| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Editorials |
1 From the Department of Radiology, University of Pittsburgh, Imaging Research, Suite 4200, 300 Halket St, Pittsburgh, PA 15213. Received July 19, 2006; final version accepted July 24. Address correspondence to the author (e-mail: gurd{at}upmc.edu).
In an editorial (1), Drs Black, Aberle, and Berg (on behalf of the Executive Committee of the National Lung Screening Trial [NLST]) reject the idea made in my previous editorial (2), which suggests a consideration for a large clinical field trial for lung cancer screening. The original editorial was written to encourage debate on this topic and it has. The authors raise three primary concerns, which are addressed below.
The first argument is that because of the lack of a control population, a large clinical field trial will not address the issue of mortality reduction, that it is subject to selection bias, and that this issue can be addressed only by the NLST. There are many ways to appropriately (even if not precisely) account for the expected mortality from lung cancer or other diseases or total mortality rates in a large cohort without the need for randomization at the diagnostic procedure level. Many of the data in support of other screening practices—including cervical and breast cancer screening, which are generally less controversial than lung cancer screening—have come from large population-based studies. The NLST is not optimally designed to determine the magnitude of the screening-attributable mortality reduction. The NLST merely compares whether there is an arbitrarily predetermined reduction in mortality between screening with computed tomography (CT) and screening with chest radiography. For the true magnitude of attributable mortality reduction to be determined, the control arm for the NLST should have been usual care (as was originally planned and unfortunately changed), and there should have been additional rounds of screening incorporated as well. The current design is highly prone to underestimation of an actual value.
In regard to the issue of eliminating selection bias, the primary purpose of a randomized controlled trial (RCT) is to control for confounding factors, rather than to eliminate selection bias. From the point of view of a traditional screening RCT, the public could have been better served in the long run from an experimental arm with both CT and chest radiography for direct comparison of technologies (similar to the Digital Mammography Imaging Screening Trial design) and a pure control arm without screening (3).
The second issue raised is the possibility that false-positive findings and overdiagnoses could cause more harm than good. In regard to the false-positive findings, the editorial mentions the recently published Mayo Clinic CT screening project where 74% of the participants had noncalcified nodules (1). What is not stated in the editorial is that a large fraction of these nodules were small, and based on the NLST's own definition, many of these examinations (eg, <4 mm) would be considered "negative screening." This should not be confused with false-positive findings.
As for overdiagnosis, our principal concern should be to investigate carefully those who have aggressive cancers and die of competing causes of death (eg, being hit by a truck or having a myocardial infarction) and those who have cancers that are exceedingly slow growing (or not progressing at all). This information will not be provided by the NLST because it cannot easily distinguish between these two extremes. If learning about overdiagnosis, and perhaps more important, overtreatment, was the primary interest of the NLST, a more appropriate design should have been implemented for this very purpose. Similar to the Extracranial-Intracranial (EC-IC) Bypass study conducted in the 1980s (4) where participants diagnosed with a "need" for additional local blood supply (perfusion) were assigned into "surgery" or "no surgery" arms, the NLST would have been better off identifying a specific category of cases (eg, participants with a solitary nodule smaller than 7 mm in size or those with nonsolid or part solid nodules smaller than 1 cm in size) and randomize them after detection into one arm undergoing interventional diagnostic procedures followed by treatment (in those with positive findings for cancer) or a second arm of participants that are followed by sequential CT and interventional diagnostic procedures and then therapeutic intervention to be ensued only when the mass increases in size (eg, by at least a factor of 1.5) with a computed doubling time of less than a predetermined duration.
As difficult as this study may be to execute (and the EC-IC Bypass study was not an easy one to perform either), such an approach would have been ideal to study specifically and directly the outcome of the very cases we expect to constitute the majority of overdiagnoses and/or overtreatments. We would learn how frequently the different subcategories of cancers progress and also how curable these may be. This approach by its very design could also include the assessment of effect of surgeries in patients with benign disease.
The third argument is related to the cost of a large clinical field trial. This commentary is not intended to challenge the decision of numerous distinguished committees that justified a cost of approximately $200 million for an RCT of 53 000 subjects. However, what is extremely interesting is that the authors extrapolate their own projected cost for estimating the possible cost of a large clinical field trial when during the past 10 years the Early Lung Cancer Action Program and the International Early Lung Cancer Action Program have performed similar "observational field trials" on more than 33 000 subjects. These programs include repeat CT scanning and a relatively rigorous long-term follow-up (as long as 10 years in some subsets), with a significantly smaller total budget that is more than an order of magnitude lower! This difference in cost is very real and is at least partially related to the underlying cost structure associated with an RCT that is not necessarily applicable to a clinical field trial.
Last, all the examples mentioned in the editorial (1) as previous field trials can hardly constitute representative large clinical field trials. Even the Health Insurance Plan of Greater New York study, an example identified as a field trial in breast cancer screening, cannot be viewed as a large representative field trial. Additionally, at the time the Breast Cancer Detection Demonstration Project study was conceptualized, it remained somewhat controversial and less than definitive in several respects. The fact is that in studying screening scenarios there are as many, if not more, examples of definitive, widely accepted results (including in terms of mortality) from field trials than from RCTs.
Not too long ago the Lung Cancer Progress Review Group, which consists of experts in all areas related to lung cancer, was assembled by the National Cancer Institute (NCI). In its summary report, the group concluded: "Several meetings co-sponsored by NCI and the American Cancer Society have determined that a number of study designs in addition to a mortality endpoint-randomized trial (the gold-standard approach) are important and valid. NCI must continue to take a strong leadership role in facilitating the initiation and completion of a number of trials evaluating spiral CT as a means of detecting lung cancer early and reducing mortality" (5). The leaders of the NLST should acknowledge the validity of other study designs and welcome such studies. Most study designs have limitations as does the NLST study design. The results from the NLST will be largely limited to a single very unique and controlled scenario. In order for optimal public health policy to be developed, there will ultimately be a need to perform both modeling and extrapolation of data, including the use of external sources such as the Surveillance Epidemiology and End Results database (which the authors themselves had recognized in their own protocol but now argue against in their editorial).
 |
FOOTNOTES |
|---|
See also the editorial by Black et al in this issue.
|
References |
|---|
 TOP References |
|---|
Related Article
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
