Preliminary Meeting to Discuss Trial Issues
April 29, 1999

The meeting was opened by Christine Berg, M.D., Acting Chief, Lung and Upper Aerodigestive Cancer Research Group, Division of Cancer Prevention, National Cancer Institute (NCI) who thanked the participants for attending this preliminary meeting to discuss the potential need for a study to assess the role of spiral computed tomography (CT) in the screening of early lung cancer. Welcoming comments were also made by Peter Greenwald, M.D., Dr. P.H., Director and Barry Kramer, M.D., M.P.H., Deputy Director, Division of Cancer Prevention, NCI. Both confirmed the importance of the discussions being held.

Presentation of Information on Spiral CT

Dr. James Jett of the Mayo Clinic opened by saying that a tremendous amount of interest exists regarding the subject of spiral CT and lung cancer detection. Each year, 170,000 new cases of lung cancer are diagnosed. This disease has the highest cancer mortality, approximately 160,000 deaths per year occur in men and women in the United States. Only a slight increase (2%) has occurred in the five-year survival rate in the last 30 years. However, if detected in the presymptomatic state, the survival rate for lung cancer is twice that of symptomatic disease. The survival rate is approximately 60% when lung cancer is detected in the early stages. However, only approximately 15% of lung cancers are diagnosed at an early stage with the disease still localized.

Several articles in the literature report on previous studies conducted on lung cancer screening. In the U.S., decreased lung cancer mortality was not shown to result from screening with standard chest radiography. Japan is currently the leader in testing spiral CT as a lung cancer screening tool. Some studies have demonstrated that spiral CT can detect small nodules poorly visible on chest X-ray (CXR) in the lung. An advantage of spiral CT is the speed of the test compared to traditional CT. An entire scan can be completed in one breath hold, approximately 15 seconds.

Dr. Jeff Sloan, also of the Mayo Clinic, discussed the level of effect size for typical screening trials. The end point in the majority of studies is mortality. Data were presented to illustrate the statistical power reached in a study depending on the sample size to be compared to a control group with a mortality rate of four deaths per 1,000 person/years. Since lung cancer has such a high incidence and case-fatality rate, even a small mortality reduction would be significant, and could be approximately equivalent to curing some cancers or greatly reducing the mortality rate of others.

The Mayo Clinic group presented a proposal of a randomized comparison of Spiral CT versus Chest X-ray for lung cancer screening. The study would be conducted on a high-risk group, i.e., participants age 50 and older who are current smokers or people who stopped smoking within the past ten years. The study would need to screen 38,000 participants. Sample size was based on the results providing a 90% statistical power to detect a 20% reduction in lung cancer mortality.

Claudia Henschke, M.D., Ph.D. of the New York Hospital/Cornell Medical Center presented information about the Early Lung Cancer Action Project (ELCAP). Early studies in lung cancer detection have not shown reductions in mortality using traditional screening tools. However, ELCAP screened 1,000 smokers and former smokers and found that spiral CT could detect lung cancers that were not visible on chest X-ray and over 80% of these detected cancers were at Stage IA.

The New York group plans to screen a total of 5,000 high-risk individuals with spiral CT at baseline with a one-year repeat screening. (By the end of 1999, the one-year rescreening will be completed on all subjects.) Similar studies are underway in Germany, Israel, Florida and Minnesota. Eligibility criteria include: age 60 and above, 10 pack-years of smoking, and no currently diagnosed cancers. Nodules detected in participants at screening will be examined by high resolution CT or biopsy depending on size of the nodule.

Biomarkers and Sputum Cytology as Screening Methodologies

Dr. James Mulshine, Head, Experimental Intervention Section, Cell and Cancer Biology Department, Medicine Branch, Division of Clinical Sciences, NCI discussed biomarkers and sputum cytology as possible screening tools. At present lung cancer is mainly treated in the advanced disease state. Identifying molecular markers that could detect the disease while it is still confined to the bronchial epithelium would potentially allow cures with local modalities.

An NCI-sponsored large-scale early lung cancer screening trial combined sputum cytology with chest radiograph. No survival outcome benefit was found. Sputum samples of the participants, which showed dysplastic changes, in the Johns Hopkins Lung Project, were archived. These samples were used to attempt to identify molecular markers useful in early detection. Some work has been done in sputum immunostaining using two monoclonal antibodies that has shown a correlation between positive staining and later development of lung cancer in the sampled population of subjects with dysplastic cytologic abnormalities. Some preliminary data has demonstrated that over-expression of a tumor-associated antigen, identified as heterogeneous nuclear ribonucleoprotein (hnRNP) is expressed in many lung cancer cases throughout normal appearing lung epithelium. This could be a promising area for further investigation. Once definite predictive markers are identified, they provide a noninvasive procedure for early detection. Further investigation is needed regarding the specificity of the sputum biomarkers testing in a screened population.

If biomarkers-based screening is determined to be accurate, there are some favorable implications. For example, this could lead to identifying disease at a much more treatable stage. The problems of treating possibly multiple lesions in a field carcinogenesis state could be solved by exploring the possibility of the aerosol delivery of retinoids. Aerosol delivery would be a direct delivery mechanism to the tracheobronchial tree. However, further investigation is needed regarding issues about dose, schedule and duration of therapy.

The American College of Radiology Imaging Network (ACRIN) Proposal

Daniel Sullivan, M.D., Associate Director, Diagnostic Imaging Program, Division of Cancer Treatment and Diagnosis, NCI introduced Denise Aberle, M.D., chairperson of the Lung Committee of the Diagnostic Imaging Network. She presented an overview of lung cancer incidence and prognosis. Risk factors include smoking (85% of lung cancers are found in smokers), advanced age, ethnicity (African American have high incidence), and air flow obstruction. Previously conducted screening trials which include Johns Hopkins, Memorial-Sloan Kettering, Mayo Clinic Lung Projects, and a Czechoslovakian trial did not show a decrease in cancer mortality. Although chest X-ray has not definitively been shown to be an accurate screening tool, helical CT is indicating advances in lung nodule detection. Concerns exist about cost, interpretation requirements, and false positives with this technology. Sputum analysis has the ability to identify genetic abnormalities associated with premalignancy in lung cancers.

Dr. Aberle presented a proposal on behalf of the American College of Radiology Imaging Network to study contemporary screening tools for lung cancer. The objectives of the study include determining if contemporary screening techniques can detect early stage cancer, whether early detection can reduce surgical stage and tumor size, and to examine quality of life and screening costs. The five-year, multi-site study would randomize participants to either a screen or control arm. Participants would be high-risk individuals age 55-70; current or prior heavy smokers; or individuals with a previous curatively resected Stage I lung cancer or an upper respiratory neoplasm. Screening participants would receive chest radiographs, low-dose helical CT and sputum analysis for lung cancer-associated molecular markers at six month intervals. Both groups would complete annual questionnaires on health status and interval behavioral practices, i.e. smoking. Strategies for the management of participants with abnormal screening tests were outlined.

Concerns related to the study were discussed. The first issue is accrual which could be problematic since the study design requires consent prior to randomization and participants may be unwilling to consent to not receive screening. Another concern is determining the appropriate sample size needed to produce sufficient numbers of cancers for meaningful outcome measurement. Contamination in the control arm may also be an issue, as these technologies become more available. Since the study would be conducted in multiple sites, standardization of staging and treatment could be an issue. Also, using several screening tools could possibly obscure the benefit of the individual tests.

Dr. Anthony Miller led a general discussion about the timeliness of undertaking a study of spiral CT. He stressed that this meeting was preliminary. Spiral CT is becoming more available and therefore any opportunity for conducting a study of screening effectiveness would be more difficult in the future once dissemination of the technology had occurred. There was discussion about the gaps in information regarding the specificity of spiral CT.

Statistical Methodologies

Dr. Jack Cuzick explained the classical approach to clinical trials. In the classical approach design is rigid and the power of the study is determined by a chosen end point. Some strengths of a randomized population-based trial include providing unbiased results and being directly applicable to the general population. However, such trials are expensive to conduct and require a high level of compliance to be valid.

Various modifications can be applied to the classical approach. However, some modifications may lead to biases, which could affect the results of the study. Examples include using nonrandomized populations such as historical controls, conducting a comparison of compliers versus noncompliers, as well as conducting case control studies. Other examples of approaches that could lead to bias are using survival or stage changes to measure effectiveness, or conducting a subgroup analysis of those individuals who responded positively to the screening techniques.

Modifications to the classical approach, which may have more favorable results, include selecting a compliant population prior to randomization, by indicating in the recruitment the need for randomization, thereby selecting those individuals willing to participate in such a study. Another approach would be to conduct a method of analysis that could potentially adjust for non-compliance and contamination (see Cuzick, J, Edwards, R and Segnan, N Statistics in Medicine 16:1017-1029,1997). In the classical model, a randomized study of mortality is the recommended design.

An alternative view was presented by Dr. Olli Miettinen. Rather than undertaking a randomized, controlled study, a non-comparative study design could be utilized. Instead of comparing a control group to a screened group, all subjects would be screened and those individuals with tumors could be studied. The study would look at tumor size at initial prevalence screen detection and at subsequent screens. The patients are then followed to establish the cure rate. This design would require a smaller cohort group in comparison to the sample size needed in a randomized controlled trial and therefore, cost would be reduced. Nonadherence would also be a lesser threat to the study validity. A requirement for validity of the design is that overdiagnosis did not occur. In his opinion, this alternative type of study would be meaningful and efficient.

Dr. William Black presented a decision model for screening studies. Randomized, controlled trials have some limitations such as non-compliance, and problems with generalizability of results. A large sample size is required depending on the questions to be addressed by the study. In randomized controlled trials, some decisions are made about multiple study factors such as timing; treatment complications, the importance and weights assigned to study factors. A simple natural history model was presented. Modeling becomes more complicated as factors are added to make the model more comprehensive. Modeling can be used to generalize results from randomized controlled trials and to design studies. In addition, modeling may be useful to assess the potential impact in terms of years of life gained with a screening intervention.

Dr. Anthony Miller summarized some of the items from the meeting. Spiral CT appears to provide advances in sensitivity and in the early detection of lung cancer. At present, uncertainty exists about its specificity, which may be no better or possibly worse than chest radiography. The technology seems to advance the time of diagnosis however there are no estimates of the long-term impact on survival. The issue of biomarkers is still a question. The group felt that the collection and archiving of samples such as sputum and blood for future testing of biomarkers in any study undertaken would be useful. Evaluation of the treatment of early Stage I disease that may be detected with spiral CT would be needed to begin to determine appropriate treatment guidelines.

Designs for possible studies were discussed. The importance of quality control in any methodology used would be essential. Issues regarding the necessity of testing screening methods were discussed. Lung cancer is a disease with a high mortality rate. Therefore evaluation of new technologies that may detect it at an earlier, more curable stage is essential.

Conclusions

The group was very enthusiastic about the concept of spiral CT. NCI will explore the possibilities for further study of this potentially promising technology.