| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
1 Department of Radiology, Dartmouth-Hitchcock Medical Center, Lebanon,
NH.
2 Department of Radiology Research, University of Arizona, Tucson, AZ.
3 Department of Radiology, University of Washington, Box 354755, 4245 Roosevelt
Way NE, Seattle, WA 98105.
Received May 10, 2007;
accepted after revision November 7, 2007.
Address correspondence to F. S. Chew
(fchew{at}u.washington.edu).
Abstract
Keywords: clinical trials methodology • radiology research study design • screening by imaging
REQUIRED READING (available at www.arrs.org)
 TopAbstractINTRODUCTIONEDUCATIONAL OBJECTIVES REQUIRED READING (available at... RECOMMENDED READINGINSTRUCTIONSQUESTION 1QUESTION 2QUESTION 3QUESTION 4QUESTION 5QUESTION 6QUESTION 7QUESTION 8QUESTION 9QUESTION 10References |
|---|
|
Solution to Question 1
Randomized controlled trials are considered the highest level of evidence
because they evenly distribute the potentially confounding factors that can
affect outcome
[1–4].
Option D is the best response. Randomization is necessary to reliably
determine the effectiveness of medical interventions, including treatment,
screening, and prevention, because these evaluations involve both an
intervention group and a control group. However, evaluations of diagnostic
accuracy do not involve an experimental group and a control group; therefore,
a randomized control trial is not appropriate (although randomization may be
used to vary the order of imaging examinations in a comparison study of
accuracy). Prospective cohort studies are considered the highest level of
evidence for the evaluation of diagnostic accuracy and the second highest
level, after randomized controlled trials, for the evaluation of
effectiveness. Option A is not the best response. Case-control studies are
considered one level below cohort studies because they are retrospective and
more susceptible to bias. Case series have a lower level of evidence than
cohort studies and case-control studies because the selection of cases is
subject to greater sampling bias. Option C is not the best response. Expert
opinion is considered the lowest level of evidence because it is prone to
subjectivity and bias [1,
3]. Neither option B nor option
E is the best response. The relative levels of evidence are summarized in
Table 2.
|
Solution to Question 2
In a cohort study, subjects are selected on the basis of an exposure and
followed up prospectively for occurrence of a disease or other outcome of
interest [2]. The observed
relationship between the outcome and the disease is reported as a relative
risk, which can be calculated as the ratio of the incidence of the disease
among the exposed and the incidence of the disease among the unexposed. Using
Table 3, the mathematic
relationship is [a /(a + b)] /[c
/(c + d)]. A well-known cohort study related to radiology is
the study of IV contrast media by Katayama et al.
[5]. In this study, begun in
1986 while ionic contrast media were being replaced by nonionic contrast media
throughout Japan, more than 300,000 patients received one of these forms of
contrast media, as determined by the local radiology departments' policies,
and were closely monitored for adverse drug reactions (ADRs). The incidence of
ADRs was 3.1% in the group receiving nonionic contrast material and 12.7% in
the group receiving ionic contrast material, for a risk ratio of 0.25. Because
the risk ratio was less than 1.0, nonionic contrast material was considered
protective against ADRs. A cohort study design is most useful for evaluating a
new exposure that may be associated multiple outcomes, as in the preceding
study of the new nonionic contrast media. If the subjects could have been
randomly assigned to receive ionic or nonionic contrast media, an even
stronger randomized controlled study might have been performed.
|
In a case-control study, cases and controls are selected on the basis of having and not having a disease or other outcome of interest, respectively [2]. The exposures of the cases and controls are then retrospectively assessed to determine their associations with the outcome. There is no way of knowing the incidence of disease among exposed and unexposed subjects in a case-control study because the number of subjects with and without disease is determined by the investigator's selection criteria. Therefore, relative risk cannot be calculated. A different but similar number, the odds ratio, is used instead. Using Table 3, the odds ratio equals ad /bc. The stronger the association between the exposure and the disease, the higher the odds ratio; an odds ratio less than 1.0 implies a negative association. When the actual incidence of disease is low (<< 1%), the odds ratio and the relative risk are essentially the same.
Blackmore et al. [6] used a case-control study design to determine the relationship between cervical spine fractures and clinical predictors (exposures) of these fractures. In that study, the clinical records of 472 patients, 168 with fractures and 304 without fractures, were retrospectively reviewed for the presence of 20 potential clinical predictors of these fractures. The strongest predictor was a focal neurologic deficient, which had an odds ratio of 58. A case-control study design is most useful for evaluating a rare outcome that may be associated with many different types of exposures (or predictors), as in the preceding study of cervical spine fractures. A case-control study design is often the only practical design when there is a long latency between the exposure and the outcome and when the outcome is rare, as in the case of radiation-induced malignancy [7].
One major advantage of the cohort study design is that outcomes after exposure are assessed prospectively using well-defined and reproducible methods [2]. In a case-control study design, an outcome, such as death from a certain disease, is used to select the cases, and the exposure is assessed retrospectively through records that have already been assembled. Because of the temporal relationship between exposure and outcome, cohort studies may suggest more strongly a causal relationship between exposure and outcome than case-control studies. Option C is the best response.
Because cohort studies are prospective and investigators must wait for the outcomes to occur, they take longer to complete than case-control studies, which are retrospective, and typically require searching records. Option A is not the best response. Both cohort and case-control studies are observational in nature and thus susceptible to bias. Option B is not the best response. In addition, both types of studies involve human subjects and therefore require institutional review board approval [8]. Option D is not the best response. One disadvantage of the cohort study design is that only one type of exposure can be studied because subjects are selected on the basis of their exposure. Option E is not the best response.
Solution to Question 3
The major advantage of the case-control study design over a cohort study
design is that the former is more efficient for investigating rare outcomes
than the latter because the cases are selected retrospectively on the basis of
already having had experienced the rare outcome
[2]. Option A is the best
response.
However, a major disadvantage of the case-control study design is that the exposures are also determined retrospectively and may be recalled (or recorded) differently among the cases than among the controls, thus creating recall bias [9]. Neither option B nor option D is the best response. Both types of studies involve human subjects and thus require institutional review board approval [8]. Option C is not the best response. Both types of studies are observational in nature and susceptible to confounding. Option E is not the best response.
Solution to Question 4
Randomization is the most effective means of evenly distributing both known
and unknown potentially confounding variables
[2,
4]. Option E is the best
response.
For observational studies, stratification, adjustment, and matching can all be used to help reduce the effect of potential confounders but cannot account for unknown confounders. None of options A, B, or C is the best response. Similarly, investigators cannot evenly distribute the confounders that are unknown among the study groups, and they would likely introduce bias if they could influence group assignments. Option D is not the best response.
Solution to Question 5
In a case-control study, the relevant measure of association is the odds
ratio [2]. In this example, the
odds ratio equals 100 x 100 /5 x 200, which equals 10. This means
that smokers are about 10 times more likely to develop lung cancer than are
nonsmokers. Option D is the best response.
The prevalence of smoking in this case-control study is 74%. However, the prevalence of an exposure in the general population cannot be reliably determined using data from a case-control study because the cases and controls are selected on the basis of their outcomes and availability. Option A is not the best response. Similarly, the lung cancer rates among smokers and nonsmokers cannot be determined from this study. Outcome rates in case-control studies are generally much higher than are outcome rates in the general population because, by design, the cases are overrepresented in the study population Neither option C nor option E is the best response. Finally, a case-control study can establish an association between exposure and outcome, but it cannot establish a causal link. Option B is not the best response.
Solution to Question 6
Because the purpose of screening is to reduce mortality from the target
disease, the appropriate metric for the evaluation of its effectiveness is the
ratio of the disease-specific mortality in the screened group to that in the
control group. Comparisons of 5-year survival overestimate screening
effectiveness because of lead-time, length, and overdiagnosis biases
[9–13].
Advancing the time of diagnosis will prolong the survival from the time of diagnosis, even if it does not delay the time of death. The time interval between when the diagnosis would have occurred with screening and when it would have occurred without screening is known as the lead time. Option A is not the best response. Length bias pertains to the length of the detectable preclinical phase of the disease. Because slowly progressing forms of disease have a longer detectable preclinical phase than rapidly progressing forms, screening preferentially detects the former. Consequently, screening-detected cases of disease would be expected to survive longer than clinically detected cases, even after adjustment for lead-time. Option E is not the best response. Finally, screening may detect some preclinical cases of disease that would not have become clinically significant had they not been detected; such cases would inflate the 5-year survival. This phenomenon is known as overdiagnosis. Option D is the best response. A prospective study would not be affected by recall bias. Option B is not the best response. Selection bias occurs when subjects in the comparison are different for reasons other than the intervention under investigation. Randomization is the only reliable method of ensuring that two different groups are comparable. Furthermore, the screening-detected cases in this comparison occur only in those complying with the screening program, and compliant patients tend to have better outcomes [14]. Option C is not the best response.
Comparisons of 5-year survival are appropriate for determining treatment effectiveness in randomized clinical trials of treatment because in treatment trials, unlike screening trials, diagnosis occurs before the time of randomization [13]. Comparisons of 5-year survival are inappropriate for determining screening effectiveness regardless of study design. Randomization by itself does not address the biases associated with early detection.
Solution to Question 7
The major purpose of randomization is to minimize the differences between
subjects in the screened group versus those in the control group
[2,
4,
9,
13]. Option D is the best
response.
Ascertainment bias results from misclassification in the cause of death. If deaths from the disease of interest in the screened group are misclassified as death from another cause, then the risk ratio is biased in favor of screening. However, if deaths from the disease of interest in the control group are misclassified as death from another cause, then the risk ratio is biased against screening [9, 13]. Option A is not the best response. Once subjects are randomized, they should be analyzed according to their assignment, regardless of whether they comply with their assignment. This is known as the "intent to treat" principle [4]. Option E is not the best response.
Lack of compliance with the randomized assignment in either the screened group or the control group results in an underestimation of the screening effect. For example, if subjects who are randomly assigned to the screening group are noncompliant and do not obtained the screening, the screening effect will be underestimated. Conversely, if subjects in the control group decide to get screening, it will also tend to reduce the differences between the screening and control groups, resulting in an underestimation. Neither option B nor option C is the best response.
Solution to Question 8
Overdiagnosis is the diagnosis of a condition that would not have become
clinically significant if it had not been detected by screening
[9,
15]. Two types of
overdiagnosis have been recognized. Type I refers to conditions that do not
progress. For example, many cases of carcinoma in situ do not progress to
invasive cancer (the "gold standard" of pathology is not truly
golden). Type II refers to conditions that would have progressed if the
individual had not died from another cause. For example, elderly men with
microscopic prostate cancer usually die from other causes long before their
cancers would have become clinically significant. Either type of overdiagnosis
increases the observed 5-year survival rate because it adds cases without real
disease to the numerator and denominator of the survival statistic
[9,
13,
15]. Option C is the best
response.
Overdiagnosis leads to an increase in the calculated sensitivity and positive predictive value of the screening test because it adds cases without real disease to the numerator and denominator of these accuracy statistics. Neither option A nor option B is the best response. Overdiagnosis leads to an increase in the observed incidence rate of the target disease because overdiagnosed cases are falsely counted as real cases of disease. Option D is not the best response. Overdiagnosis does not affect the real mortality rate from the target disease, and this is one major justification for using disease-specific mortality as the major end point in randomized controlled trials of screening. Option E is not the best response.
Solution to Question 9
For many diseases, there is a critical point in time beyond which therapy
is less effective [9,
13]. For most cancers, the
critical point occurs when the primary tumor metastasizes. For screening to be
effective, the critical point must occur within the detectable preclinical
phase of disease [9,
13]. Option C is the best
response. If the critical point occurs before the detectable preclinical
phase of disease, then screening is futile. Options A and E are not the best
responses. If the critical point occurs after the detectable preclinical phase
of disease, then screening is unnecessary. Options B and D are not the best
responses.
Solution to Question 10
Slowly progressing cases of disease are more likely to be detected during
screening because they exist in the detectable preclinical phase for a longer
period of time than do rapidly progressing cases
[9,
11,
13]. Option B is the best
response.
The rapidity of disease progression is inversely related to the length of the detectable preclinical phase. Rapidly progressing cases of disease are less likely to be detected during screening than are slowly progressing cases because the former exist in the detectable preclinical phase for a shorter period of time. None of options A, C, and D is the best response. Interval cases, which surface clinically after a negative screening examination, tend to be more rapidly progressing than are screening-detected cases. Option E is not the best response.
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
