. Thus, a medication's ability to produce
an effect on a surrogate endpoint is relatively meaningless if that medication
causes an increase in the more distant, more important clinical event to which the
surrogate is related. The demonstration of the harmful effect of treating mild
cardiac rhythm problems demonstrates the difficulty of the surrogate endpoint
approach. Investigators using new therapy to reduce the occurrence of mild cardiac
arrhythmias were surprised to learn that the therapy did, in fact, treat the mild
rhythm problem, but induced an arrhythmia that was far worse[19].
55 Fenfluramines' absence of efficacy: It is accepted that fenfluramine use is
associated with some weight loss[17] In the treatment of chronic obesity, however,
weight loss must be sustained. The researcher Guy-Grand[18] studied the effect of
chronic administration of dexfenfluramine in obese subjects. The weight loss was
not progressive after six months of dexfenfluramine administration.
56 Transient weight loss: Does the effect of fenfluramine on weight persist after
the drug is discontinued? This is important since the treatment of chronic obesity
most often requires chronic therapy. If the weight loss effect of dexfenfluramine
persisted after its discontinuation, it would be possible for the patient to
receive a long-term weight loss benefit from a short-term administration of the
drug. This is not the case. The effect of withdrawal and re-introduction of
dexfenfluramine in the treatment of obesity was studied by Ditschuneit[20]. In
this study, 25 patients who had completed the INDEX trial [18] were reexamined two
months after withdrawal of the study medication. The observed weight gain after
withdrawal of study medication was higher in patients treated beforehand with
dexfenfluramine (5.7 +- 2.64 lbs.) than in patients who had received the placebo
(1.76 +- 2.20 lbs.).
57 Purported Benefits: The purported benefit of fenfluramine exposure is weight
loss, i.e., a change in the surrogate endpoint. However, this weight loss is
modest, and the information available demonstrates a weight plateau is reached in
less than one year. Also, discontinuation of fenfluramine leads to rebound weight
gain, so patients must be continually exposed to the medication to receive its
benefit, a benefit which Ditschuneit's work suggests decreases over time. Thus,
the use of fenfluramine produces modest, short- term weight loss with return to
pretreatment weights upon its discontinuation.
58 Absence of Efficacy Evidence: It is critical to note that there have never been
any fenfluramine or dexfenfluramine studies proving a reduction in long-term
morbidity or mortality. The treatment of a surrogate endpoint, e.g., weight
reduction, should be followed by the demonstration and not the presumption that no
harm will befall the patient due to unknown effects of the drug. However, such a
study would, in all likelihood, be impossible to carry out for the fenfluramines.
For one reason, the required study would be huge, involving large numbers of
patients followed for a long period of time. This would be required to insure the
collection of enough long-term clinical events (e.g., diabetes,
hypercholesterolemia, cause specific mortality) to allow researchers to identify
an effect with adequate control of statistical errors. Second, there are ethical
considerations. There is a growing body of evidence and literature, which, in
total, demonstrate that the fenfluramines cause cardiovascular and pulmonary
disease.
59 Amphetamines: It is important to note that experts in the pharmacologic basis
of therapeutics have classified the fenfluramines as amphetamines[21].
60 The use of agents known to increase the risk of primary pulmonary hypertension
and cardiac valve disease, in the face of alternative therapies (dietary
discipline in combination with moderate, balanced exercise) that do not increase
these risks, would, in the eyes of many in the scientific community, be unethical.
61 FDA Comments: Dr. Michael A. Friedman of the FDA stated, "The data we have
obtained indicate that fenfluramine, and the chemically closely related
dexfenfluramine, present an unacceptable risk at this time to patients who take
them" (HHS News September 15, 1997).
62 Chronicity of obesity: Obesity is a chronic disease requiring chronic
interventions. The treatment of this chronic disease requires the demonstration of
long-term efficacy and long- term safety, i.e., a positive risk benefit balance.
The fenfluramines produce only a short- term weight loss, which is reversed upon
its discontinuation, i.e. a self-limited benefit. In addition, the fenfluramines
are associated with the occurrence of primary pulmonary hypertension and
valvulopathy. We do not know the long-term effects of fenfluramines. This would
require a long-term clinical study that would take years to design, execute and
analyze. However, I believe that a balanced assessment of the risks and benefits
of fenfluramines as treatment for obesity, with the available evidence,
demonstrates that the risks of the fenfluramines for weight reduction outweigh the
benefits.
Fenfluramines and Primary Pulmonary Hypertension (PPH)
63 Chronologic Summary: The chronological examination of the relationship between
exposure to the fenfluramines and primary pulmonary hypertension (PPH) is a fine
example of the evolution of scientific thought from observation to hypothesis
generation to hypothesis confirmation. From 1968-1972, an epidemic of primary
pulmonary hypertension was observed in Europe, associated with diet drugs. These
reports were followed by observations describing PPH in young women taking
fenfluramine for weight reduction, published in 1981 and 1982.[6] Others
followed[22], and Brenot and coworkers in 1993[5] provided the first retrospective
analysis that linked PPH with fenfluramine use. These researchers showed that 20%
of their patients with PPH had taken fenfluramine; these patients survival times
were as poor as those who have PPH without exposure to these drugs. With the
increased consumption of dexfenfluramine in Europe, case reports also appeared
linking it to PPH[23,24,25]
64 Brenot[5]was amazingly prescient given its early appearance, and therefore, its
observations require careful attention. In this manuscript, Brenot states that 20%
of their patients with PPH had in fact taken fenfluramine; these patients
experienced a survival rate as poor as that of patients who have PPH from other
causes. He also noted that there was no relationship between the duration of the
use of appetite suppressants and the degree of pulmonary hypertension, implying
that short durations of exposure may also be associated with the occurrence of
PPH. In fact, Brenot stated that fenfluramine may not only precipitate PPH, but
might also hasten the course or primary pulmonary hypertension.
65 Latency. Brenot recognized that not every patient exposed to fenfluramine who
developed elevated pulmonary arterial pressures died. Taking advantage of this
observation, he chose to evaluate whether patients exposed to fenfluramine
developed difficulty later in life from pulmonary disease related to the use of
fenfluramine. In Brenot's evaluation, half of the survivors were restudied several
years after initial diagnosis. It is important to note that each of these
survivors had been exposed to fenfluramine, and then had the fenfluramine
discontinued. These survivors with essentially normal pulmonary vascular pressures
were then tested several years later.
66 Danger after discontinuation: Remarkably, Brenot noted that, upon testing years
after the exposure to fenfluramine was discontinued, each of these same patients
demonstrated abnormal responses during exercise catheterization. To Brenot, this
suggested that the underlying lung injury, induced by fenfluramine, persisted in a
subclinical state as underlying pulmonary arteriolar disease for years. Even
though the diet drug had been discontinued, and even after the patients had normal
responses to pulmonary vascular testing after the drug was discontinued, exercise
testing years later demonstrated that a vascular injury remained.
67 This important finding begged the question as to whether early exposure to and
discontinuation of fenfluramine could produce PPH years later, i.e. could there be
a latency period, beginning during fenfluramine exposure and continuing after the
end fenfluramine exposure which induces the occurrence of PPH years after
fenfluramine exposure ceased. Brenot's careful examination of this patient cohort
suggests an answer. Fifteen patients in Brenot's study had taken fenfluramine and
then developed PPH. However, six of these fifteen patients were not taking the
drug when the diagnosis of PPH was made - in these patients, the drug had been
discontinued well before the symptoms of PPH appeared. In four of these six
patients, the symptoms of PPH appeared 1-11 months after exposure to fenfluramine
was discontinued. This is critical. Patients who had been exposed to fenfluramine
and then had the drug discontinued developed PPH up to a year after they stopped
taking the fenfluramine. In one other patient, PPH symptoms appeared three years
after exposure to fenfluramine was discontinued. Another patient lived for over
twenty years after being exposed to fenfluramine before symptoms of PPH occurred.
Although by definition, consideration of a latency period requires the observer to
consider alternative intervening causes for the disease, the rarity and
specificity of PPH makes it unlikely that there is a cause other then fenfluramine
for this patient's illness. This manuscript, which appeared in 1993 provided
specific data that bolsters the argument that early fenfluramine exposure can
cause the late occurrence of pulmonary hypertension years after the medication has
been discontinued. This is the solid evidence that there can be a prolonged
latency period between the cessation of diet drugs e.g. fenfluramine and the
occurrence of PPH.
68 Unfortunately, with the increased consumption of dexfenfluramine in Europe,
case reports additional reports appeared, further cementing the fenfluramine-PPH
relationship.
69 IPPHS. If the only cause of PPH was anorexigen exposure, the case series of
Brenot would have been sufficient to demonstrate that fenfluramines caused PPH.
However, since PPH has more than one cause, a study must be done which has two
groups, one exposed to fenfluramines while the other is not, in order to compare
the frequency of occurrence of PPH in these two groups. This was the IPPHS study.
This multi-national prospective case- control study was published in an
authoritative medical journal and has been subject to peer-reviewed.
70 Acceptance: The scientific community generally accepts the International
Primary Pulmonary Hypertension Study's (IPPHS)[4] results as true, reliable, and
valid.
71 The purpose of this study was to assess the incidence of PPH and investigate
the causative roles of various suspected risk factors, especially anorexic agents.
This study included 95 patients with PPH matched to 355 controls. Dexfenfluramine
and fenfluramine constituted 90% of all anorexigens used by the PPH patients, a
strong presence which defeats the claim that the fenfluramines were not adequately
represented.
72 Conclusions: The IPPHS study concluded that the risk of PPH was 6.3 times
higher in patients who were exposed to anorexigens (primarily fenfluramine and
dexfenfluramine) for all definite users of these anorexigens compared to
non-anorexigen users. Evaluations of the risk of PPH in patients exposed to
anorexigens for less than three months, or greater than three months are of
interest. The data provided in the manuscript on these ancillary findings are of
interest. However, since these subgroup analysis on the patients exposed to
anorexigens for less than three months are unreliable, we cannot rely on these
assessments that affect only a fraction of the overall cohort. Fortunately, modern
epidemiological thought illuminates a useful path for us to follow, and the peer
reviewed, published manuscript of Yusuf[13] tells us how to proceed. When
extending the findings from a sample to a population, the safest interpretation of
the subgroup is to assume the finding for the entire sample applies to the
subgroup. This is a critical point. If the researcher is not interested in
extending sample results to the population (i.e. the researcher is only interested
in the sample) this rule is not invoked. However, as soon as the researchers
attempt to generalize their findings (i.e. as soon as the researchers attempt to
extend their subgroup results from the sample to the larger population) Yusuf's
rule should be invoked. This well established thought process dictates that in the
large population, the fenfluramines are associated with PPH, regardless of the
duration of exposure. Therefore the best evidence from Abenhaim is that exposure
to the fenfluamines for less than three months, and exposures greater than three
months, are associated with the occurrence of primary pulmonary hypertension. No
exposure is safe.
73 Causation and IPPHS: The IPPHS, in concert with previous published work,
provides important evidence that there is a causal link between fenfluramine and
dexfenfluramine and PPH. PPH, however the diagnosis is made, (via the older
technique of catheterization or the new noninvasive techniques of
echocardiography) was caused by exposure to fenfluramine. Dexfenfluramine and
fenfluramine were the most commonly used anorexigens: 22 patients (23.2 percent)
and 23 controls (6.5 percent) had used at least one of them. Of these subjects, 16
patients (16.8 percent) and 18 controls (5.1 percent) reported not using any other
anorexic drug.
74 The IPPHS study design was based upon generally accepted statistical and
epidemiological methods. Specifically, the study did provide a well-distributed
cohort. Patients were accepted from four countries in Europe (France, Belgium, the
United Kingdom, and the Netherlands). Such a sample provides justification for
generalizing the findings in the research effort to the public at large.
75 Statistical Analysis: Finally, the statistical analysis used by IPPHS was a
standard, generally accepted method and is reliable. In addition, the statistical
technique of adjustment was used appropriately. Adjustment is simply a way to
remove the effect of variables, which are associated with the risk factor and or
the disease (e.g. obesity, which is associated with fenfluramine use and pulmonary
disease). In this example, adjustment is basically identifying and isolating the
relationship between obesity and fenfluramine use, and obesity and PPH, removing
these relationships, and examining the relationship between obesity adjusted
fenfluramine use and obesity adjusted PPH.
76 Sufficiency: One can always ask for yet one more study to be completed.
However, every new study has its own limitations, which will be the justification
for yet an additional study. Primary pulmonary hypertension is a serious disease
with long-term morbidity and survival implications. An objective view of the data
and evaluation of the weight of evidence indicates that there is a relationship
between dexfenfluramine/fenfluramine use and primary pulmonary hypertension and
that this relationship is causal.
77 Result of Tool Application Applying these tools to the available scientific
literature, it is clear that exposure to the fenfluramines causes primary
pulmonary hypertension. Wyeth demonstrated consistent and willful malfeasance in
its obligation to notify both the medical community and the Food and Drug
Administration (F.D.A.) of the association between use of the fenfluramines and
pulmonary hypertension, and use of the fenfluramines and valvular heart disease.
In violation of the code of federal regulations, Wyeth did not provide timely and
accurate information to the medical community when they learned of the association
between fenfluramine and PPH, waiting years after they had evidence of this
association before they notified the medical community of the relationship. In
addition, Wyeth withheld requested and vital information from the repeated verbal
queries from the Advisory Committee to the F.D.A. during the discussion of the
approval of Redux. Wyeth undermined the approval process of Redux, blunted the
warning to the medical community about the damage to hearts and lungs caused by
fenfluraine and subjugated the public health for profit.
Conclusions
78 To a reasonable degree of scientific and medical certainty, the fenfluramines
cause primary pulmonary hypertension.
79 Fenfluramines are not effective in producing health improving weight loss in
the obese
80 The overwhelming epidemiologic evidence, using state of the art scientific
methodology, concludes that the fenfluramines cause primary pulmonary
hypertension.
81 Fenfluramines cause primary pulmonary hypertension even for short durations of
exposure (less than three months).
82 PPH can be induced by the fenfluramines months or years after the fenfluramines
have been discontinued.
83 The risks of use of fenfluramines exceed the benefits.
84 Wyeth knew that fenfluramines caused primary pulmonary hypertension.
85 Wyeth did not keep the FDA, physicians, or patients apprised of the risks
associated with the fenfluramines.
86 Wyeth mislead the FDA Advisory Committee about the risks associated with Redux
and Pondimin.
87 The labels for Redux and Pondimin are both defective. In fact no label could be
written to describe the safe usage of these compound.
88 Redux and Pondimin are defective products.
Appetite-Suppressant Drugs and the Risk of Valvulopathy
89 Introduction: The role of science is to begin with a multitudinous collection
of data, itself teeming with associations, and distill this down to a compact body
of knowledge, governed by a few master principles. This is what the combination of
the Connolly[26], Jick[27], Khan[28], and Weissman[29] studies do. They represent
a progression of the knowledge gained from the first published association of the
fenfluramine exposure and valvulopathy and have refined it. Each has the important
feature of identifying a different design.
Background rate of valvulopathy
90 Importance: Before serious discussion on the role of fenfluramines in inducing
valvulopathy (abnormal changes in heart valves) can commence, a clear
understanding of the respective roles of prevalence, incidence and background
rates is required. These are straightforward concepts. The incidence rate is a
quotient, measuring the rate of new cases of the disease. It is always for a
specified time period. In the incidence computation, the numerator is the number
of patients who have the disease for the first time during this time period. The
denominator is the number of patients at risk for the disease during this time
period.
91 Definition: The background rate is, as the descriptor implies, the rate of
disease which is not new, but has been existing in the population for a prolonged,
perhaps unknown period of time. The prevalence of the disease is the sum of the
background rate and the incidence rate, i.e. the quantitative measure of the
disease's presence in the population regardless of whether the disease is newly
occurring in the population or not.[FNd] The concepts reflected in this
nomenclature have important research design implications.[FNe]
FNd. As a hypothetical example, consider a community of 1000 people, 80 of
which are known to have diabetes for several years. During the year 2000, of
the remaining 920 people at risk, 5 are newly diagnosed to have diabetes.
The background rate of diabetes in this community is 80/1000 = 0.08. The
incidence rate for the year 2000 =5/920= 0.005. The prevalence of diabetes
in the community is 0.08.005 = 0.085.
FNe. One of the concepts which considerable implications for the design of a
research effort attempting to relate the exposure to a disease is how to
measure the disease during the exposure. Clearly a new exposure is not going
to effect the background rate of a disease. The new exposure will effect the
incidence rate. Therefore, the clearest measure of the effect of exposure of
the occurrence of a disease is to compare the incidence rate of the disease
of the exposed to the incidence rate of the unexposed. This creates an
incidence ratio, the best measure of the impact of disease on exposure.
Prevalence ratios and odds ratios are used when incidence ratios are
unavailable, but they are not as direct an assessment of the effect of the
exposure as are incidence ratios.
92 Measurement: To begin to understand the role of the fenfluramines in inducing
valvulopathy, we begin with the recognition that there is a nontrivial background
rate[FNf] for valvular heart disease. Singh was able to measure the background
rate of valvulopathy in the Framingham study[30], which is a prospective
epidemiologic study established in 1948 to evaluate potential risk factors for
coronary heart disease. Singh identified the background proportions of patients
with valvular heart disease[FNg].
FNf. Some diseases have no background rate. One example is pancreatic cancer
which kills its patients so rapidly (usually within six months of the
diagnosis) that cases cannot be accrued over time. Another example is the
occurrence of a disease which is new and unique to the exposure. A fine
example of the latter is radiation poisoning. With no exposure to radiation,
there is no iliness. In these rare circumstances the background rate is
zero, and the prevalence rate is equal to the incidence rate.
FNg. Valvular heart disease is defined using the FDA criteria. Aortic
regurgitation is defined as regurgitation in the aortic valve which is
either mild, moderate or severe in degree. Mitral regurgitation is defined
as regurgitation in the mitral valve which is either moderate or severe.
TABLE
93 Observations: There are three observations that we can make from this table
which are of direct relevance to this affidavit. The first is that the proportion
of patients with background valvulopathy is not zero, and certainly not
insignificant. The second is that the proportion of patients with background
cardiac valvulopathy depends on the age and gender of the patients under
consideration. For example, men and women in their twenties and thirties would not
be expected to have (FDA criteria) valvulopathy at all. On the other hand, more
than 16% of women over 70 years of age have (FDA criteria) aortic regurgitation.
Therefore, if a group of patients is found to have (FDA criteria) mitral valve
disease or (FDA criteria) aortic valvular disease, their age and gender must be
considered before the level of disease can be deemed excessive. If a research
effort has its own control group then the rates of patients with valvulopathy in
the exposed group can be compared to the rates of patients in the unexposed group
directly - the unexposed group will have valvulopathy at the background rates.
However, if the research effort does not have a control group, the investigators
in that group will often search for an "external control group". If that external
control group is to be the Framingham analysis of Dr. Singh, then the
comparison[FNh] must take into account the age and gender distribution of the
research group, and compare that to that of Dr. Singh's. If the two are not the
same, then age and gender adjusted valvulopathy rates must be produced.
FNh. In the context of this report, that would be patients who are unexposed
to the fenfluramines.
94 Case Series of Connolly:
The first published evidence, case series of Connolly[26] Earlier this affidavit
discussed the evolution of scientific inquiry from observation to hypothesis
testing to conclusions and implications. The first examination of a relationship
between an exposure and disease often begins with the simplest of scientific tools
- observation. In 1997, Connolly reported a cluster of patients who clearly
demonstrated cardiac valvulopathy after a course of fenfluramine. At times, case
series have been sufficient to demonstrate a causal link between a risk factor and
a disease.[FNi] Connolly's article, by itself, did not prove that exposure to
fenfluramine caused heart valve disease. This is because heart valve disease has
other causes. However, Connolly's article served the indispensable purpose of
"sounding the alarm" which was subsequently confirmed by others. It is regrettable
that Wyeth-Ayerst did not themselves sound the alarm years earlier.
FNi. Two recent infamous examples were provided earlier in this affidavit,
1) the link between Thalidomide ingestion and congenital defects1 and 2) the
link between the combined use of Seldane and the antibiotic Ketoconazole and
the occurrence of fatal heart rhythms.2
95 The Epidemiological Study of Khan
The epidemiological study executed by Khan et. aL[28] provided important new
evidence for a causative link between anorexigen ingestion and cardiac
valvulopathy. Using well established and accepted methods of epidemiological
research, these researchers examined patients who had taken dexfenfluramine alone,
dexfenfluramine and phentermine, or fenfluramine and phentermine for various
periods. This methodology is consistent with the careful choices of cases and
controls, judicious use of matching, and adjustment of effect size measures. The
association between the use of any appetite suppressant and cardiac- valve
abnormalities was analyzed in a final matched group of 233 pairs of patients and
controls. The odds ratio for such cardiac-valve abnormalities was 12.7 (95%
confidence interval, 2.9 to 56.4) with the use of dexfenfluramine alone. This
means that valvulopathy patients were 12.7 times more likely than controls to have
been exposed to Dexfenfluramine and the results in the population from which this
sample came could be as low as 2.9 or as high as 56.4. The odds ratio was 24.5
(5,9 to 102.2) with the use of dexfenfluramine and phentermine and 26.3 (7.9 to
87.1) with the use of fenfluramine and phentermine. This work has generated the
criticism that bias has not been ruled out as a possible explanation of the
association between fenfluramine or dexfenfluramine and cardiac valvular
insufficiency. However, the striking magnitude of the odds ratio makes is
extremely unlikely that bias remaining after his covariate-adjusted analysis would
substantially reduce these findings.
96 Prospective endpoint choice. In addition, Khan should be applauded, not
criticized, for staying with the analysis of his prospectively-chosen primary
endpoint (cardiac valvulopathy). He recognized that the size of his sample was too
small to extend sample findings to the population for the relationship between
fenfluramines and valve disease for individual valves (e.g. the mitral valve, or
the tricuspid valve). He had to use what is termed a combined endpoint. It is
notable that Khan did not report the relationship between anorexigen exposure and
aortic valvulopathy separately from the relationship between anorexigen exposure
and mitral valvulopathy. The manuscript reports only the relationship for the
combined endpoint. It is useful to ask, why use the combined endpoint (HVD) and
not its individual endpoints (aortic valve component and mitral valve component).
The answer is prevalence.
97 Combined Endpoints: Combined Endpoint Rationale: It is a truism that, in order
to identify a meaningful relationship between exposure and a clinical event, the
investigator must identify enough clinical events in his sample to be able to draw
a conclusion. For example, a study attempting to relate anorexigen use to cardiac
valve disease would be unpersuasive if there were only two cases of valve disease
in the sample. Even if those two cases occurred in the exposed group, the
investigator would not be able to persuasively argue that conclusions drawn on
just two cases can be extended to the population.
98 Combined Endpoint Definition: Investigators recognize this problem and
therefore decide before the execution of the research program to build a combined
endpoint up from individual components, each of which may have low prevalence. A
combined endpoint includes patients who meet not just one criterion for disease,
but patients who meet any one of the different criteria. The advantage of the
combined endpoint is that it provides an increased prevalence of the endpoint of
measure, making it easier to discern differences in its prevalence between exposed
and unexposed patients. In order for the combined endpoint to make sense, its
individual components must each be plausibly linked to exposure. For example, an
unreasonable combined endpoint would be the occurrence of either valvular heart
disease or patient injury in an automobile accident. Although this combined,
endpoint would have high prevalence and would be dysfunctional since anorexigen
exposure is not thought to be linked with automobile accidents.
99 Implications of Combined Endpoint: A foreseeable implication of the use of
combined endpoints is that decisions made for each component of the combined
endpoint are made only with difficulty. After all, if one could draw
straightforward conclusions about the individual components (tricuspid or mitral
regurgitation) of the endpoint, the combined endpoint (HVD) would not have been
created. The fact that a combined endpoint is chosen, essentially means that the
low prevalence of its components precludes any possibility of a meaningful
inference based on the findings for that component alone. Therefore, in an
intelligently chosen combined endpoint, the findings for the combined endpoint are
extended to its individual components.
100 Standard practice. A recent example is the CARE study[31]. The investigators
examined the role of cholesterol reduction therapy in 4,159 patients. Patients
were randomized and followed for five years. The combined endpoint was fatal heart
attack or nonfatal heart attack, i.e. a patient was considered a "case" if either
they had died of a heart attack, or they survived but had an attack. The p value
for the combined endpoint in this randomized study was 0.003 with a risk reduction
of 24% (95% CI 9% to 36%). However, if we look at one of the components of the
combined endpoint, examining the evidence for benefit, we see that, for example,
for the fatal component of this endpoint (fatal heart attack), the p value is 0.10
and the risk reduction is 20% (95% CI -5 to 39). Is it reasonable to conclude that
the drug does not reduce fatal heart attacks? Certainly not. It's much more likely
that the investigators were simply unable to infer a difference in the fatal event
rate in the population, based on the very small number of fatal heart attacks in
the sample.
Comments