Screening for Lung Cancer
Claudia I. Henschke
David F. Yankelevitz
Olli S. Miettinen
WHY CONSIDER SCREENING
Lung cancer makes, at present, for quite a sad chapter in pulmonary medicine. In the United States, some 170,000 cases are diagnosed annually1; the average cost of care per patient is about $50,0002,3,4; and yet, the annual number of deaths from this disease is almost as high as the number of cases diagnosed, some 160,000.1 This is to say that, despite the very costly care, the case-fatality rate—the proportion of cases that are fatal—is near 95%.
While the overall case-fatality rate remains dismal, cases diagnosed in stage I—before clinically manifest metastases—are quite commonly curable and, thus, nonfatal. Depending on the size of the tumor at diagnosis, the curability rate ranges from some 50% to more than 90%5,6,7,8 (see Chapters 30 and 32). Thus, the problem with lung cancer now is that those stage I diagnoses remain quite uncommon, representing only some 15% of all diagnoses of lung cancer.9
The solution to this problem potentially is the pursuit of early, latent-state diagnosis in persons at relatively high risk for lung cancer; that is, screening high-risk people for lung cancer, before any overt, clinical manifestations of the disease.
That some increase in stage I diagnoses can be achieved by means of a suitable regimen of screening is obvious; but the question is whether the attainable increase is substantial enough to justify the screening, especially for people with only moderately elevated risk for lung cancer. It is thus important to know the magnitude of the stage shift, notably the increase in the proportion of stage I diagnoses, and this in reference to a well-thought-out, realistic regimen of the screening. For, the attainable rate of curability is, to a close approximation, the proportion of stage I diagnoses multiplied by the curability rate of stage I cases, specifically such stage I cases that are diagnosed in the context of the screening—asymptomatic and with the tumor typically smaller than in stage I diagnoses in the absence of screening.
Screening for lung cancer has been of interest not only because of the generally dismal prognosis in the absence of screening but also for two other reasons: highly discriminating risk assessment is possible to identify those at high risk, and the small, latent-state lung cancers tend to be relatively well identifiable against the backdrop of the airy parenchyma of the lungs in radiographic imaging, particularly in computed tomography (CT) imaging.
Interest particularly heightened when several studies demonstrated the considerable superiority of CT over traditional radiography (chest x-ray [CXR]) in the identification of small pulmonary nodules.10,11,12,13 What is more, research on CT screening for lung cancer has already led to quite a well-established regimen for it,14,15 and the research also has produced evidence indicating the attainability of quite a high rate of curability of this, thus far, near-uniformly fatal disease.16,17,18 Thus, the time has come for physicians to consider CT screening for lung cancer of persons at high risk for the disease or for persons who are asking their doctors about it. These decisions should be made on a case by case with the individual at issue suitably informed by his or her doctor.
Research on Screening for Lung Cancer In the early 1970s, the National Cancer Institute (NCI) funded a screening trial for lung cancer.19 In this trial, sputum cytology was the screening test and half of the 30,000 high-risk participants were to be randomly assigned to the “intervention” (sputum cytology every 4 months for 6 years) and half to the “control” (no screening). All participants were to have annual CXR. This study evolved into three separate ones, each having about 10,000 participants: the Memorial Sloan-Kettering Lung Project (MSKLP),20,21 the Johns Hopkins Lung Project (JHLP),22,23 and the Mayo Lung Project (MLP).24,25 The MSKLP and JHLP performed the study as planned for 6 years. The MLP investigators, however, wanted to test both sputum cytology and the CXR and thus developed a different protocol. They first screened all of the 10,933 participants using both sputum cytology and CXR, and then assigned 9211 people with no evidence of cancer on the baseline round to either receive sputum cytology and CXR every 4 months for 6 years
or to no screening; all received the usual Mayo Clinic advice of having annual screening.
or to no screening; all received the usual Mayo Clinic advice of having annual screening.
At the completion of these studies in the late 1970s, none showed a reduction in lung cancer mortality because of screening using sputum cytology. Beyond this, because of the results of the MLP, CXR was also deemed not to be useful, even though the investigators themselves26,27,28 as well as independent experts29,30,31 judged the MLP to be inconclusive and seriously flawed. Further, the International Union Against Cancer (UICC) workshop on screening for cancer32 in 1984 concluded that “the effectiveness of annual CXR in reducing lung cancer mortality was not evaluable in these trials. A case-control study was proposed to attempt a relatively quick evaluation, to be followed by a randomized trial if indicated. A search for new screening procedures is warranted.”
Subsequently, in Japan where screening with CXR continued as a matter of national policy, five case-control studies were performed.33,34,35,36,37 These five studies considered the timing of diagnosis with respect to the screening test, accumulated many more lung cancers than have been found in the randomized trials, and provided convincing evidence that deaths from lung cancer were less common if the diagnosis was achieved within 12 months of having CXR, but not when the delay was longer.
Ultimately in the United States, the NCI funded another screening trial to assess the benefit screening for prostate, lung, colorectal, and ovarian cancer, known as the PLCO trial.38 For lung cancer, participants were randomly assigned to receive the “intervention” (baseline and two annual CXRs) or no screening. It was started in 1993, but the conclusions of this trial have not yet been reported.
Interestingly, at the same time as the lung cancer trial was being planned in the early 1970s, the NCI also funded a trial to assess the benefit of screening for colorectal cancer.39 This trial enrolled 45,000 participants, 15,000 received annual screening, 15,000 biannual screening, and 15,000 no screening, but even with the greater number of participants, it did not show a mortality reduction after 5 years of screening and further follow-up. But for this trial, different from the lung cancer trials, the decision was made to provide another 5 years of screening and ultimately some 20 years after the trial started, it demonstrated a significant mortality reduction, improved survival rate, and a decrease in the incidence of late-stage cancers due to screening.39,40 Since then, many advances in screening for colon cancer have been introduced, and they are accepted and reimbursed without any further evidence from randomized trials.
Research in the CT Era In 1993, we initiated the Early Lung Cancer Action Project (ELCAP) for research on CT screening for lung cancer.10,11,41 To us, different from the prior randomized trials, screening is not a single test nor an intervention, rather it is a sequential process of pursuing early, latent-stage diagnosis of the cancer in order to provide for early treatment of it.42 From this vantage, we saw it necessary to first endeavor to develop a justifiable regimen for the diagnostic process; that is, a suitable definition of the initial test, of its positive result, and of the workup that is to follow that positive result, possibly leading to diagnosis of the cancer.42 Thereupon, the principal concern was the diagnostic performance properties of the regimen, and updating of this regimen based on emerging evidence and new technologies. Secondarily, there was going to be need for prognostic research, focusing on the curability of screen-diagnosed cases of lung cancer, stage I cases in particular.
Initially, the first version of the CT screening regimen was compared with its CXR counterpart, applying both to all participants in the study.11 Each regimen’s diagnostic performance was addressed in terms of the proportion of stage I diagnoses among all diagnoses and the proportion of screen-diagnoses among all diagnoses. In the baseline round, 29 cases of lung cancer were diagnosed, 27 of them screendiagnosed, two interim-diagnosed. Of the 29, 25 were in clinical stage I, all of them screen diagnosed.11 CXR screening identified only seven of the 27 cases of screen diagnosed by the CT regimen, and only 4 of the 23 stage I cases. Consequently, only CT was used in the 1184 repeat screenings.41 In these repeat screenings, seven cases of lung cancer were diagnosed, and there were no interim diagnoses. Of the seven, six were in clinical stage I.
Subsequent studies of the expanded ELCAP in New York State43 and then throughout the world44 showed that the proportion of diagnoses in clinical stage I has remained high, around 85%, for both baseline and annual repeat rounds of screening, with rare interim diagnoses, confirming the initial ELCAP results.
Prognostic research as to the curability of screendiagnosed cases of lung cancer was provided after long term follow-up of the diagnosed cases of lung cancer. This research demonstrated an estimated curability rate of 80% for all those diagnosed with lung cancer, regardless of stage and treatment. If diagnosed in clinical stage I and promptly resected, the estimated rate was 92%.44 The high curability rates are not surprising as others previously had shown that when lung cancer is diagnosed when it is still small and in stage I, it is highly curable.5,6,7,8
The high proportion of stage I diagnoses and the high estimated curability rate have raised the concern by some that these may be due to “overdiagnosed” lung cancers,45 meaning that CT screening identifies slow-growing cancers which, if not resected, would not lead to death. The I-ELCAP protocol aims to minimize “overdiagnosis”15 by requiring documentation of growth of small nodules prior to biopsy, biopsy diagnosis prior to resection, review of the resected specimens by a panel of expert pulmonary pathologists, and by addressing the outcomes of patients diagnosed in stage I but not treated. The review by pathologists who are experts in pulmonary pathology has confirmed that all diagnosed with lung cancer had genuine lung cancers whose pathologic criteria met the World Health Organization (WHO) criteria of malignancy,46 and patients diagnosed with stage I lung cancer who had no treatment, all died of it.44
Other studies had also shown that if left untreated, stage I lung cancers identified in the absence of screening, are usually fatal47,48 and so are stage I cases diagnosed by CXR screening.49,50,51 If, nevertheless, substantial concern about overdiagnosis still exists, then a randomized trial could ethically be performed by randomly assigning patients diagnosed with potential overdiagnosed lung cancers to either immediate treatment or delayed treatment. For early prostate cancer, such a trial was performed and it demonstrated that even for a cancer with a much lower fatality rate, surgical resection was significantly better.52
Slower-growing cancers are proportionately more common among cases diagnosed in the baseline round of screening, a phenomenon that has been termed length bias.45 This bias is reflected by the pathologic subtypes of cancers diagnosed in the baseline round in which a higher proportion of adenocarcinomas and lower proportion of squamous and small cell carcinomas were identified as compared with the repeat rounds.46 Further insight as to which cancers might be slower growing was provided by the initial 1000 ELCAP participants as we had identified a higher proportion of cancers manifesting as subsolid (nonsolid and part-solid) nodules in the baseline rounds than in the repeat rounds.53,54 The significance of finding subsolid nodules, previously termed ground-glass opacities, relative to lung cancer had not been fully appreciated (see Chapter 33). Remarkably, the rate of malignancy was significantly higher for part-solid nodules than for either solid or nonsolid ones.53 We also found that the distribution by type of malignancy was very different, with the malignancies manifesting as subsolid nodules either being adenocarcinomas with bronchioloalveolar features or adenocarcinomas-mixed subtype, whereas malignancies manifesting as solid nodules included the entire spectrum of the cell types of lung cancer with the exception of adenocarcinoma with bronchioloalveolar features.46 Analyses of the growth rates of adenocarcinomas has allowed us to identify those manifesting as nonsolid nodules as being slower growing lung cancers.54