Lung Cancer Screening, Overdiagnosis Bias, and Reevaluation of the Mayo Lung Project

The role of screening for lung cancer has been debated for more than 50 years. In 1949, the first lung cancer survey study including more than 530 000 persons living in Boston was initiated ( 1 ) , and in 1951, the first prospective screening trial, known as the Philadelphia Pulmonary Neoplasm Research Project, began ( 2 ) . After the National Cancer Act in 1971, there was renewed interest in lung cancer screening with chest radiographs and sputum cytology that gave rise to three large randomized trials in the United States, including the Mayo Lung Project ( 3 – 5 ) .

When the data from the Mayo Lung Project were analyzed in the 1980s, the screened group, as compared with the usual-care arm, had more resectable tumors, more early-stage tumors, and increased survival rates, but no difference was found in overall mortality—the ultimate goal of a screening program. It is of interest that 206 cases of lung cancer were diagnosed in the screened group, compared with 160 cases in the usual-care arm. The most likely explanation for the 22% (46 of the 206 cases) excess cases of lung cancer in the screened group was that some screened detected tumors were indolent, so that these participants could have died with the disease rather than from the disease (i.e., overdiagnosis bias). In other words, these tumors may never have become clinically apparent or detected if these patients were not in the screened arm of the trial.

Critics argued that the Mayo Lung Project did not have a sufficient follow-up period to confirm that overdiagnosis bias was responsible for the additional lung cancers in the screened group. They predicted that the number of lung cancer patients in the control group would eventually catch up, because the two arms of this randomized study should be identical. This hypothesis, however, is not supported by the data from Marcus' article in this issue of the Journal ( 6 ) . Marcus et al. performed a remarkable extended follow-up study through 1999 of patients enrolled in the original Mayo Lung Project. Instead of the theoretical convergence of the number of lung cancer cases in the screened and control groups, they found an increasing discrepancy between the two trial arms. The explanation for the increase is uncertain, although this compelling evidence supports the author's conclusion that overdiagnosis bias must be considered in lung cancer screening trials.

Although these trials have taught us about the process of screening and associated biases, the radiographic appearance of lung cancer, and the patterns of metastasis, we must pause to reconsider whether anatomic imaging studies are the optimal approach for early detection of lung cancer. This doubt was foreshadowed in the preliminary report of the Philadelphia Pulmonary Neoplasm Research Project ( 2 ) as follows:

“There is an impressive contrast in well conducted surveys between the detection of curable tuberculosis and that of curable lung cancer. A significant amount of survey-found tuberculosis is minimal in extent and, therefore, amenable to present-day treatment, whereas, despite immediate emergency referral of persons whose photofluorograms were read ‘suspect neoplasms’ at two official Philadelphia chest x-rays units, resection could be done in only 30% of a series of 100 consecutive cases.”

The issue of overdiagnosis bias is also relevant to the trial design and to data interpretation of more recent low-dose spiral computed tomography (CT) screening studies ( 7 – 9 ) . The original premise for screening with CT was that, compared with conventional chest radiographs, it would detect smaller, earlier-stage lung cancers and that this early detection would translate into a dramatic improvement in patient outcomes. Unfortunately, initial trials were designed as single-arm observational studies; CT scans were performed in all participants without a control group. Researchers argued that a randomized study would be too expensive, take too long, and require too many participants. However, after a decade of enrolling tens of thousands of individuals, there are still no published data demonstrating that CT is effective in improving outcomes for patients with lung cancer.

In addition to the presumed practical limitations of a randomized trial, and central to the argument for an observational study, was the hypothesis that all lung cancers were clinically significant and uniformly fatal. It was assumed that there was no need to consider overdiagnosis bias in CT screening trials.

Given the results of the Mayo Lung Project, it is not clear why the debate over overdiagnosis developed because there are convincing data to show that this bias could not be dismissed. In addition, autopsy series unquestionably confirm that some individuals have undiagnosed lung cancer at postmortem and that the cause of death is unrelated to cancer ( 10 – 14 ) . Some individuals, even those enrolled in screening trials, will live and die with their lung cancer, although the degree and impact on screening remains uncertain.

It is intriguing to analyze the preliminary results with respect to overdiagnosis emerging from a present-day observational CT screening trial, because these data parallel the original observations from the Mayo Lung Project. The investigators of an observational CT study recently reported the radiologic and pathologic findings for its 69 patients with stage I lung cancer at baseline and the potential for overdiagnosis bias ( 15 ) . The authors ( 15 ) stated that “overdiagnosis of the disease is uncommon, with the cases presenting as a nonsolid nodule a possible exception to this.” Previous reports ( 16 – 19 ) support the concept that lung cancers with the appearance of a nonsolid nodule in a CT scan may have an indolent phenotype and that the tumor may not affect the individuals' lifespan. Indeed, 15 (22%) of the 69 lung cancers from the observational CT screening trial presented as a nonsolid nodule, again implying that overdiagnosis bias is clinically significant and cannot be overlooked. The 22% of potential excess cases in the current observational CT trial and identical confirmed 22% excess lung cancers in the original Mayo Lung Project support the theory that clinical indolent tumors are detected in screening trials.

To date, substantial resources have gone into screening for lung cancer. Despite the use of these various imaging modalities, trial designs, and the inclusion of hundreds of thousands of individuals, none of these studies have shown a reduction in lung cancer mortality. The ongoing National Lung Cancer Screening Trial has randomly assigned more than 50 000 participants to CT or control chest radiograph arms and continues to follow patients. Preliminary results will be available in the next several years ( 20 , 21 ) . As the debate over lung cancer screening continues, it has become apparent that we must learn more about the biology of this disease and integrate this knowledge with early diagnostic strategies. Although we all want to reduce mortality from lung cancer, the theoretical benefits of screening must be validated by appropriate, rigorous clinical trials before being introduced into routine practice.

References