{Jobs}0688jw/makeup/688ch34.

3d

34
The Use of Pharmacoepidemiology to
Study Beneficial Drug Effects
BRIAN L. STROM
Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania School of Medicine,
Philadelphia, PA, USA
KENNETH L. MELMON
Department of Medicine, Stanford University School of Medicine, Stanford, CA, USA

INTRODUCTION DEFINITIONS

In order to be approved for marketing in the There are at least four different types of measur-
United States, drugs must be proven to be safe able drug effect of interest to a prescriber.
and effective using ``adequate and well-controlled Unanticipated harmful effects are the unwanted
investigations.'' Earlier chapters in this book have effects of drugs that could not have been predicted
shown that this premarketing information often is on the basis of their preclinical pharmacologic
insufficient to provide some of the information profile or the results of premarketing clinical
about drug toxicity which is clinically most studies. These effects are most often Type B
important. The same applies to information about adverse reactions, as defined in Chapter 1. For
drug efficacy. example, chloramphenicol was not known to cause
In this chapter we will begin by clarifying the aplastic anemia at the time it was marketed,1 nor
different definitions of various types of beneficial was the skeletal muscle pain associated with use of
drug effect. Then we will discuss the need for HMG-CoA reductase inhibitors. A major research
postmarketing studies of drug effectiveness. Next, challenge is to discover medically important
we will present the unique methodologic problems unanticipated harmful effects as soon as possible
raised by studies of beneficial drug effects, as well after drug marketing. Quantitation of the inci-
as potential solutions to these problems. Finally dence of these effects is medically useful as well.
we will evaluate the frequency with which these Anticipated harmful effects are unwanted effects
proposed solutions might be successful. Specific of drugs that could have been predicted on the
examples of approaches to the study of efficacy basis of preclinical and premarketing studies. They
also will be presented. can be either Type A reactions or Type B reactions

Pharmacoepidemiology, Third Edition. Edited by B. L. Strom.

# 2000 John Wiley & Sons, Ltd.
{Jobs}0688jw/makeup/688ch34.3d

554 PHARMACOEPIDEMIOLOGY

(see Chapter 1). One example is the syncope that Studies of drug effectiveness generally would best
sometimes occurs after patients take their first be conducted using nonexperimental study de-
dose of prazosin.2 Although this effect was known signs. However, these raise special methodologic
to occur at the time of marketing, a major question problems, which are discussed below.
remaining to be answered was how often the event Lastly, a study of efficiency is a study of whether
occurred. The dominant research challenge that a drug can bring about a desired effect at an
this type of drug effect presents is establishing its acceptable cost. This type of assessment falls in the
incidence. province of health economics, and is discussed in
Unanticipated beneficial effects are desirable Chapter 35.
effects of drugs that were not anticipated at the Note that the outcome variable for any of these
time of drug marketing. Although these effects studies can be of multiple different types. They can
may be medically useful, they are nevertheless side- be clinical outcomes (diseased=undiseased), or so-
effects, if they are not the purpose for which the called ``outcomes research,'' as defined by health
drug was given. An example of an unanticipated services researchers (see Chapter 39 for a discus-
beneficial effect is aspirin's ability to decrease the sion of the validity issues involved in measuring
probability of a subsequent myocardial infarction such outcomes); they can be measures of quality of
in patients who were given the drug for its life (see Chapter 36), often referred to in the
analgesic or anti-inflammatory action.3 Only pharmaceutical industry as ``outcomes research;''
recently has this been confirmed as a valid new they can be measures of utility, i.e., global
indication for the use of aspirin. A major research measures of the desirability of certain clinical
challenge is to discover this type of drug effect. For outcomes (see Chapters 35 and 36); they can be
example, it currently remains an open question economic outcomes (see Chapter 35); etc. Regard-
whether non-aspirin nonsteroidal anti-inflamma- less, the same methodologic issues apply to each.
tory drugs have the same beneficial effects.
Secondarily, it is useful to quantitate the frequency
of the event. CLINICAL PROBLEMS TO BE
Anticipated beneficial effects are the desirable ADDRESSED BY
effects that are known to be caused by the drug. PHARMACOEPIDEMIOLOGY
They represent the reason for prescribing the drug. RESEARCH
The study of anticipated beneficial effects has three
aspects. A study of drug efficacy is a study of In order to make optimal clinical decisions about
whether a drug has the ability to bring about the whether to use a drug, a prescriber needs to know
intended effect. In an ideal world, with perfect whether, and to what degree, the drug actually
compliance, no interactions with other drugs or is able to produce the intended effect (see
other diseases, etc., could the drug achieve its Table 34.1).4 Premarketing randomized clinical
intended effects? Drug efficacy usually is studied trials generally provide information on whether a
using a randomized clinical trial. drug can produce at least one beneficial effect.
In contrast, a study of drug effectiveness is a Specifically, premarketing studies generally inves-
study of whether, in the real world, a drug in fact tigate the efficacy of drug relative to a placebo,
achieves its desired effect. For example, a drug when both are used to treat a particular illness.
given in experimental conditions might be able to These premarketing studies of efficacy tend to be
lower blood pressure but, if it causes such severe conducted in very atypical clinical settings, com-
sedation that patients refuse to ingest it, it will not pared to those in which the drug ultimately will be
be effective. Thus an efficacious drug may lack used. Compliance during these studies is assured,
effectiveness. Studies of drug effectiveness usually and the patients included are similar to each other
are performed after a drug's efficacy has been in age and sex, do not have other diseases, and are
established. In contrast, if a drug is demonstrated not taking other drugs. Such restrictions maximize
to be effective, it also is obviously efficacious. the ability of premarketing studies to demonstrate
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 555

Table 34.1. Clinically important information about information available to the FDA at the time of its
intended beneficial effects of drugsa
regulatory decisions about the marketing and
1. Can the drug have the desired effect? labeling of the drugs involved in these uses.4 The
2. Does the drug actually achieve the desired effects comparison restricted itself to drugs approved
when used in practice? after 1962, when the Kefauver ±Harris Amend-
3. Can and does the drug have other beneficial effects, ments introduced a requirement for the submission
including long-term effects for the same indication?
4. Can the drug achieve these desired effects better than of data about drug efficacy prior to approval of
other alternative drugs available for the same the drug for marketing.
indication? Of the 100 common drug uses, 31 had not been
5. For each of the above, what is the magnitude of the approved by the FDA at the time of initial
effect in light of the many different factors in medical marketing, and 18 still had not been approved at
practice that might modify the effect, including
(a) variations in drug regimen: dose per unit time, the time of the comparison. Eight of the 18
distribution of dose over time, duration of unapproved uses were probably medically and
regimen, therapeutically inappropriate. For example, the
(b) characteristics of the indication: severity, use of antibiotics is not justified for the treatment
subcategories of the illness, changes over time, of viral infections, but such use was common.
and
(c) characteristics of the patient: age, sex, race, Other unapproved drug± indication pairs could
genetics, geographic location, diet, nutritional well have been quite appropriate, but the regula-
status, compliance, other illnesses, drugs taken for tory process does not need to and did not reflect
this or other illness (including tobacco and the current medical practice.
alcohol), etc. Of the 100 common drug uses, eight were based
a
Modified from reference 4. on the assumption that a drug had a particular
long-term effect, but only an intermediate effect
had been studied prior to marketing. For exam-
a drug's efficacy, if the drug actually is efficacious. ple, antihypertensive drugs are used for their
Additional information may then be needed on presumed ability to prevent long-term cardiovas-
whether, in the world of daily medical practice, the cular complications, but are approved for mar-
drug actually achieves the same beneficial effects keting on the basis of their ability to lower blood
and whether the drug can and does have other pressure. Five of the 100 common drug uses may
beneficial effects. In addition, at the time of have been for either the intermediate effect or the
marketing there may be little data on a drug's long-term effect of the drugs, but only the
efficacy relative to other medical or surgical intermediate effect was studied prior to market-
alternatives available for the same indication. ing. For example, hypoglycemic agents may be
Finally, a number of factors that are encountered used to control the symptoms of diabetes or to
in the practice of medicine can modify a drug's prevent the vascular complications of diabetes,
ability to achieve its beneficial effects. Included are but only the former were studied before drug
variations in the drug regimen, characteristics of marketing.
the indication for the drug, and characteristics of Drugs other than those in the list of 100
the patient receiving the drug, including demo- common uses were sometimes prescribed as treat-
graphic factors, nutritional status, the presence of ment for each of the 52 indications included in
concomitant illnesses, the ingestion of drugs, and those 100 uses. Yet, eight of the uses involved
so on. Many, if not most, of these factors that can drugs whose effects relative to alternative drugs
influence the effects of drugs are not fully explored had not been studied prior to marketing.
prior to marketing. The 100 common drug uses also included a
In order to quantitate the need for postmarket- number of examples of clinical factors that
ing studies of the beneficial effects of drugs, a are able to modify the effects of the drug, but
comparison was made of the 100 most common these were not discovered until after drug market-
drug uses of 1978 (drug ±indication pairs) to the ing. Some are listed in Table 34.2. In addition,
{Jobs}0688jw/makeup/688ch34.3d

556 PHARMACOEPIDEMIOLOGY

Table 34.2. Examples of factors determining drug efficacy that were demonstrated after marketing, selected from the
100 most common drug uses of 1978a

Factors Drug Indication Comments Reference

Regimen
Dose per unit Ibuprofen Rheumatoid Daily dosage initially approved 5
time arthritis, proved to be suboptimal
osteoarthritis
Distribution of Furosemide Congestive heart Efficacy improved by more 6
dose over time failure frequent, smaller doses
Duration Clonidine Hypertension Tolerance develops in the 7
absence of a diuretic
Hypoglycemics (e.g., Diabetes mellitus Tolerance develops in many 8
acetohexamide and patients
tolazamide)

Indication
Severity Metaproterenol Asthma Patients with severe illness do 9
not have a response without
additional, supplementary
therapy
Subcategories Desipramine Depression May vary with endogenous 10
versus exogenous depression
Changes over Ampicillin Otitis media No longer the drug of choice in 11, 12
time some geographic areas
due to bacterial resistance

Patient
Age Diazepam Anxiety A given regimen is more 13
effective in the aged than in the
young
Metabolism varies markedly 14
from premature infants (half-life
54 hours), to full term infants, to
older children (half-life 18
hours); young children can have
paradoxic reactions
Other illness Gentamicin Infection Lower doses required in renal 15
failure

Other
Drugs Lithium Manic± depressive Clearance impaired by 16
illness diuretics, e.g., furosemide
Acetohexamide Diabetes mellitus Many drugs interfere, by 17
causing hyperglycemia (e.g.,
diuretics), displacing drug from
binding sites (e.g., nonsteroidal
anti-inflammatory drugs), etc.
Diet Diuretics (e.g., Hypertension A decrease in sodium intake 18
metolazone, can improve efficacy
furosemide)
Lithium Manic± depressive Significant sodium depletion or 16
illness excess can modify renal
excretion
a
From Reference 4.
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 557

additional prescriptions accompanied 62% of METHODOLOGIC PROBLEMS TO BE

the prescriptions studied, and 41% of the ADDRESSED BY
prescriptions were for patients who had illnesses PHARMACOEPIDEMIOLOGY
other than just the one the drug was being used to RESEARCH
treat. Of the 100 common drug uses, the mean
number of concomitantly administered drugs Chapter 2 introduced the concept of a confound-
ranged from 0.04 to 2.1. The mean number of ing variable, that is a variable other than the risk
concomitant diagnoses ranged from 0.1 to 1.2. factor and outcome variable under study which is
Yet, for none of the uses was the potential for related independently to each of the other two and,
modification of the drug effect by concomitant thereby, can create an apparent association or
drugs or concomitant diagnoses fully explored mask a real one. Studies of intended drug effects
before marketing. present a special methodologic problem of con-
The proportion of prescriptions for patients less founding by the indication for therapy.22, 23 In this
than age 20 ranged from 0.0%, for 43 of the uses, case the risk factor under study is the drug and the
to 97%. Yet, many of these uses had not been outcome variable under study is the clinical
tested in children prior to marketing. Analogously, condition which the drug is supposed to change
only three of the drugs were approved for use in (cure, ameliorate, or prevent). In clinical practice,
pregnant patients, yet we know that drug use in one would expect treated patients to differ from
pregnancy is common.19 ± 21 untreated patients, as the former have an indica-
Thus, this study revealed considerable gaps tion for the treatment. To the extent that the
in the information about beneficial drug effects at indication is related to the outcome variable as
the time of drug marketing. These deficiencies in well, the indication can function as a confounding
the available information should not be surpris- variable.
ing, nor should they be considered inadequacies For example, if one wanted to evaluate the
that should prevent the release of the drug to the effectiveness of a -blocker used after a myocar-
marketplace. The data needed for clinical deci- dial infarction in preventing a recurrent myocar-
sions are frequently and understandably different dial infarction, one might conduct a cohort study
from those needed for regulatory decisions. comparing patients who were treated with the -
Studies performed prior to marketing per force blocker as part of their usual post-myocardial
are focused predominantly on meeting appropri- infarction medical care to patients who were not,
ate regulatory requirements, and only secondarily measuring the incidence of myocardial infarction
on providing a basis for optimal therapeutic in both groups. However, patients with angina,
decisions. The physician also should keep in mind arrhythmias, and hypertension, all indications
that the FDA is not allowed to regulate physi- for -blocker therapy, all are at increased risk of
cians but, rather, pharmaceutical manufacturers. subsequent myocardial infarction. As such, one
This regulation is not aimed at telling a physician might well observe an increase in the risk of
precisely how an agent should be used. In myocardial infarction, rather than the expected
addition, the FDA does not initiate its own decrease. Thus, even if use of the drug was
studies of drug effects, but generally evaluates beneficial, it might appear to be harmful!
those submitted to it by manufacturers. Finally, Confounding by the indication for the treatment
there are reasonable logistical limitations on what generally is not a problem if a study is focusing on
can be expected prior to marketing, without unexpected drug effects, or side-effects, whether
undue cost in time and resources, as well as they are harmful or beneficial. In this situation, the
delaying the availability of a chemical entity with indication for treatment is not usually related to
a proven potential for efficacy. Thus, it seems that the outcome variable under study. For example, in
more studies of beneficial drug effects are needed, a study of gastrointestinal bleeding from nonster-
perhaps as a routine part of postmarketing drug oidal anti-inflammatory drugs, the possible
surveillance. indications for treatment, such as arthritis,
{Jobs}0688jw/makeup/688ch34.3d

558 PHARMACOEPIDEMIOLOGY

dysmenorrhea, and acute pain, have little or no the effects of the narcotic antagonist wear off, the
relationship in and of themselves to the risk of patient returns to coma. Another injection of the
gastrointestinal bleeding.24 Nevertheless, some- naloxone results in awakening once more, and
times the problem of confounding by indication then later the coma returns again. This sequence of
can emerge even in studies of unexpected drug events represents a convincing demonstration of
effects (beneficial or harmful). For instance, in a the drug's ability to have its desired effect. No
study of hypersensitivity reactions associated with elaborate studies are needed to make this point.
the use of nonsteroidal anti-inflammatory drugs, The same would be true for a case series of patients
the increased risk of hypersensitivity reactions given penicillin to treat pneumococcal pneumonia.
evident in patients taking nonsteroidal anti-in- However, in applying this approach, the course
flammatory drugs was higher in those using the of a patient's disease must be sufficiently pre-
drugs for acute pain than in those using the drugs dictable that one can differentiate a true drug
for osteoarthritis and other chronic conditions. effect from spontaneous improvement. In particu-
This probably was because of the intermittent lar, one must be able to exclude regression to the
ingestion of the drug by those receiving it for acute mean as the mechanism of the observed change:
pain.25 individuals selected to participate in a study based
Although confounding by the indication is a upon the severity of their disease spontaneously
relatively uncommon problem for studies of side and usually will tend to improve. One example
effects, this is not the case for studies of anticipated would be a patient with recurrent headaches. The
beneficial effects. In these studies one would expect patient would most likely seek medical attention
the indication to be more closely related to the when the headaches are most severe or most
outcome variable. In fact, the problem presented frequent. A spontaneous return to the baseline
by confounding by the indication has been thought pattern of headaches generally could be expected.
by some to invalidate nonexperimental approaches However, if the patient were treated in the interim,
to studies of the beneficial effects of drugs. Some then the treating physician likely would view the
have felt that questions of beneficial drug effects return to normality as evidence of successful
can be addressed only by using randomized clinical therapy, no matter what treatment was used or
trials.26 Yet, although postmarketing randomized whether it contributed anything to the recovery.
clinical trials certainly can be very useful, they are Second, some questions about beneficial drug
vexed by many of the same logistical problems, effects can be answered using formal nonexperi-
ethical restrictions, and artificial medical settings mental studies, because there is no confounding by
found in premarketing clinical trials. the indication. If the decision about whether to
treat is not based on a formal indication, but on
some other factor that may not be related to the
CURRENTLY AVAILABLE SOLUTIONS outcome variable under study, such as limited
availability of the drug in question, then there is no
Not all studies of beneficial drug effects need be opportunity for confounding by the indication.
randomized clinical trials (see Table 34.3).22 First, This situation occurs most commonly in studies of
some questions do not require any comparative primary prevention. The use of measles vaccine,
(analytic) research for their answer. For these, routinely administered to healthy infants, is one
simple clinical observations, as reported in a case example.
report or case series, can be sufficient. For Third, there are several settings in which
example, the efficacy and effectiveness of nalox- confounding by the indication may exist but
one, used as a narcotic antagonist, is demonstrable theoretically can be controlled. When the indica-
simply through the observation of a single patient. tion can be measured sufficiently well, the tradi-
Consider a patient comatose from an overdose of tional epidemiologic techniques of exclusion,
methadone. An injection of naloxone results in his matching, stratification, and mathematical model-
prompt awakening. However, 30 minutes later, as ing can be applied. The indication clearly can be
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 559

Table 34.3. Classification of research questions according to their problems of confounding by the indication for therapya

Situation Example

1. Comparative studies unnecessary

(a) Drug effect obvious in the individual patient, or Naloxone used for methadone overdose
(b) drug effect obvious in a series of patients Penicillin used for pneumococcal pneumonia
2. Confounding by the indication nonexistent: there is Measles vaccine given routinely to healthy infants
no indication
3. Confounding by the indication exists but is
controllable
(a) The indication is dichotomous
(i) Gradations in the indication do not exist, or Anti-Rh (D) immune globulin given to Rh (D) negative
mothers who deliver Rh (D) positive newborns to prevent
future erythroblastosis fetalis
(ii) Gradations in the indication are unrelated to Penicillin used for endocarditis prophylaxis in patients
the choice of treatment, or with congenital aortic stenosis who are undergoing tooth
extraction
(iii) Gradations in the indication are unrelated to Penicillin used to prevent tertiary syphilis, given to
expected outcome, or patients with an asymptomatic positive serologic test for
syphilis
(iv) Special clinical settings Anticoagulants used after myocardial infarctions to
prevent death
(b) The indication is sufficiently characterizable Isoniazid used for tuberculosis prophylaxis in a patient
with an asymptomatic positive PPD
(i) Complete characterization of the indication
as it relates to choice of therapy or as it relates
to expected outcome, and
(ii) Characterization must continue after
initiation of therapy
4. Confounding by the indication exists and is not Ampicillin used to treat urinary tract infection
controllable
a
From reference 22.

sufficiently measured if it is dichotomous or by mathematical modeling using, for example,

binary. In this situation, the indication either is multiple regression. Then confounding by the
present or absent, but has no gradations in indication can be controlled. Recently, researchers
severity. The indication also can be sufficiently have begun to use Propensity Scores towards this
measured if any gradations in severity either are end.27, 28 This is an approach that uses mathema-
unrelated to the choice of whether or not to treat tical modeling to predict exposure, rather than the
or are unrelated to the expected outcome. Alter- traditional approach of predicting outcome. This
natively, sometimes one can find special clinical is, essentially, a direct measure of indication. One
settings in which the gradations are not related can then use the propensity score to create
to the choice of therapy. For example, if the categories of probability of exposure, and control
availability of drugs is limited or there are for those categories in the analysis. While this
consistent philosophical differences among pre- approach has many attractive features, especially
scribers for using or not using the drug, then as a direct way to control for confounding by
gradations in the indication will not be related to indication, it is important to point out that it is still
the choice of therapy. dependent on identifying and measuring those
Finally, if an indication is graded but can be variables that are the true predictors of therapeutic
sufficiently precisely measured, it can be controlled choice.
{Jobs}0688jw/makeup/688ch34.3d

560 PHARMACOEPIDEMIOLOGY

When questions of intended drug effects do not most recently approved new molecular entities as
fall into any of the preceding categories, confound- of December 1978 was studied to determine what
ing by the indication cannot be controlled. Non- types of nonexperimental study design, if any,
experimental study designs cannot then be used, or could be used to evaluate drug effectiveness.29 Of
they can only be used to demonstrate qualitatively these 100 ``drugs,'' seven were used in contact
some degree of beneficial effect. Specifically, if lenses and were excluded. The remaining 93 drugs
confounding by the indication is such that treated were examined for all potential indications and
patients would have a worse clinical outcome than clinical outcomes that could be used to evaluate
untreated patients, yet the outcome observed in intended drug effects. Ultimately we assessed 131
treated patients is better than that observed in drug uses, that is 131 drug± indication pairs. Each
untreated patients, some degree of confidence that drug use was categorized as to whether a study
the drug has a beneficial effect can be built. As an evaluating the effectiveness of that drug for that
example, patients treated with corticosteroids for indication would present the problem of con-
status asthmaticus would be expected to be sicker founding by the indication and, if so, whether one
than those not so treated. If patients receiving of the approaches described above would be
corticosteroids stop wheezing sooner than those adequate to address it. Eighty-nine (67.9%) of
not receiving corticosteroids, corticosteroids the drug uses could have been evaluated using
would indeed seem to have a beneficial effect. simple clinical observations, without formal com-
However, if the patients receiving corticosteroids parative research. A very few of these drugs were,
do not stop wheezing sooner than those not in fact, approved by FDA on the basis of such
receiving corticosteroids, the results of the study studies, e.g., nitroprusside (approved for malig-
are uninterpretable. It is possible that the corti- nant hypertension) and bretylium (approved for
costeroids in fact have no beneficial effect. How- life threatening arrhythmias, in patients refractory
ever, it is also possible that a beneficial effect was to all other antiarrhythmics). The remaining 42
present but was being masked by the difference in drug uses required comparative research for their
severity between the two treatment groups. evaluation, because they all presented the problem
The qualitative approach illustrated above must of confounding by the indication. In seven of the
be used with caution. First, the effect of the 42 (5.3% of the total), this confounding was not an
confounding by indication must be opposite in obstacle to valid nonexperimental research. Most
direction to the expected effect of the drug. often the validity of the approach rested on the
Second, the effect of the confounding by indication observation that any given physician usually used
must be absolutely predictable in its direction. the drug to treat either all or none of his patients
Third, the effect of the confounding by indication with the indication.
must be sufficiently large so as to exclude In the remaining 35 of the 42 uses (26.7% of the
regression to the mean as an explanation for the total), confounding by the indication was judged
results. Even if all of these conditions are met, the to be uncontrollable using currently available
results must be interpreted only qualitatively, not nonexperimental techniques.
quantitatively. To place these findings in perspective, of the 42
Examples of each of these situations are drug uses that required comparative research to
presented in Table 34.3 and discussed further in evaluate their effectiveness, 30 could not ethically
reference 22. be addressed using a randomized clinical trial and
a placebo control. Most of these 30 involved the
use of drugs to treat infections or malignancies. In
APPLICABILITY OF THE PROPOSED
these situations, patients could not ethically be left
APPROACHES
``untreated,'' that is assigned to the placebo group.
How commonly are the nonexperimental ap- Studies of the effects of one drug relative to
proaches we have described applicable for the another active drug, of course, gave different
study of beneficial drug effects? A list of the 100 results. Formal comparative research was necessary
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 561

for all 131 drug uses. Nonexperimental studies Instead of a randomized clinical trial, a series
theoretically could be conducted validly for 94 of of nonexperimental studies were performed. Both
the 131 drug uses (71.8%). Experimental studies case ± control and cohort designs were used.36 ± 53 In
would be ethical for all of them. general, these studies were rigorous and well done.
Of course, judging theoretically that a question Unfortunately, however, the question of con-
of effectiveness is ``studiable'' by a given technique founding by the indication was not addressed in
is not the same as proving that a valid outcome most of the studies.30 In particular, most of the
would emerge from such a study. There are many studies failed to address why some of the women
specifics in the actual conduct of such studies that received the postmenopausal exogenous estrogens
must be addressed on a case by case basis. It is, and others did not. Given the data already
therefore, instructive to examine some specific available on the effects of estrogens on bone
examples of nonexperimental research into bene- density and endometrial cancer, it is reasonable to
ficial drug effects. assume that some physicians might preferentially
routinely use the drugs and others might routinely
avoid them. In such a setting, nonexperimental
SPECIFIC EXAMPLES
techniques could yield valid results, unaffected by
confounding by the indication (category 3.a.iv) in
Estrogens for Prevention of Osteoporotic
Table 34.3). However, many physicians might try
Fractures
to selectively prescribe the drugs for patients who
One of the first series of studies of drug have undergone hysterectomy, because these pa-
effectiveness using rigorous nonexperimental tients are at no risk of endometrial cancer.
study designs examined whether exogenous estro- Alternatively, some physicians may try to use the
gens could prevent fractures in postmenopausal drugs only on patients who they feel are at high
women with osteoporosis.30 Biochemical studies risk of fractures or are at high risk of complica-
had documented that the menopause resulted in a tions from fractures. These situations would
negative calcium and phosphorus balance, and represent uncontrollable confounding by the in-
that the balance returned towards normal with dication Ð category 4 in Table 34.3. Finally, one
the ingestion of exogenous estrogens.31 Studies of might expect that the direction of the confounding
bone density documented that exogenous estro- by indication might be opposite to that of the drug
gens prevented the loss of bone density that was effect, allowing one to use these data to make at
associated with the menopause,32 for as long as least qualitative conclusions. This assumes, how-
the estrogens were continued.33 It seemed plau- ever, that physicians can accurately predict who is
sible that the use of estrogens might prevent at high risk of fracture. Such a presumption was
fractures from osteoporosis, but no data directly not borne out by the available data.47
addressed that question. On the other hand, In fact, the three studies that closely examined
postmenopausal estrogens had been shown to the comparability of the study groups were able to
cause endometrial cancer.34, 35 document that they were not comparable.36, 47, 49
A randomized clinical trial would have been the Specifically, one study was a case ±control study
ideal way to address the effect of estrogen on within an orthopedic service, and documented that
fractures. However, such a study was impractical cases with fractures of the hip or radius weighed
for many reasons. This is prophylactic therapy. less than controls matched for age and race, had a
Although postmenopausal fractures are common, later menopause, and more frequently were alco-
they are experienced by a sufficiently small holics.36 A second was a cohort study of patients
proportion of the population during any defined with known estrogen deficiency. In this study,
time period that an extremely large sample size those who were treated with estrogens differed
would be needed. Also, the study would need to be from those who were not in age, age of meno-
carried on for many years before a beneficial effect pause, duration of followup, height, weight, blood
could begin to be seen. pressure, marital status, race, economic status, and
{Jobs}0688jw/makeup/688ch34.3d

562 PHARMACOEPIDEMIOLOGY

gravidity, as well as in the frequency of the As with the question of the effect of estrogens
following diagnoses: atrophic vaginitis, bilateral on bone fractures from osteoporosis, a rando-
oophorectomy, premature ovarian failure, hypo- mized clinical trial would be the ideal design to
pituitarism, gonadal dysgenesis, endocrine disease, address the question of the optimal duration of
hypertension, and osteoporosis.47 anticoagulation after venous thromboembolism,
A third study used a case ±control design to but such a study is impractical. After patients
investigate patients admitted to surgical services.49 have been anticoagulated in the hospital and
It compared cases with hip fractures to a control followed for a short time as outpatients, the risk
group of surgical patients, divided into those with of recurrence is sufficiently small that an en-
trauma and those without trauma. Cases were ormous population would be needed to detect a
noted to be older, taller, and to have a lower body difference in outcome due to differences in
weight than the controls. The cases more fre- therapy. Until recently, the only randomized
quently had undergone ovariectomy, breastfed clinical trial in the literature that addressed this
fewer times and for fewer months, and were question compared six weeks of outpatient treat-
hypothyroid less frequently than the controls. ment to six months of treatment. No difference in
When these factors were controlled for as con- recurrence rate between two groups of patients
founding variables, the effect of estrogens was still was observed.62 However, only 186 subjects were
apparent. However, as in the other studies, there included, yielding a total of only seven recur-
was no information on how or why the decision rences. In addition, over half the study subjects
was made to treat with or withhold estrogens. had known short-term risk factors for venous
thromboembolism. These included pregnancy,
use of oral contraceptives, and recent surgery.
Anticoagulants for Prevention of Recurrent
Patients with these transient underlying risk
Venous Thromboembolism
factors might be expected to be less likely to
The use of intravenous anticoagulants reduces the benefit from longer-term anticoagulant therapy
risk of recurrent venous thromboembolism,54 and than patients with idiopathic disease.
the addition of oral anticoagulants to intravenous The question of the optimal duration of antic-
anticoagulants probably reduces the risk even oagulation was addressed in a cohort study using
further.55 However, how long oral anticoagulant data from the Northern California Kaiser Per-
treatment should be continued had not been well manente Medical Program.63 The study required
studied. Most explicit advice from experts on the the use of ten years of data from this population
optimal duration of anticoagulation therapy was of 1.6 million, or a total of 16 million patient
and is based on anecdotal experience.56, 57 Most of years of experience. There were a total of 3384
the data available that are used to suggest the individuals identified as being hospitalized for
appropriate duration of therapy are derived from venous thromboembolism. Of these, 2473 suf-
clinical observations in a single medical center.58± 61 fered from idiopathic venous thromboembolism.
They represent an accumulating case series. Over Their clinical outcomes were evaluated, according
time, gradually patients' treatment has been to how long they had been treated with oral
prolonged. Thus, changes in the duration of anticoagulants. Using those treated with six
treatment are intermingled with other changes in weeks of therapy or less as a control group,
medical care over decades. In addition, the studies prolongation of therapy beyond that point was
do not compare patients receiving treatments of found to increase the risk of major bleeding
different length, but simply observe when most dramatically, but to have no effect on recurrence
recurrences tend to occur. The investigators have rates.
assumed that treatment should be prolonged The feature of this study that allowed the
sufficiently to include that time when recurrences investigators to overcome the problem of con-
can be expected. Problems with these studies have founding by indication was that physician beha-
been detailed.56, 57 vior regarding how long therapy was continued
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 563

as essentially random (category 3.a.ii in Ta- results do coincide with those of a randomized
ble 34.3). The choice of how long to treat became clinical trial evaluating the efficacy of lidocaine in
random, because there was no prior information preventing primary ventricular fibrillation.65 How-
on how long one should treat. In fact, the duration ever, while the drug prevented the arrhythmia in
of treatment was relatively uniformly distributed that randomized clinical trial, it did not alter
across the years of followup, and the results were mortality.
no different when one restricted the analysis to
those who had their anticoagulation stopped
Anticoagulants for Prevention of Death from
because of hemorrhage, rather than at the option
Myocardial Infarction
of their physician.
Whether anticoagulants can prevent death from
myocardial infarction had been addressed using
Lidocaine for Prevention of Death from
randomized clinical trials.66 However, the results
Myocardial Infarction
had been inconsistent and inconclusive, possibly
In another study, the efficacy of lidocaine in because of problems of sample size. Thus, this
preventing death from myocardial infarction was question would appear to be a good candidate for
studied using a case ±control design.64 Among a case ± control study. Such a study was done,67
patients admitted to a coronary or intensive care with the investigators treating this research
unit for acute myocardial infarction, those who question as if it were a category 3.b question in
died were compared to an equal number of Table 34.3. However, as with the study of the
patients who survived. The controls were matched effects of lidocaine on myocardial infarction, it is
to the cases for age, gender, race, and date of doubtful whether one can measure and quantitate
hospitalization. Overall, lidocaine did not protect precisely the risk of dying from a myocardial
against death. Lidocaine was effective only when infarction at the time of the acute episode. This
deaths attributable to ventricular arrhythmia were study might have been more convincing if the
analyzed separately. investigators had identified the patients of practi-
In this careful study, the investigators obviously tioners who always used anticoagulants for their
were well aware of the risk of confounding by patients with myocardial infarctions, and then
indication. They attempted to control for this compared them to a control group of patients of
confounding by using the epidemiologic technique practitioners who never used anticoagulants for
of stratification, that is classifying patients accord- their patients with myocardial infarctions. Inas-
ing to their risk of dying from myocardial much as the choice of therapy in these patients
infarction, in order to control for this inequality would not have been made on the basis of any
of risk as a confounding variable. Thus, they perceived difference among the patients in their
treated the study as a category 3.b question in risk of dying from myocardial infarction, con-
Table 34.3. Unfortunately, however, it is doubtful founding by the indication would not be a
that one can accurately and fully measure the basis problem. Of course, if the investigators had
for physicians' judgments about who they think is designed the study as we suggest, they then would
at high risk of death from myocardial infarction. have had to consider whether the physicians
Similarly, it is unlikely that each individual's risk themselves were somehow a predictor of out-
of dying from a myocardial infarction can be come, and whether this was consistently related to
predicted, especially death by ventricular arrhyth- their philosophy of using anticoagulants, across
mia. Certainly a classification according to just the multiple physicians. Thus, randomized trials are
presence or absence of congestive heart failure, as really needed to provide the answer to this
was used, is overly simplistic. In fact, the rates of question, and of course in recent years, with the
death attributed to ventricular arrhythmia were advent of low molecular weight heparin and
virtually identical in those patients with and thrombolytic therapy, many have been forth-
without congestive heart failure. Nevertheless the coming.
{Jobs}0688jw/makeup/688ch34.3d

564 PHARMACOEPIDEMIOLOGY

A few pharmacoepidemiology studies on the

Generic versus Brand Name Drugs
relative effectiveness of different preparations used
Another potential use of nonexperimental study for the same purpose have been performed by
designs to study the beneficial effects of drugs Strom, using the COMPASS1 database. These
arose with the passage of the 1984 Waxman ± studies compared patients who began on a brand
Hatch Act in the US. Generic drugs can now be name product and switched to a generic product
marketed after simple demonstration of bioequi- when it became available to patients who remained
valence i.e., equivalent bioavailability, in 18 to 24 on the brand name product. The drugs studied
normal adults.68 However, it is not clear whether were thioridazine, chlorpropamide, and slow
bioequivalence assures clinical equivalence, that absorption theophylline. These studies naturally
is equivalent efficacy and toxicity.69 Clinical raise concerns about the ability to identify the
inequivalence is more likely to be evident as a actual product dispensed. Very few of the phar-
difference in beneficial effects than as a difference macoepidemiology approaches described in Part
in adverse effects. In developing a drug, dosages III of the book are able to identify the specific
are sought which optimize drug efficacy. Toxi- product dispensed. Often the approach does not
city, other than idiosyncratic or allergic reactions, even distinguish whether it is a brand name
usually occurs at higher doses and concentrations product or a generic product that is being used.
than efficacy. Modest variations in the plasma Even when the distinction is made, for example
concentration of the active drug, created by most Medicaid datasets use the National Drug
receiving the same dose in different preparations, Code to identify specifically the drug, the manu-
are most likely, therefore, to be a problem for facturer, the dosage form, and the dose, one is
drug efficacy than for drug toxicity. Variations in inevitably left with questions about whether a
plasma concentration are even more likely to be a brand name is being billed for, while a generic drug
problem for drug effectiveness and cost-effective- is dispensed. In addition, such studies raise
ness. Even a simple change in the physical concerns about how to define the clinical outcome
appearance of the drug could conceivably lead variable. For example, how is drug efficacy
to a decrease in compliance and, thereby, effec- reflected in a claims database? The studies
tiveness. described above used outcomes like number of
Studies designed to evaluate differences in physician visits, number of hospitalizations, and
efficacy among different preparations of the use of adjunctive therapy to obtain an estimate of
same drug require enormous sample sizes, as drug efficacy.
one would be searching for relatively small Using these outcomes, the investigators first
differences. However, such sample sizes can be analyzed the baseline data, comparing the experi-
achieved relatively easily and efficiently as part ence, prior to switching, of those who ultimately
of nonexperimental pharmacoepidemiology stu- switched to generic products to the experience of
dies. Thus, the suggestion has been made that those who did not later switch to a generic
studies of clinical equivalence could possibly product. In each of the three studies, the future
be carried out as postmarketing surveillance switchers were different from the future non-
studies.69 Confounding by the indication is switchers, prior to the switch. Thus, it appears
unlikely to be a problem because, as far as the that patients who were to be switched to generic
physician is concerned, he or she is dealing with products were different than patients who stayed
different products of the same drug, products on the brand name products: confounding by
which are theoretically interchangeable. The indication was indeed operating. Because of this,
choice among the alternative therapies is not no analyses of efficacy after the switch were
being made by the prescriber on the basis of performed. Parenthetically, because of this, and
patient characteristics, but by the pharmacist on questions about the uncertain interpretability of
the basis of product availability Ð category 3.a.ii the clinical outcomes, it was elected not to publish
in Table 34.3. the results of these papers.
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 565

Cost-Effectiveness Studies vaccines are designed to prevent, particularly in

populations which are partly vaccinated, make use
An important new category of studies of beneficial of this design difficult, although not impossible. In
drug effects includes studies of their cost-effective- fact, in one situation, a new Chinese manufactured
ness. These studies measure the resources neces- Japanese encephalitis vaccine was studied for
sary to achieve a particular beneficial outcome, efficacy using a case ± control design,90 despite a
and thus have two main study variables Ð one that study of its safety conducted by the same authors
is clinical and one that is economic. For example, using a randomized clinical trial design.91 In
one could perform a cohort study comparing considering the applicability of nonexperimental
treated patients to untreated patients, and deter- study designs, the relatively indiscriminate use of
mine whether the clinical outcomes they experience such vaccines places the study in category 2 of
and the cost of the medical care they subsequently Table 34.3. Patients who receive these vaccines
receive is different. In such a study, one would differ from those who do not in their socio-
need to consider the possibility of confounding by economic status, their access to medical care, and
the indication for both the clinical outcome and their physicians' attitudes towards vaccines. How-
the cost variables. It should be noted that the ever, for most vaccines, an individual physician is
indication may have different effects on the clinical not likely to give only some of his eligible patients
outcomes and the costs. Thus, while performing the vaccine, withholding it from other eligible
the clinical outcome assessment, one needs to patients. Thus, patients receiving vaccines are not
consider and, potentially, quantify the implications likely to differ from those who do not get the
of the indication for the treatment on the clinical vaccine, at least in their physicians' perceptions
outcome variable. In contrast, while performing about the patients' risk of contracting these
the cost assessment, one needs to consider and, diseases. Nonexperimental studies of such ques-
potentially, quantify the cost implications of the tions should produce valid results, therefore. We
indication on both the clinical outcomes and the refer the interested reader to some methodologic
costs. The subject of health economics as applied papers on the subtleties of designing nonexperi-
to drug use is discussed in more detail in mental studies of vaccine efficacy.104 ± 110
Chapter 35.

Cancer Screening
Vaccines
Another recent and frequent use of nonexperi-
In the last several years, nonexperimental study mental study designs is to evaluate the efficacy of
designs have been widely used to evaluate the cancer screening programs. Although this does not
efficacy of vaccines. Specifically, case ±control directly relate to drugs, the methodological im-
studies have been used to explore the efficacy of plications are the same, and have been better
pneumococcal vaccine,70, 71 rubella vaccine,72, 73 enunciated than in the pharmacoepidemiology
measles vaccine,74 ± 77 Haemophilus influenzae type literature. The use of nonexperimental study
b polysaccharide vaccine,78± 86 oral poliovirus designs to evaluate the efficacy of cancer screening
vaccine,87, 88 meningococcus vaccine,89 Japanese programs will be briefly discussed here, therefore.
encephalitis vaccine,90, 91 and BCG vaccine in pro- Once again, ideally questions about the value of
tecting against tuberculosis92± 99 and leprosy.100, 101 screening would be addressed using randomized
Cohort studies have been used to explore the clinical trials. However, most diseases that are
efficacy of Haemophilus influenzae type b poly- screened for are relatively uncommon. Only a very
saccharide vaccine,79 measles vaccine,80 and per- small fraction of those in a broad screening
tussis vaccine.102, 103 program could be expected to benefit from the
Again studies like these should ideally be screening program. Thus, randomized clinical
conducted as randomized clinical trials. However, trials of screening can be expensive and may
the relative infrequency of the diseases the above require years to complete. Even more importantly,
{Jobs}0688jw/makeup/688ch34.3d

566 PHARMACOEPIDEMIOLOGY

once a screening procedure is widely accepted, limitations of nonexperimental studies, some

even without data documenting its efficacy, scientists have insisted that ``the randomized
recruiting patients into a randomized clinical trial clinical trial (RCT) is the only scientifically
can be impractical and possibly truly unethical. reliable method for assessment of the efficacy
Instead, investigators have used nonexperimental (and risks) of most clinical treatments.''26 Sackett
designs. Screening procedures that have been et al. argue ``... to keep up with the clinical
evaluated repeatedly in this fashion include the literature ... discard at once all articles on therapy
value of ``Pap'' smears for cervical cancer111±122 and that are not randomized trials.''161 In light of the
mammography and self-examination for breast analysis presented above, this posture seems too
cancer.123±136 Other studies investigated screening simplistic and far-reaching. If overbearing, it
measures for lung cancer122, 137, 138 and gastric results in clinically necessary and potentially
cancer.139 All of these were case±control studies. available information being uncollected and un-
Again, they raise similar methodologic considera- used. The proper balance in attitude about the
tions of confounding by indication. Specifically, value of these approaches probably lies some-
why do some women choose to have the screening where between the two extremes. To quote Sir
procedure and others do not? One randomized Austin Bradford Hill, one of the developers of the
clinical trial documented that women who attended randomized trial: ``Any belief that the controlled
screening sessions were at higher risk of developing trial is the only way (to study therapeutic efficacy)
breast cancer than women who were offered would mean not only that the pendulum had
screening but did not attend.140 In addition, case± swung too far but that it had come right off its
control studies of screening present additional hook.''162 Many investigators are now applying
thorny methodologic problems regarding how to nonexperimental designs to studies of beneficial
define cases, how to define controls, the time period drug effects. However, careful attention needs to
to choose for the study, etc.141±156 be paid to the possibility of confounding by the
indication. Some approaches to this problem are
now available, and hopefully more will be
Other Examples
available in the future. However, when confound-
Other analogous work using case ±control study ing by indication can be addressed, clinical
designs has explored the effectiveness of bicycle observations and nonexperimental research can
safety helmets in preventing face injuries,157 anti- be used. The results of nonexperimental research
biotic prophylaxis in preventing post-dental in- are unlikely to be as powerful or as convincing as
fective endocarditis,158 -blockers in preventing those of experimental research. We are not
mortality in patients with acute myocardial infarc- suggesting that nonexperimental studies be used
tion,159 -blockers and incident coronary artery as replacements for experimental studies. How-
events,160 etc. ever, when an experimental study is felt to be
unnecessary, unethical, infeasible, or too costly
relative to the expected benefits, there frequently
THE FUTURE is a good alternative.

Clinicians have long recognized the value of

clinical observations and nonexperimental re- REFERENCES
search. Much of our current knowledge about
the usefulness of medical interventions is based 1. Wallerstein RO, Condit PK, Kasper CK et al.
on information that is nonexperimental. Yet the Statewide study of chloramphenicol therapy and
data and conclusions from the information are fatal aplastic anemia. J Am Med Assoc 1969; 208:
2045± 50.
useful and valid. However, the information that 2. Graham RM, Thornell IR, Gain JM, et al.
observational techniques generate cannot be Prazosin: the first dose phenomenon. Br Med J
accepted uncritically. Perhaps in reaction to the 1976; 2: 1293± 4.
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 567

3. Boston Collaborative Drug Surveillance Group. 22. Strom BL, Miettinen OS, Melmon KL. Postmar-
Regular aspirin intake and acute myocardial keting studies of drug efficacy: when must they be
infarction. Br Med J 1974; 1: 440±3. randomized? Clin Pharmacol Ther 1983; 34: 1 ± 7.
4. Strom BL, Melmon KL, Miettinen OS. Post- 23. Psaty BM, Koepsell TD, Siscovick D, et al. An
marketing studies of drug efficacy: Why? Am J approach to several problems in using large
Med 1985; 78: 475± 80. databases for population-based case± control stu-
5. Anonymous. Ibuprofen (Motrin) Ð a new drug for dies of the therapeutic efficacy and safety of
arthritis. Med Lett Drugs Ther 1974; 16: 109±10. anti-hypertensive medicines. Stat Med 1991; 10:
6. Wilson TW, Falk KJ, Labelle JL, et al. Effect of 653± 62.
dosage regimen on natriuretic response to furose- 24. Carson JL, Strom BL, Soper KA, et al. The
mide. Clin Pharmacol Ther 1975; 18: 165± 9. association of nonsteroidal anti-inflammatory
7. Anonymous. Clonidine (Catapres) for hyperten- drugs with upper gastrointestinal tract bleeding.
sion. Med Lett Drugs Ther 1975; 17: 45 ±6. Arch Intern Med 1987; 147: 85 ± 8.
8. Shen SW, Bressler R. Clinical pharmacology of 25. Strom BL, Carson JL, Morse ML, et al. The effect
oral antidiabetic agents. New Engl J Med 1977; of indication on hypersensitivity reactions asso-
296: 787± 93. ciated with zomepirac sodium and other nonster-
9. Webb-Johnson DC, Andrews JL. Bronchodilator oidal antinflammatory drugs. Arthritis Rheum
therapy. New Engl J Med 1977; 297: 758± 64. 1987; 30: 1142± 8.
10. Hollister LE. Tricyclic antidepressants. New Engl J 26. Juhl E, Christensen E, Tygstrup N. The epidemiol-
Med 1978; 299: 1106± 9. ogy of the gastrointestinal randomized clinical
11. Syriopoulou V, Scheifele D, Howie V, et al. trial. New Engl J Med 1977; 296: 20 ± 2.
Incidence of ampicillin-resistant Hemophilus influ- 27. Rosenbaum PR. Springer Series in Statistics:
enzae in otitis media. J Pediatr 1976; 89: 839± 41. Observational Studies. New York: Springer, 1995;
12. Schwartz R, Rodriquez W, Khan W, et al. The 200± 24.
increasing incidence of ampicillin-resistant Hemo- 28. D'Agostino RB Jr. Propensity score methods for
philus influenzae. J Am Med Assoc 1978; 239: bias reduction in the comparison of a treatment to
320± 3. a non-randomized control group. Stat Med 1998;
13. Vestal RE. Drug use in the elderly: a review of 17: 2265±81.
problems and special considerations. Drugs 1978; 29. Strom BL, Miettinen OS, Melmon KL. Post-
16: 358± 82. marketing studies of drug efficacy: how? Am J
14. Morselli PL, Principi N, Tognoni G, et al. Med 1984; 77: 703± 8.
Diazepam elimination in premature and full term 30. Strom BL. Are estrogens effective in preventing
infants, and children. J Perinat Med 1973; 1: fractures from post-menopausal osteoporosis? In:
133± 41. Melmon KL, ed, Drug Therapeutics: Concepts for
15. Chan RA, Bennes EJ, Hoeprich PD. Gentamicin Physicians. New York, Elsevier ± North-Holland,
therapy in renal failure: a nomogram for dosage. 1982, 67 ±80.
Ann Intern Med 1972; 76: 773± 8. 31. Lagrelius A. Treatment with oral estrone sulphate
16. Hansten PD. Drug interactions, 4th edn Philadel- in the female climacteric. III. Effects on bone
phia, PA: Lea and Febiger, 1979. density and on certain biochemical parameters.
17. Physicians' desk reference, 33rd edn Oradell, NJ: Acta Obstet Gynecol Scand 1981; 60: 481± 8.
Medical Economics, 1979. 32. Lindsay R, Hart DM, Aitken JM, et al. Long-term
18. Parijs J, Joossens JV, Van der Linden L, et al. prevention of postmenopausal osteoporosis by
Moderate sodium restriction and diuretics in the oestrogen. Lancet 1976; 1: 1038± 41.
treatment of hypertension. Am Heart J 1973; 85: 33. Lindsay R, Hart DM, MacLean A, et al. Bone
22 ± 34. response to termination of oestrogen treatment.
19. Doering PL, Stewart RB. The extent and character Lancet 1978; 1: 1325± 7.
of drug consumption during pregnancy. J Am Med 34. Ziel HK, Finkle WD. Increased risk of endometrial
Assoc 1978; 239: 843± 6. cancer among users of conjugated estrogens. New
20. Boethius G. Recording of drug prescriptions in the Engl J Med 1975; 293: 1167± 70.
county of JaÈmtland, Sweden. II. Drug exposure of 35. Antunes CMF, Stolley PD, Rosenshein NB, et al.
pregnant women in relation to course and outcome Endometrial cancer and estrogen use. New Engl J
of pregnancy. Eur J Clin Pharmacol 1977; 12: Med 1979; 300: 9 ± 13.
37 ± 43. 36. Hutchinson TA, Polansky SM, Feinstein AR. Post-
21. Degenhardt KH, Kerken H, KnoÈrr K. Drug usage menopausal oestrogens protect against fractures of
and fetal development: preliminary evaluations of hip and distal radius. Lancet 1979; 2: 705± 9.
a prospective investigation. Adv Exp Med Biol 37. Weiss NS, Ure CL, Ballard JH, et al. Decreased
1972; 27: 467±79. risk of fractures of the hip and lower forearm with
{Jobs}0688jw/makeup/688ch34.3d

568 PHARMACOEPIDEMIOLOGY

postmenopausal use of estrogen. New Engl J Med S. Hormone replacement therapy and risk of hip
1980; 303: 1195± 8. fracture: population based case ±control study. The
38. Johnson RE, Specht EE. The risk of hip fracture in Swedish Hip Fracture Study Group. Br Med J
postmenopausal females with and without estrogen 1998; 316: 1858± 63.
drug exposure. Am J Public Health 1981; 71: 138± 54. Barritt DW, Jordan SC. Anticoagulant drugs in the
44. treatment of pulmonary embolism. Lancet 1960; 1:
39. Wallach S, Henneman PH. Prolonged estrogen 1309± 12.
therapy in post-menopausal women. J Am Med 55. Coon WW, Willis PW. Recurrence of venous
Assoc 1959; 171: 1637± 42. thromboembolism. Surgery 1973; 73: 823± 7.
40. Gordon GS, Picchi J, Roof BS. Antifracture 56. Acheson L, Speizer FE, Tager I. Venous throm-
efficacy of long-term estrogens for osteoporosis. boembolism [letter]. New Engl J Med 1979; 293:
Trans Assoc Am Phys 1973; 86: 326± 32. 879.
41. Burch JC, Byrd BF, Vaughn WK. The effects of 57. Salzman EW. Venous thrombosis. Duration of
long-term estrogen on hysterectomized women. Am anticoagulant therapy [letter]. New Engl J Med
J Obstet Gynecol 1974; 118: 778± 82. 1975; 293: 879.
42. Burch JC, Byrd BF, Vaughn WK. The effects of 58. Coon WW, Mackenzie JW, Hodgson PE. A critical
long-term estrogen administration to women fol- evaluation of anticoagulant therapy in peripheral
lowing hysterectomy. Front Horm Res 1975; 3: venous thrombosis and pulmonary embolism. Surg
208± 14. Gynecol Obstet 1958; 106: 129± 36.
43. Nachtigall LE, Nachtigall RH, Nachtigall RD, et 59. Coon WW, Willis PW, Symons MJ. Assessment of
al. Estrogen replacement therapy I. A 10-year anticoagulant treatment of venous thromboembo-
prospective study in the relationship to osteoporo- lism. Ann Surg 1969; 170: 559± 68.
sis. Obstet Gynecol 1979; 53: 277±81. 60. Coon WW, Willis PW. Thromboembolic compli-
44. Nachtigall LE, Nachtigall RH, Nachtigall RD, et cations during anticoagulant therapy. Arch Surg
al. Estrogen replacement therapy II. A prospective 1972; 105: 209± 12.
study in the relationship to carcinoma and 61. Coon WW, Willis PW. Recurrence of venous
cardiovascular and metabolic problems. Obstet thromboembolism. Surgery 1973; 73: 823.
Gynecol 1979; 54: 74 ± 9. 62. O'Sullivan EF. Duration of anticoagulant therapy
45. Nordin BEC, Horsman A, Crilly RG, et al. in venous thrombo-embolism. Med J Aust 1972; 2:
Treatment of spinal osteoporosis in postmenopau- 1104± 7.
sal women. Br Med J 1980; 1: 451± 4. 63. Petitti DB, Strom BL, Melmon KL. Duration of
46. Paganini-Hill A, Ross RK, Gerkins VR, et al. warfarin anticoagulant therapy and the probabil-
Menopausal estrogen therapy and hip fractures. ities of recurrent thromboembolism and hemor-
Ann Intern Med 1981; 95: 28 ± 31. rhage. Am J Med 1986; 81: 255±9.
47. Hammond CB, Jelovsek FR, Lee KL, et al. Effects 64. Horwitz RI, Feinstein AR. Improved observa-
of long-term estrogen replacement therapy I. tional method for studying therapeutic efficacy.
Metabolic effects. Am J Obstet Gynecol 1979; Suggestive evidence that lidocaine prophylaxis
133: 525±36. prevents death in acute myocardial infarction. J
48. Riggs BL, Seeman E, Hodgson SF, et al. Effect of Am Med Assoc 1981; 246: 2455± 9.
the fluoride=calcium regimen on vertebral fracture 65. Lie KI, Wellens HJ, van Capelle FJ, et al.
occurrence in postmenopausal osteoporosis. New Lidocaine in the prevention of primary ventricular
Engl J Med 1982; 306: 446± 50. fibrillation. New Engl J Med 1974; 291: 1324± 6.
49. Krieger N, Kelsey JL, Holford TR, et al. An 66. Hirsh J. Effectiveness of anticoagulants. Semin
epidemiologic study of hip fracture in postmeno- Thromb Hemost 1986; 12: 21 ± 37.
pausal women. Am J Epidemiol 1982; 116: 141±8. 67. Horwitz RI, Feinstein AR. The application of
50. Hooyman JR, Melton LJ, Nelson AM, et al. therapeutic-trial principles to improve the design of
Fractures after rheumatoid arthritis. Arthritis epidemiologic research: a case± control study sug-
Rheum 1984; 27: 1353± 61. gesting that anticoagulants reduce mortality in
51. Ettinger B, Genant HK, Cann CE. Long-term patients with myocardial infarction. J Chron Dis
estrogen replacement therapy prevents bone loss 1981; 34: 575± 83.
and fractures. Ann Intern Med 1985; 102: 319± 24. 68. Mattison N. Pharmaceutical innovation and
52. Wasnich RD, Ross PD, Heilbrun LK, et al. generic drug competition in the USA: effects of
Differential effects of thiazide and estrogen upon the Drug Price Competition and Patent Term
bone mineral content and fracture prevalence. Restoration Act of 1984. Pharmaceut Med 1986;
Obstet Gynecol 1986; 67: 457± 62. 1: 177 ± 85.
53. Michaelsson K, Baron JA, Farahmand BY, Johnell 69. Strom BL. Generic drug substitution ± revisited.
O, Magnusson C, Persson PG, Persson I, Ljunghall New Engl J Med 1987; 316: 1456± 62.
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 569

70. Shapiro ED, Clemens JD. A controlled evaluation 83. Osterholm MT, Rambeck JH, White KE, Jacobs
of the protective efficacy of pneumococcal vaccine JL, Pierson LM, Neaton JD, Hedberg CW,
for patients at high risk of serious pneumococcal MacDonald KL, Granoff DM. Lack of efficacy
infections. Ann Intern Med 1984; 101: 325± 30. of Haemophilus b polysaccharide vaccine in
71. Rheingold AL, Broome CV, Hightower AW, et al. Minnesota. J Am Med Assoc 1988; 260: 1423± 8.
Age-specific differences in duration of clinical 84. Black SB, Shinefield HR, Hiatt RA, Fireman BH,
protection after vaccination with meningococcal and the Kaiser Permanente Pediatric Vaccine
polysaccharide A vaccine. Lancet 1985; 2: 114± 21. Study Group. Efficacy of Haemophilus influenzae
72. Greaves WL, Orenstein WA, Hinman AR, et al. type b capsular polysaccharide vaccine. Pediatr
Clinical efficacy of rubella vaccine. Pediatr Infect Infect Dis J 1988; 7: 149± 56.
Dis 1983; 2: 284± 6. 85. Bower C, Condon R, Payne J, Burton P, Watson
73. Strassburg MA, Greenland S, Stephenson TG, et C, Wild B. Measuring the impact of conjugate
al. Clinical effectiveness of rubella vaccine in a vaccines on invasive Haemophilus influenzae type
college population. Vaccine 1985; 3: 109± 12. b infection in Western Australia. Aust N Z J Public
74. Killewo J, Makwaya C, Munubhi E, Mpembeni R. Health 1998; 22: 67 ± 72.
The protective effect of measles vaccine under 86. Muhlemann K, Alexander ER, Weiss NS, Pepe M,
rountine vaccination conditions in Dar Es Salaam, Schopfer K. Risk factors for invasive Haemophilus
Tanzania: a case ±control study. Int J Epidemiol influenzae disease among children 2 ± 16 years of
1991; 20: 508±14. age in the vaccine era, Switzerland 1991± 1993. The
75. Mahomva AI, Moyo IM, Mbengeranwa LO. Swiss H. Influenzae Study Group. Int J Epidemiol
Evaluation of a measles vaccine efficacy during a 1996; 25: 1280± 5.
measles outbreak in Mbare, City of Harare 87. Balraj V, John TJ, Thomas M, Mukundan S.
Zimbabwe. Cent Afr J Med 1997; 43: 254±6. Efficacy of oral poliovirus vaccine in rural com-
76. Mudzamiri WS, Peterson DE, Marufu T, Biellik munities of North District, India. Int J Epidemiol
RJ, L'Herminez M. Measles vaccine efficacy in 1990; 19: 711± 4.
Masvingo district, Zimbabwe. Cent Afr J Med 88. Deming MS, Jaiteh KO, Otten MW Jr, et al.
1996; 42: 195±7. Epidemic poliomyelitis in The Gambia following
77. Wenger JD, Pierce R, Deaver KA, Plikaytis BD, the control of poliomyelitis as an endemic disease.
Facklam RR, Broome CV and the Haemophilus Am J Epidemiol 1992; 135: 393± 408.
Influenzae Vaccine Efficacy Study Group. Efficacy 89. Rosenstein N, Levine O, Taylor JP, Evans D,
of Haemophilus influenzae type b polysaccharide ± Plikaytis BD, Wenger JD, Perkins BA. Efficacy of
diphtheria toxoid conjugate vaccine in US children meningococcal vaccine and barriers to vaccination.
aged 18 ± 59 months. Lancet 1991; 338: 395± 8. J Am Med Assoc 1998; 279: 435± 9.
78. Greenberg DP, Vadheim CM, Bordenave N, 90. Hennessy S, Lui Z, Tsai TF, Strom BL, Wan CM,
Ziontz L, Christenson P, Waterman SH, Ward JI. Liu HL, Wu TX, Yu HJ, Liu QM, Karabatsos N,
Protective efficacy of Haemophilus influenzae type Bilker WB, Halstead SB. Effectiveness of live-
b polysaccharide and conjugate vaccines in chil- attenuated Japanese encephalitis vaccine (SA14 ±
dren 18 months of age and older. J Am Med Assoc 14 ± 2): a case-control study. Lancet 1996; 347:
1991; 265: 987± 92. 1583± 6.
79. Kaiser Permanente Pediatric Vaccine Study 91. Liu ZL, Hennessy S, Strom BL, Tsai TF, Wan
Group. Efficacy of Haemophilus influenzae type CM, Tang SC, Xiang CF, Bilker WB, Pan XP,
b capsular polysaccharide vaccine. Pediatr Infect Yao YJ, Xu ZW, Halstead SB. Short-term safety
Dis 1988; 7: 149± 56. of live-attenuated Japanese Encephalitis vaccine
80. Sutcliffe PA, Rea E. Outbreak of measles in a (SA14 ± 14 ± 2): results of a 26,239-subject ran-
highly vaccinated secondary school population. domized clinical trial. J Infect Dis 1997; 176:
Can Med Assoc J 1996 155: 1407± 13. 1366 ± 9.
81. Harrison LH, Broome CV, Hightower AW, Hoppe 92. Miceli I, Colaiacovo D, Kantor IN, et al. Evalua-
CC, Makintubee S, Sitze SL, Taylor JA, Gaventa tion of the effectiveness of BCG vaccination using
S, Wenger JD, Facklam RR, and the Haemophilus the case ± control method. Dev Biol Stand 1986; 58:
Vaccine Efficacy Study Group. A day care-based 293± 6.
study of the efficacy of Haemophilus b poly- 93. Shapiro C, Cook N, Evans D, et al. A case± control
saccharide vaccine. J Am Med Assoc 1988; 260: study of BCG and childhood tuberculosis in Cali,
1413± 8. Colombia. Int J Epidemiol 1985; 14: 441± 6.
82. Shapiro ED, Murphy TV, Wald ER, Brady CA. 94. Young TK, Hershfield ES. A case± control study
The protective efficacy of Haemophilus b poly- to evaluate the effectiveness of mass BCG vaccina-
saccharide vaccine. J Am Med Assoc 1988; 260: tion among Canadian Indians. Am J Public Health
1419± 22. 1986; 76: 783± 6.
{Jobs}0688jw/makeup/688ch34.3d

570 PHARMACOEPIDEMIOLOGY

95. Houston S, Fanning A, Soskolne CL, Fraser N. 112. Aristizabal N, Cuello C, Correa P, et al. The
The effectiveness of bacillus Calmette ± Guerin impact of vaginal cytology on cervical cancer risks
(BCG) vaccination against tuberculosis. Am J in Cali, Colombia. Int J Cancer 1984; 34: 5 ±9.
Epidemiol 1990; 131: 340± 8. 113. LaVecchia C, Franceschi S, Decarli A, et al. ``Pap''
96. Filho VW, de Castilho EA, Rodrigues LC, Huttly smear and the risk of cervical neoplasia: quantita-
SRA. Effectiveness of BCG vaccination against tive estimates from a case± control study. Lancet
tuberculous meningitis: a case± control study in 1984; 2: 779± 82.
Sao Paulo, Brazil. Bull WHO 1990; 68: 69 ± 74. 114. MacGregor JE, Moss SM, Parkin DM, et al. A
97. Patel A, Schofield F, Siskind V, et al. Case ± control case ± control study of cervical cancer screening
evaluation of a school-age BCG vaccination in north east Scotland. Br Med J 1985; 290:
programme in subtropical Australia. Bull WHO 1543 ± 6.
1991; 69: 425± 33. 115. van der Graaf Y, Zielhuis GA, Peer PGM et al.
98. Sirinavin S, Chotpitayasunondh T, Suwanjutha S, The effectiveness of cervical screening: a popula-
et al. Protective efficacy of neonatal Bacillus tion-based case± control study. J Clin Epidemiol
Calmette ± GueÂrin vaccination against tuberculosis. 1988; 41: 21 ± 6.
Pediatric Infect Dis J 1991; 10: 359± 65. 116. Clarke EA, Hilditch S, Anderson TW. Optimal
99. Sharma RS, Srivastava DK, Singh AA, et al. frequency of screening for cervical cancer: a
Epidemiological evaluation of BCG vaccine efficacy Toronto case ±control study. IARC Sci Publ
in DelhiÐ1989. J Commun Dis 1989; 21: 200±6. 1986; 76: 125± 31.
100. Fine PEM, Ponninghaus JM, Maine N, et al. 117. Raymond L, Obradovic M, Riotton G. Additional
Protective efficacy of BCG against leprosy in results on relative protection of cervical cancer
Northern Malawi. Lancet 1986; 2: 499±502. screening according to stage of tumour from the
101. Muliyil J, Nelson KE, Diamond EL. Effect of Geneva case± control study. IARC Sci Publ 1986;
BCG on the risk of leprosy in an endemic area: a 76: 107± 10.
case control study. Int J Lepr 1991; 59: 229± 36. 118. Geirsson G, Kristiansdottir R, Sigurdsson K, et al.
102. Isomura S. Clinical studies on efficacy and safety Cervical cancer screening in Iceland: a case±
of an acellular pertussis vaccine in Aichi prefecture, control study. IARC Sci Publ 1986; 76: 37 ± 41.
Japan. Dev Biol Stand 1991; 73: 37 ± 42. 119. Berrino F, Gatta G, d'Alto M, et al. Efficacy of
103. Report from the PHLS Epidemiological Research screening in preventing invasive cervical cancer: a
Laboratory and 21 area health authorities. Efficacy case± control study in Milan, Italy. IARC Sci Publ
of pertussis vaccination in New England. Br Med J 1986; 76: 111± 23.
1982; 285: 357± 9. 120. Celentano DD, Klassen AC, Weisman CS, Rosen-
104. Smith PG. Retrospective assessment of the effec- shein NB. Cervical cancer screening practices
tiveness of BCG vaccination against tuberculosis among older women: results from the Maryland
using the case± control method. Tubercle 1982; 63: cervical cancer case± control study. J Clin Epide-
23 ± 35. miol 1988; 41: 531± 41.
105. Smith PG, Rodrigues LC, Fine PEM. Assessment 121. Zhang ZF, Parkin DM, Yu SZ, et al. Cervical
of the protective efficacy of vaccines against screening attendance and its effectiveness in a rural
common diseases using case± control and cohort population in China. Cancer Detect Prev 1989; 13:
studies. Int J Epidemiol 1984; 13: 87 ± 93. 337±42.
106. Clemens JD, Shapiro ED. Resolving the pneumo- 122. Sobue T, Suzuki T, Fujimoto I, et al. Population-
coccal vaccine controversy: are there alternatives to based case± control study on cancer screening.
clinical trials? Rev Infect Dis 1984; 6: 589±600. Environ Health Perspect 1990; 87: 57 ± 62.
107. Orenstein WA, Bernier RH, Dondero TJ, et al. 123. Collette JHA, Day NE, Rombach JJ, de Waard F.
Field evaluation of vaccine efficacy. Bull WHO Evaluation of screening for breast cancer in a
1985; 63: 1055± 68. non-randomized study (the DOM project) by
108. Smith PG. Evaluating interventions against tropi- means of a case ± control study. Lancet 1984; 1:
cal diseases. Int J Epidemiol 1987; 16: 159± 66. 1224 ± 6.
109. Farrington CP. Quantifying misclassification bias 124. Foster RS Jr, Costanza MC. Breast self-examina-
in cohort studies of vaccine efficacy. Stat Med tion practices and breast cancer survival. Cancer
1990; 9: 1327± 37. 1984; 53: 999± 1005.
110. Joseph KS. A potential bias in the estimation of 125. Greenwald P, Nasca PC, Lawrence CE, et al.
vaccine efficacy through observational study de- Estimated effect of breast self-examination and
signs. [letter]. Int J Epidemiol 1989; 18: 729± 31. routine physical examinations on breast cancer
111. Clarke EA, Anderson TW. Does screening by mortality. New Engl J Med 1978; 299: 271± 3.
``Pap'' smears help prevent cervical cancer? Lancet 126. Smith EM, Francis AM, Polissar L. The effect of
1979; 2: 1 ± 4. breast self-exam practices and physical examina-
{Jobs}0688jw/makeup/688ch34.3d

THE USE OF PHARMACOEPIDEMIOLOGY TO STUDY BENEFICIAL DRUG EFFECTS 571

tion on extent of disease at diagnosis. Prev Med 142. Morrison AS. Case definition in case± control
1980; 9: 409±17. studies of the efficacy of screening. Am J Epidemiol
127. Huquely CM, Brown RL. The value of breast self- 1982; 115: 6 ± 8.
examination. Cancer 1981; 47: 989± 95. 143. Weiss NS. Control definition in case± control
128. Senie RT, Rosen PP, Lesser ML, Kinne DW. studies of the efficacy of screening and diagnostic
Breast self-examination and medical examination testing. Am J Epidemiol 1983; 118: 457±60.
related to breast cancer stage. Am J Public Health 144. Sasco AJ, Day NE, Walter SD. Case ± control
1981; 71: 583±90. studies for the evaluation of screening. J Chronic
129. Feldman JG, Carter AC, Nicastri AD, et al. Breast Dis 1986; 39: 399± 405.
self-examination, relationship to stage of breast 145. Frommer DJ. Case± control studies of screening. J
cancer at diagnosis. Cancer 1981; 47: 2740± 5. Clin Epidemiol 1988; 41: 101.
130. Dubin N, Pasternack BS. Breast cancer screening 146. Sasco AJ. Lead time and length bias in case±
data in case± control studies. Am J Epidemiol 1984; control studies for the evaluation of screening. J
120: 8 ± 16. Clin Epidemiol 1988; 41: 103± 4.
131. Dubin N, Friedman DR, Toniolo PG, et al. Breast 147. Cole P, Morrison AS. Basic issues in population
cancer detection centers and case± control studies screening for cancer. J Natl Cancer Inst 1980; 64:
of the efficacy of screening. J Chronic Dis 1987; 40: 1263± 72.
1041± 50. 148. Morrison AS. Case definition in case± control
132. Verbeek ALM, Hendriks JHCL, Holland R, et al. studies of the efficacy of screening. Am J Epidemiol
Mammographic screening and breast cancer mor- 1982; 115: 6 ± 8.
tality: age-specific effects in Nijmegen project, 149. Baum M, MacRae KD. Screening for breast
1975± 82. Lancet 1985; 1: 865± 6. cancer. Lancet 1984; 2: 462.
133. Palli D, Del Turco MR, Buiatti E, Carli S, et al. A 150. Flanders WD, Longini IM Jr. Estimating benefits
case± control study of the efficacy of a non- of screening from observational cohort studies.
randomized breast cancer screening program in Stat Med 1990; 9: 969± 80.
Florence (Italy). Int J Cancer 1986; 38: 501± 4. 151. Weiss NS, McKnight B, Stevens NG. Approaches
134. Palli D, Del Turco MR, Buiatti E, Ciatto S, et al. to the analysis of case± control studies of the
Time interval since last test in a breast cancer efficacy of screening for cancer. Am J Epidemiol
screening programme: a case±control study in Italy. 1992; 135: 817± 23.
J Epidemiol Community Health 1989; 43: 241±8. 152. Knox G. Case± control studies of screening proce-
135. Friedman DR, Dubin N. Case± control evaluation dures. Public Health 1991; 105: 55 ± 61.
of breast cancer screening efficacy. Am J Epidemiol 153. Weiss NS. Case ±control studies of the efficacy of
1991; 133: 974± 84. screening tests designed to prevent the incidence of
136. Verbeek ALM, Holland R, Sturmans F, et al. cancer. Am J Epidemiol 1999; 149: 1 ±4.
Reduction of breast cancer mortality through mass 154. Weiss NS. Analysis of case± control studies of the
screening with modern mammography. Lancet efficacy of screening for cancer: how should we
1984; 1: 1222±4. deal with tests done in persons with symptoms? Am
137. Ebeling K, Nischan P. Screening for lung cancer Ð J Epidemiol 1998; 147: 1099± 102.
results from a case± control study. Int J Cancer 155. Cronin KA, Weed DL, Connor RJ, Prorok PC.
1987; 40: 141±4. Case ±control studies of cancer screening: theory
138. Sobue T, Suzuki T, Naruke T, and the Japanese and practice. J Natl Cancer Inst 1998; 90: 498± 504.
Lung-Cancer-Screening Research Group. A case ± 156. Weiss NS, Lazovich D. Case±control studies of
control study for evaluating lung-cancer screening screening efficacy: the use of persons newly diag-
in Japan. Int J Cancer 1992; 50: 230± 7. nosed with cancer who later sustain an unfavorable
139. Oshima A, Hirata N, Ubukata T, et al. Evaluation outcome. Am J Epidemiol 1996; 143: 319±22.
of a mass screening program for stomach cancer 157. Thompson DC, Thompson RS, Rivara FP, Wolf
with a case ±control study design. Int J Cancer ME. A case± control study of the effectiveness of
1986; 38: 829±33. bicycle safety helmets in preventing facial injury.
140. Shapiro S, Venet W, Strax P, et al. Selection, Am J Public Health 1990; 80: 1471± 4.
follow-up, and analysis in the Health Insurance 158. Imperiale TF, Horwitz RI. Does prophylaxis
Plan Study: a randomized trial with breast cancer prevent postdental infective endocarditis? A con-
screening. J Natl Cancer Inst Monogr 1985; 67: trolled evaluation of protective efficacy. Am J Med
65 ± 74. 1990; 88: 131± 6.
141. Connor RJ, Prorok PC, Weed DL. The case ± 159. Horwitz RI, Viscoli CM, Clemens JD, Sadock RT.
control design and the assessment of the efficacy Developing improved observational methods for
of cancer screening. J Clin Epidemiol 1991; 44: evaluating therapeutic effectiveness. Am J Med
1215± 21. 1990; 89: 630± 8.
{Jobs}0688jw/makeup/688ch34.3d

572 PHARMACOEPIDEMIOLOGY

160. Psaty BM, Koepsell TD, LoGerfo JP, et al. - 161. Sackett DL, Haynes RB, Tugwell PT. Clinical
blockers and primary prevention of coronary heart Epidemiology. A Basic Science for Clinical Med-
disease in patients with high blood pressure. J Am icine. Boston: Little Brown, 1985.
Med Assoc 1989; 261: 2087± 94. 162. Hill AB. Reflections on the controlled trial. Ann
Rheum Dis 1966; 25: 107± 13.