Baseline Findings of a Randomized Feasibility Trial of Lung Cancer Screening With Spiral CT Scan vs Chest Radiograph^*: The Lung Screening Study of the National Cancer Institute FREE TO VIEW

Lung cancer is the leading cause of cancer-related mortality in both men and women in the United States.¹ Although the overwhelming majority of lung cancer cases have a known and potentially preventable cause (ie, tobacco use), it is estimated that currently about one half of all lung cancer cases in the United States occur in former tobacco smokers, who remain at elevated risk for lung cancer years after quitting smoking.² Thus, although the control of tobacco use is a critical component in the fight against lung cancer (and other diseases), in the foreseeable future successful chemoprevention or early detection of lung cancer in persons with a history of tobacco use or other high-risk exposure will be necessary if lung cancer morbidity and mortality rates are to decline substantially.

Although several approaches for the early detection of lung cancer have been investigated in clinical trials,³⁴⁵⁶ to date none has been shown to be effective in reducing lung cancer mortality. The lung component of the ongoing Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial⁷ is currently evaluating the effect of posteroanterior chest radiograph (CXR) screening vs usual care on lung cancer mortality, however, the final results may not be available for several years. Low-radiation-dose spiral CT (LDCT) scanning, an advance in CT scan technology that was introduced in the mid-1990s, offers rapid image acquisition at radiation doses substantially below that of standard CT scanning, although at higher doses than a CXR.⁸ Findings from the Early Lung Cancer Action Project (ELCAP)⁹ published in 1999 and from a mass screening program in Japan reported in 1998¹⁰ demonstrated that LDCT scanning could detect lung cancer in asymptomatic subjects with greater sensitivity than could CXRs. The National Cancer Institute (NCI) began deliberations on the feasibility and design of a randomized controlled trial (RCT) to assess the impact on mortality of LDCT scan lung screening in 1999.

To assess the feasibility of conducting a definitive RCT of LDCT scanning, the Division of Cancer Prevention at the NCI conducted the Lung Screening Study (LSS), a pilot RCT of LDCT vs CXR. The primary goals of this project were to determine the following: (1) the feasibility of rapidly accruing high-risk participants who were not actively being screened with spiral CT scans into a study of lung cancer screening, (2) the willingness of study participants to be randomized to either an LDCT scan or a CXR arm and to undergo the appropriate examinations, (3) the likelihood that participants randomized to a CXR would subsequently receive a spiral CT scan examination on their own (and vice versa), (4) the prevalence of abnormal findings on baseline screening, and (5) the extent of diagnostic follow-up subsequent to abnormal screening findings.

The LSS began as a 1-year special project of the PLCO cancer trial.⁷ The PLCO cancer trial provided an established infrastructure, resulting in an acceleration of LSS activities and rapid data collection. The following 6 of 10 PLCO screening centers were chosen, in a competitive fashion, to participate in the LSS: Georgetown University Medical Center/Lombardi Cancer Center (Washington, DC); Henry Ford Health System (Detroit, MI); Marshfield Medical Research and Education Foundation (Marshfield, WI); University of Minnesota School of Public Health/Virginia Piper Cancer Center (Minneapolis, MN); Washington University School of Medicine (St. Louis, MO); and the University of Alabama at Birmingham (Birmingham, AL). Screening centers were responsible for all day-to-day study activities, including recruitment, screening, follow-up, and quality assurance. The NCI was responsible for scientific oversight and data analysis, while Westat, Inc (Rockville, MD), the coordinating center, was responsible for data management, including forms development, data entry, and systems development.

The LSS sought to recruit, randomize, and screen 3,000 persons who were at elevated risk for lung cancer in a short period of time. Recruitment was accomplished primarily by mass mailings, although other sources, such as posters, advertisements, and recommendations from practitioners, were employed. To be eligible for the LSS, an individual had to be between 55 and 74 years old at the time of randomization, to be either a current cigarette smoker or a former smoker who had quit within the previous 10 years, and to have at least a 30-pack-year history of cigarette smoking. Exclusion criteria included a history of a spiral CT scan examination of the lungs or thorax in the previous 24 months, a history of lung cancer, currently receiving treatment for any cancer other than non-melanoma skin cancer, removal of a portion of a lung or an entire lung, and participation in another cancer screening trial (including the PLCO cancer trial) or a primary cancer prevention trial other than a smoking cessation study. In addition, all persons were required to sign an institutional review board-approved consent form before undergoing randomization.

Once eligibility was established and consent was obtained, participants were randomized, using a secure web-based system that was maintained by the coordinating center, to one of the two study arms, LDCT scan or CXR. Randomization was stratified by age (in 5-year categories), sex, and screening center, using blocks of varying sizes.

LDCT scan examinations were obtained using the following technical parameters: 120 to 140 kilovolt peak, 60 mA, scan time of 1 s, 5-mm collimation, a pitch of 2 or equivalent (depending on the model and type of scanner), and contiguous reconstructions. In the case of obese participants, the milliampere setting could be raised as high as 120 mA. Images were obtained using a standard algorithm or were reconstructed using a high-resolution bone or lung algorithm. Lung windows (standard width, 1500 Hounsfield units [HU]; standard level −650 HU) and mediastinal windows (standard width, 400 to 500 HU; standard level, 10 to 30 HU) were provided for review, although standard settings could be adjusted to optimize viewings. The use of filters was optional, and, when filmed, a 15 to 20 on one format was used. CXR examinations consisted of a single posteroanterior view and were obtained using equipment with a high kilovolt potential (ie, 110 to 150 kilovolt peak) at a tube-to-receiver distance of 6 to 10 feet. Wide-latitude film with a 12:1 standard grid or higher was used. Screening film and computed radiography systems were used to perform the CXR examinations.

Radiologic technologists were responsible for performing all imaging examinations and were required to be certified by the American Registry of Radiologic Technologists. Scans were read by radiologists at each center who were board-certified by the American Board of Radiology or were board-eligible.

Standardized forms, one specific to each screening examination, were used to record screening abnormalities and results. Each LDCT scan examination was assigned one of the following six results categories: positive screen, abnormality suspicious for lung cancer; negative screen, no abnormality; negative screen, minor abnormality; negative screen, smooth noncalcified nodule ≤ 3 mm in diameter; negative screen, significant abnormality not suspicious for lung cancer; and inadequate. Abnormalities observed on LDCT scans that were suspicious for lung cancer included, but were not limited to, noncalcified nodules or masses > 3 mm in diameter, spiculated noncalcified nodules ≤ 3 mm in diameter, focal parenchymal opacification (ie, consolidation or ground glass attenuation), and endobronchial lesions. Each CXR examination was assigned one of the following five results categories: positive screen, abnormality suspicious for lung cancer; negative screen, no abnormality; negative screen, other abnormalities (referral optional); negative screen, other abnormalities (referral required); and inadequate. Abnormalities observed on a CXR that were suspicious for lung cancer included, but were not limited to, a nodule or mass, hilar or mediastinal lymph node enlargement (excluding calcified nodes), major atelectasis or lobar collapse, infiltrate, consolidation or alveolar opacity, and pleural mass.

The LSS protocol specified that participants be mailed their test results within 3 weeks of undergoing their examination, and that participants with positive screens or other clinically significant abnormalities be contacted initially by telephone and be urged to receive medical follow-up. Additional phone calls were made at 4 weeks (and at 8 weeks if follow-up had not begun at the time of the 4-week phone call) after the screening to ensure that diagnostic evaluation was underway. Referrals to specialists for the follow-up of positive screening results were provided if requested by the participant. The LSS protocol did not employ a common diagnostic algorithm, but most screening centers provided participants and their physicians with a description of the usual course of action recommended by their clinic if asked for guidance.

Diagnostic evaluation of positive screening findings was tracked by the collection and abstraction of medical records. Medical records collection began shortly after the notification of a positive screening finding and continued until a conclusive diagnosis was made or 12 months had passed after the positive screening finding. Information was collected on diagnostic procedures for lung cancer and on the possible complications of those procedures. For those persons in whom primary lung cancer was diagnosed, the tumor stage, grade, and histologic type also were abstracted.

To assess the crossover contamination (ie, the phenomenon whereby a study participant independently obtains, for screening purposes, the examination to which that person was not randomized), a health assessment questionnaire was mailed to a random sample of 240 CXR participants and 120 LDCT scan participants who had negative screening findings about 6 months after undergoing their examination. The questionnaire requested information on five medical tests, including CXR and spiral CT scanning. Participants were asked whether they had undergone that test since their LSS examination. If so, they were asked to indicate whether the test was received because of a specific health problem, as a follow-up to a previous health problem, or as part of a routine physical or screening examination. A screening examination was defined for the participants as a medical test used to detect a disease before symptoms have occurred. Additional details on the LSS can be found in the Manual of Operations and Procedures (http://www3.cancer.gov/prevention/lss/mooptoc.html).

Figure 1 shows the Consolidated Standards of Reporting Trials (or CONSORT)¹¹ flow diagram for the trial. The six LSS screening centers mailed 653,417 information packages, beginning approximately September 1, 2000. Eligibility was assessed for the 12,270 persons who contacted a screening center. Of these, 4,828 persons (39%) were deemed to be eligible. Only a small number of participants (148; 1.2%) were ineligible due to having undergone a spiral CT scan examination in the previous 24 months. Randomization began on September 5, 2000. By early November 2000, the target goal of 3,000 participants had been met, but randomization was extended through November 15, 2000, in order to accommodate eligible persons who had already been invited to participate. The final randomization total for the LSS was 3,409 participants. Ninety-one of these 3,409 participants (LDCT scan arm, 46 participants; CXR arm, 45 participants) were subsequently found to be ineligible, mostly (LDCT scan arm, 85%; CXR arm, 93%) because of participation in the PLCO cancer trial. The analysis below is limited to the 3,318 randomized participants who were eligible (LDCT scan arm, 1,660 participants; CXR arm, 1,658 participants).

The two study arms were essentially identical with regard to age, sex, and smoking history (Table 1 ). About 60% of participants were men and just < 60% were current smokers. Ninety-six percent of participants (1,586 participants) in the LDCT scan arm and 93% of participants (1,550 participants) in the CXR arm received their appropriate examination. Screening was completed on January 31, 2001.

Table 2 displays the following positivity rates for each randomization arm: overall; by age; by sex; by pack-years of smoking; and by smoking status. The overall positivity rate was 20.5% (325 of 1,586 participants) in the LDCT scan arm and 9.8% (152 of 1,550 participants) in the CXR arm (difference, 10.7%; 95% confidence interval, 8.2 to 13.2%). Positivity rates on LDCT scanning were higher in older participants and current smokers, and were slightly higher in men. A multivariable logistic regression model employing age, sex, pack-years of smoking, and smoking status as independent variables showed that older age (odds ratio [OR], 1.7) and current smoking (OR, 1.4) conveyed statistically significantly increased odds for a positive LDCT scan finding. In the CXR arm, the logistic model showed that pack-years of smoking (ie, ≥ 50 vs < 50 pack-years) and current smoking each conveyed borderline statistically significantly increased odds for a positive screening finding (OR for both, 1.4; p = 0.053 and p = 0.08, respectively).

Information on diagnostic follow-up was obtained for 316 of the 325 eligible LDCT scan arm participants with a positive examination finding (97%) and for 146 of the 152 eligible CXR arm participants with a positive examination finding (96%). Table 3 displays follow-up diagnostic procedures by study arm for those with follow-up information. Almost all participants with positive examination findings, 98% in the LDCT scan arm and 96% in the CXR arm, had at least one follow-up procedure. Clinical evaluation was performed in 77% of LDCT scan arm subjects and in 47% of CXR arm subjects, while comparison with a prior CT or radiograph was performed in 49% of subjects in each arm. A total of 73% of LDCT scan arm subjects and 52% of CXR arm subjects had at least one follow-up chest CT scan. The corresponding percentages for follow-up CXR were 29% (LDCT scan arm) and 47% (CXR arm). Pulmonary function tests were the next most common procedure performed in each arm of the study. Overall, 17% of subjects in the LDCT scan arm and 10% of those in the CXR arm with positive screening findings underwent at least one invasive procedure (see Table 3 for list of invasive procedures). In each study arm, roughly half of the subjects who had undergone invasive procedures ultimately received a diagnosis of lung cancer.

Lung cancer was diagnosed in a total of 30 LDCT scan arm participants and 7 CXR arm participants within 1 year of receiving their positive screening findings, giving a cancer detection rate among screened subjects of 1.9% in the LDCT scan arm and 0.45% in the CXR arm. Table 4 displays the rate of lung cancer diagnosis for different findings on LDCT scans among LDCT arm subjects with positive screening findings. The rate was 34.5% for a nodule or mass of ≥ 20 mm in diameter and 21.3% for a nodule 10 to 19 mm in diameter. In contrast, the lung cancer diagnosis rate was < 5% for a nodule < 10 mm in diameter and 7% for focal parenchymal opacification (Table 4). Overall, the lung cancer diagnosis rate among subjects with positive screening findings was 9.2%.

Table 5 shows the distribution of stage and histopathologic type for the lung cancers diagnosed following a positive screening finding. In the LDCT scan arm, 16 of the 30 cancers (53%) were at stage I (stage IA, 9 cancers; stage IB, 7 cancers), while in the CXR arm 6 of the 7 cancers (86%) were at stage I (stage IA, 3 cancers; stage IB, 3 cancers). Three stage II, six stage III, three stage IV, and two unstaged cancers also were seen in the LDCT scan arm, and one unstaged cancer was seen in the CXR arm. Nineteen adenocarcinomas (including bronchoalveolar and adenosquamous carcinoma) [63%], five squamous cell carcinomas (17%), three non-small cell carcinomas not otherwise specified (10%), two large cell carcinomas (7%), and one small cell carcinoma (3%) were observed in the LDCT scan arm of the study. In the CXR arm, four squamous cell carcinomas (57%) and three adenocarcinomas (43%) were observed.

Six participants in the LDCT arm (2%) and six participants in the CXR arm (4%) who received positive screening findings had medical complications that possibly were related to their diagnostic follow-up. Nine of these 12 participants (LDCT scan arm, 4 participants; CXR arm, 5 participants) received lung cancer diagnoses. The total number of distinct medical complications was 7 in LDCT scan arm participants and 11 in the CXR arm participants. Three participants in each arm of the study experienced pneumothorax, four CXR and two LDCT scan participants experienced infection or fever requiring antibiotics, one subject in each arm experienced atelectasis, two CXR participants experienced cardiac arrhythmia, one LDCT scan subject experienced a stroke, and one LDCT scan subject experienced acute respiratory failure. An additional LDCT scan (noncancer) subject died about 5 months after receiving a positive screening finding; however, this death does not appear to have been a result of the diagnostic follow-up. A further review of this subject’s records revealed that he had undergone a CT scan about a month after the screening and that comparison of the subject’s recent CT scans with a CT scan from 1998 revealed no radiographic changes and apparently no concern for malignancy. About 5 months after the screening, the subject was admitted to the hospital for increasing shortness of breath, and after thoracentesis and reaccumulation of pleural fluid, the subject became unresponsive and died.

Crossover contamination surveys were returned by 233 of the 238 eligible CXR participants (98%) and all 115 of the eligible LDCT scan participants to whom surveys had been sent. A total of 0.9% of CXR arm respondents reported having undergone a spiral CT scan as part of a routine medical or screening examination since their LSS examination, and 2.6% reported having undergone a spiral CT scan for any reason since their LSS examination. In the LDCT scan arm, 5% reported having undergone a CXR as part of a routine medical or screening examination since their study examination, and 13% reported having undergone a CXR for any reason.

The LSS demonstrated convincingly the feasibility of an RCT of lung cancer screening with LDCT scanning. Substantial enrollment and minimal crossover contamination provide compelling evidence that randomization was acceptable to persons who were at elevated risk for lung cancer and that background use of LDCT scanning among these individuals was quite low. Furthermore, the LSS demonstrated that accrual could be accomplished at a rate rapid enough to ensure that a definitive trial could be completed in a timely fashion.

In addition to demonstrating the feasibility of an RCT, the LSS also provided one of the first sets of baseline screening positivity and detection rates of an RCT of LDCT scanning. Our LDCT scan positivity rate of 20.5% is similar to that observed in one US (New York City) single-arm study (ELCAP, 23.3%)⁹ but is quite different from that observed in another US single-arm study (Mayo Clinic, 51%).¹² The LSS LDCT scan cancer detection rate on the baseline screening of 1.9% is similar to the baseline screening cancer detection rates from both the ELCAP study⁹ and Mayo Clinic study¹² (2.7% and 1.4%, respectively). Preliminary findings from a small ongoing RCT,¹³ also being conducted in the United States (Colorado), have been published as well. The authors report a positivity rate of 33% and a detection rate of 3.1%, based on 92 LDCT scan screenings. LSS CXR positivity and detection rates of 9.8% and 0.45%, respectively, also are similar to those observed in the ELCAP study⁹ (6.8% and 0.7%, respectively), the only other US LDCT scan study to incorporate CXR in any manner. Differences in positivity rates on LDCT scanning may have arisen from differences in lung cancer risk profiles, differences in radiologist reading practices, and geographic differences in other lung conditions.

In this study, 19% of screened subjects in the LDCT scan arm and 9% of screened subjects in the CXR arm received apparently false-positive screening findings (ie, a positive screen with no resulting lung cancer diagnosis). Since only a small fraction of these subjects underwent biopsies, it is possible that some may have had latent lung cancer. However, it is likely that the majority were truly free of lung cancer at the time of the screening. The high false-positive rate represents a public health burden for screening in terms of costs and medical complications of diagnostic follow-up, as well as possible emotional stress. This burden will have to be weighed against any benefits of LDCT scan screening (relative to CXR or to no screening) in reducing lung cancer mortality and/or morbidity in order to fully evaluate the usefulness of any lung cancer-screening program.

Lung cancer stage findings in the LSS were somewhat different from those reported in other studies. The percentage of cancers found on baseline screening that were at stage I was lower in the LSS LDCT scan arm (53%) than in the Mayo Clinic study (68%)¹² or the ELCAP study (85%).⁹ Differences also exist for histology. In the LSS LDCT scan arm, adenocarcinoma accounted for 63% of lung cancers, while the comparable figures for the Mayo Clinic study¹² and the ELCAP study⁹ were 71% and 93%, respectively. Note that all of these studies involved relatively few lung cancer cases (ie, 20 to 30 cases), so some of the above differences may be due merely to chance variation.

Six of the 7 lung cancers diagnosed in the CXR arm (86%) were classified as stage I, compared to 16 of 30 of the lung cancers diagnosed in the LDCT scan arm (53%). Although the higher proportion of stage I cancers in the CXR arm may seem puzzling, the absolute number of cases is small, and the difference in the proportions is not statistically significantly different from zero (p = 0.2 [two-sided Fisher exact test]).

Since the LSS was a feasibility study for a larger definitive RCT, the basic study design of the LSS, including the choice of a control group, was conceived to mirror as much as possible what was thought to be an optimal design for a definitive RCT. The choice of a control-arm regimen for a definitive RCT (and thus for the LSS) was not obvious. A usual-care approach, one in which participants randomized to the control arm receive neither an intervention nor a screening recommendation, was considered. However, we chose to use CXR for several reasons.¹⁴ Although previous RCTs of screening CXR did not show a reduction in lung cancer mortality, these studies were limited by their size and design and were thus unable to determine whether annual CXR (vs no screening) could reduce lung cancer mortality by a modest yet clinically relevant amount.⁵⁶¹⁵ Because the question of efficacy of an annual CXR is currently under investigation in the PLCO cancer trial, it was necessary to design a definitive RCT so that its results would be meaningful, regardless of the PLCO cancer trial outcome. Using CXRs for the control arm provided that assurance. If a benefit of CXRs over usual care is observed in the PLCO cancer trial, CXR would become the standard of care and an RCT of LDCT scanning vs CXR would then test the new modality (ie, LDCT scanning) against the standard of care. A lack of benefit for CXR in the PLCO cancer trial would allow equivalence of CXR with a usual-care regimen, thus making the inclusion of CXR inconsequential, unless CXR significantly increases the risk of lung cancer-related deaths, which is an unlikely situation. A three-arm trial (ie, LDCT scanning, CXR, and usual care) would have provided the most definitive comparisons, but the incremental gain in rigor was not thought to warrant the additional resources necessary for such an undertaking.

The LSS was highly successful. We did, however, experience one minor operational difficulty due to our use of the established PLCO cancer trial screening network. A small number of participants (< 3%) were found to be ineligible, primarily because of enrollment in the PLCO cancer trial. Although potential LSS participants were asked whether they were participating in the PLCO cancer trial, some thought that they were not participating because they either had completed their screening examinations or had been randomized to the control arm. Although their inclusion was unfortunate, data analyses both with and without the ineligibles produced similar results.

The success of the LSS resulted in the establishment of the National Lung Screening Trial (NLST),¹⁴ a large RCT of LDCT scanning vs CXR screening for lung cancer. The NLST will enroll 50,000 high-risk, heavy smokers (and former heavy smokers who quit within 15 years from randomization), aged 55 to 74 years. Participants will receive an initial screening and two subsequent annual screenings, and will be observed for a minimum of 4.5 years. The trial is powered to detect a modest but clinically relevant reduction of 20% in lung cancer mortality with LDCT scan screening. Randomization for the NLST began in September 2002 and is scheduled to conclude in August 2004. Interim analyses are proposed to begin in 2005, with the final analyses currently scheduled for 2009.

Susan Ascher, William Bailey, Brenda Brewer, Timothy Church, Deborah Engelhard, Richard Fagerstrom, Mona Fouad, Matthew Freedman, Edward Gelmann, John Gohagan, William Hocking, Subbarao Inampudi, Brian Irons, Christine Cole Johnson, Arthur Jones, Barnett Kramer, Gena Kucera, Paul Kvale, Karen Lappe, Lisa McFarland, William Manor, Pamela Marous, Alisha Moore, Hrudaya Nath, Sarah Neff, Martin Oken, Paul Pirsky, Michael Plunkett, Helen Price, Philip Prorok, Douglas Reding, Thomas Riley, Martin Schwartz, Richard Slone, David Spizarny, Roberta Yoffie, and Carl Zylak.