HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Reich, J. M. Search for Related Content PubMed PubMed Citation Articles by Reich, J. M. Related Collections Related Article

Assessing the Efficacy of Lung Cancer Screening¹

¹ From the Earl A. Chiles Research Institute, Portland, Ore. Received April 21, 2005; revision requested June 14; revision received June 23; final version accepted August 15. Address correspondence to the author, 7400 SW Barnes Rd, A622, Portland, OR 97225 (e-mail: Reichje{at}dnamail.com).

Lung cancer would seem, a priori, to be an ideal screening candidate: It is the most common lethal neoplasm in the United States; it is often detectable in a preclinical phase; radiographs are noninvasive and readily acceptable; individuals at highest risk—older current and former smokers—are readily identifiable; when it is detected at an early stage, the "5-year cure" rate is approximately 70%; and it is highly lethal when unresectable (1). Although survival is substantially higher in patients with screening-identified disease than in patients in whom disease was identified by other means, screening is not recommended by recognized authorities because results of two large randomized controlled trials demonstrated that although 5-year survival was far higher in the screened than in the unscreened cohort, lung cancer and all-cause mortality were higher as well (1). This controversy has been reignited by the development of low-dose computed tomography (CT), a far more sensitive method than conventional radiography for identifying resectable lung cancers (1). Screening advocates assert that improvement in lung cancer survival, predicated on stage at diagnosis, suffices to demonstrate efficacy (2,3). Others have pointed out that cause-specific survival is an invalid metric of screening efficacy because it discounts other causes of mortality and it introduces biases—lead-time bias, length-biased sampling, and overdiagnosis—that generate a spurious improvement (4). The purpose of this communication is to delineate the limitations of outcome metrics used to assess lung cancer screening efficacy. Definitive evidence of efficacy requires the demonstration of a reduction in lung cancer and all-cause mortality in prospective controlled trials.

	BIASES INTRODUCED BY SCREENING

TOP INTRODUCTION BIASES INTRODUCED BY SCREENING MEASUREMENTS OF SCREENING... LUNG CANCER SURVIVAL VERSUS... OUTCOME OF CONTROLLED TRIALS... NET BENEFIT OF LUNG... IMPACT OF LOW-DOSE CT... SUMMARY AND CONCLUSIONS References

 
Survival was originally and appropriately used as an outcome metric in the assessment of efficacy in controlled trials of antineoplastic agents (5). In such trials, (a) all participants have the disease; (b) diagnosis precedes intervention and the interval between diagnosis and intervention is similar for all participants; and (c) all participants are matched as closely as possible in terms of disease stage, age, sex, and other risk factors. Survival, as a metric of efficacy, is inapplicable to the intervention of screening because screening entrains three interrelated biases: lead-time bias, length-biased sampling, and overdiagnosis, each of which produces an illusory improvement in outcome (4).

Lead-time bias refers to earlier detection of disease, in which intervention, as long as it is not harmful, will lengthen survival without necessarily improving longevity. Consider a hypothetical screening test for motor neuron dysfunction capable of advancing by 5 years the diagnosis of amyotrophic lateral sclerosis, a highly lethal and untreatable condition. Theoretically, if screening were widely applied, it would, due solely to a shift of time frame, increase 5-year survival from near zero to virtually 100%, while mortality, after the onset of clinical disease, would be unchanged.

Length-biased sampling refers to the preferential screening identification of a more phenotypically favorable population of lung cancers because they are slower growing and therefore have a lengthier preclinical detectable phase. Aggressive lung cancers are less likely to be screening identified because of their briefer preclinical detectable phase. Reflecting this bias, survival for individuals with screening-identified lung cancers, particularly lung cancers identified at prevalence screening examinations, is predictably higher than in the overall population of individuals with lung cancer.

Overdiagnosis (or "pseudodisease") refers to the condition in which screening-identified cancers would not have shortened an individual's longevity (6). Overdiagnosis occurs in two circumstances: (a) Some cancers (eg, some prostate and thyroid cancers, which are frequent, incidental autopsy findings) have such limited biologic potential that they prove to be nonlethal and (b) the course of potentially lethal cancers may be overtaken by competing lethal morbidities. For example, lung cancer frequently occurs in persons with other smoking-related disorders—chronic obstructive lung disease and coronary artery disease.

These biases, which are invariably entrained by screening, are interrelated: A protracted lead-time bias increases the likelihood of an intervening lethal comorbidity (resulting in overdiagnosis), and overdiagnosis bias may be considered the extreme example of length-biased sampling, in which the preclinical detectable phase of the neoplasm is so protracted that its course is overtaken by that of a lethal comorbidity.

The influence of overdiagnosis on screening outcomes is critically important: Overdiagnosis imparts a spurious improvement in outcome as measured by survival, and, while both lead-time–biased and length-biased sampling generate a spurious outcome improvement, it is reasonable to assume that some individuals otherwise destined to die of lung cancer benefit from earlier surgical intervention. In contrast, overdiagnosed individuals cannot benefit from earlier intervention because they are destined to die of causes other than lung cancer. Moreover, invasive diagnostic procedures and surgical intervention may shorten their life expectancy. Also, although lead-time bias may be roughly gauged by measurement of growth rates and length-biased sampling may be estimated by the proportion of interval cancers (ie, those that arise between screenings) in one-arm (ie, uncontrolled) investigations, it is impossible to quantify overdiagnosis in such investigations. The magnitude of overdiagnosis and its assessment remain highly controversial and complex subjects that are beyond the scope of this editorial.

	MEASUREMENTS OF SCREENING OUTCOME

TOP INTRODUCTION BIASES INTRODUCED BY SCREENING MEASUREMENTS OF SCREENING... LUNG CANCER SURVIVAL VERSUS... OUTCOME OF CONTROLLED TRIALS... NET BENEFIT OF LUNG... IMPACT OF LOW-DOSE CT... SUMMARY AND CONCLUSIONS References

 
Potentially confusing outcome measurement terms are used in the screening arena. Survival (case survival) addresses only those individuals given a diagnosis of disease. Survival may be reported as a cause-specific or an all-cause outcome and is expressed as a percentage or probability—for example, 70% 5-year lung cancer survival. The terms cure and 5-year cure have been applied in a confusing and misleading fashion, as if they were synonymous with survival. The complement of survival (1 – survival) is fatality (case fatality). Mortality, by contrast, is a measure of longevity, and, unlike survival, it pertains to all screenees, not just those identified as having disease. It is a rate, usually age-adjusted, expressed as the number of deaths over time in a population or sample—for example, 25 deaths per 1000 person-years. (Person-years designates the sum of each individual's time being observed.) Like survival, mortality can be provided as a cause-specific or an all-cause value. Lung cancer survival and lung cancer mortality are cause-specific measures. Intuitively, it would seem that survival improvement would necessarily be accompanied by a reduction in mortality. That this is not the case for many solid cancers—including lung cancer—was demonstrated by Welch et al (7) who employed a sequential comparison of Surveillance, Epidemiology, and End Results survival versus mortality data.

Lung cancer mortality, because it reflects death only from lung cancer, has three limitations: (a) Correctly assigning the cause of death may be impossible in some persons with lung cancer. A careful review of medical records, as was undertaken in the Mayo Lung Program (see below), will more accurately ascertain whether progressive lung cancer was the cause of death than will reference to death certificates (8). However, consider an individual with obstructive airway disease who at surgery is found to have bilateral micronodal metastases. After resection, the patient develops an air leak that necessitates prolonged chest tube drainage, which induces pleural fibrosis; the patient succumbs to respiratory failure months later. Would this event be consistently or correctly classified as a lung cancer death? (b) Lung cancer mortality fails to reflect fatalities caused by lethal comorbidities (which are frequently present in individuals with lung cancer), the courses of which are likely to be foreshortened by invasive diagnostic or therapeutic interventions. (c) Lung cancer mortality omits deaths in the screened cohort caused by intervention after false-positive test results—for example, a lethal myocardial infarction that occurs because of an invasive diagnostic or therapeutic intervention for a benign nodule. Black et al (9) coined the term slippery linkage to describe the increased risk of death that is not captured by cause-specific survival or cause-specific mortality statistics. All-cause mortality circumvents these limitations. Were it not for the disadvantage that all-cause mortality requires impractically large samples (because of the dilution of the effect of organ-specific screening by other causes of mortality that it cannot influence), it would be the reference standard for appraising screening efficacy. The usefulness of all-cause mortality lies in augmenting cause-specific mortality, where it may signal unrecognized or underestimated adverse effects of screening and intervention that are not reflected by cause-specific mortality. The reader will find a comprehensive discussion of the importance of all-cause mortality in the article by Black et al (9).

	LUNG CANCER SURVIVAL VERSUS ALL-CAUSE SURVIVAL

TOP INTRODUCTION BIASES INTRODUCED BY SCREENING MEASUREMENTS OF SCREENING... LUNG CANCER SURVIVAL VERSUS... OUTCOME OF CONTROLLED TRIALS... NET BENEFIT OF LUNG... IMPACT OF LOW-DOSE CT... SUMMARY AND CONCLUSIONS References

 
If deaths within 30 days of surgery are excluded or if a disproportionate number of individuals allocated to surgical intervention die of comorbidities and are excluded from the analysis (ie, are treated as withdrawals [censored data]) the consequence will be a spurious increment in 5-year survival. To make this distinction clear, consider an extreme hypothetical example of persons with stage IA lung cancer. (Stage IA, classified as T1N0M0, designates peripheral tumors that are 3 cm in size and are surrounded by lung or visceral pleura, with bronchoscopic evidence of spread no more proximal than the lobar bronchus. N0 signifies no hilar or mediastinal metastases; M0 indicates no discernible extranodal metastases.) A cohort of 100 patients with stage IA lung cancer undergo lobectomy; their projected 5-year lung cancer survival rate is 70%. If 50 of the patients died of surgically foreshortened comorbidities at 3 years, while eight lung cancer deaths occurred at 1 year and seven occurred at 3 years, the Kaplan-Meier estimate of 5-year lung cancer survival would be 78%; inclusion of the censored data would result in an all-cause survival rate of 35%.

	OUTCOME OF CONTROLLED TRIALS OF LUNG CANCER SCREENING

TOP INTRODUCTION BIASES INTRODUCED BY SCREENING MEASUREMENTS OF SCREENING... LUNG CANCER SURVIVAL VERSUS... OUTCOME OF CONTROLLED TRIALS... NET BENEFIT OF LUNG... IMPACT OF LOW-DOSE CT... SUMMARY AND CONCLUSIONS References

 
There have been two large-scale prospective controlled studies in which, after radiographic and cytologic prevalence screening examinations were performed to exclude persons with identifiable lung cancer, subjects were randomly allocated to a screened versus an unscreened cohort. In the incidence portion of the Mayo Lung Program (10), which included 9211 participants, 206 cases of lung cancer were identified in the screened cohort, versus 160 in the unscreened cohort. The case excess in the screened cohort was therefore (206 – 160)/206 = 22%. Five-year lung cancer survival was 35% in the screened cohort versus 15% in the unscreened cohort, suggesting that early detection combined with surgical intervention was efficacious. However, lung cancer mortality (3.2 versus 3.0 deaths per 1000 person-years) and all-cause mortality (24.8 versus 24.6 deaths per 1000 person-years) were both higher in the screened than in the unscreened cohort, although the differences did not achieve a P value of less than or equal to .05. The eightfold difference between lung cancer mortality and all-cause mortality was ascribable principally to ischemic cardiovascular disease. The 46 excess cases were characterized as "missing cases" because a comparable number of lung cancer cases did not appear in the unscreened cohort after extended follow-up (11). Eddy (12) postulated that the "missing cases" were overdiagnosed—that is, that the number of lung cancers was approximately equal in both cohorts and the control-arm counterparts with unidentified lung cancer died of competing morbidities.

Kubïk et al (13) undertook a similar study in Czechoslovakia. After a prevalence screening examination, 6346 participants were randomly allocated to a cohort that underwent semiannual radiographic and cytologic screening or to an unscreened control cohort. At 3 years and annually thereafter for 3 years, both cohorts were radiographically screened. Twenty-six more lung cancers (108 vs 82) were identified in the initially screened cohort than in the control cohort. The majority of excess cases were identified during the initial 3-year screening. The case excess in the screened cohort was therefore (108 – 82) ÷ 108 = 24%, a value close to the 22% in the Mayo Lung Project. Moreover, in the Czech study as in the Mayo Lung Project, 5-year lung cancer survival markedly favored the screened cohort. Although lung cancer was identified at an earlier stage, lung cancer deaths were higher in the initially screened cohort (n = 45) than in the control cohort (n = 40). After an additional 3 years of follow-up limited to patients with lung cancer, there were 85 lung cancer deaths in the initially screened cohort versus 67 such deaths in the control cohort. All-cause mortality was substantially higher in the initially screened cohort than in the control cohort—17.9 versus 15.4 deaths per 1000 person-years (9).

Manser et al (14) pooled the updated Mayo (11) and Czech (13) study results and found the risk of death from lung cancer to be 11% higher in the intervention groups, with a relative risk (RR) of 1.11 (95% confidence interval [CI]: 1.00, 1.23) and a P value of .05. For all-cause mortality, they reported an intervention cohort RR of 1.16 (95% CI: 1.00, 1.35) in the Czech study and 1.03 (95% CI: 0.93, 1.14) in the Mayo study.

Screening, if it is to be effective, must intercept the growth of cancer at an early stage, while it remains resectable. A reduction in the absolute number (ie, not the proportion) of advanced (unresectable) cancers, referred to as "stage shift," is the fundamental underlying premise of screening. Neither the Mayo nor the Czech study found a reduction in the number of advanced lung cancers in the screened versus the control groups. Both the Mayo and the Czech studies revealed an excess number of lung cancers and an excess of lung cancer and all-cause mortality in the intervention cohort, although the differences were greater in the Czech study. Black et al (9) emphasized the disparity between lung cancer and all-cause mortality in both studies (values are deaths per 10 000 person-years): In the extended follow-up of the Mayo trial, lung cancer mortality in the intervention cohort exceeded that in the control cohort by 4.4, while all-cause mortality was higher by 6.3. In the Czech trial, lung cancer mortality in the intervention cohort exceeded that in the unscreened cohort by 8.9, while all-cause mortality was higher by 25.3.

	NET BENEFIT OF LUNG CANCER SCREENING

TOP INTRODUCTION BIASES INTRODUCED BY SCREENING MEASUREMENTS OF SCREENING... LUNG CANCER SURVIVAL VERSUS... OUTCOME OF CONTROLLED TRIALS... NET BENEFIT OF LUNG... IMPACT OF LOW-DOSE CT... SUMMARY AND CONCLUSIONS References

 
The net benefit of lung cancer screening can best be gauged by a comparison of all-cause mortality in screened versus unscreened cohorts. This metric encompasses the benefit conferred by earlier intervention as a consequence of screening and the offsetting harm done to persons with overdiagnosed lung cancer who are subjected to foreshortened lethal comorbidities, to persons with occult metastatic disease who undergo resection, and to persons with false-positive test results (which constitute the majority of positive test results in most studies), some of whom may be adversely affected by invasive diagnostic and therapeutic procedures and a few of whom will elect to undergo surgery due to fear inspired by an indeterminate result (15). All-cause mortality was higher in the screened than in the control cohort in both the Mayo and Czech trials. Given the assumption that vital status is ascertained in all participants after an adequate period of observation—an assumption that appears to be correct (11,13)—this higher all-cause mortality in the screened cohorts suggests that lung cancer screening with cytologic examination and conventional radiography exerts a negative net benefit. One hypothesis that has been advanced to account for the seeming paradox of higher mortality despite improved survival is that lobectomy diminishes life expectancy (16).

	IMPACT OF LOW-DOSE CT SCREENING

TOP INTRODUCTION BIASES INTRODUCED BY SCREENING MEASUREMENTS OF SCREENING... LUNG CANCER SURVIVAL VERSUS... OUTCOME OF CONTROLLED TRIALS... NET BENEFIT OF LUNG... IMPACT OF LOW-DOSE CT... SUMMARY AND CONCLUSIONS References

 
Owing to its far greater sensitivity in identifying small, stage I lung cancers, the use of low-dose CT will increase lead-time bias. One consequence of increased lead-time bias is an increase in overdiagnosis due to the increased time during which individuals with otherwise undiagnosed lung cancer would have been at risk from their lethal comorbidities. It is unknown whether low-dose CT screening will achieve a stage shift, and, if so, whether such a shift will be sufficient to offset the adverse effect of surgery imposed by increased overdiagnosis. Two preliminary analyses (17,18) of low-dose CT trials demonstrated no reduction in the number of advanced lung cancers. Swensen et al (19) reported no significant difference (P = .43, Poisson regression) in incidence lung cancer mortality (2.8 vs 2.0 deaths per 1000 person-years) among men screened with low-dose CT who were matched for age and duration of follow-up with men who underwent conventional radiographic screening in the earlier Mayo Lung Program (10).

	SUMMARY AND CONCLUSIONS

TOP INTRODUCTION BIASES INTRODUCED BY SCREENING MEASUREMENTS OF SCREENING... LUNG CANCER SURVIVAL VERSUS... OUTCOME OF CONTROLLED TRIALS... NET BENEFIT OF LUNG... IMPACT OF LOW-DOSE CT... SUMMARY AND CONCLUSIONS References

 
Lung cancer survival is a misleading metric. It leads to a spurious overestimate of screening efficacy because of lead-time bias, length-biased sampling, and overdiagnosis and because it fails to capture the adverse long-term consequences of invasive procedures and pulmonary resection performed after both true- and false-positive results.

In conclusion, use of the lung cancer mortality metric circumvents the three biases introduced by screening, but, unlike all-cause mortality, this metric fails to fully capture the adverse consequences of invasive procedures, including resection.

To achieve an improvement in lung cancer outcome, low-dose CT, which will predictably increase the proportion of overdiagnosed lung cancer, would have to achieve a reduction in lung cancer mortality sufficient to offset this incremental adverse effect.

The efficacy of low-dose CT screening can be ascertained only by a comparison of lung cancer mortality—augmented by all-cause mortality—in both arms of a large randomized prospective controlled trial.

	ACKNOWLEDGMENTS

 
I am greatly indebted to Pamela Marcus, MS, PhD, epidemiologist, National Cancer Institute, for her critical reading of the manuscript and for her numerous incisive comments and contributions.

	FOOTNOTES

 
The author has served as a consultant or given testimony in class action suits in which the plaintiffs were requesting monitoring for lung cancer.

See also the editorial by Gur in this issue.

	References

TOP INTRODUCTION BIASES INTRODUCED BY SCREENING MEASUREMENTS OF SCREENING... LUNG CANCER SURVIVAL VERSUS... OUTCOME OF CONTROLLED TRIALS... NET BENEFIT OF LUNG... IMPACT OF LOW-DOSE CT... SUMMARY AND CONCLUSIONS References

 

1. Humphrey LL, Teutsch S, Johnson M. Lung cancer screening with sputum cytologic examination, chest radiography, and computed tomography: an update for the U.S. preventive services task force. Ann Intern Med 2004;140:740–753.
2. Strauss GM, Dominioni L, Jett JR, Freedman M, Grannis FW. Como international conference position statement: lung cancer screening for early diagnosis 5 years after the 1998 Varese conference. Chest 2005;127:1146–1151.[Abstract/Free Full Text]
3. Henschke CI, Naidich DP, Yankelevitz DF, et al. Early lung cancer action project: initial findings on repeat screening. Cancer 2001;92:153–159.[CrossRef][Medline]
4. Patz EF, Goodman PC, Bepler G. Screening for lung cancer. N Engl J Med 2000;343(22):1627–1633.[Free Full Text]
5. Peto R, Pike MC, Armitage P, et al. Design and analysis of randomized clinical trials requiring prolonged observation of each patient. Br J Cancer 1976;34:585–612.[Medline]
6. Black WC. Overdiagnosis: an underrecognized cause of confusion and harm in cancer screening [editorial]. J Natl Cancer Inst 2000;92:1280-1282.[Free Full Text]
7. Welch HG, Schwartz LM, Woloshin S. Are increasing 5-year survival rates evidence of success against cancer? JAMA 2000;283(22):2975–2978.[Abstract/Free Full Text]
8. Brown BW, Brauner C, Minnotte MC. Noncancer deaths in white adult cancer patients. J Natl Cancer Inst 1993;85:979–987.[Abstract/Free Full Text]
9. Black WC, Haggstrom DA, Welch HG. All-cause mortality in randomized trials of cancer screening. J Natl Cancer Inst 2002;94(3):167–173.[Abstract/Free Full Text]
10. Fontana RS, Sanderson DR, Woolner LB, Taylor WF, Miller WE, Muhn JR. Lung cancer screening: the Mayo program. J Occup Med 1986;28:746–750.[Medline]
11. Marcus PM, Bergstralh E, Fagerstrom RM, et al. Lung cancer mortality in the Mayo Lung Project: impact of extended follow-up. J Natl Cancer Inst 2000;92:1308–1316.[Abstract/Free Full Text]
12. Eddy DM. Screening for lung cancer. Ann Intern Med 1989;111:232–237.[Abstract/Free Full Text]
13. Kubïk A, Parkin DM, Khlat M, Erban J, Polak J, Adamec M. Lack of benefit from semi-annual screening for cancer of the lung: follow-up report of a randomized controlled trial on a population of high-risk males in Czechoslovakia. Int J Cancer 1990;45:26–33.[Medline]
14. Manser RL, Irving LB, Byrnes G, Abramson MJ, Stone CA, Campbell DA. Screening for lung cancer: a systematic review and meta-analysis of controlled trials. Thorax 2003;58:784–789.[Abstract/Free Full Text]
15. Reich JM. More on lung cancer screening [letter]. Chest 1995;108(1):294.[Free Full Text]
16. Reich JM. Improved survival and higher mortality: the conundrum of lung cancer screening. Chest 2002;122:329–337.[Free Full Text]
17. Patz EF, Swensen SJ, Herndon JE II. Estimate of lung cancer mortality from low-dose spiral computed tomography screening trials: implication for current mass screening recommendations. J Clin Oncol 2004;22:2202–2206.[Abstract/Free Full Text]
18. Gohagan JK, Marcus PM, Fagerstrom RM, et al. Final results of the Lung Screening Study, a randomized feasibility study of spiral CT versus chest x-ray screening for lung cancer. Lung Cancer 2005;47(1):9–15.[CrossRef][Medline]
19. Swensen SJ, Jett JR, Hartman TE, et al. CT screening for lung cancer: 5-year prospective experience. Radiology 2005;235:259–265.[Abstract/Free Full Text]

This article has been cited by other articles:

				J M Reich A critical appraisal of overdiagnosis: estimates of its magnitude and implications for lung cancer screening Thorax, April 1, 2008; 63(4): 377 - 383. [Abstract] [Full Text] [PDF]

				E. J. Wilck Computed Tomography Screening for Lung Cancer Ann. Thorac. Surg., February 1, 2008; 85(2): S699 - S700. [Full Text] [PDF]

				J. Reich, J. Asaph, D. J. Raz, and D. M. Jablons Natural History of Stage I Lung Cancer Chest, December 1, 2007; 132(6): 2062 - 2063. [Full Text] [PDF]

				W. C. Black, D. R. Aberle, C. D. Berg, and For the Executive Committee of the National Lung S Large Field Trial for Lung Cancer Screening: Putting the Wrong Cart before the Horse? Radiology, May 1, 2007; 243(2): 314 - 316. [Full Text] [PDF]

				H. Schoder and M. Gonen Screening for Cancer with PET and PET/CT: Potential and Limitations J. Nucl. Med., January 1, 2007; 48(1_suppl): 4S - 18S. [Abstract] [Full Text] [PDF]

This Article Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Reich, J. M. Search for Related Content PubMed PubMed Citation Articles by Reich, J. M. Related Collections Related Article