Fig. 19.1
Lead time represents the interval of time between screen detection and symptom-driven detection of cancer
Fig. 19.2
Length time bias is introduced by the propensity of screening to identify cancers (red) more indolent than their clinically detected counterparts (blue). Overdiagnosis (yellow) is an extreme form of length time bias and is due to cancers detected by screening that never would have become apparent otherwise
These concepts emphasize the importance of considering mortality as an endpoint rather than survival in lung cancer screening; this will be discussed further in a following chapter (limitations of LDCT screening for lung cancer; see below).
Low-Dose Computed Tomography: The Way of the Future?
Conventional chest radiographs have a poor sensitivity for lung cancer when compared to high-resolution computed tomography (HRCT) , particularly with early lung cancers. The probability of identifying stage I lung cancer with CXR has been estimated around 16%. The use of HRCT for the screening of lung cancer has long been hampered by the excessively stringent technical requirements of HRCT (acquisition times of several minutes with multiple breath holds using single-row detector machines) and the amount of cumulative radiation exposing patients to possible long-term risks of secondary malignancy (7 mSv, the equivalent of 2 years of background radiation). Low-dose computed tomography, however, has a sensitivity for lung nodules similar to that of conventional HRCT, but with a fraction of the radiation (1.5 mSv, equivalent to 6 months of background radiation). In addition, multi-row detector CT scans now allow full chest scans in less than 15 s with a single breath hold.
Numerous single-arm noncontrolled observational prospective studies using LDCT for lung cancer screening have been performed to date and have been reviewed elsewhere [21]. These studies have consistently reported a high detection rate of lung cancer at early stages, with excellent curability and survival rates. One of the most influential studies in that regard combined the results of the Early Lung Cancer Action Project (ELCAP) initiated in 1993 with those of the International Early Lung Cancer Action Project (I-ELCAP), an ongoing multicenter collaborative effort distributed across North America, Europe, Israel, and East Asia [22]. A total of 31,567 asymptomatic subjects at risk for lung cancer (including a minority of non-smokers at risk from occupational exposure or secondhand smoke) were screened from 1993 to 2005. A clearly defined protocol was made available to participating centers to guide the management of screen-detected lung abnormalities. A total of 484 subjects were diagnosed with lung cancer based on positive screening LDCT. The vast majority of these subjects (412 subjects, 85%) had clinical stage I lung cancer (77% were pathologic stage I), and the estimated 10-year survival for these patients was 88% and 92% for those undergoing surgical treatment within 1 month of diagnosis. The actual median follow-up was 40 months. It is noteworthy that 84% of these 412 subjects had lung cancer diagnosed on the first screening round (i.e., prevalent cancer) and that the vast majority of these cancers belonged to the adenocarcinoma spectrum of the disease (71%), a subset of lung cancer known to include more indolent tumors than in other cell types. Nonetheless, these encouraging results were in line with previously reported similar, though smaller in size, observational noncontrolled studies and supported the notion that LDCT may indeed represent an attractive strategy for lung cancer screening.
Another influential report published shortly after the I-ELCAP study reached apparently opposite conclusions. Using two validated lung cancer prediction models, Bach and colleagues collated the results of three other single-arm observational prospective studies on LDCT screening for lung cancer and compared overall lung cancer diagnoses, lung cancer surgical resections, advanced lung cancer diagnoses, and lung cancer deaths observed in these studies to what could have been expected in the same population in the absence of screening [23]. Similar to what had previously been described in the CXR studies, more lung cancers were identified (three times more) leading to ten times more surgical resections, but the numbers of advanced lung cancers and lung cancer deaths were not significantly different. Limitations of this study included a relatively short follow-up (median 3.9 years), a relatively wide 95% confidence interval (allowing for up to a 30% relative reduction in lung cancer mortality), and the reliance of prediction models rather than true control groups. This study suggested that overdiagnosis may indeed be a potential limitation of LDCT screening and that at least some of the screen-detected lung cancers could be fundamentally different than their clinically detected counterparts. If anything, these results reemphasized the importance of waiting for the long-anticipated NLST results before considering profound changes in recommendation for lung cancer screening.
Interestingly, another prediction model-based study using the Mayo LDCT screening trial data based on a different prediction model, the Lung Cancer Policy Model, reached conclusions very similar to those observed in the NLST [24]. This model differed from those used in the previous study in that it simulated survival based on individual disease characteristics, explicitly modeled benign disease, and, perhaps more importantly, incorporated competing causes of death, an important consideration as described above. Using the screening regimen used in the Mayo LDCT screening trial (five annual LDCT), the predicted relative reduction in lung cancer mortality was 28% at 6 years (number needed to screen to save one patient from lung cancer, or NNS = 205), while the relative reduction in overall mortality, including lung cancer mortality, was 3.6% at 6 years (NNS = 262). The discrepancy between lung cancer and overall mortality was attributed to the frequent coexistence of severe comorbidities, potentially lessening the impact of lung cancer screening in a population at risk not only for lung cancer but also for a host of other life-threatening conditions.
Two contemporary studies to the NLST—the Detection and Screening of Early Lung Cancer by Novel Imaging Technology (DANTE) and the Danish Lung Cancer Screening Trial (DLCST) —did not show a mortality benefit to lung cancer screening. The DANTE trial included subjects aged 60–74 years old with a 20 pack-year smoking history or more with 10 years or less since smoking cessation. The DLCST had the same tobacco exposure with age limited to 50–70 years old—with subjects having less tobacco exposure than those enrolled in the NLST. Still, significantly more stage I lung cancers were detected in the screening arm of the DANTE trial with 47 versus 16 cancers over the 5-year period. Similarly, significantly more early-stage lung cancers (stage Ia–IIb) were detected in the screening arm of the DLCST, 48 versus 21 over their 8-year follow-up period, raising the concern of lead time bias. Each trial has substantially less power to detect a mortality benefit with a cumulative of 6554 subjects between the two trials compared with the 53,454 subjects in the NLST. There is hope to pool multiple smaller European trials to better understand the benefits of LDCT screening for high-risk patients [8, 9].
The National Lung Screening Trial
The NLST was a large randomized controlled trial that enrolled 53,454 subjects at high risk for lung cancer at 33 US medical centers [5]. At-risk subjects were defined as being between 55 and 74 years of age with a significant smoking history (30 pack-years, having quit less than 15 years prior to enrollment for former smokers). Subjects were randomized to an experimental arm consisting of three annual screening LDCT or a control arm, consisting of three annual screening CXR, with a median follow-up of 6.5 years. The rationale for using CXR in the control group rather than the “standard of care” (i.e., no screening) was that the yet-to-be-released results of the PLCO could potentially show some benefit of CXR screening, in which case a study comparing LDCT to no screening would have been of lesser value.
A positive LDCT scan consisted of lung nodules of 4 mm or more (adenopathy or pleural effusion could also be considered a positive result), while any visible nodule or mass on CXR was considered positive. Overall, LDCT yielded positive results in 24.2% of cases, while CXR was considered suspicious for lung cancer in 6.9% of cases. A total of 1060 lung cancers were diagnosed in the LDCT group, 649 of which were diagnosed after a positive screening. In the CXR group, 941 cancers were diagnosed, 279 of which were identified after a positive CXR screen. Early-stage lung cancers were more frequent after a positive screening test, in both the LDCT and CXR groups. Stages I and II represented 70% of LDCT-detected lung cancers. Perhaps more importantly, stage IV lung cancers were less frequent in the LDCT than in the CXR group, supporting real stage shift, a sine qua non attribute of effective screening.
A total of 356 lung cancer-related deaths were observed in the LDCT group vs. 443 in the CXR group, representing a 20% relative reduction in lung cancer-specific mortality (NNS = 1/320). A statistically significant relative reduction in overall mortality of 6.7% (including lung cancer mortality) was also observed. Contrary to the explanation advanced in the Lung Cancer Policy Model study described above, the calculated NNS for overall mortality is only 220. While the significance of this observation deserves further study, one possible explanation is that LDCT may have additional health benefits that remain to be characterized.
While no clear guidelines for management of suspicious lesions detected on LDCT or CXR were provided to participating centers, the frequency of invasive investigations was low, most of the follow-up consisting of further imaging studies, and the complications from invasive investigations or surgery were rare (1.4% with at least one complication in the LDCT group and 1.6% in the CXR group). A total of 16 patients died within 60 days of an invasive procedure, six of whom did not have evidence of lung cancer. Bronchoscopy was performed in 76 of the 649 lung cancers identified by LDCT, resulting in four deaths, and 227 of the 17,053 subjects without lung cancer but abnormal LDCT, also resulting in four deaths.
This arguably represents the most significant advance in lung cancer management achieved over the past 20 years, and LDCT screening is being implemented in clinical practice based on these results with slight modifications from both the USPSTF and Medicare in terms of screening window and reimbursement [6, 7]. Clearly, LDCT screening allows for at least some clinically relevant lung cancers to be identified and treated earlier, resulting in significant improvement in mortality. However, a number of problems and unanswered questions will need to be addressed before the full benefits of LDCT screening may be appreciated.
Implementation of Lung Cancer Screening
For the reasons outlined above, the USPSTF formally recommended lung cancer screening in high-risk individuals in the late 2013. Based on modeling, they expanded the age criteria from that studied in the NLST to reach from ages 55 to 80 years old. The ATS/ACCP guidelines adhere strictly to the NLST criteria. The USPSTF recommends annual LDCT screening based on the NLST inclusion criteria: current or former smokers within 15 years of cessation aged 55–80 years old with a minimum of 30 pack-year tobacco exposure. Recommendations to cease screening include 15 years or more of tobacco abstinence for ages greater than 80 years old or inability to benefit from definitive therapy for lung cancer. The USPSTF expanded the upper limit of age eligibility based on modeling data, and Medicare has settled in between, reimbursing lung cancer screening for appropriately selected patients aged 55–77 years old [7].
The NLST depicts the theoretical “best-case” scenario of lung cancer screening, conducted mostly at major academic centers with multidisciplinary approaches to lung cancer and expertise in both thoracic radiology and oncology. Given that the preferred definitive management is thoracic surgery, access to this specialty will be essential to ensure the mortality benefit seen in the NLST [25]. There is concern that widespread implementation of LDCT screening across the community may not realize the same mortality benefit seen in the NLST. For these reasons, the ATS and ACCP released a joint statement outlining guidelines for implementation of a “quality” lung cancer screening program, many of which may become needed metrics for reimbursement: specifically a need for structured radiology reporting, a registry of screened patients, counseling regarding smoking cessation, and practitioners’ ability to adequately discuss the risks and benefits of lung cancer screening.
Practical concerns exist regarding cost-effectiveness and the ability to best target the patient population. The NLST included a higher proportion of former smokers with higher education status than seen in the general population of current smokers in the USA [26]. Current US smokers are more likely to be less educated, report poorer health status, and not have access to regular medical care. Additionally, current smokers tended to adopt a more fatalistic attitude toward lung cancer screening [27]. This raises logistic challenges regarding how to best reach and counsel the high-risk group in which screening is being implemented.
The psychological effects of lung cancer screening are unknown including the impact on mental health of incidental findings as well as the impact on tobacco use; however, the data available suggest that without clear communication between ordering providers and patients, there is the real possibility for psychological distress on behalf of the patient [28]. With 39.1% of patients in the NLST experiencing at least one positive screen, the psychological burden of screening will need attention [5].
Limitations of Screening with Low-Dose Computed Tomography and Clinical Implications
The relative reduction in lung cancer-specific mortality of 20% reported in the NLST is a compelling argument for integrating LDCT in a screening program for lung cancer, particularly when contrasted with existing mass screening programs such as mammogram for breast cancer, associated with conservative estimates for number needed to screen 1/2500 (vs. 1/320 in the NLST) [29]. A number of important questions not directly answered by the NLST will however need to be addressed as LDCT screening is adopted on a larger scale. These include the questions of cost-effectiveness, the high frequency of false-positive studies, the persistent question of overdiagnosis, and the long-term risk of exposure to ionizing radiations. The question of false-positive management and overdiagnosis will be discussed here.