Screening by Low-Dose Computed Tomography
False-positive exams
False-positive exams are particularly problematic in the context of lung cancer screening. The individuals most likely to be screened for lung cancer, (i.e., heavy smokers) have comorbidities, such as chronic obstructive pulmonary disease and heart disease, that make them poor candidates for certain diagnostic procedures.
False-positive test results must be considered when lung cancer screening with low-dose computed tomography (LDCT) is being evaluated. A false-positive test may lead to anxiety and invasive diagnostic procedures, such as percutaneous needle biopsy or thoracotomy. The percentage of false-positive findings varies substantially among studies and is primarily attributable to differences in how a positive scan is defined (the size criteria), the thickness of the slice used between cuts (smaller slice thicknesses lead to detection of more nodules), and whether the subject resides in a geographic location where granulomatous disease is highly prevalent.
In the National Lung Screening Trial (NLST), the false-positive rate was 24% at baseline, and 27% and 16% for the two subsequent screening rounds. In a systematic review of 20 studies (including the NLST), the median false-positive rate was 20.5% (range, 1%–49%) on baseline screens and 9.5% (range, 1%–42%) on postbaseline screens. False-positive rates are generally lower on postbaseline screens because a nodule’s growth rate can be assessed when there is a previous screen available, and stable (nongrowing) nodules are often denoted as negative screens. The Lung-RADs criteria for assessing LDCT findings, which are in wide use in the United States, are stricter than the NLST criteria for defining a positive screen, and have the potential to lower the false-positive rate from that seen in the NLST.
Diagnostic evaluations and downstream complications
A systematic review of the benefits and harms of computed tomography (CT) screening for lung cancer summarized 21 studies with respect to various diagnostic outcomes, although not all studies reported on all outcomes. The rate of diagnostic CT imaging after a reported nodule varied from 0% to 45% of all individuals who were screened. Positron emission tomography scanning was performed in 2.5% to 5.5% of individuals who were screened. The frequency of nonsurgical biopsies or procedures ranged from 0.7% to 4.4% of individuals who were screened, with the finding of a benign result on biopsy ranging from 6% to 79%. The rate of surgical resection for screen-detected nodules was between 0.9% and 5.6% of individuals who were screened; the proportion among these with a benign result ranged from 6% to 45%.
In the NLST, most major complications were related to invasive procedures and surgeries performed on patients diagnosed with lung cancer, with a major complication rate of 11.8%. The rates of complications from the NLST may not be generalizable to a community setting; participants in the NLST were younger, better educated, and less likely to be current smokers (therefore, healthier) than the population of smokers and former smokers in the general U.S. population who would be eligible for screening. Of note, 82% of the participants were enrolled at large academic medical centers, and 76% of the participants were enrolled at National Cancer Institute–designated cancer centers. However, diagnostic follow-up did not necessarily occur at the NLST screening centers and could have been carried out in community settings. This may account for the low complication rate and surgical mortality rate (1%) found in the NLST. These findings led the multisociety position paper to strongly recommend that screening be carried out at centers with the same patient-management resources as those in the NLST.
A retrospective cohort study of community practices indirectly estimated the complication rates and downstream medical costs of invasive diagnostic procedures performed for lung abnormalities identified through lung cancer screening. The observed complications rates of 22.2% (in patients aged 55–64 years) and 23.8% (in patients aged 65–77 years) were more than twice that reported in the NLST (8.5%–9.8%). The mean costs of managing complications ranged from $6,320 (minor complication) to $56,845 (major complication). These data suggest that the NLST, which was conducted in the context of a controlled clinical trial, may have underestimated the potential for adverse events and high downstream costs in the community setting. Study limitations include a lack of information about patient eligibility for lung cancer screening, the fact that the diagnostic procedures were not generally performed as follow-up to screening, and the extent to which complications were affected by poorer patient health and lower quality of care. Despite limitations, these results reinforce the need for the discussion about risks, benefits, and shared decision making.
Overdiagnosis
A less familiar harm is overdiagnosis, which means the diagnosis of a condition that would not have become clinically significant had it not been detected by screening —that is, had the patient not been diagnosed with the cancer, the patient would have died of other causes. In the case of screening with LDCT, overdiagnosis could lead to unnecessary diagnosis of lung cancer requiring some combination of therapy (e.g., lobectomy, chemotherapy, and radiation therapy). Autopsy studies suggest that a significant number of individuals die with lung cancer rather than die of lung cancer. In one study, about one-sixth of all lung cancers found at autopsy had not been clinically recognized before death. This may be an underestimate; depending on the extent of the autopsy, many small lung cancers that are detectable by CT may go unrecorded in an autopsy record. Studies in Japan provided additional evidence that screening with LDCT could lead to a substantial amount of overdiagnosis.
One approach to assessing overdiagnosis involves examining the volume-doubling time of lung tumors detected on LDCT. In one study, the volume-doubling times of 61 lung cancers were estimated by using an exponential model and successive CT images. Lesions were classified into the three following types: type G (ground glass opacity), type GS (focal glass opacity with a solid central component), and type S (solid nodule).
The mean volume-doubling times were 813 days, 457 days, and 149 days for types G, GS, and S, respectively. In this study, annual CT screening identified a large number of slowly growing adenocarcinomas that were not visible on chest x-ray, suggesting overdiagnosis.
In a screening cohort with more than 5,000 participants, volume-doubling time was used as a surrogate for overdiagnosis. Patients with a calculated volume-doubling time of more than 400 days before surgical resection were considered to have a slow-growing or indolent cancer. The investigators discovered that 25% of incident cancers (31 of 120) met the criteria of a slow-growing or indolent tumor. This rate is consistent with previous chest radiograph screening studies and for other solid tumors.
Another approach to assessing overdiagnosis is to compare lung cancer incidence rates across arms in randomized trials of LDCT screening. Data from the NLST showed a gap of about 120 excess lung cancer cases in the LDCT group compared with the chest radiograph group after a medium follow-up of 6.5 years (i.e., 4.5 years after the last scheduled screen). This suggests that 18% of screen-detected lung cancers (N = 649) were overdiagnosed. However, an extended follow-up analysis of the NLST based on a median of 11.3 years of follow-up for incident cancer found a much smaller, and nonstatistically significant, excess of only 20 cancers in the LDCT group, resulting in an estimate of the percentage of overdiagnosed LDCT screen-detected cancers of 3%. Note that the NLST control group was screened with chest x-rays, so technically the above overdiagnosis estimates were in comparison with what would have been diagnosed with chest x-ray screening, not with what would have been diagnosed with no screening.
Additional evidence of overdiagnosis with LDCT screening was observed in the randomized Danish Lung Cancer Screening Trial. At 10 years of follow-up (5 years after the last screening exam), almost twice as many lung cancers had been diagnosed in the screening group as in the control group: 5.1 vs. 2.7 cases per 1,000 person-years or 100 vs. 53 lung cancer cases in 4,104 total participants, respectively. Most of the lung cancers were early-stage adenocarcinomas, with no statistically significant difference in the number of stage III and IV cancers between the two groups. Overdiagnosis was estimated at 67%. In three other small trials of LDCT screening, one showed a borderline significant increase in lung cancer incidence in the LDCT versus the control arm (P = .04), suggesting overdiagnosis, while there was no significant difference in lung cancer incidence across arms in the other two trials. In the NELSON trial (Nederlands–Leuvens Longkanker Screenings Onderzoek), with 4.5 years follow-up after the last screen, the overdiagnosis rate was 19.7% (95% confidence interval , -5% to 42%).
The overdiagnosis estimates from the NLST are compared with what would have been diagnosed with chest x-ray screening; therefore, in order to interpret them, it is necessary to have an estimate of the level of overdiagnosis using chest x-ray screening, preferably, covering a time period and population similar to those in the NLST. Such an estimate comes from the U.S. Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial of chest x-ray screening versus usual care, specifically in the subset of PLCO trial participants who met the NLST eligibility criteria. These data showed no evidence of overdiagnosis, with essentially equivalent numbers of diagnosed lung cancers in the chest x-ray and usual-care arms after 3 years of follow-up after the last scheduled screen (rate ratio, 1.00).
A meta-analysis of overdiagnosis from six randomized controlled trials, including the NLST and the NELSON, showed an aggregate overdiagnosis rate of 0.30 (95% CI, 0.06–0.55). The overdiagnosis rate was defined as the difference across arms in incident lung cancers divided by the number of screen-detected cases in the LDCT arm. However, there was significant heterogeneity (P = .0001) in the overdiagnosis rate across trials, with two small trials showing rates around 0.65 and the NLST showing a low rate of 0.04.
Radiation exposure
Another potential risk from screening with LDCT is radiation exposure. The average exposure is low; the mean effective dose for LDCT in the NLST was 1.4 (SD = 0.5) mSv. It is estimated that over a 3-year period of screening, NLST participants were exposed to an average of 8 mSv of radiation (which accounts for radiation from screens and additional imaging for screen-detected nodules). A study of LDCT screens that were performed on more than 12,000 patients from 2016 to 2017 at 72 U.S. institutions found a mean effective dose of 1.2 (SD = 1.1) mSv. Almost two-thirds (65%) of the institutions had a median effective dose higher than the American College of Radiology guideline of 1 mSv. Modeling from previous work on radiation exposure and the development of cancer suggests that there could be one death per 2,500 screens in those participating in a screening program such as the NLST, although the benefit of screening of about one death avoided per 960 screens substantially outweighs the risk. Younger individuals and those without a significant risk of lung cancer may be more likely to suffer a radiation-induced lung cancer from screening than to be spared a lung cancer death.
Screening by Chest X-ray and/or Sputum Cytology
False-positive exams
In the PLCO Cancer Screening Trial, the false-positive rate with chest x-ray screening ranged from 6.8% to 8.7% per exam over the four screening rounds. In the NLST chest x-ray arm, false-positive rates were generally similar (range, 4.7%–8.7% over three rounds).
Diagnostic evaluation and downstream complications
In the NLST chest x-ray arm, among subjects with positive screens at baseline, 86% received imaging as diagnostic follow-up, 5% received a bronchoscopy, and 5% underwent a surgical procedure. Diagnostic imaging rates were modestly lower after postbaseline positive screens, while bronchoscopy and surgery rates were similar. A total of 0.3% of false-positive screens were associated with a complication of an invasive diagnostic procedure.
In the PLCO trial, 0.4% of participants with at least one false-positive screen who had a diagnostic evaluation had a complication associated with a diagnostic procedure. The most common of the 69 complications were pneumothorax (29%), atelectasis (15%), and infection (10%).
Overdiagnosis
In the Mayo Lung Project trial of screening with chest x-ray and sputum cytology, after 5 years of follow-up after the last scheduled screen, 206 cancers were diagnosed in the screening arm compared with 160 cancers in the control arm. Based on 90 screen-detected cancers in the screened arm, the overdiagnosis rate would be computed as 51% (i.e., /90). After 13 years of follow-up in the PLCO trial, 1,696 lung cancers had been diagnosed in the intervention arm as compared with 1,620 cancers diagnosed in the usual-care arm, suggesting that about 25% of the 307 chest x-ray screen-detected cancers in the trial were overdiagnosed. However, the incidence of lung cancer was not statistically different between the intervention and usual-care arms in the PLCO trial (rate ratio, 1.05; 95% CI, 0.98–1.12), indicating that the null hypothesis of no overdiagnosis could not be rejected.