Lung cancer is the leading cause of death due to malignancy among men and women in the USA. More Americans die from lung cancer than from the next three most common cancer-related causes of death (colorectal, breast and prostate).1 Lung cancer is fatal in more than 85% of the affected individuals and is often referred to as a ‘silent killer’.1 Worldwide, lung cancer remains a considerable public health problem with more than 1.3 million deaths estimated in the year 2000.2
Lung cancer screening has been a controversial issue as the effectiveness of various screening tools remains uncertain. No studies have yet proven a mortality benefit with screening and there have been concerns about the potential harm from unnecessary invasive diagnostic procedures and treatments. Many studies of screening for lung cancer have been conducted in different populations using a range of modalities. At the present time, the existing evidence is not sufficient to support lung cancer screening. In 2004, The US Preventive Services Task Force concluded that ‘the evidence is insufficient to recommend for or against screening asymptomatic persons for lung cancer with low-dose CT, CXR, sputum cytology, or a combination of these tests’.3 However, reappraisal of long-term results from previous studies4,5 and evidence indicating that low-dose CT detects cancer at a smaller size and earlier stage,6 have led to renewed interest in screening for lung cancer.
In this report, we review the basic principles of screening and results from randomized-controlled trials (RCT) of CXR and sputum cytology. We also present an update on the existing and ongoing studies, along with the controversy surrounding the use of CXR and low-dose CT in lung cancer screening.
Principles and rationale for lung cancer screening
A screening test is designed to identify cancer in asymptomatic individuals who are at risk. It is not diagnostic of cancer and further tests are usually needed to confirm or exclude malignancy. For a screening test to be considered appropriate for general use, it should meet the following conditions. First, screening should detect a cancer in its preclinical stage. Second, an effective treatment should be available in the preclinical stage. Third, early intervention in the preclinical stage should change the course of the disease and result in a decrease in the cancer-specific mortality. Fourth, accessibility, cost and morbidity associated with a screening test should be reasonable.7,8
In the case of the third condition, an effect on mortality rather than survival is required to validate a screening test, since survival from the time of diagnosis is often misleading and subject to bias. Survival, a common outcome reported in clinical trials, can be affected by at least three biases (lead-time, length time and overdiagnosis bias). Lead-time bias is an overestimation of survival duration among screen-detected cases (relative to those detected by signs and symptoms) when survival is measured from diagnosis. Length-time bias is an overestimation of the survival among screen-detected cases caused by a relative excess of slow growing tumours. Detection of indolent tumours through screening may result in an apparent increase in survival (relative to those detected by signs and symptoms). Overdiagnosis bias is another overestimation of survival among screen-detected cases caused by identification of subclinical cancer, which otherwise would have not become apparent before an individual dies of other causes. Overdiagnosis bias is also considered to be an extreme for length-time bias. Since the extent of these biases is never clear in a particular study, survival outcome is not an appropriate measure of a screening test.7–9
With regard to lung cancer, despite the ongoing advancement in diagnosis and treatment, the overall 5-year survival from this cancer is dismal and remains around 15%.1 Smoking is generally accepted to be the most powerful risk factor and an estimated 85% of lung cancers are attributed to smoking.10 In addition, former smokers continue to be at an increased risk for developing lung cancer although the risk goes down with increased duration of abstinence.11,12 Thus, current and former smokers constitute a high-risk population to which screening is offered. Of the various histological types of lung cancer, non-small-cell (NSCLC) is the most common, accounting for 70% to 75% of all new cases. It is well-known that the prognosis of lung cancer is a function of its stage. Surgically resectable NSCLC (stage I and II) are potentially curable, while the treatment of advanced stages (stage III and IV) results in only a modest improvement in survival.13 Unfortunately, in the absence of an effective screening tool, the majority of NSCLC at the time of diagnosis are symptomatic either due to the locally advanced tumour or systemic metastasis.14 In contrast, small-cell lung cancer is usually widespread when it presents, and even those with seemingly localized disease are presumed to have occult metastatic involvement.15 Based on this, it has been hypothesized that ‘detection of NSCLC at an earlier stage could potentially improve the overall survival and disease-specific mortality from lung cancer.
In the context of screening, the highest level of evidence is that obtained from several well-designed and well-conducted RCTs. However, a RCT is not always practical and population-based screening and cohorts are often valuable in establishing efficacy. For example, mammography has been shown to be effective in eight RCTs (over 500 000 patients) to cause a 20–35% decrease in mortality among women between the ages of 50 and 70 years.16 For colorectal cancer, there is no direct evidence from a RCT that screening colonoscopy reduces mortality in people without symptoms or known polyps. The evidence for its use comes from a large study of patients with polyps (not a screened population) in which, by colonoscopy, there was a large reduction in the development of advanced lesions (76–90%) compared with patients who did not have polyps removed or compared with people in the general population.17 Also, Papanicolaou test for cervical cancer screening has never been examined in a RCT. Observational studies have shown that both incidence and mortality from cervical cancer have sharply decreased in a number of large populations following the introduction of well-run screening programs and these studies have established efficacy beyond a reasonable doubt.18,19 As for lung cancer, trials designed to study CXR and/or sputum cytology include randomized, non-randomized, case-control and cohort studies. Trials studying low-dose CT are mainly case-control or cohort studies.
Results from CXR and sputum cytology trials
CXR and sputum cytology are among the earliest tools used for lung cancer screening. These tests are inexpensive, relatively safe and readily available. In the early 1970s, four major RCTs were conducted studying CXR and sputum cytology for lung cancer screening (Table 1). The National Cancer Institute sponsored three trials: Johns Hopkins Lung Project (JHLP),20 Memorial Sloan-Kettering Lung Project (MSKLP)21 and the Mayo Lung Project (MLP);22 the fourth trial was conducted in Czechoslovakia.23 Approximately 37 000 individuals were included in these trials; the target population was male smokers above the age of 45 years. The following trials should be interpreted with caution, since considerable variability existed in the design, screening protocols and control groups used. The JHLP and MSKLP had a similar design and compared dual screening with CXR annually and sputum cytology every 4 months to a group in which only annual CXR screening was performed. The intervention and control groups in these two studies were followed up for total of 5–8 years. The MLP compared an intensive screening with CXR and sputum cytology every 4 months for a total of 6 years to a less intensely screened control group who were recommended to have the same tests annually. In the Czechoslovak study, screening with CXR every 6 months was compared to no screening in a control group for 3 years. Subsequently, both groups underwent a CXR annually at the end of the fourth, fifth and sixth years.
Table 1. Outcomes of randomized controlled trials with CXR and sputum cytology
|Study||Study arm||No.||No. cancer detected||Mortality (per 1000 person-year)|
|John Hopkins Lung Project, 197320||All||10 386|| || || |
|Experimental|| 5 336||39||194||3.4|
|Control|| 5 161||40||202||3.8|
|Memorial Sloan-Kettering Lung Project, 197421||All||10 040|| || || |
|Experimental|| 5 072||30||114||2.7|
|Control|| 4 968||23||121||2.7|
|Mayo Lung Project, 197122||All||10 933||91†|| || |
|Experimental|| 4 618||—||206||3.2|
|Control|| 4 593||—||160||3.0|
|Czechoslovakia, 197523||All|| 6 364||18†|| || |
|Experimental|| 3 172||—|| 36‡||1.7|
|Control|| 3 174||—|| 19‡||1.5|
|Experimental|| 3 172||—||108§||3.6|
|Control|| 3 174||—|| 82§||2.6|
Mayo Lung Project
The MLP was conducted in 9211 male smokers between 1971 and 1983. A total of 10 933 participants underwent a cancer prevalence screen which consisted of CXR and sputum cytology. The prevalence screen revealed 91 cases of lung cancer (0.83%), 59 of which were detected by CXR, 17 by sputum cytology and 15 by both. Subsequently, those free of cancer were randomized (4618 individuals were included in the intervention group and 4593 individuals included in the control group). At the conclusion of the study, the incidence of cancer was 22% higher in the intervention group (206 vs 160 cases of cancer were detected in the intervention and control groups, respectively). The proportion of early stage lung cancer and resectable disease was higher in the intervention group. Also, the actuarial 5-year survival was significantly higher in the intervention group (33% vs 15%). However, the lung cancer mortality rates were 3.2 per 1000 person-years in the intervention group and 3.0 per 1000 person-years in the control group, and this was not statistically different.22
Following an initial prevalence screen, 6364 men were randomized (3172 in the screening group and 3174 in the control group). During the initial 3-year experimental period (excluding the scheduled screening at the end of the third year), 36 lung cancers were diagnosed in the study group compared with 19 in the control group. Similar to the MLP, the incidence of early stage cancer, resectibility and 5-year survival was higher in the study group. However, mortality was higher in the intervention group (28 vs 18 lung cancer deaths), although not significant. Over the 6-year study period, there were 85 lung cancer deaths in the study group and 67 in the control group. Despite a trend towards increased mortality in the intervention group, the mortality was not statistically different in both arms.23
Discussion of CXR and sputum cytology trials
In light of their design, the JHLP and MSKLP studied the beneficial effects of sputum cytology screening (experimental and control groups received CXR). These two trials demonstrated no additional benefit from the addition of sputum cytology to an annual CXR screen. Furthermore, screening in the JHLP failed to identify a substantial number of developing lung cancers. Almost half of the individuals in this trial were identified due to clinical manifestations prior to the next scheduled follow-up study. This observation argues that early detection efforts using CXR may not be successful, since certain tumours could be biologically aggressive and have the potential for rapid growth and metastasis.
The MLP and Czechoslovak trials were designed to study more directly the effects of CXR screening. Analysis of data from these studies showed no favourable effect on lung cancer stage distribution. Although the number of early stage cases was higher with screening (240 vs 212), the number of advanced cases were similar (303 vs 304).24 Thus, the anticipated ‘stage shift’ was not observed, and the additional cases of cancer detected seem to be not clinically relevant (i.e. overdiagnosis bias).
The conclusions drawn from the MLP were questioned by many, especially by those who are proponents for CXR screening. A much debated matter is the hypothesis of overdiagnosis proposed in the MLP. To date, there is no convincing evidence that certain lung cancers may never progress to cause symptoms or death in that individual's lifetime. Autopsy studies provide some data in support of overdiagnosis. In one large series from Australia, 47 incidental cases of lung cancer were detected among individuals who died of natural causes. Of these incidental cases 28 were NSCLC and 86% were stage I.25 Others have also claimed that the impact of overdiagnosis bias may be more important than previously conceived, since autopsies may occasionally be unsuccessful in identifying small pulmonary nodules.26 However, if the hypothesis of overdiagnosis is true, the survival of some untreated cases of early stage lung cancer should not be effected by their malignancy. The effect of surgery on the survival was evaluated in stage I (T1N0M0, T2N0M0) NSCLC individuals identified in the JHLP, MSKLP and MLP. Among 331 cases, 286 underwent surgical resection and 45 did not undergo surgery either because of refusal or medical contraindication. The 5-year survival was considerably higher in the group who underwent surgery (70% vs 10%).27 In another study of stage IA (T1N0M0) NSCLC, the 8-year fatality rate of untreated cases of cancer (<3 cm in diameter) was more than 87%. Almost all cases of stage IA lung cancer in this study had a fatal course if the patient was not treated.28 These data are not consistent with the concept of overdiagnosis of indolent lung cancer.
The MLP has also been subjected to various limitations. First, there was poor compliance with the scheduled testing (compliance of 75% in the intervention group). Second, the absence of a completely unscreened arm and misclassification of the cause of death was thought to underestimate the mortality. Third, the study was designed to detect a 50% reduction in lung cancer mortality and had an inadequate statistical power to identify a more modest reduction. Fourth, the follow-up time in the trial (average of 3 years after the last screening) to show a screening benefit was believed to be brief. Some of these limitations have been answered in an extended follow-up of the MLP participants that was performed through to the end of 1996. In this analysis, a total of 337 and 303 lung cancer deaths were detected in the intervention and usual-care arms, corresponding to lung cancer mortality rates of 4.4 and 3.9 per 1000 person-years, respectively.5 Hence, even with an extended follow-up the intense screening utilized in the MLP caused no reduction in lung cancer mortality.
In view of all the controversy in interpretation of results from completed studies, the role of CXR in lung cancer screening remains uncertain. This led the National Cancer Institute (NCI) to sponsor the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial, which is a large, multicentre, randomized controlled cancer screening trial of more than 150 000 men and women between the ages 55 and 75 years. The primary endpoint of this trial is cancer-specific mortality and the trial is sized to detect a 10% lung cancer mortality reduction (power of 89%). For lung cancer, smokers will undergo a baseline CXR and then annually for 3 years and non-smokers will undergo CXR annually for 2 years. The control group will receive routine medical care. Following recruitment, the participants will be followed for at least 13 years to ascertain all lung cancers, as well as deaths from all causes.29,30 The results of this trial may not be available for several years, but we are hopeful that this trial will provide a more complete knowledge of the utility of CXR for lung cancer screening.
Results from low-dose CT trials
Advances in imaging technology in the 1990s led to the introduction of the multislice low-dose CT scanner, which improved the ability of the spiral CT to detect and provide better resolution and diagnostic details of pulmonary nodules. This technology allows rapid acquisition of images during a single breath hold (20 s), requires no contrast media, and the radiation exposure is approximately one-eighth of the radiation dose in conventional CT. However, the use of low-dose CT remains controversial and none of the studies were able to demonstrate a true increase in survival or a decrease in lung cancer-related mortality as a result of CT screening (Table 2).
Table 2. Outcomes of selected screening trials with low-dose CT
|Study||Testing interval (months)||No. screening test performed||Positive test results n (%)||Detected cases of lung cancer, n (%)||Detected cases of stage I cancer (%)|
|Sone et al.31||12||5483||8303||676 (12)|| 518 (6)||22 (0.40)||34 (0.41)||100|| 86|
|Nawa et al.32||12||7956||5568||NR||NR||36 (0.45)|| 4 (0.07)|| 86||100|
|Sobue et al.33|| 6||1611||7891||186 (11.5)|| 721 (9.1)||14 (0.87)||22 (0.28)|| 77|| 82|
|Diederich et al.34||NR|| 817||NR||350 (43)||NR||11 (1.3)|| 1|| 58||NR|
|Henschke et al.35,36||6–18||1000||1184||233 (23)|| 63 (5)||27 (2.7)|| 7 (0.59)|| 81|| 85|
|Swensen et al.37,38||12||1520||5609||780 (51)||1118 (74)||31 (2.0)||34 (2.2)|| 71|| 61|
Over the past decade, several reports of low-dose CT screening were published from Japan and data are available from three large trials.31–33 These trials included a total of 10 049 individuals age 40 years or older, who were at risk for developing lung cancer. Two of these trials were prospective cohorts and the third was a case-control study in which CXR and sputum cytology were done along with low-dose CT screening. In the latter study by Sone and colleagues, CXR failed to identify 79% of the lung cancers that were ≤ 2 cm, and only 11 out of 44 low-dose CT-detected tumours were visualized by CXR.31 Among these three trials, a total of 15 050 prevalence and 21 762 incidence screening tests were done. At baseline, 72 cases of lung cancer were detected (prevalence 0.47%) of which 63 (89%) were stage I tumours. Incidence screening identified 60 cases of lung cancer, of which 45 were stage I tumours. No mortality data are available from these studies.
An additional trial from Germany involved 817 asymptomatic individuals above 40 years of age, with a minimum of 20 pack-year smoking history. Prevalence data from this trial identified a total of 858 non-calcified pulmonary nodules in 350 (43%) individuals, of which only 32 lesions (29 individuals) were larger than 10 mm in size. Thirteen of the 29 individuals had biopsies and 12 lung cancers were identified in 11 individuals. Fifty-eight percent (7/12) of the tumours detected were stage I and 92% (11/12) were resectable at diagnosis. After 27 months of follow-up, six of the 11 individuals were alive with no evidence of recurrent tumour, one developed local recurrence and four died (three of lung cancer, one of unknown cause).34
The first of the two cohort studies of low-dose CT conducted in the USA is the Early Lung Cancer Action Project (ELCAP). This trial performed annual low-dose CT and CXR screening in 1000 asymptomatic patients, aged 60 and above and with ≥10 pack-year smoking history. At baseline, low-dose CT detected 233 (23%) non-calcified pulmonary nodules and CXR detected only 68 (7%). Biopsies were done in 28 of the 233 CT-detected nodules. This led to the detection of 27 (2.7%) cases of lung cancer by low-dose CT and seven (0.7%) cases by CXR. The proportion of stage I lung cancer was significantly higher with the use of low-dose CT (85% vs 60%), and 83% of the CT detected stage I cancers were not visualized by CXR. Low-dose CT detected more nodules that ultimately proved benign (20.6% vs 6.1%), but no patient had a thoracotomy for a benign nodule. Subsequently, 1184 annual repeat screening examinations led to positive results in 63 individuals and eight required lung biopsy. Cancer was diagnosed in seven individuals, of which six were stage I tumours.35,36
The second cohort of low-dose CT was conducted by the Mayo Clinic in conjunction with the NCI. This trial included 1520 asymptomatic current or former smokers, aged 50 years or older and with a ≥ 20 pack-year smoking history. All participants underwent a prevalence CT and annual incidence scans and sputum cytology. One year after baseline screening, 2244 non-calcified nodules were identified in 1000 individuals (782 individuals having one or more nodule). Among the 25 cases of lung cancer detected (22 prevalence, three incidence), CT alone detected 22 cases and sputum cytology alone detected two cases.37 After five annual CT examinations, 3356 non-calcified lung nodules were identified in 1118 (74%) participants and 68 lung cancers were diagnosed (31 prevalence, 34 incidence, three interval) in 66 participants. Twenty-eight subsequent cases of NSCLC were detected, of which 17 (61%) were stage I tumours.38
For the true effects of CT screening to be determined, it was essential that the results of these previous trials be validated and evaluated in a large RCT. Accordingly, the NCI conducted the Lung Cancer Screening Study (LSS), a pilot RCT of CT compared with CXR, to assess the feasibility of conducting a future large-scale RCT of low-dose CT. The LSS provides the only data obtained from a RCT of CT screening. Six PLCO contract screening centres recruited 3318 individuals, who were not participants in the PLCO trial, and randomized them to screening with low-dose CT (n = 1586) and CXR (n = 1580). The rates of positive screening were higher with low-dose CT; 325 (20.5%) individuals in the low-dose CT and 152 (9.8%) individuals in the CXR arm had a positive screen for lung cancer. A total of 30 cases of lung cancer were identified in the low-dose CT arm and seven cases in the CXR arm. The proportion of detected stage I tumours was higher with CXR screening (86% vs 53%), however, this difference was not statistically significant. Compliance rates were more than 90% in both groups and the rate of false positive examination and its associated morbidity was low. The results of this pilot study confirmed the feasibility of conducting the National Lung Cancer Screening Trial (NLST).39
The NLST has enrolled >50 000 heavy current and former smokers (quit within 15 years from randomization) aged 55–74 years at 30 sites throughout the USA and randomized them to low-dose CT or CXR screening. Enrollment of new patients has closed at participating centres, and the participants will receive an initial and two subsequent annual screens and will be followed-up for a total of 5 years. The study is powered to detect a 20% reduction in lung cancer mortality. The results of this trial will not be available for several years.
Discussion of low-dose CT trials
For CT screening of lung cancer to be effective, the key assumption is that the smaller lesions detected by CT have a different prognosis than larger lesions. Compared to CXR, low-dose CT scan has a superior sensitivity in detecting lung cancers. Two of the studies comparing low-dose CT with CXR have clearly demonstrated the capability of low-dose CT to detect tumours that are smaller size cancers and are at an earlier stage.33,35 However, the existence of a correlation between tumour size and survival is an intensely debated topic. Many believe that there are biological properties of lung cancer that limit the success of screening. Experimental studies have shown that a 1-cm tumour sheds 3–6 million tumour cells into the blood every 24 h, suggesting the occurrence of early micrometastasis.40 Also, metastasis is believed by many to depend on tumour genetics and angiogenesis rather than size. Clinical support for this hypothesis comes from a study of 510 patients with stage IA lung (≤3 cm) lung cancer and found no correlation between tumour size and 5-year survival.41 Further analysis of the same registry showed no correlation between size of the tumour and the stage at presentation.42 Conversely, there are data to suggest that smaller tumour size within stage I is associated with improved curability.43 In addition, a recent large population-based registry by Henschke and her colleagues argued that tumour size at diagnosis is related to stage; the smaller the tumour size, the more likely that it was stage I (even in tumors ≤ 3 cm in diameter).44 In the midst of these contradictory results, the impact of detecting smaller tumours by CT screening on mortality from lung cancer remains to be answered.
With respect to stage distribution, results from the current CT trials do not support the occurrence of a ‘stage shift’ necessary to reduce mortality from lung cancer. Similar to previous CXR screening trials, low-dose CT has certainly caused an increase in the number of stage I tumours; however, there seems not to be an accompanying decrease in advanced and/or inoperable stages of lung cancer. In fact, the investigators of the Mayo CT screening trial compared the follow-up incidence results with the MLP (similar subset of age and gender). Despite all the limitations associated with this analysis, stage distribution was similar in these two trials and no difference was detected in the proportion of stage I, advanced-stage lung cancer and the incidence of lung cancer mortality.45 The occurrence of a shift from advanced to early stage tumours by CT screening needs further investigation.
One concern with the use of low-dose CT is the potential for harm due to the large number of false positive test results. The vast majority of CT-screened individuals will have indeterminate nodules detected. At prevalence screen, 23% of subjects in the ELCAP trial and 51% of the subjects in the Mayo Clinic trial were detected to have non-calcified pulmonary nodules.35,37 The difference in these rates is attributed to differences in risk profiles of the screened populations, the radiologist's interpretations and geographic location. Fortunately, more than 95% of these abnormal findings ultimately proved to be benign and cancer detection rates were significantly lower. Nevertheless, false positive studies may result in considerable anxiety and morbidity from unnecessary invasive diagnostic and surgical treatments. Interestingly, a recent report showed that the American public is enthusiastic about cancer screening in general and their commitment to screening is not influenced by false positive tests.46 Ongoing advancement in CT imaging and development of new technologies, such as software volumetric analysis, show some promise in this regard.47 Preliminary experience suggests that volumetric estimates of the doubling time may help in determining the benign versus malignant nature of indeterminate nodules and possibly avoid unnecessary testing.47
Another concern with CT screening is the potential for overdiagnosis due to unnecessary diagnosis of indolent tumours. As previously discussed, the clinical significance of such tumours detected by various screening modalities is not clear and the impact of this bias on mortality needs further investigation.
Finally, data on the cost-effectiveness of CT screening is limited. Using a decision model, one study suggested that the cost associated with low-dose CT screening is substantial. In this model, the screening cost among active smokers was $116 300 per quality-adjusted life-year gained. Screening was increasingly less cost-effective among just-quit smokers and former smokers.48 However, results from such mathematical models may not be valid, especially since some of the assumptions used to reach this conclusion are not verified in clinical studies as yet. Additional analyses on the cost-effectiveness of CT use will be sought once this screening modality is proven to be beneficial.
A cascade of chromosomal abnormalities precedes the development of lung cancer. Examination of sputum samples for these molecular alterations may permit early detection of lung cancer, particularly ones that are centrally located. The advantage of molecular testing is the opportunity for cancer detection before histological changes occur. Overexpression of a ribonuclear protein (hnRNP) A2/B1 has shown some promise in the early detection of lung cancer.53,54 Preserved samples from the JHLP were analyzed for upregulation of this protein. Among abnormal sputum samples (dysplasia), this biomarker showed a sensitivity of 91% and a specificity of 88% for the diagnosis of lung cancer within 2 years.55 In another study, detection of this biomarker in metaplastic bronchial epithelium was highly predictive of neoplasm.53 Other potential markers that may have a role in early cancer detection are ras mutations, P53 mutation/expressions and abnormal methylation and genomic instability (loss of heterozygosity and microsatellite alterations).54 Further studies are needed to validate the role of these biomarkers in lung cancer screening.
Recent advancements in molecular techniques have also made it possible to detect small amounts of tumour-related genes in the blood.56,57 However, the detection of these genes has not been correlated with tumour stage and its sensitivity for detecting early lung cancer is low. Cancer cells also produce aberrant proteins and the serum proteomic profile has been exploited in the detection of lung cancer.58 No data are available on the identification and validation of serum proteomic markers for early detection of lung cancer patients.