| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
* From the Center for the Biology of Natural Systems (Drs. Markowitz and Morabia, and Ms. Manowitz)), Queens College, City University of New York, Flushing, NY; Division of Pulmonary Medicine (Dr. A. Miller), Catholic Medical Center, Queens, NY; Department of Radiology (Drs. Sider and J. Miller), Beth Israel Medical Center, New York, NY; and United Steelworkers (Ms. Kieding), Pittsburgh, PA.
Correspondence to: Steven B. Markowitz, MD, Queens College, Fourth Floor, 163–03 Horace Harding Expressway, Flushing, NY 11365; e-mail: smarkowitz{at}qc.cuny.edu
Abstract
Study objectives: Low-dose helical CT scanning identifies early stage lung malignancies and also a large proportion of lung nodules of uncertain diagnostic and prognostic significance (ie, indeterminate nodules). The sensitivity, specificity, and predictive value of these indeterminate nodules detected by CT scanning as part of a lung cancer screening program is largely unknown. We therefore calculated the sensitivity, specificity, and predictive values of CT-detected lung nodules that were followed up at least 18 months.
Design: Single-arm screening trial with longitudinal follow-up.
Setting: Rural areas of United States, from 2000 to 2004.
Participants: Former and current nuclear weapons workers, 
45 years old, including smokers and never-smokers, with variable exposure to occupational lung carcinogens.
Interventions: A total of 4,401 participants were CT scanned for lung cancer with an initial full chest low-dose CT scan, interval CT scans at 3, 6, and 12 months for indeterminate lung nodules (eg, nodules not immediately suspicious for lung cancer), and a 18-month, full-chest, low-dose incidence CT scan.
Results: We achieved follow-up for a minimum of 18 months for > 95% of 807 participants with indeterminate or suspicious lung nodules. Only 3 of 727 indeterminate nodules were identified as being malignant during the subsequent 18 months. The radiologist’s designation of a nodule as suspicious had a sensitivity of 84.2% and a specificity of 96.6%. Given a prior probability of lung cancer of 2.4%, positive and negative predictive values were 37.2% and 99.6%. Overall, we detected 33 primary lung cancers, including 19 stage I cancers, 5 stage II cancers, 7 stage III-IV cancers, and 3 limited-stage small cell cancers.
Conclusions: Helical CT scanning detects many indeterminate nodules, but few are malignant. CT scanning has high sensitivity and specificity to detect early lung cancer. The problem of false-positive results in helical CT scanning is limited and can be rationally managed. Current CT follow-up recommendations are supported.
Key Words: CT scan • indeterminate nodules • lung cancer • screening
Recent studies in Japan, the United States, and Italy123456 have established that the low-dose helical CT scan can detect early stage lung cancer that is amenable to resection. The large majority (60 to 85%) of lung cancers detected by CT screening were stage I or IIa at diagnosis. These studies occurred at academic medical centers and principally evaluated high-risk populations, defined as people at least 50 or 60 years old who had smoked cigarettes.
A salient problem in the use of low-dose CT scan for early lung cancer detection is the uncertainty that attends the high proportion of participants who have indeterminate nodules (eg, nodules not immediately suspicious for lung cancer). Most indeterminate nodules are not malignant, and there is a critical need to avoid potential harm that may derive from excess radiation, anxiety, and invasive workup of false-positive CT findings. Due to this uncertainty, initial CT screening protocols recommended frequent follow-up CT scans.278 Indeed, these initial screening protocols echoed recommendations for follow-up of lung nodules detected incidentally during clinical workup (ie, nonscreening context).91011
More recently, investigators have factored in nodule size when recommending frequency of follow-up CT scans; Henschke and colleagues12 no longer recommend any interval scan (ie, prior to the annual 12-month scan) for nodules < 5 mm. Similarly, a committee of the Fleischner Society made the same recommendation for small lung nodules (< 5 mm) that are detected in nonscreening settings.13 However, studies1214 that specify the probability that nodules < 5 mm are malignant are few.
Lung nodule morphology viewed on CT scan is a second key determinant of likelihood of malignancy.12141516 Morphologic characteristics that are suggestive of malignancy are well recognized,1516 although the CT screening trials completed to date provide little detail about this aspect of detected nodules, especially the relation between size and morphology.
The current study employs frequent CT scans and near complete follow-up of the cohort to provide valuable information about the nature of CT-detected indeterminate nodules. The screening population consists of rural uranium plant workers, with a broader range of lung cancer risk factors than heretofore studied in the United States, including a greater age span, nonsmokers and former smokers who had quit many years previous to examination, and potential occupational exposures to lung carcinogens.
Materials and Methods
Study Population
The Early Lung Cancer Detection (ELCD) program applied helical CT scanning to 4,401 active or retired workers at three Department of Energy uranium gaseous diffusion plants in Tennessee, KY, and Ohio between 2000 and 2004. Workers were variably exposed to lung carcinogens, including uranium, plutonium, asbestos, nickel, and beryllium.17 The United Steelworkers (and its predecessor union, the Paper, Allied-Industrial, Chemical, and Energy International Union) was a major collaborator.
Participants were previously screened with spirometry and posteroanterior and lateral chest radiographs. If a chest radiograph showed a lung mass, the participant was referred to the personal physician for diagnosis. Chest radiographs were obtained a mean of 10.1 months (SD, 7.0 months) prior to the initial CT scan.
Eligibility Criteria for Helical CT Scan
Gaseous diffusion plant workers were eligible for helical CT scanning if they met risk criteria (Table 1
). Seven people whose lung cancer was first identified by chest radiograph were not eligible for the ELCD program. Institutional Review Boards of Queens College and the Oak Ridge Associated Universities approved the study. All participants signed an informed consent.
|
Nodule Definition
Opacities with a length/width ratio < 3 were considered nodules, permitting differentiation from blood vessels. Lung nodules with uniform or ring calcification were considered benign. Noncalcified nodules with irregular margins, regardless of size were deemed suspicious. Nodules 3 mm in maximum diameter with no or partial calcification or smooth walls were considered indeterminate. Areas of atelectasis, scar, inflammation, or rarely, blood vessels that could not definitively be identified as such were also considered indeterminate nodules.
Scanning Strategy
The initial scan was a low-dose, full-chest CT scan. Individuals with one to nine indeterminate nodules underwent a repeat standard radiation dose, thin-section CT of the nodule(s) within 6 weeks of the initial scan.
Interval scans were performed if the 6-week follow-up scan confirmed the presence of one or more indeterminate nodules. These scans consisted of a standard radiation dose, thin-section CT at the level of the nodule only. For indeterminate nodules 5 mm, interval scans at 3, 6, and 12 months were performed. Indeterminate nodules that were 3 to 4.9 mm in diameter received interval scans at 6 months and 12 months. Initial scans with more than five nodules were followed up with a low-dose full chest CT scan at 6 months.
The incidence scan consisted of a low-dose full chest CT scan at 18 months (range, 15 to 21 months) following the initial scan offered to participants between 50 and 79 years of age, without previously CT-detected lung cancer. The ELCD study is ongoing, and not all participants have had an incidence CT scan.
Diagnostic Evaluation
Individuals with suspicious nodules or other significant CT scan findings were referred to their personal physician for diagnostic evaluation. We made recommendations but did not direct the diagnostic workup, which was determined by the personal physician. This was similar to the study design used in the Cornell Early Lung Cancer Action Project by Henschke and colleagues.2 We obtained medical records and pathology slides from participants with suspicious and indeterminate nodules who underwent biopsy and/or surgery. Histology slides were reviewed by an expert panel of the International Early Lung Cancer Action Program at Cornell Medical College.
Data and Statistical Analysis
Data from medical questionnaires, radiographs, and spirometry were combined with ELCD-specific data collection. A never-smoker smoked 
100 cigarettes lifetime. Nonmalignant asbestos-related fibrosis was present if the initial screening CT scan showed typical pleural and/or bilateral lower zone parenchymal scarring. Data analysis was performed with standard software.
The performance of the CT scan in detecting lung cancer was estimated by its sensitivity, ie, percentage of lung cancers detected by the CT scan, or true-positive test results divided by the sum of true-positive results and false-negative results; specificity, ie, percentage of nonmalignant nodules correctly identified by the CT scan or true-negative results divided by the sum of true-negative results and false-positive results; and predictive values, ie, probability of lung cancer given CT scan nodule classification. Positive predictive value is the percentage of positive results in patients with a diagnosis with lung cancer. The negative predictive value is the percentage of negative findings in patients who do not have lung cancer. Of importance to the interpretation of predictive values is the prevalence of lung cancer among the population with indeterminate or suspicious nodules, which was 2.4%
Maximum nodule dimension is reported from the scan that triggered referral for diagnosis. The presence of lung cancer was established histologically. The absence of lung cancer was based on either a negative histology or the nonevolution of an indeterminate or a suspicious nodule after at least 18 to 24 months of follow-up. A receiver operator characteristic curve was drawn as a graph of sensitivity vs false-positive proportions, each point representing a cutoff for positivity of the CT scan based on nodule size. We calculated the area under the curve.
Results
Table 2 provides the demographic and lung cancer risk factors of the study population.
|
Most detected lung cancers (29 of 33) were 10 mm at the time of referral for diagnosis. Sizes of the remaining four cancers were 7 mm (n = 1), 8 mm (n = 2), and 9 mm (n = 1). Nearly one half (16 of 33 cases, 48.5%) of lung cancers were 20 mm at the time of CT detection.
Most lung cancers (29 of 33, 87.9%) were detected on the initial (n = 19) or incidence (n = 10) scans. The stage distribution of the lung cancers detected on the initial and the incidence scans was similar. Four of the 33 lung cancers (12.1%) were detected due to growth observed of an indeterminate nodule on interval scans; all 4 lung cancers were stage I.
Follow-up of Indeterminate Nodules Identified at Initial CT Scan
The initial CT scan (including 6-week follow-up scan) identified 1,535 indeterminate nodules among 982 of the 4,401 baseline participants (22.3%). We restrict the following analysis to indeterminate nodules that were detected on the initial scan prior to March 1, 2003, in order to allow for completion of a minimum of 18 months of follow-up for the 764 participants who had one or more indeterminate nodule. Follow-up was completed for 95.2% (n = 728) of these 764 participants. 80% (or 609) underwent a full-chest, low-dose, incidence CT scan at 18 months. We contacted the remaining 155 people (or their next of kin) who did not have the 18-month incidence scan. We were unable to reach 36 of 764 participants (4.7%) with indeterminate nodules on initial scan.
Only 3 of the 728 contacted people (0.41%) who had one or more indeterminate nodules on the initial scan had lung cancer diagnosed during the subsequent 12 months. These lung cancers occurred in the nodules that were identified on the baseline scan. All three nodules were > 5 mm in size on initial scan. All three of the lung cancers that were detected during the 12-month follow-up period were detected on the 3-month or 6-month interval scans. All three lung cancers were stage I at the time of diagnosis.
Follow-up of Lung Nodules Suspicious for Malignancy
Of the 4,401 baseline participants, 102 patients (2.3%) had a suspicious nodule on one of their scans, as of March 1, 2004. The mean proportion of suspicious nodules on any one type of scan (eg, initial, interval, or incidence) was 1.2 per 100 scans.
Thirty-three of these 102 suspicious nodules (one third) were diagnosed as primary lung cancer. Twenty patients with suspicious nodules underwent a surgical biopsy that showed benign disease. In five patients with suspicious nodules, metastatic cancer from another primary site was diagnosed. Three patients died prior to follow-up (none from lung cancer).
There were 41 remaining patients with suspicious nodules. A positron emission tomography (PET) scan result negative in 15 patients, and the remaining 26 patients were monitored by their personal physicians. We followed up these 41 people for a mean of 46.5 months (minimum, 24 months; 31 of 41 patients [76%], > 36 months). We successfully contacted 38 of the 41 patients (or their next of kin); none had received a diagnosis of lung cancer.
Sensitivity, Specificity, and Predictive Values
To assess the screening test characteristics, we achieved virtually complete (95.2%) 18-month follow-up of the subset of participants who completed an initial scan prior to March 1, 2003. This subset included 807 participants who had at least one indeterminate (n = 764) or suspicious nodule (n = 43) on the initial scan. This follow-up permits analysis of the extent to which indeterminate and suspicious nodules were subsequently found to be malignant.
Table 3
provides the sensitivity, specificity, and predictive values for nodule suspiciousness, various nodule size thresholds, and a combination of suspicious morphology and nodule size. This analysis is restricted to the 807 participants cited above. Suspicious nodule morphology of nodules yielded the highest combination of sensitivity (84.2%) and specificity (96.6%), which did not improve when combined with various nodule size thresholds. Three of 19 lung cancers in this subcohort were not designated suspicious by the radiologist (false negatives). Only 27 of 788 participants without lung cancer were designated as having nodules of suspicious morphology (false positives). Almost two of every five nodules labeled suspicious were malignant (positive predictive value, 37.2%). Among the lung nodules that the radiologist labeled indeterminate (that is, not suspicious), lung cancer was present in only 3 of 764 nodules (negative predictive value, 99.6%). Using nodule size threshold 7 mm alone without regard to suspicious morphology increased sensitivity to 94.7% but lowered specificity to 62.7% (Table 3).
|
|
This study provides evidence that low-dose helical CT scan can detect early stage lung cancers in a population at varied lung cancer risk (including people < 50 years old, nonsmokers, and former smokers who quit 10 years) and having access to typical medical care in nonmetropolitan areas of the United States. Most (80%, n = 24) of the nonsmall cell lung cancers were stage I or II disease, and all three small cell lung cancers were limited in extent. This distribution is within the range reported in other studies.123456
Streamlining the CT follow-up of indeterminate nodules is important in order to reduce costs, anxiety, and radiation exposure. Based on virtually complete 18-month follow-up of initial scans, we found that the lung cancer yield of the 3-, 6-, and 12-month interval scans was very limited. Nearly 90% of all lung cancers detected were detected on the initial or incidence scan. Thus, in our study population, eliminating all interval scans would have minimal impact (ie, a 10% decrease) on the overall lung cancer yield. Since none of the lung cancers detected on the interval scans were 5 mm at baseline, eliminating the interval scans of such small nodules would have no impact on lung cancer detection. Such a protocol change would be consequential, since 51% of participants (387 of 764 patients) with indeterminate nodules on the baseline scan had nodules 5 mm in size. Notably though, all four lung cancers detected on interval scans were stage I. In the absence of interval scans, a subsequent incidence scan would have detected these cancers. Other studies24 had similar findings with respect to the fate of nodules 5 mm. These data support the recommendation by Henschke et al12 and MacMahon and colleagues13 that follow-up of indeterminate nodules 5 mm can be restricted to an annual low-dose CT scan.
Frequent false-positive results are a major concern about CT scanning for lung cancer. All studies123456 to date have found a high rate of indeterminate nodules (19 to 51%) on the baseline scan. Test positivity can be defined differently according to degree of suspiciousness (suspicious vs indeterminate), size of nodule, nodule growth, or other nodule characteristics. The goal is to maximize the proportion of lung cancers detected (high sensitivity) while minimizing the harm to participants with benign lesions (high specificity).
The medical and lay literature variably uses the terms test positive (and concomitantly, false positive) to refer to indeterminate nodules of low malignant potential or, alternatively, to truly suspicious nodules that require PET scan, biopsy, or similar careful attention. This aggregation of these two types of positive tests obscures the promises and problems of CT scanning for early lung cancer detection. While it is true that both indeterminate and suspicious nodules represent positive tests, their separation is important, because they occur in very different proportions, have a highly varied probability of being malignant, and require different follow-up protocols.
Our results suggest that indeterminate nodules can be managed with minimal harm and expense. Indeterminate nodules 5 mm are unlikely to be malignant and can await re-evaluation during a periodic incidence scan. Indeterminate nodules > 5 mm could undergo a single, full-chest interval scan at 6 months, with limited associated risk and cost. To reduce anxiety, participants can be informed of the low likelihood that indeterminate nodules are malignant.
Suspicious nodules represent the more important type of test positivity. Using this variable, we achieved 84% sensitivity and 97% specificity. In terms of people, among the cohort subset with complete 18-month follow-up, designation of suspicious nodules resulted in the detection of 16 of 19 lung cancers, but assigned 37 of 788 tested people without lung cancer to further testing for benign lesions. Overall, one fifth of participants with suspicious but ultimately benign lesions underwent biopsy. Increased use of PET scanning over the last 12 months in the study communities resulted in fewer invasive interventions for benign disease. Additional experience with PET scanning following helical CT scanning will further reduce the rate of needless intervention.4 Nevertheless, limited false positivity is inevitable, since a positive PET scan and nodule growth may accompany inflammatory conditions.1819 In addition, false-negative PET findings of nodules < 10 mm are a problem.19
The radiologist’s determination of suspiciousness maximizes test accuracy. Not surprisingly, it surpasses nodule size alone in accuracy. It is of note that even the lower bound of the 95% CI of the area under the curve for suspicious nodules only (0.89) was higher than the upper bound of the 95% CI of the area under the curve for all nodules (0.86), indicating a substantial improvement. The difference between the two areas cannot be formally tested, because they are not independent. However, this approach quantifies the extent to which the radiologist’s judgment maximizes the balance between detecting as many lung malignancies as possible while minimizing potential harm to people without malignancy. The judgment exercised by radiologists may be subjective but is critical to the screening process.
A word of caution about relevance of our findings to other studies is in order. Predictive values vary according to the prior probability, ie, prevalence of disease. In the current study, there was a 2.4% prevalence of lung cancer in the population with indeterminate or suspicious nodules. Other populations with different prevalences of lung cancer will experience different predictive values, even given the same sensitivity and specificity.
There were numerous important limitations in this study. We used a single-slice CT scanner that provided 7-mm sections, which was state-of-the-art at the year of study initiation and comparable to the technique used in the ELCD project by Henschke et al2 and in the Italian spiral CT scanning trial.4 Single-slice scanners have since been replaced by multislice scanners, which yield a greater number of indeterminate nodules.3612 Such multislice scanners detect more nodules with an expected concomitant increase in sensitivity and decrease in specificity of lung cancer detection. The current finding of a sensitivity of 84.2% using a single-slice scanner provides an acceptable floor from which multislice scanners can improve. Follow-up comparable to that achieved in the current study is not yet available in studies using multislice scanners.
We also did not address the problem of interobserver variability in that a single thoracic radiologist read most of the study CT scans. Third, we did not quantify the extent to which nodules were seen retrospectively on initial CT scans when these scans were re-reviewed at the time of the incidence scans 18 months later. Hence, the prevalence of nodules reported is lower than in numerous other reports that added nodules seen only in retrospect to the total number of nodules present. However, nodules seen only in hindsight provide information on nodule frequency and limitations of radiologic reading, but are not useful in the decision making at the time of the initial CT scan about whether a nodule requires follow-up or leads to the diagnosis of lung cancer.
CT screening studies completed to date vary in important respects. We found equal numbers of adenocarcinoma and squamous cell type among the CT-detected lung cancers, whereas most other studies23456 have shown a clear predominance of adenocarcinomas. Mean size of malignant nodules in the current study is 20.9 mm (with only 12% of lesions < 10 mm), which is slightly greater than in most other studies.3456 Three fourths of the lung malignancies detected in the National Cancer Institute study5 were 10 mm. By contrast, in the Cornell study,2 56% of CT-detected cancers were 10 mm at diagnosis. The reasons for the differences among studies are not clear but are not explained by the ratio of initial to incidence and/or growth cancers.
We found very similar proportions of lung cancer on the initial and incidence CT scans (4.3 and 4.9 per 1,000 participants, respectively). The Italian study4 had a similar finding, although with higher proportions of lung cancer detection. By contrast, the Mayo Clinic study3 found 13.8 and 1.4 cancers per 1,000 participants on the initial and incidence scans, respectively. Our results and those of the Italian study4 contradict conventional wisdom, which predicts that initial CT scanning will yield a higher proportion of lung cancers than a subsequent incidence scan. This suggests that the number of baseline-detected slow-growing lung cancers may be small.
In conclusion, our findings support three ways to mitigate the problem of large numbers of CT-detected indeterminate nodules and the problem of false-positive suspicious nodules. These include the following: (1) limited CT follow-up required for indeterminate nodules (a single 6-month interval scan); (2) high specificity associated with the radiologist’s determination of suspicious nodule morphology; and (3) use of the PET scan to help identify false-positive CT results, acknowledging the limitations of the PET scan, especially for nodules < 10 mm in diameter.1819
Acknowledgements
We thank M. Griffon for assistance in categorizing occupational job titles; L. Brannon for serving as the CT technician; Mike Church and Gerold W. Wilken for transporting the CT mobile unit throughout the study; P. Cooney for database design and revision; and J. Stuckey, R. Melendez, H. Athanasiou, and F. Feeley for their work in scheduling and communicating with patients and inputting and correcting data; and the International Early Lung Cancer Action Program Expert Pathology Review Panel for reviewing pathology slides.
Footnotes
Abbreviations: CI = confidence interval; ELCD = Early Lung Cancer Detection; PET = positron emission tomography
This work was performed principally at Queens College, City University of New York.
Funded by the United States Department of Energy.
The authors have no conflicts of interest to disclose.
Received for publication November 22, 2005. Accepted for publication November 10, 2006.
References
1 cm) detected at population-based CT screening for lung cancer: reliable high-resolution CT features of benign lesions. AJR Am J Roentgenol 2003;180,955-964
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
