| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Opinion |
1 Department of Radiology, Massachusetts General Hospital and the Harvard
Medical School, Boston, MA 02114.
2 Avon Foundation Comprehensive Breast Evaluation Center, Wang Ambulatory Care
Center, 15 Parkman St., Ste. 240, Boston, MA 02114.
Received July 11, 2003;
accepted after revision August 12, 2003.
Address correspondence to D. B. Kopans.
Despite the fact that efforts began four decades ago to show that mammography screening can decrease the rate of death from breast cancer, questions persist [2]. The recent issues have turned out to be specious [3], but the episode is an example of the complexitics involved in showing a benefit from screening. Because negative effects are associated with screening, efficacy must be shown scientifically before a new breast cancer screening test is introduced into clinical practice. Not only is this requirement good medical practice, but it will be demanded by our clinical colleagues. All the arguments about the efficacy of mammography screening were not merely academic but were based on the need for scientific proof of benefit that should and will be demanded by the medical community.
The level of proof needed to validate a screening test is much higher than that required to introduce a test used to care for ill patients. Screening involves the evaluation of healthy individuals. Certainly most individuals who are screened for breast cancer do not and will never have the disease. If a test with false-positive findings had no consequences, then there would be little concern. However, tests with false-positive findings are unavoidable, and they result in anxiety, recall of patients for additional evaluations, biopsies, useless and unnecessary treatment, inconvenience, and time away from work or home. These consequences have been called the "harms" of screening. Because there are negative consequences from screening healthy people, these tests should be more rigorously evaluated and a favorable benefit-to-risk ratio should be established. One needs to consider how many will benefit from the test in light of the number of healthy individuals who may be made ill as a consequence of the test. If a test does not reduce the death rate from breast cancer or have some other important benefit, then it can only cause harm from the consequences of studies with false-positive findings.
It is not always true that finding cancer earlier results in a benefit for the individual. Two situations account for this. The lethality of the cancer may have been established before it could be detected by any means. Conversely, some cancers will never be lethal. Breast cancer kills by metastatic spread. Although metastatic spread, on average, is related to the size of the primary cancer at the time it is diagnosed and treated [4], some cancers become metastatic long before the primary cancer is detectable. By the time the cancer is detected in the breast, it has already spread to other organs and is essentially incurable; therefore, mammography screening does not save all women even though their cancers are detected before they become clinically evident. In fact, some women present with metastatic disease with either microscopic cancer in the breast or a primary that is not detectable even by the pathologist. Detecting cancers that have already spread to other organs will have no beneficial effect on the individual. The opposite can also be true because some cancers become extremely large yet have not spread outside the breast and the patient eventually dies from some other cause. Finding these cancers earlier will also not affect the outcome.
Breast cancer is actually a fairly unique malignancy in that most women who develop breast cancer do not die from it. This characteristic strongly suggests that many breast cancers have a low level of lethality. If, for example, a test is developed that finds predominantly lesions that have already spread, then detecting these cancers earlier only results in the patient being aware of her fate sooner with no benefit. She will be subjected to toxic treatments that may not alter the course of the disease but may make her ill and even lead to her premature death. On the other hand, the new test may find only the nonlethal cancers. Because it is not possible to predict the precise lethality of any cancer, detecting nonlethal cancer may also lead to toxic treatment and death when the patient actually has a harmless cancer; therefore, before a new test is introduced, it must be shown that earlier detection has greater benefit than harm.
Proving efficacy is not a simple task. Finding a cancer earlier may provide a false impression of benefit [5], even though it does not result in a decrease in the death rate from breast cancer; therefore, carefully performed clinical trials are needed to show true benefit. The randomized controlled trial is the only way to eliminate the biases that can be misleading when data from other studies are used. For example, if a group of women are screened with a new test and they appear to live longer than women who are not screened, the test may merely advance the time of diagnosis but do nothing to alter the time of death [5]. We may learn about cancers earlier in their growth and know about their presence for a longer period, but the time of death may not be altered by early discovery of the cancer. These women would appear to have a longer survival rate when, in fact, the date of diagnosis was earlier, but they did not live any longer than they would have without the screening test. This is called "lead-time bias."
The randomized controlled trial has been devised to eliminate this bias and many others. It takes a large population of women and divides them randomly into two groups. If the number of women is sufficiently large and the allocation is truly random, the groups are identical in their demographic features and every woman in one of the groups has her "twin" in the other group [6]. Essentially, every woman who develops breast cancer in one group has her counterpart in the other group who develops a breast cancer at the same time that progresses at the same rate so that if nothing is done to alter its course (and it is destined to be lethal), both women die at approximately the same time. Of course we do not have any idea how the women are paired, but this information is not necessary. The percentage of women dying from breast cancer each year is the same in both groups. If the screening test is then introduced to one of the groups and there are fewer deaths from breast cancer over time in that group and if the differences are large enough to be statistically significant, it can be deduced that the screening test actually reduced the death rate from breast cancer. Because randomized controlled trials only evaluate the percentage of women who die each year from breast cancer, they are not biased by such variables as lead time, the number of cancers detected, or the lethality of the cancers. Multiple randomized controlled trials performed with mammography have shown a statistically significant decrease in breast cancer deaths among the women who were offered screening [7].
Currently, the randomized controlled trial, with differences in deaths from breast cancer as the measured end point, is the only way to prove that a test can reduce the number of deaths from breast cancer. Because it may take years or even decades to show a decrease in deaths, it is hoped that surrogate end points can be used instead. Surrogate end points are measurable factors that can predict whether the individual dies from breast cancer so that the results can be analyzed without waiting for the woman to die. If, for example, the size of the cancer could accurately predict outcome and a test found many more smaller cancers in a tested population than in a similar-sized untested population, it might be deduced that there would be fewer deaths among those tested, and the new test might be validated. Because many hidden differences between populations might bias a comparison, a randomized controlled trial would still be needed to ensure that the test did not merely detect additional nonlethal cancers that were going undetected in the unscreened control group. This group would be monitored to ensure that the number of cancers was the same in both groups over time. If the total number of cancers detected in both groups over time was the same (all biologically relevant cancers eventually become clinically evident) and the size of the cancers in the screened group was significantly smaller than that of the cancers eventually detected in the unscreened control group, the new test might be shown to lead to earlier detection and fewer deaths. There is ongoing debate to determine whether this approach is acceptable and whether surrogate end points can be used to avoid long and expensive trials to prove that a screening test is efficacious. However, currently a randomized controlled trial is the only way to prove the efficacy of a screening test.
Everything discussed previously is pertinent in trying to decide whether an imaging test such as sonography or MRI has any efficacy for screening. Many researchers have claimed that sonography [8–14] and MRI [15] can detect breast cancers that are not detectable on mammography or at clinical breast examination. None of the reported studies of sonography have been performed in a manner that can substantiate this claim. All the reported series are purely anecdotal. Either the same individual who interpreted the mammogram also obtained and interpreted the sonogram or, as in one report in which sonograms were interpreted by a different radiologist, the sonography was performed by a researcher who had seen the mammograms. As a result, the sonography evaluations were biased by the knowledge obtained from the mammograms. The person interpreting the sonogram cannot help but be influenced by the findings on the mammogram. Kolb et al. [16] argued that knowledge of the mammography is immaterial because a radiologist would never interpret the findings of sonography without the findings of mammography and clinical evaluation. This may well be true, but it is not the way that scientific validation occurs. The only way to know the true contribution of sonography to screening is to evaluate it first without any other information [1, 17]. Such a regimen provides a pure understanding of the true contribution of sonography, even if the technique is never used in isolation. After the reviewer has logged in his or her blinded analysis of the sonogram (or MRI or another test), then a second review is performed using the other information that is available. This procedure is the only way to really understand the true contribution of the new test and to understand how it might act synergistically with other tests. Because none of the studies, thus far, has been blinded, no scientifically valid data show that any cancers can be detected on sonography or MRI, or any other test alone, that are not detectable either on mammography or at clinical breast examination.
The importance of blinded interpretation can be seen with the following example: Numerous structures such as fibrosis and Cooper's ligaments cause acoustic shadowing on a sonogram. These must be dismissed fairly often because histologic investigation would lead to a large number of biopsies with benign results. A blinded interpretation of the sonogram would tend to ignore these structures. However, suppose that there was the suggestion of architectural distortion in the upper outer portion of one breast on the mammogram. The observer who, at least in these initial reports, is likely to be an advocate for sonography, might dismiss the subtle finding (which really should undergo additional evaluation) and report the findings of mammography as negative but would be more concerned about the upper outer quadrant on sonography. With knowledge of the findings on mammography, the acoustic shadowing in that area (which might have been dismissed on a blinded evaluation) might be considered significant on the unblinded evaluation, and sonography would be given credit for having detected the cancer when it had really been detected because of the findings of mammography. The immediate reaction is that investigators are honest and would never intentionally bias a study. Certainly I would hope that they are, but the primary reason for blinding is to avoid any question of bias that might influence the result. Blinded interpretation of a new test is critical to avoid biases.
All the articles that have been written in support of sonography screening claim that sonography detected the cancers. Language becomes important. "Detection" means that the lesion was discovered by the test. If the person performing the sonography was aware of an abnormality in the breast and could identify it using sonography, this process should not be called detection, but rather the "identification" of a lesion that had already been detected by the other tests. On the other hand, if a blinded sonography study depicted the lesion without the use of other information, then this can be deemed true detection.
The published studies strongly suggest that some cancers are detectable only on sonography. However, because the interpretation of the sonogram was not blinded to the clinical and mammographic information, none of the published studies proves that there are cancers that are detectable on sonography and occult to mammography. Because its protocol includes blinded sonographic evaluation, the proposed ACRIN (American College of Radiology Imaging Network) and Avon study [18] will be the first to determine the percentage of cancers that are actually detectable on sonography alone. It will provide information on the percentage of cancers actually detectable on sonography alone. Unfortunately, this study still cannot answer the question of whether screening with sonography, even in conjunction with mammography, has any efficacy. Until it can be shown that adding sonography screening will result in decreased death rates, sonography screening may result only in increased "harms." Consequently, despite the anecdotal information in the literature, sonography screening should not be performed as a clinical examination. It should be used only in carefully designed and executed scientific trials attempting to show whether it has any efficacy. The same is true for MRI and any other screening test that is proposed for breast cancer detection.
Mammographic screening is not perfect. It does not find all cancers nor does it find all cancers early enough to result in cures. Strong support is needed to develop additional screening tests to further reduce the death rate from breast cancer. Given the potential risks, however, proper scientific validation is mandatory to prove that a new test is efficacious. Currently, X-ray mammography remains the only screening test that has been shown to lower the death rate in randomized controlled trials and that, when introduced to the general population, has resulted in a decrease in deaths from breast cancer [19, 20]. It is the only imaging study that has been shown scientifically to be efficacious for screening and that should be used for general screening.
 Top References |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  M. J. Kim, E.-K. Kim, J. Y. Kwak, B.-W. Park, S.-I. Kim, J. Sohn, and K. K. Oh Role of Sonography in the Detection of Contralateral Metachronous Breast Cancer in an Asian Population Am. J. Roentgenol., February 1, 2008; 190(2): 476 - 480. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 O. Graf, T. H. Helbich, G. Hopf, C. Graf, and E. A. Sickles Probably Benign Breast Masses at US: Is Follow-up an Acceptable Alternative to Biopsy? Radiology, July 1, 2007; 244(1): 87 - 93. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Strano, P. Crystal, and D. Kopans Adjunct Sonography and Not Screening in Cancer Detection Am. J. Roentgenol., August 1, 2004; 183(2): 539 - 539. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
