Applied Radiology

Dr. Gotway is an Assistant Professor-In-Residence, Diagnostic Radiology and Pulmonary/Critical Care Medicine, Director, Radiology Residency Training Program, University of California, San Francisco, Director, Body Imaging Practicum Program, Director, Thoracic Imaging, San Francisco General Hospital, San Francisco, CA. Dr. Webb is a Professor, Diagnostic Radiology, Chief, Thoracic Imaging, Department of Radiology, University of California, San Francisco, CA.

Despite the recent attention in both the radiologic literature and press, screening for lung cancer is not a new idea. Lung cancer screening has been attempted as far back as the 1960s, ¹ and four large lung cancer screening trials ^2-5 were carried out in the 1970s. However, with the advent of helical computed tomography (CT), particularly multidetector-row helical CT, the possibility of accurate mass screening of at-risk populations has been revisited.

Past lung cancer screening trials

Three major lung cancer screening studies sponsored by the National Cancer Institute (NCI), have been performed in the United States in the last 3 decades: Johns Hopkins trial, ² The Mayo Lung Project (MLP), ^4,6 and Memorial Sloan Kettering Cancer Institute trial. ^3,7 The Mayo Clinic trial was specifically designed to assess the utility of screening radiography for the detection of lung carcinoma, whereas the other two trials were directed more at evaluating the use of sputum cytology for the early detection of lung carcinoma.

Why screen patients for lung carcinoma?

According to Patz et al, ⁸ "The purpose of screening is to prevent, or delay, by means of earlier detection, the development of advanced disease and its adverse effects." The major reasons to screen populations for diseases such as lung carcinoma are to improve the quality of healthcare and to reduce heathcare costs. To implement such objectives on a large scale, the disease being screened for should be a worthy target for screening. Lung carcinoma certainly fits that criterion. In 1999, approximately 175,000 new cases of lung carcinoma were diagnosed in the United States; more than 150,000 of these patients will die of lung carcinoma in the same year. More than one million lung-cancer­related deaths occurred worldwide in 2000. ^9,10 The death rate from lung carcinoma exceeds that of breast, colon, and prostate carcinomas combined. ^8,10 Given the fact that screening programs for these latter three diseases are in place, it certainly makes sense to examine the utility of screening for lung carcinoma, particularly in light of recent advances in helical CT technology.

What is required for a successful screening program?

There are numerous parameters that impact the effectiveness of screening for disease, and these must be addressed before significant resources are committed to a screening program. ¹¹ These considerations were outlined elegantly by Obuchowski et al ¹¹ in a recent article in the American Journal of Roentgenology . This article will discuss these parameters and the evidence for, or against, the use of CT for lung cancer screening.

Is the disease being screened for serious? ¹¹

Any screened-for disease should be capable of producing significant morbidity and mortality, thus affecting patients from an emotional and financial standpoint, as well as impacting society in general from a productivity loss and general expense standpoint. Lung carcinoma certainly fits this criterion. Lung carcinoma is the leading cause of cancer-related deaths, and caring for cancer patients contributes to enormous losses in productivity as well as incredible healthcare costs. The 5- to 10-year mortality of untreated stage I lung carcinoma is 80% to 100%, ¹¹ which would tend to suggest that indolent lung carcinomas are either incredibly rare or nonexistent (this controversy will be discussed shortly). Because 50% to 60% of newly diagnosed lung carcinomas are first detected when they are already stage IV, many of the costs associated with caring for lung carcinoma patients are accrued in the last few months or years of life.

Is the disease being screened for treatable? ¹¹

Patient outcomes must be improved by whatever treatment exists for the disease being screened for, otherwise there would be no point in detecting the disease early. As mentioned above, the 5- to 10-year mortality rate of untreated stage I lung carcinoma is 80% to 100%; however, surgery for stage I lung carcinoma improves 5-year survival to 50% to 70%. ¹¹ Therefore, although lung cancer is certainly a deadly disease, proper treatment markedly improves patient survival.

Does the population being screened have a high prevalence of detectable preclinical disease? ¹¹

The preclincal disease phase is defined as the period from the point of onset of the illness being screened for to the time when symptoms first appear. If this phase were extremely short, the possibility of detecting disease early (before the patient would have presented with symptoms anyway) would be remote, and screening would simply cost money but could not improve patient outcomes. Within the preclinical disease phase, the detectable preclinical phase is the period of time when the disease is detectable, but before the patient develops signs or symptoms and presents to medical attention.

If the screened population has a low disease prevalence during the detectable preclinical phase, screening will not be cost effective. Additionally, if the disease prevalence is low, the likelihood that a positive test will actually represent true disease is also correspondingly low. This concept represents the positive predictive value of a test. Unlike sensitivity and specificity, predictive values are influenced by the prevalence of disease. This would mean if the disease prevalence is low, most patients who have positive results may not actually have lung cancer, and would be subjected unnecessarily to additional (likely expensive) diagnostic testing as well as the possibility of unnecessary surgery.

How prevalent is lung carcinoma in at-risk populations during the detectable preclinical phase? The answer to that question depends on the definition of the at-risk population. Certainly lung carcinoma screening programs will not be applied to just anyone; rather, at-risk populations are targeted for screening based on such criteria as age, history of smoking, and perhaps, as exclusionary criteria, a history of prior lung carcinoma or extrathoracic malignancy. Current estimates of the prevalence of detectable preclincal lung carcinoma in at-risk populations range from 2% to 4%. ¹¹ An appropriate detectable preclinical lung carcinoma prevalence in a screening population is 5%. This prevalence results in a positive predictive value of 50% if the screening test is good (95% sensitive and specific). ¹¹

How is data regarding preclinical lung carcinoma prevalence obtained? One way is to use information gained from prior lung cancer screening trials that used chest radiography and sputum cytology, or to analyze the prevalence of asymptomatic lung carcinomas detected incidentally during lung transplantation for pulmonary emphysema (nearly 5%). Also, several lung carcinoma CT screening trials are currently well under way, and data from the first year of prevalence scans obtained from these trials is available (Table 1). ^2-4,9,12-20

The detectable preclinical prevalence of lung carcinoma in the at-risk populations in these trials is below the ideal, which will result in a lower positive predictive value of CT scanning, as well as potentially decreased cost effectiveness of screening. The criteria for selecting patients for enrollment in lung cancer screening studies are listed in Table 2. ^{2-4,9,12-18,20}

Increasing the screening requirements (older patients, more extensive smoking history), would increase the prevalence of lung carcinoma and thereby make screening more specific and cost effective, but at the risk of decreasing the sensitivity of the screening test. It is known that the age-specific prevalence of lung carcinoma rises abruptly at age 50 to 55, so this may be a reasonable target age for lung cancer screening. Other reasonable inclusion criteria for lung cancer screening programs is to require that the patient be fit for surgical resection and that the patients can hold their breath for the scan duration.

Does the screening test have a high accuracy for the detection
of disease during the detectable preclinical phase?

This question is somewhat difficult to answer directly for CT scanning. The usual approach investigators take for answering this question is to compare the ability of CT to detect lung carcinoma with that of chest radiography, and CT is clearly better than chest radiography for nodule/cancer detection. ⁸ Recent prevalence data from CT lung cancer screening trials, compared with detection rates in chest radiography trials, confirm this assertion: CT detected 27 cancers per 1000 high-risk screened subjects versus 9.1 to 9.7 cancers detected by chest radiography per 1000 screened subjects in the Early Lung Cancer Action Project (ELCAP) trial. ^8,12

Compared with chest radiography, CT will detect lung carcinomas at a significantly smaller size, and therefore (presumably) at a more curable stage. The average sizes of lung carcinomas detected with screening chest radiography in the major trials to date are presented in Table 3. ^{2-4,12,16-19,22}

CT clearly detects lung carcinoma at a smaller size and a greater rate than chest radiography. Whether or not CT detects lung carcinomas at a small enough size such that patient outcomes are improved with the use of CT screening remains a question; there is no guarantee that smaller size at detection will translate to decreased mortality. It is already known that screening with chest radiography will result in the detection of more lung cancers at an earlier stage, and this appears to improve survival. In the Johns Hopkins chest radiograph and sputum cytology study, ² 67% of the prevalence cancers detected in the screened group were stage I (for cancers that present symptomatically, only 20% of patients will present with stage I disease). However, survival is not considered to be the proper end point for determining the effectiveness of a screening program--mortality is the proper end point for this purpose (see biases, below, for more discussion on this point). In the Mayo Lung Project, ⁴ 206 cancers were diagnosed in the intensive screening arm ("intervention arm," which consisted of chest radiography and sputum every 4 months) versus 160 lung carcinomas diagnosed in the less intensive arm ("usual care arm," consisting of yearly chest radiography at the clinician's discretion). In the intervention arm, 40% of the detected cancers were stage I, but only 25% of cancers in the usual care arm were stage I. ²³ Despite these encouraging results, no reduction in lung cancer mortality was noted. Long-term data from the Mayo Lung Project has shown that the lack of reduction in lung cancer mortality with intensive chest radiographic screening has persisted. ²⁴ A total of 337 lung cancer deaths were noted in the intervention arm, whereas 303 lung cancer deaths were noted in the usual care arm. This data translates to lung cancer mortality rates of 4.4 and 3.9 per 1000 person-years, respectively. ^23,24 Part of the issue is that screening with chest radiography has failed to provide a "stage shift": although screening has detected more early stage cancers, radiographic screening did not detect more advanced stage cancers. Comparing screened groups with control groups, investigators found a similar number of advanced stage cancers, indicating that screening detected the indolent cancers, but had no impact on the more aggressive cancers. Because screening detects more cancers compared with control groups, but does not necessarily detect more advanced cancers, the possibility that screening is detecting indolent cancers (perhaps even inconsequential lesions) must be considered. However, data regarding the impact of stage presentation on survival are subject to biases (particularly lead-time bias), which will be discussed below.

Does the screening test detect the disease being screened for at a point in the detectable preclinical phase when treatment is effective? ¹¹

For many serious diseases, especially malignancies, there is a point during the disease history after which treatment is largely ineffective; this point has been termed the "critical point." ^11,25 For malignancies, this critical point usually occurs when the cancer becomes metastatic. Therefore, for a screening test to be effective, the screening test should be able to detect the disease prior to the critical point. This issue is extremely important because it reflects the biologic behavior of the tumors. The question is, "Does screening detect more cancers because it detects them at an earlier stage or because they are slow-growing and of low biologic potential?" Answering this question is difficult. Advocates of CT screening for lung carcinoma cite data that patients screened with CT have a much greater prevalence of lower stage tumors than do patients who present with lung carcinoma. The stages of presentation of lung carcinomas detected in lung cancer screening studies are given in Table 4.

The data presented in Table 4 certainly represent an improvement compared to the historical lung carcinoma early stage presentation rate of 10% to 20%. However, data regarding the stage of presentation are subject to biases (particularly lead time bias) will be discussed below.

There is compelling evidence that lung carcinomas are heterogeneous lesions with significant variability in biologic behavior. This implies that smaller tumors are not always synonymous with early stage disease. In a recent study examining this issue, in patients with lung carcinoma with what appeared to be limited, local disease (T1N0M0), extrathoracic metastases were found upon further work-up in 13% of cases. ²⁶ In the remaining patients who were thought to be stage I, an additional 11% of patients developed extrathoracic metastases 1 year later. Furthermore, no significant difference was found between the prevalence of metastases at diagnosis and at 1-year follow-up for tumors <2 cm and those >2 cm. Finally, as many as 60% of patients with stage I disease die of lung carcinoma within 5 years despite appropriate therapy.

Another way of examining this issue is to look at patient outcomes with respect to tumor size. A recent investigation examined the survival of patients with T1N0M0 lung carcinoma as a function of the tumor sizes, and found that there was no difference in survival among patients with tumors of various sizes <3 cm (Table 5). ²⁷

The type of data presented in Table 5 has been noted previously. For example, it had been demonstrated already that there is no significant difference in survival among patients with stage I disease. ²⁸ Thus, patients with T1 disease (tumor <3 cm) all have about the same 5-year survival. Given the assumption that chest radiography is an ineffective screening method for lung carcinoma because no mortality reduction has been noted, and the average size of tumors detected in chest radiography screening programs was ¾3 cm in several major chest radiography screening trials, the ability of CT to detect smaller tumors may not necessarily translate to a clinical benefit.

Does the screening test itself cause little morbidity? ¹¹

The patient being screened presumably is asymptomatic and therefore at little risk of dying from the disease being screened for in the immediate future. Therefore, the screening test must not have a significant frequency of adverse effects, or it may cause more damage than it prevents. Does CT fit this criterion? No intravenous contrast is required for lung cancer screening studies, so there should be no contrast-associated risks. The major risk related to lung cancer screening is the radiation dose associated with CT. ¹¹ For this reason, screening lung cancer CT should be performed using decreased tube current, thereby decreasing the radiation dose. Because the target of lung cancer screening (a lung nodule) presents as a soft-tissue density structure surrounded by low-attenuation pulmonary parenchyma, there is an intrinsically high signal-to-noise ratio between the target and background. Therefore, tube currents may be decreased to one-fifth or one-sixth of that used for routine studies without compromising diagnostic accuracy for lung nodules. ²⁹ It has been estimated that the risk of inducing a fatal cancer with this technique is approximately 1 in 25,000. ¹¹

Although CT scans may cause little morbidity themselves if conducted properly, complications resulting from overdiagnosis and additional CT scans required to follow nonspecific small nodules will increase the morbidity and costs associated with CT for lung cancer screening. ³⁰ Whether or not this effect is important depends on the rate at which CT detects indeterminate, but ultimately benign, lung nodules. Preliminary data suggest that this event may be common. A recent study of prevalence data from 817 asymptomatic smokers found noncalcified nodules in 43% of patients. ⁹ In the 1520 prevalence scans in the Mayo Clinic NCI trial, 51% of patients had noncalcified lung nodules; more than 1300 nodules were detected in these patients. ¹³

Is the screening affordable and widely available? ¹¹

Helical CT is certainly widely available throughout the United States, but the question of cost is a major consideration for the widespread adoption of lung cancer screening protocols. The cost of lung cancer screening depends on how much is charged for a CT scan, how many false-positive results occur, and how the at-risk population to be screened is defined. The National Cancer Institute has estimated that a national lung cancer screening program for current and former smokers aged 45 years or older would cost more than $39 billion. Chest radiographic screening studies ($50 per chest radiograph and $60 for sputum cytology) suggest that the cost of screening the at-risk U.S. population once per year (in 1988) would have cost $1.5 billion, not including the cost of false-positive studies.

Obviously, such a cost is prohibitively expensive. However, is the estimate accurate? One must also consider what an "appropriate" amount of money spent per life saved may be. Some investigators have asserted that $50,000 spent per life-year saved is an appropriate return. For example, current estimates suggest that screening for breast and colorectal carcinoma costs between $30,000 and $50,000 per life-year saved ¹¹ ; a similar target level may be appropriate for lung carcinoma. Clearly the NCI estimate far exceeds this value, but not all investigators agree that lung cancer screening with CT will be so expensive. Estimates from the ELCAP group suggest that CT screening for lung carcinoma may cost only $2000 per life-year saved. ³¹

Part of the discrepancy between these two estimates depends on how they account for the additional costs incurred by CT lung cancer screening. Although the first scan may be relatively inexpensive (an average of approximately $300 per scan), recommended follow-up scans may inflate the cost of screening dramatically. CT--particularly narrow collimation multidetector CT--commonly detects small nodules of unknown significance; these nodules are either followed, biopsied, or evaluated noninvasively. If followed, follow-up regimens aim to establish 2-year stability by scanning the nodule at doubling time intervals, beginning 6 weeks to 3 months after the prevalence scan. This regimen can result in one screening CT examination leading to 5 or 6 CT studies over the ensuing 2 years, with the obvious accompanying increases in costs. How common would this scenario be? In the Mayo Clinic trial, among 1520 patients undergoing prevalence scans over a year period, more than 1300 noncalcified nodules (51% of patients) were detected. ¹³ Only a few of these lesions were carcinomas, thus the rest either must be followed, biopsied, or evaluated with other noninvasive methods, such as positron emission tomography (PET) scanning or repeat CT. In the ELCAP study, ¹² 1 to 6 noncalcified nodules were detected in 233 patients among the 1000 patients scanned. Only 27 cancers were found among these 233 patients, implying that the remainder of the nodules required some sort of evaluation (with associated costs) to exclude carcinoma. In addition to these costs, the costs of surgery for false-positive CT examinations (such cases will be inevitable) and costs associated with losses in productivity for patients with false-positive results must be considered.

One cost-effectiveness analysis of CT lung cancer screening has been proposed. ³² Cost effectiveness is defined as the cost of screening divided by the effect of screening (in terms of life-years saved). This cost-effectiveness analysis assumes that screening results in a stage shift (more carcinomas detected at an earlier stage, resulting in fewer later stage carcinomas), and that this down-staging will result in decreased mortality. It is also assumed that no biases (see below) complicate the analysis. Variables assumed for this analysis include a cost of $291 for a screening CT, a diagnostic CT cost of about $340 to $416, lung biopsy costs ranging from $507 to $5071 (depending on method used and complications), a prevalence rate for abnormal CT screening studies of 25% to 50%, a prevalence of carcinoma in the at-risk screened population of 1.2% and 2.7%, and a carcinoma incidence rate of 0.7% to 1.2% in the screened population. If CT screening for lung cancer results in 20% of the detected lesions identified at stage I, then no benefit will be derived because that is roughly the percentage of patients that will present at stage I once symptoms develop (this is where we are without screening). If CT screening finds 30% of the lung carcinomas while still localized, then the cost per life-years saved (assuming the above variables) is $90,022. If CT detects 50% of the carcinomas while still localized, then the cost for screening the at-risk population in terms of life-years saved is $48,357. If CT detects 70% of lung carcinomas while still localized, the cost of screening is $33,557. Current prevalence data suggests that CT may be able to detect a significant number of carcinomas at stage I (approximately 50%), indicating that screening will cost <$50,000 per life-year saved. Compare that with other screening programs in practice: cervical carcinoma ($15,600 per life-year saved), breast cancer screening with mammography ($24,100 per life-year saved), colonoscopy ($127,700 per life-year saved), and chest radiograph and sputum cytology screening for lung carcinoma (in Japan, $93,000 per life-year saved). ^32,33

Who will pay for lung cancer screening with CT? Most investigators agree that the initial scan will be paid for by the patient; insurance will not cover such scans. Abnormalities detected on the prevalence scan are usually covered by insurance, which means that society in general (including nonsmokers) will be absorbing the costs associated with screening for a largely self-induced illness. It remains to be seen if insurance companies will "penalize" patients for screening themselves. It is possible that insurance companies may try to deny payment for CT and other costs incurred with lung cancer screening by claiming that the detection of a lesion on a self-pay scan represents a pre-existing condition.

Some investigators believe that screening for lung cancer will still be cost-effective because the costs associated with caring for late-stage lung carcinoma patients are so incredibly high. ³¹ Since the majority of patients with lung carcinoma present with unresectable disease, most patients will incur costs related to palliative care as well as losses in productivity. Potentially, such costs could exceed the costs of screening for lung cancer with CT.

Is treatment for lung carcinoma effective when disease is detected in the preclinical phase? ¹¹

Lung cancer treatment must be effective and/or less toxic when applied during the detectable preclinical phase than when applied during the symptomatic period. If the disease can be treated successfully after symptoms are present, then there is no point to screening.

Proving that treatment is effective during the preclinical phase is difficult. Five-year, disease-specific survival and mortality rates have been used as end points for determining the effectiveness of treatment for carcinoma, and have also been used to determine the effectiveness of screening programs. Although 5-year, disease-specific survival has been commonly employed to demonstrate that screening programs detect carcinomas early, and therefore lead to effective treatment at an earlier disease stage, using 5-year, disease-specific survival as an end point results in the introduction of bias. ²³ The major biases that deserve consideration include lead-time bias, length-time bias, and overdiagnosis bias.

Lead-time bias­­ For unscreened patients, disease-specific survival is defined as the length of time from the clinical diagnosis of a disease to death from the disease. ²⁵ For screened patients, disease-specific survival is defined as the time from the detection of the disease by the screening program to the time of death from the disease. ¹¹ Lead time is defined as the amount of time between the detection of the disease by the screening test and the point at which the disease first becomes symptomatic. ²³ Because the screening test may detect disease before the disease becomes symptomatic, the screening may create the impression that it has detected the disease at an earlier and, therefore, more curable stage. However, even if the disease is completely untreatable, screening will create the impression that the patient has survived longer, simply because the screening test detected the disease before it became symptomatic. ¹¹

Length-time bias­­ Length-time bias occurs because of the heterogeneity of the biologic behavior of diseases; disease in some patients will progress at a slower rate than in others. ²⁵ The probability that a carcinoma will be detected in a screening program is directly proportional to the length of the asymptomatic preclinical phase. ⁸ Indolent disease will progress at a slower rate than more aggressive lesions, and thus may have a longer detectable preclinical phase (hence the term "length-time" bias) than more aggressive lesions. ²³ Because these more indolent lesions are easier to detect, they tend to be over-represented in the screening cohort; this tends to make the screening test appear more effective than it really is because patients with more indolent lesions will appear to survive longer, whether or not treatment is effective. ¹¹ This longer survival is attributed inappropriately to the effectiveness of early detection in the screening program.

For a lung cancer screening example, length-time bias may result from the detection of relatively low-grade lesions such as some bronchioloalveolar carcinomas (BAC). Because some BACs grow slowly and behave more indolently compared to other bronchogenic carcinomas, an over-representation of BACs in a screened population would create the impression that screening prolongs survival. Recent data analyzing the behavior of peripheral adenocarcinomas supports this assertion. ³⁴ Among small (<3 cm), peripheral, surgically resected lung carcinomas, tumor volume doubling times ranged from 42 to 1486 days. If one considers that it will take 8 years for a tumor with a doubling time of 365 days to increase in size from 5 mm to 3 cm, the potential for length-time bias and overdiagnosis bias in CT lung cancer screening regimens becomes clear. ⁸

Overdiagnosis bias­­ Overdiagnosis bias refers to the possibility that some patients may have extremely indolent forms of disease, or "pseudodisease." This pseudodisease represents preclinical disease that would have never become apparent clinically had it not been discovered by screening. ⁸ Because such pseudodisease does not have significant malignant potential, when the patient "survives" for a long period after disease detection, it appears that the screening test effectively improves survival by early detection. ¹¹ Less commonly, overdiagnosis bias can result from detection of a lethal lung cancer in a screened subject, but the subject dies of another process before they would have died of the detected lung cancer. ²³ In essence, pseudodisease "dilutes" the screening disease detection cohort with patients that would have survived anyway and makes the screening test appear more effective. ⁸ For lung carcinoma, the detection of atypical adenomatous hyperplasia (AAH) may represent an example of overdiagnosis bias. Although this is controversial, AAH may represent a precancerous lesion, and a significant number, but possibly not all, of AAH lesions may never progress to frank carcinoma. If such a lesion is detected on a lung cancer CT screening study (AAH often presents as a small ground-glass nodule) and it is removed and cure is assumed, that patient will be labeled as a "long-term survivor." An over-representation of such patients in a screening cohort will create the appearance of longer patient survival in the cohort, and thus the appearance that screening is effective.

At first glance, it seems that overdiagnosis bias would be an uncommon problem in lung cancer screening with CT. However, some investigators have found clinically silent cases of lung carcinoma at autopsy, ³⁵ implying that there may be some potential for overdiagnosis bias in lung cancer screening. ^8,23,36 Clinically silent lesions are certainly known in other organ systems (especially prostate carcinoma), and some of the information surrounding the concept of indolent carcinomas has been discovered through mass screening programs. ²³ Although the high fatality rate associated with untreated lung carcinoma has often been cited as evidence that overdiagnosis bias would not be a significant problem with lung cancer CT screening, this fatality rate has been calculated based on lesions that presented symptomatically, not those that are detected with screening. ²³ Furthermore, the disease- specific mortality of lung cancer, based on chest radiographic screening studies, is about 2 to 5 per 1000 individuals. However, with prevalence data from CT screening studies, it is clear that the number of cancers detected with CT screening is much greater than would be expected to present clinically in a 1- or 2-year period. ³⁷

This suggests that CT may detect cancers that may not have aggressive biologic potential. If lung carcinoma is rapidly progressive and usually fatal (as lung carcinoma has traditionally been considered to be), there should be no "backlog" of prevalence cases waiting to be detected. Finally, if one considers that overdiagnosis bias could lead to the false identification of lung carcinoma and subsequent lobectomy, the potential for overdiagnosis bias to harm patients is apparent. ³⁰ To guard against any potential damage that could result from overdiagnosis bias, investigators in the ELCAP group have suggested that small lesions should be followed closely to first demonstrate growth before proceeding to more invasive diagnostic procedures. ^38,39 Therefore, because CT will be more sensitive for the detection of small cancers and, therefore, presumably will detect small indolent lesions (the source of overdiagnosis bias), short follow-up CTs to demonstrate growth potential should allow for the distinction of indolent lesions from small, but aggressive, lesions. ⁴⁰ For this reason, some have advocated a mandatory observation period for small nodules. ^25,30

Because of these biases, using disease-specific survival is not a valid end point to determine the effectiveness of a screening program. ^8,23 Rather, disease-specific mortality is the most appropriate end point to determine the effectiveness of screening examinations. ^8,23 Disease-specific mortality is defined as the number of deaths due to the disease in question divided by the number of patients at risk. Disease-specific mortality is not a perfect measure because it may require a very large number of patients to prove that a screening program is actually effective in reducing mortality. For example, a randomized, controlled trial in which the disease-specific mortality is only 3 to 5 cases per year per 1000 individuals would require nearly 100,000 individuals in each study arm to detect a small, but statistically significant, difference between screened and control populations. Also, disease-specific mortality will not be able to detect whether a treatment program or screening program prolongs survival without preventing death. Clearly, prolonging survival would be potentially meaningful to oncology patients.

Does the treatment have a high rate of significant adverse effects?

If the disease being screened for has only toxic or highly risky treatments available, then screening for the disease may be problematic if the screening method has a significant false-positive rate. In this case, the treatment may cause nearly as much damage as the disease, or may be perceived as causing so much damage that patients may not want treatment. Obviously both situations would render screening useless. In the case of lung carcinoma, treatment is thoracotomy with lobectomy or pneumonectomy. Thoracotomy for lung carcinoma carries a 30-day mortality of 2% to 4%, ¹¹ with additional major and minor complications, and thus the treatment is associated with significant potential for damage. This means that the screening test must be very specific for carcinoma, which CT is not. Therefore, additional costs will be incurred because patients will need to undergo confirmatory procedures in many cases, including PET scans, repeat CT, and biopsy. Given that the target of lung cancer screening is a small nodule, these additional methods (biopsy and PET) may have a significant false-negative rate and costs associated with them as well.

How should lung cancer screening with CT be offered?

There are several elements that should be considered in a lung cancer CT screening regimen, including: patient selection, patient consent, patient referral, scanning technique, image interpretation, and communication and reporting of results. The Society of Thoracic Radiology has developed a valuable consensus statement regarding the use and implementation of helical CT lung cancer screening protocols; anyone developing a lung cancer CT screening regimen would be well-advised to read these recommendations. ⁴¹

Patient selection

As discussed earlier, screening is more likely to be effective, both in terms of cost and yield, if there is a substantial pre-test probability of disease. A reasonable target disease prevalence is 5%. How does one achieve such a prevalence? The pre-test probability of a CT scan discovering a lung carcinoma is increased when the factors that influence the likelihood of carcinoma are more prevalent. The two most important factors increasing the likelihood of carcinoma are patient age and history of smoking. Current lung cancer screening trials enroll patients 50 years old or older, with at least a 10- to 20-pack per year history of smoking. The Lung Screening Study, a randomized controlled trial being conducted at 6 medical centers in the United States, is enrolling men or women aged 55 to 74 years with at least a 30-pack per year history of smoking. If less stringent criteria are used, the prevalence of carcinoma will be lower, and it will become more likely that nodules discovered will represent false-positive cases. This situation will result in increased costs and increased complication rates. Many trials have an upper age limit because the benefit of screening takes years to be realized. After a certain age, it becomes as likely or more likely that a patient will die of something else even if they have a lung cancer.

Most trials to date have excluded patients with extrathoracic malignancies. ^9,12 Including such patients clearly would affect survival statistics, but potentially such patients could be screened in routine practice. If such patients are scanned, the detection of a nodule is a very serious matter. There is evidence to suggest that the prevalence of malignancy within a nodule in a patient with a known extrathoracic malignancy exceeds 40%, and may range as high as 80%, even for very small lesions. ⁴²

Of course, for the study to be truly a screening study, the patient should be asymptomatic. ^9,11 Some investigators have excluded patients with recent febrile illness in an effort to decrease false-positive studies. ⁹

Patient consent

The potential for misconceptions regarding the utility of lung cancer screening with CT makes discussion of the risks and benefits of such screening a prudent idea. ⁴¹ This is particularly true now that numerous radiology groups are advertising the benefits of CT screening, as well as the fact that the lay press has run stories that cast a favorable light on CT lung cancer screening, ⁴³ although the utility of CT lung cancer screening is far from proven. This situation results in one of the most common reasons patients will sue: false expectations. Patients may elect to undergo a CT screening study under the assumption that a negative CT represents a "clean bill of health," only to develop advanced stage pulmonary carcinoma sometime in the future. Alternatively, a patient may undergo a CT screening study that is interpreted as positive, only to suffer a complication during the course of evaluation of the discovered nodule. If that nodule is proven ultimately not to represent carcinoma, the patient may sue for damages. Whether or not such lawsuits will be successful remains to be seen, but the potential for litigation is certainly real. The misperceptions that could result in such litigation may be dispelled with a realistic discussion of the potential risks (false-positive studies, radiation, complications incurred in the work-up of lesions discovered on CT, etc.) and benefits of CT lung cancer screening. ^9,41

Patients should be made aware that CT is an unproven technology for screening for lung carcinoma; and that it is therefore possible that a patient could have a negative CT screening study and still develop lung cancer and die from it. Furthermore, patients should understand that radiologists are fallible, and despite best efforts, small carcinomas will be missed. Finally, patients should understand that, given the nature of CT, lung cancer screening with CT targets small peripheral cancers (Figure 1). We fully expect to miss some centrally located lesions. ^44-48 Also, the target of lung cancer screening is primarily small peripheral adenocarcinomas, which happen to be the cell type least associated with smoking. In fact, the cell type most associated with smoking, small-cell carcinoma, usually presents as a central lesion and may frequently be missed on most CT screening studies. This type of information will alter (appropriately) patient's expectations, and some patients may forego the test altogether in light of such information. ^25,30

Patient referral

Anyone establishing a lung cancer screening regimen must decide whether or not they will accept self-referred patients. Self-referred patients should probably sign informed consent forms (as above), but, more importantly, a radiologist that accepts self-referred patients essentially establishes a doctor-patient relationship with that patient. This means that the radiologist will assume the responsibility of being certain that the results of the study are communicated to the patient, ⁴¹ and, more importantly, the radiologist will assume the responsibility of explaining the significance of the results, discussing the proposed treatment plan, and ensuring that the patient follows up appropriately. Such consultations may become the ethical standard for self-referred patients, and may even become commonplace for physician-referred patients. Therefore, if the patient has a positive study, it may become the radiologist's responsibility to remind the patient to report for follow-up studies at the assigned intervals, and the radiologist will have to assume the responsibility of counseling the patient regarding findings and further evaluation. This situation is analogous to mammography registries, and has already become the standard in several of the large screening trials.

Scanning technique

The scanning technique used depends on the available equipment. Single-detector helical CT studies may be conducted with 5-mm collimation and a pitch of 2, with mA 20% of routine. ^9,41 Narrower collimation may be employed, but it is important to obtain the study in a single breath hold to avoid slice misregistration. ⁴¹ Multidetector CT studies are currently being conducted with very narrow collimation (2.5 mm) and decreased (40 to 50) mA. ⁴¹ Using multidetector CT, patients may also be scanned with narrow collimation with the images reconstructed at wider increments to facilitate more rapid review. This method has the advantage that the images can be reconstructed at the narrow collimation should a nodule be discovered. Narrow collimation provides improved spatial resolution, which is essential for the detection of small nodules. However, the narrower the collimation, the more a vessel in cross section may simulate a nodule. Therefore, these studies must be interpreted very cautiously. Some investigators have used sliding thick-slab maximum intensity projection images to assist in nodule detection (see below).

Image interpretation

It is critically important to know the target of lung cancer screening with CT--the peripheral nodule. Central lesions, particularly endobronchial lesions, may be missed easily. Before embarking on a lung cancer screening regimen, it is useful to review the results of studies examining spiral CT lung cancer detection failures. ^44-46 The overall detection failure rate is low, but lesions that are missed may have one or more of the following features: 1) are located in an area of scarring; 2) are located in an endobronchial location; 3) have faint high attenuation, suggesting calcification; 4) are very small nodules; 5) are nodules immediately adjacent to vessels; 6) are central lesions (Figure 2).

Accurate diagnosis is facilitated with workstation review. The ability to display the images sequentially in rapid format and the ability to rapidly switch back and forth between different windows are invaluable advantages, and should be considered essential for the review of lung cancer screening studies. ⁴⁹

As mentioned above, sliding thin-slab maximum intensity projection images (MIPs) are useful for demonstrating small nodules. MIPs can make nodules more apparent while simultaneously rendering the appearance of vessels more obvious. In a recent study, when using 3.75 mm collimation MDCT images (reconstructed at 3-mm intervals), Gruden et al ⁵⁰ found that 10-mm MIPs detected more small nodules than did review of the axial data alone. Eventually, computer-aided diagnosis may play a significant role in lung cancer screening studies.

How should small nodules discovered during the course of a routine lung cancer screening study be addressed?

While there is no definite consensus, the management schemes of the investigators currently conducting lung cancer screening trials may provide a management guide ^{9,12-15,38,39} The following represents a combination of the algorithms used by the ELCAP group and the Mayo Clinic trial, as well as a recent survey of the nodule management schemes used by members of the Society of Thoracic Radiology. ⁵¹

Nodules that are clearly benign (fat or benign pattern of calcification) or negative CT studies result in annual follow-up (not following the nodule, but having the patient return for repeat annual screening). If the nodule is ¾3 to 4 mm, thin-section CT is performed to further characterize the nodule's morphology. Some investigators may perform thin-section imaging at the time of nodule detection, while others may perform the thin-section CT at the time of the first follow-up study, which is usually approximately 3 months after the nodule was detected initially. If the nodule remains stable, it is followed with CT at successive intervals: 3 months, 6 months, 12 months, and 24 months. If the nodule remains stable for more than 2 years, it is generally considered benign. Any growth raises concern for malignancy, and prompts investigation. Obviously, great care must be taken to measure the nodule accurately to detect even slight growth. While some investigators have employed sophisticated segmentation algorithms for this purpose, such methods are not available routinely to most practitioners. Instead, repeat thin-section imaging, using identical scan parameters, with very careful comparison to (multiple) priors scans, must be performed.

If the detected nodule has no benign features and ranges in size from 5 to 10 mm, it may be resected by video-assisted thoracoscopic surgery (VATS); it may be biopsied; or, if it is in a dangerous or inaccessible location, it may be followed closely for interval growth. PET scanning may play a limited role for nodules of this size. If the nodule shows no evidence of benign characteristics and is >=10 mm in size, most investigators suggest that the lesion be biopsied or resected (via VATS or thoracotomy). The Mayo Clinic regimen suggests that 8- to 20-mm indeterminate nodules should be worked up immediately with thin-section CT to detect benign features. If the nodule remains indeterminate, the nodule may be evaluated with a CT nodule enhancement protocol. PET scanning could be used in this case, particularly if the nodule is in an inaccessible location (rendering biopsy dangerous or impossible), if the patient cannot cooperate with biopsy, or if VATS resection cannot be performed.

Communication and reporting of results

Standard reporting methods may suffice if patients are physician-referred. A call to the referring physician and some discussion of the risks and benefits of lung cancer screening may be advisable. However, as discussed above, self-referred patients must be handled directly by the radiologist. This involves patient informed consent, with discussion of risks and benefits; reporting of the results of the study to the patient, including the implication of the results; and discussion of recommendations for management (Naidich DP, personal communication, January 2001). ⁵² If follow-up studies are recommended, then it may be the responsibility of the radiologist to contact the patient to remind him/her of future CT appointments. This responsibility will require the maintenance of records, perhaps in the form of an electronic database.

Conclusion

The accuracy of helical CT for the detection of small lung nodules has prompted the use of helical CT for lung cancer screening programs. While lung cancer is a common, and often fatal, disease, the issues surrounding the use of helical CT for lung cancer screening are complex and are hotly debated. Although there are numerous points of contention between proponents and opponents of the use of CT for lung cancer screening, the debate primarily centers around whether or not a randomized, controlled trial is needed (or would be useful) to prove the efficacy of lung cancer screening with CT. Currently, most academic radiology societies have suggested that CT screening for lung cancer be reserved for patients enrolled in clinical trials. Nevertheless, the use of CT for screening purposes has been popularized by the press as well as some entrepreneurial radiologists, so it is likely that many radiology practices will eventually encounter patients interested in screening CT studies. Familiarity with the issues surrounding the debate on lung cancer screening, as well as an understanding of the implementation of proper lung cancer screening protocols, image interpretation, and the logistics of instituting a CT screening program are therefore very important. AR