| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
Dr. Black is from the Department of Radiology, Dartmouth-Hitchcock Medical Center, Lebanon, NH, and the Center for the Evaluative Clinical Sciences, Department of Community and Family Medicine, Dartmouth Medical School, Hanover, NH.
Correspondence to: William C. Black, MD, Department of Radiology, Dartmouth-Hitchcock Medical Center, 1 Medical Center Dr, Lebanon, NH 03756; e-mail: William.Black{at}Hitchcock.org
Although screening for lung cancer with chest radiography
has never been shown to be effective, there is growing enthusiasm for
screening with low-dose helical CT, which is a much more sensitive
technique. In a recently published study,1
chest
radiography detected only 4 of 23 stage I lung cancers detected by
helical CT, 15 of which were 
1.0 cm in diameter. Underlying the
enthusiasm for screening with CT is the premise that earlier detection
will increase the curability of this disease. In this issue of
CHEST (see page 1568), however, this basic premise is
challenged by a report on the survival of 510 consecutive patients with
stage IA non-small cell lung cancer. Using a Cox proportional hazards
model, Patz et al found no statistically significant relationship
between tumor size and survival. Given that stage is the single
strongest predictor of survival for lung cancer overall, these findings
are surprising and beg the following questions. Why did the authors not
find an inverse relationship between tumor size and survival, and how
does this relate to the issue of screening with helical CT?
Let me begin by offering three reasons why there should be a causal inverse relationship between tumor size and survival. First, considering any individual case, if a lung cancer is diagnosed when it is small instead of when it is large, then survival, which is measured from the time of diagnosis, should be increased by the lead time from earlier diagnosis.2 For example, at a constant doubling time of 180 days, it takes about 2.3 years for a tumor to grow from a diameter of 1.0 cm to 3.0 cm. Thus, assuming no additional benefit of earlier detection, a patient should live about 2.3 years longer when the tumor is diagnosed at a diameter of 1.0 cm vs 3.0 cm.
Second, earlier detection should increase overdiagnosis, that is, the detection of pseudodisease. Two types of pseudodisease have been distinguished.2 Type I is preclinical disease that does not progress or actually regresses. Type I pseudodisease can be thought of as a false-positive diagnosis by the pathologist, which is relevant to the sometimes subtle distinction between atypical adenomatous hyperplasia and well-differentiated adenocarcinoma.3 Type II pseudodisease is preclinical disease that does not progress rapidly enough to produce any signs or symptoms before the individual dies of other causes. This type is especially relevant to patients with advanced age or serious comorbidity, such as coronary artery disease or chronic obstructive lung disease. Although there is ongoing debate about the amount of pseudodisease detected by screening chest radiography,4 some have argued that it is rare,5 6 while others have argued that it accounts for all of the reported survival advantage among the screened detected cases in the Mayo Lung Project7 ; there is agreement that both types of pseudodisease will be more commonly detected by CT screening. Because overdiagnosis effectively dilutes lung cancer cases with normal individuals, it can markedly increase reported survival.
Finally, earlier detection and treatment might increase the probability of curing "real" lung cancer. Even if some cases metastasize when they are < 1 mm in size, others may remain curable until they reach several centimeters. Curing lung cancer would, of course, also increase reported survival considerably. If, for example, the probability of cure is 20% higher (in absolute terms) when the tumor is 1.0 cm than when it is 3.0 cm and the expected years of lost life from a fatal lung cancer is 10 years, then the increased probability of cure should extend survival an additional 2 years.
There are two likely explanations for why the expected relationship between tumor size and survival was not observed. First, the study may not have had sufficient power. Although > 500 patients were followed, only 62 deaths were observed. Consequently, the confidence intervals around the estimates for the 5-year survivals and hazard ratios for each of the four size categories are rather wide (Table 1 in Patz et al). In addition, the authors reported observed survival, which is based on deaths from all causes, rather than the conventionally reported relative survival, which effectively removes deaths from causes other than the disease of interest.8 Because about half of all deaths in patients with stage IA lung cancer are from other causes,7 the inclusion of these deaths may have further reduced the power of this study to detect differences in lung cancer survival.
A second likely explanation for the negative findings is that of confounding by tumor biology. To understand how this factor could have affected the observed relationship between tumor size and survival, let us assume that there are only two types of lung cancer, aggressive and indolent, and that they behave as follows. Aggressive tumors metastasize when they become 1 mm in size. However, because the metastases start as small clusters of cells and pathologic sampling is incomplete, they remain occult until the primary tumor has grown to 1.5 cm. In contrast, indolent tumors don’t metastasize until they are > 3.0 cm in diameter. Under these assumptions, all stage IA tumors > 1.5 cm would be indolent and curable because none would have occult metastases. However, some of the stage IA tumors < 1.5 cm would be aggressive and incurable because they would already have occult metastases. If we discard our simplifying assumptions and consider a fuller spectrum of tumor biology, we can apply the same reasoning and infer that larger stage IA tumors should tend to be less aggressive than smaller stage IA tumors. In addition, the aggressiveness of stage IA tumors should also be inversely related to survival. Consequently, tumor aggressiveness would be expected to confound the relationship between size and survival among stage IA cases in the direction opposite of what is expected on the basis of lead time, overdiagnosis, and cure. Other confounding variables could also be operating, such as patient age and year of diagnosis, although the direction of their effect is less predictable.
Returning to the issue of screening, suppose the findings of Patz et al are explained by chance and that there really is a strong inverse relationship between tumor size and survival. Could we conclude that screening with CT would be effective at reducing lung cancer mortality? No. As pointed out above, a substantial improvement in survival is expected simply on the basis of lead time and overdiagnosis. A randomized clinical trial or at least some sophisticated modeling that accounts for lead time and overdiagnosis would be needed to determine if and how much screening decreases lung cancer mortality.9 10 11
Suppose the findings of Patz et al are explained by confounding from tumor biology, as they imply in their discussion. Could we conclude that screening is futile? No, again. It could still be that early detection by CT (vs chest radiography or clinical presentation) would enable the cure of many tumors that would not have yet metastasized. However, if the variation in tumor aggressiveness were sufficient to hide the expected relationship between tumor size and survival, then it would greatly complicate the modeling that would be required to assess the effectiveness of alternative screening strategies and would bolster the argument for conducting randomized clinical trials. It is worth remembering that the major reason for conducting a randomized clinical trial is to minimize the effects of confounding variables, which might otherwise mimic or hide a causal relationship between an intervention and outcome.12
In summary, the unexpected observations on survival in stage IA lung cancer are timely and provocative. Although they can probably be explained by some combination of chance and confounding, these findings nevertheless force us to think hard about screening with CT and remind us that survival statistics can be very misleading. As the authors caution, we should not rush headlong into screening before its effectiveness has been demonstrated by randomized clinical trials or mathematical models that properly account for lead time, overdiagnosis, and variations in tumor biology.
References
This article has been cited by other articles:
|
|
 |
|
  D. B. Flieder, J. L. Port, R. J. Korst, P. J. Christos, M. A. Levin, D. E. Becker, and N. K. Altorki Tumor Size Is a Determinant of Stage Distribution in T1 Non-Small Cell Lung Cancer Chest, October 1, 2005; 128(4): 2304 - 2308. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. W. Grannis Overestimation of Lung Cancer Mortality in a Computed Tomography-Screened Population J. Clin. Oncol., April 1, 2005; 23(10): 2439 - 2440. [Full Text] [PDF]
|
|
|
|
 |
|
 B. Mahesh, C. Forrester-Wood, A. Yunus, R. Ahsan, K. Amer, A. Morgan, and R. Ascione Value of wide-margin wedge resection for solitary pulmonary nodule: a single center experience Eur. J. Cardiothorac. Surg., September 1, 2004; 26(3): 474 - 479. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. P. Wisnivesky, D. Yankelevitz, and C. I. Henschke The Effect of Tumor Size on Curability of Stage I Non-small Cell Lung Cancers Chest, September 1, 2004; 126(3): 761 - 765. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. L. Port, M. S. Kent, R. J. Korst, D. Libby, M. Pasmantier, and N. K. Altorki Tumor Size Predicts Survival Within Stage IA Non-Small Cell Lung Cancer Chest, November 1, 2003; 124(5): 1828 - 1833. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Lopez-Encuentra, J. L. Duque-Medina, R. Rami-Porta, A. G. de la Camara, and P. Ferrando Staging in Lung Cancer: Is 3 cm a Prognostic Threshold in Pathologic Stage I Non-small Cell Lung Cancer? : A Multicenter Study of 1,020 Patients Chest, May 1, 2002; 121(5): 1515 - 1520. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. Crocetti, E. Paci, E. F. Patz Jr., and P. C. Goodman Stage IA Non-small Cell Lung Cancer A Small Proportion of Cases in the General Population Chest, January 1, 2001; 119(1): 313 - 315. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
