|Home | About | Journals | Submit | Contact Us | Français|
The 5-year overall survival rate of lung cancer patients is approximately 15%. Most patients are diagnosed with advanced-stage disease and have shorter survival rates than patients with early-stage disease. Although screening for lung cancer has the potential to increase early diagnosis, it has not been shown to reduce lung cancer mortality rates. In 1993, the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial was initiated specifically to determine whether screening would reduce mortality rates from PLCO cancers.
A total of 77464 participants, aged 55–74 years, were randomly assigned to the intervention arm of the PLCO Cancer Screening Trial between November 8, 1993, and July 2, 2001. Participants received a baseline chest radiograph (CXR), followed by three annual single-view CXRs at the 10 US screening centers. Cancers were classified as screen detected and nonscreen detected (interval or never screened) and according to tumor histology. The positivity rates of screen-detected cancers and positive predictive values (PPVs) were calculated. Because 51.6% of the participants were current or former smokers, logistic regression analysis was performed to control for smoking status. All statistical tests were two-sided.
Compliance with screening decreased from 86.6% at baseline to 78.9% at the last screening. Overall positivity rates were 8.9% at baseline and 6.6%–7.1% at subsequent screenings; positivity rates increased modestly with smoking risk categories (Ptrend < .001). The PPVs for all participants were 2.0% at baseline and 1.1%, 1.5%, and 2.4% at years 1, 2, and 3, respectively; PPVs in current smokers were 5.9% at baseline and 3.3%, 4.2%, and 5.6% at years 1, 2, and 3, respectively. A total of 564 lung cancers were diagnosed, of which 306 (54%) were screen-detected cancers and 87% were non–small cell lung cancers. Among non–small cell lung cancers, 59.6% of screen-detected cancers and 33.3% of interval cancers were early (I–II) stage.
The PLCO Cancer Screening Trial demonstrated the ability to recruit, retain, and screen a large population over multiple years at multiple centers. A higher proportion of screen-detected lung cancers were early stage, but a conclusion on the effectiveness of CXR screening must await final PLCO results, which are anticipated at the end of 2015.
Previous lung cancer screening trials showed that screening did not decrease mortality. However, several shortcomings in these trials may have failed to detect a small screening benefit.
Participants in the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, who were recruited at 10 screening centers across the United States between 1993 and 2001, were randomly assigned to chest radiograph screening vs a usual-care control group. Compliance with screening, screen positivity rates, and positive predictive values (PPVs) were also calculated to assess the effectiveness of screening. All values were adjusted for tobacco smoking. The stages and histological types of screen-detected lung cancers were also determined.
The compliance rate with screening was good. Screen positivity rates from the baseline to the final annual screening modestly decreased. Among participants with positive screens, the PPVs were associated with smoking and type of radiological abnormality (mass vs nodule). More than half of the lung cancers diagnosed during the screening period were screen detected; non–small cell lung cancers (NSCLCs) were detected at a higher frequency than other types. A substantially high proportion of NSCLCs were early-stage cancers.
PLCO maintained high participant compliance throughout the study period, and screening robustly detected potentially curable early-stage lung cancers.
Whether lung cancer screening reduces mortality is not yet known; the PLCO trial final results are anticipated in 2015.
From the Editors
Lung cancer accounts for 12% of all cancers and has the highest annual mortality rate in men and women, resulting in greater than a million deaths worldwide (1). In the Unites States, approximately 160000 patients (29% of all cancer deaths) die from lung cancer every year, exceeding deaths from colorectal, breast, pancreatic, and prostate cancers combined (2). Lung cancer is often diagnosed at an advanced stage, and only 30% of lung cancers are potentially resectable at the time of diagnosis. Approximately 85% of lung cancer patients die from the disease (2).
Lung cancer, when diagnosed at an early stage, has a 5-year survival rate as high as 60%–70% (3–5). Therefore, early detection of lung cancer may reduce mortality; however, nonrandomized studies of lung cancer screening in the United States and United Kingdom using chest radiograph (CXR) alone or combined with sputum cytology (6–12) showed no reduction in mortality. Four randomized controlled trials (13–16) that each included 6300–11000 participants evaluated the benefits of lung cancer screening using CXR alone or in combination with sputum cytology in male smokers and found no reduction in mortality rate. A more recently updated systematic review of lung cancer screening trials by the Cochrane Library (17) concluded that there is currently no evidence supporting lung cancer screening, and data from previous trials even suggested a slightly higher mortality associated with frequent CXR screening.
In each of the above-mentioned trials (13–16), participants in the control group received CXRs. This contamination of the control group, as well as relatively small samples sizes, raised concerns that a small but clinically important benefit from screening may have gone undetected. In 1993, the National Cancer Institute (NCI) initiated the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial (18) to determine whether the screening programs reduce mortality in PLCO cancers. The lung cancer component of the PLCO trial differed in several important respects from previous lung cancer screening trials, specifically with respect to the inclusion of women and never-smokers, the absence of scheduled CXR in the control group, and the large sample size. Previously, we published the results of the baseline lung cancer screening of the PLCO Screening Trial (19). In this study, we describe the results of the completed lung cancer screening through the final round of annual screening. To ascertain the effectiveness of PLCO screening, we analyzed the compliance with lung cancer screening, screen positivity rate, positive predictive value (PPV), diagnostic follow-up, and characteristics of all diagnosed lung cancers.
The trial registration number (on ClinicalTrials.gov) is NCT00002540. The components of the two-arm PLCO trial that includes a detailed description of study design are reported elsewhere (18). At 10 screening centers across the United States, men and women, aged 55–74 years, who reported no prostate, lung, colorectal, or ovarian cancer, or undergoing treatment for any cancer other than nonmelanoma skin cancer, were included in the study. The flow of patients into the trial and through four rounds of screening is shown in Figure 1. Between November 8, 1993, and July 2, 2001, a total of 154934 participants were enrolled. Participants were randomly assigned to the screening (intervention) arm (n = 77464) or the control (usual care) arm (n = 77470). Because of a duplicate randomization, there is one less intervention arm participant than in our publication of the baseline screening results (19). Exclusion criteria were participation in another cancer screening or primary prevention trial; having taken finasteride in the 6 months before entry; previous surgical removal of the entire prostate, one lung, or the entire colon; men who have had more than one prostate-specific antigen test; and men and women who have had any colonoscopy, sigmoidoscopy, or barium enema examination in the past 3 years. Recruitment was targeted to healthy volunteers primarily through direct mail. All participants signed informed consent documents approved by both the National Cancer Institute and their local institutional review board. The participants were to be followed for 13 years from the time of random assignment, ending the trial in 2015.
The screening evaluation for lung cancer was a single-view posterior–anterior CXR. Current and former smokers underwent an initial CXR screening for lung cancer at baseline (T0), followed by annual screens through year 3 (T1–T3). At some screening centers, the baseline T0 screen was conducted immediately following random assignment but at others was scheduled at a subsequent time. Never-smokers initially followed the same protocol, but a protocol amendment effective December 7, 1998, changed the eligibility for CXR at T3 to current or former smokers only. However, about one-third of never-smokers also received the third annual CXR because they were previously scheduled or requested the T3 examination. CXRs were defined as positive when the radiologist identified a mass, nodule, infiltrate, or any other abnormalities considered “suspicious” for cancer. A mass was defined as a lesion greater than 3 cm; a nodule was less than 3 cm. Screens at later rounds were interpreted independently, without comparison to earlier examinations. Following a positive screen, participants and their physicians were notified of the result and the participants were referred to their primary health-care provider for the management of cancer. The PLCO trial did not specify or recommend any diagnostic algorithm. Subjects received annual study update forms that asked about any cancer diagnoses occurring in the absence of a positive screen. Medical records were obtained to document diagnostic follow-up and the characteristics of any diagnosed lung cancers. Medical records were abstracted by Certified Tumor Registrars with minimum 2 years experience. Quality assurance measures include reabstraction of 50 lung cancer diagnoses per year. Cancers were classified as non–small cell lung cancers (NSCLCs), small-cell lung cancers, or carcinoid tumors. The NSCLCs were subclassified as adenocarcinomas, squamous carcinomas, large-cell carcinomas, or other histologies. Staging of cancers was based on the fifth edition of the American Joint Committee on Cancer Staging Manual (20).
All lung cancers diagnosed through the end of the last eligible screening year (T2 for never-smokers per revised protocol; T3 for everyone else) were included in this analysis. Screen-detected cancers were defined as those diagnosed within a window extending 9 months from a positive screen or 9 months from a diagnostic evaluation that was linked to the positive screen. Non–screen detected cancers were classified as interval if the participant had at least one previous PLCO screen; otherwise, they were denoted as never-screened cancers.
We calculated the screen positivity rate as the number of positive screens divided by the number of subjects screened. The PPV of a screen was the proportion of subjects with a positive screen who had screen-detected cancer in the same screening round. SAS software (SAS Systems for Windows 9.2, Cary, NC) was used for statistical calculations. We used χ2 tests to assess the statistical significance of differences between proportions. All statistically significant tests were two-sided. Because of changes in the protocol, mentioned above, the screened population at T3 was more heavily weighted with smokers than at T0–T2 screening rounds. Because smoking also influences the screening outcomes, it is a potential confounder in the relationship between screening outcomes and screening round; thus, overall comparisons between findings at baseline (T0) vs follow-up (T1–T3) screening rounds were assessed using a logistic regression model that controlled for smoking using groups never-smokers and ever-smokers, that included all of the smoking categories.
A total of 77464 participants were randomly assigned to the intervention arm of lung cancer screening. The baseline characteristics of the participants are shown in Table 1. Approximately 50.5% (n = 39115) of total participants were women and 49.5% (n = 38349) were men. A high proportion, approximately 64%, of men and women were aged 55–64 years at enrollment. More than half of the participants were current or former smokers (ever-smokers), but among women, more than half were never-smokers. The proportion of current smokers was about half the rate of active smokers among the adult population in the United States in 2010. Participants were predominantly White, and Blacks and Hispanics were underrepresented in comparison with the US population. Asian participants were primarily from the screening center in Hawaii, but inclusion was proportional to the Asian population of the United States. Participants had a higher level of education than the general population in the United States.
Compliance with screening by round and demographic characteristics is shown in Table 2. Compliance showed a modest but statistically significant decrease from baseline to later screening rounds (P < .001). Compliance at T0–T3 was higher among men than among women (P < .001). With respect to smoking status, compliance was lowest among current smokers and highest among former smokers who quit more than 15 years ago, with intermediate rates among never-smokers, and former smokers who quit within 15 years of enrollment (P < .001). Compliance with completion of annual study update was greater than 92% in each screening round. Medical records were obtained for all (100%) patients with lung cancers.
Results of the screening examinations are summarized in Table 3. At each screening round, the lowest positivity rate was observed in the never-smoker group, and the rates increased modestly with smoking risk category (Ptrend < .001). In addition, within each smoking category, positivity rates decreased modestly from T0 to T3. The rate decreased from 7.9% at T0 to 6.1% at T3 in never-smokers and 11% at T0 to 9.4% at T3 in current smokers and other categories of smokers showing a range of intermediate rates (Ptrend < .001). Across screening rounds, positivity rates were also statistically significantly higher in men (P < .001), in older subjects (Ptrend < .001), and in those with a family history of lung cancer (P < .001). These differences by sex, age, and family history were more pronounced at the earlier screening rounds, as compared with the latter ones. The level of education was not associated with screen positivity.
The screen positivity rates were influenced by the previous screening experience of the subjects (data not shown in Table 3). At T1, T2, and T3, 68%, 58%, and 52% of positive screens, respectively, were in participants with no prior positive screens. In participants with only previous negative screens, a subsequent screen was positive 4.7% of the time, as compared with 23% of the time in participants with at least one previous positive screen (P < .001). Overall, 18.5% of screened participants had at least one positive screening test (data not shown).
The number and percentage of cancers diagnosed by different diagnostic procedures following positive screens by screening round and smoking status are summarized in Table 4. In general, the rates of computed tomography (CT) scans increased modestly across the hierarchy of smoking risk categories from never- to current smokers (Ptrend < .001). Overall, the CT scan rates at T0 were slightly higher than the rates at T1–T3 (P < .001).
The biopsy rates at each screening round increased sharply across the smoking risk categories (Ptrend < .001). Because the proportion of biopsies that were positive also increased with smoking risk, the PPVs increased even more sharply across the smoking risk categories, as compared with biopsy rates. Across all categories of current and former smokers, both biopsy rates (P < .001) and PPVs (P = .004) were statistically significantly greater at T0 than in subsequent screening rounds combined, the differences being greatest among current and recent former smokers. The trends in cancer yield (per 1000 screened) were similar to those of PPV (P < .001).
In addition to smoking status and study year, PPV was also examined as a function of sex, family history of lung cancer, and type of radiographic abnormality (data not shown in Table 4). Overall, across all screening rounds, although PPV was borderline statistically significantly higher in men than in women (1.9% in men vs 1.5% in women) (P = .08), this relationship no longer existed when a logistic regression model that controlled for smoking was applied (P = .8).
A statistically significantly higher PPV was also observed among subjects with a family history of lung cancer (2.8%) than among those without such a history (1.5%). This association between family history and lung cancer persisted even when adjusted for smoking (P ≤ .001).
The type of abnormality—mass vs nodule or any other abnormality—had a large effect on PPV. Although a mass was observed on only 6.8% of all positive screens, 35% of all screen-detected cancers occurred in participants with a mass detected in the preceding screening examination. Across all screening rounds, the PPV for a mass was 8.7%, compared with 1.2% for a nodule or other abnormality (P < .001).
Of 77464 participants in the intervention arm of the trial, a total of 564 (0.7%) participants were diagnosed with lung cancer during the screening period (Table 5). Out of 564 lung cancers, 306 (54%) were screen-detected cancers, 183 (32%) were interval cancers, and 75 (13%) were diagnosed in participants who were never screened. The histological types of lung cancers are described in Table 5. An additional 13 carcinoid tumors are not discussed further in this publication. Screen-detected lung cancers were statistically significantly more likely to be NSCLC as opposed to small-cell lung cancers than either interval cancers or never-screened cancers (P < .001 and P = .006, respectively). Furthermore, among the NSCLCs, adenocarcinomas comprised a statistically significantly higher proportion of screen-detected cancers than either the interval cancers or the never-screened cancers (P = .003 and P = .004, respectively).
Among the screen-detected NSCLCs, a statistically significantly higher proportion was stage I cancers, as compared with either interval cancers or never-screened cancers (P < .001), and 59.6% of screen-detected cancers were stage I–II, compared with 33.3% of interval cancers (Table 6). The stage distribution among screen-detected NSCLCs was similar in T0 and T1–T3. Among small-cell lung cancers, screen-detected cancers were less likely to be stage IV, compared with interval cancers (P = .01).
Of 564 cancers, 220 cancers were diagnosed in women and 344 cancers in men, corresponding to raw (unadjusted for smoking) rates of 15.5 and 23.6 per 10000 person-years, respectively (P < .001). After controlling for smoking categories (never, quit <15 years, quit >15 years, and current), men had a relative risk of 1.28 (95% confidence interval = 1.08 to 1.53, P = .004) for lung cancer, compared with women. Although a similar proportion of cancers in men and women were NSCLCs (86.0% and 89.5%, respectively), among the NSCLCs, adenocarcinomas were more frequent in women than in men (62.9% vs 46.6%, respectively, P < .001). Also among NSCLCs, women had a modestly greater rate of stage I disease, compared with men (45.2% vs 37.5%); however, this difference was not statistically significant (P = .09).
The distribution of lung cancers by smoking risk category is shown in Table 7. The incidence rates in never-smokers, smokers who quit more than 15 years ago, smokers who quit less than 15 years ago, and current smokers were 2.8, 12.2, 40.6, and 70.9 per 10000 person-years, respectively (P < .001). The proportion of cancers that were NSCLC decreased slightly across risk categories from 94% in never-smokers to 84% in current smokers (P = .04). Among NSCLCs, adenocarcinomas were statistically significantly less frequent among former smokers who quit less than 15 years ago and current smokers (45.8% and 48.7%, respectively) than among smokers who quit more than 15 years ago and never-smokers (67.8% and 90.3%, respectively) (P < .001). The stage of NSCLCs was similar in all ever-smoker categories (current and former smokers), approximately 39%–40% in each category being stage I; in contrast, never-smokers had a statistically significantly greater proportion (58%) of NSCLCs being stage I (P = .04). A family history of lung cancer was associated with an increased risk of lung cancer (relative risk [RR] = 2.04, 95% confidence interval = 1.66 to 2.56, P < .001); this association persisted when controlling for smoking (P < .001).
This report describes the findings of lung cancer screening with CXR in the intervention arm of the PLCO trial and extends our observations from the baseline screen (19) to three subsequent annual screens. With additional screening rounds and follow-up time, we are now able to assess changes over time in screening results, diagnostic follow-up, and characteristics of screen-detected cancers as well as compare screen-detected cancers with interval cancers. Screening compliance remained high throughout the study and decreased only slightly from baseline (T0) to the fourth annual screening round (T3). Screen positivity rates, which decreased modestly from baseline to T3, showed a high degree of dependence on past results, evidenced by the fact that subjects with a previous positive screen were much more likely to have a positive screening result (23%) than subjects with no previous positive screens (4.7%). At screening round T3, almost one half of positive screens were in subjects with previous positive screens. Yet, for the entire screening period, the majority of subjects with a positive screen had a negative screen the following year.
Our results showed that among all participants with positive screens, the PPV of a positive screen was 1.7%; among current and former smokers who quit within 15 years, PPVs were 4.7% and 3.4%, respectively. In the T0 round, PPVs for all, current, and former smokers who quit within 15 years were 2.0%, 5.9%, and 4.7%, respectively; this round was most comparable to the Mayo Lung Project prevalence screen (13). In the Mayo Lung Project prevalence screen, the PPV for a suspicious screen was 59%, but if “indeterminate” screens were also considered positive, the PPV decreased to 18.6% (13). In the Johns Hopkins Lung Cancer Screening Trial, the PPV for screens that were considered to be “cancer” or suspicious was 11.2% (15). In the Memorial Sloan-Kettering Lung Cancer Screening Trial, the PPV was 3.8% for indeterminate plus suspicious screens (14). These previous trials included only smokers or former smokers. In PLCO, suspicious and indeterminate abnormalities were considered positive. In recent spiral CT scan lung cancer screening trials, which also included primarily smokers, PPVs ranged from 1.7% to 12.8% (21). The PPV for prostate-specific antigen screening for prostate cancer is 10.5%–32% (22,23), for digital rectal examination 2.9%–21% (22,23), for mammography 3%–12.7% (24,25), and for stool occult blood (for detection of colorectal cancer) 2.2%–17.2% (26). Variability in PPVs is dependent on the definition of a positive screen, disease prevalence in the screened population, and the sensitivity and specificity of the screening test. In this context, the PPV for CXR in the PLCO trial seems consistent with other studies.
We observed that the PPV of an abnormal screen decreased moderately from the baseline to later annual screening rounds. The presence of repeat positive screens at later rounds contributed to this decrease but only explained a relatively small portion of it. Breaking down the post-baseline positive screens into first and repeat positives, the PPV was 4.0% among first positives compared with 3.0% among repeat positives. Thus, even among first positive screens, the PPV was still about 25% less at subsequent rounds than at baseline. Across all screening rounds, type of radiological abnormality and smoking status had the greatest association with PPV. The PPV for a mass was 8.7%, which is approximately seven times higher than that (1.2%) for a nodule or other abnormality; the PPV for current smokers was 4.7%, which is approximately 11 times higher than that (0.41%) for never-smokers.
Roughly half of the screen-detected cancers were potentially curable stage I or stage II NSCLC. This stage distribution was generally similar to that seen among screen-detected lung cancers in the Mayo Clinic, Memorial Sloan-Kettering, and Johns Hopkins randomized screening trials (13–15). In both reports by the Memorial Sloan-Kettering and Mayo Clinic, early-stage lung cancers represented a statistically significantly greater proportion of later vs baseline round screen-detected cancers. In contrast, in the PLCO trial, the stage distribution of screen-detected cancers at baseline and at subsequent screening rounds was essentially identical. This difference between the PLCO findings and the Mayo and Memorial Sloan-Kettering findings may relate to differences in the PLCO population, including inclusion of never-smokers and women. In addition, 53% of PLCO participants had a CXR within 3 years of random assignment, which might have reduced the number of advanced-stage cancers discovered at T0. As in PLCO, the interval cancers in all studies had a less favorable stage distribution than the screen-detected cancers. Because of well-known potential screening-related biases of overdiagnosis and length-biased sampling, comparison with an unscreened control group is necessary to conclusively determine whether a true reduction in the rate of advanced-stage and increase in early-stage lung cancers (stage shift) have occurred. Such a comparison in PLCO must await the final outcome of the trial.
In this study, the histological distribution of lung cancers was generally consistent with that observed in nonscreening populations in the United States. Adenocarcinoma was the most common NSCLC histology, consistent with the reported trends (27,28). Also consistent with the trends in the United States was the increased relative frequency of adenocarcinoma in women, as compared with men (27).
Tobacco smoking is a major risk factor for lung cancer. Over 4 years in this PLCO Screening Trial, where just over half of the cancers were detected by screening, the observed relative risks for current, recent former, and distant former smokers, as compared with never-smokers, were 25, 14.5, and 4.4, respectively, based on person-year incidence rates. The relative risks are generally concordant with, perhaps a bit higher than, those observed in nonscreening populations (29–31). These findings suggested that screening with CXR did not preferentially detect lung cancers in never-smokers. In contrast, there was some evidence that screening with low-dose CT scans preferentially detected cancers, perhaps overdiagnosed, in never-smokers. In a mass screening program with CT in Japan (32), the detection rate in women never-smokers (45/10000) was essentially equivalent to that in male smokers (49/10000), but there were too few women smokers and male never-smokers to make direct within-gender comparisons.
The limitations of the PLCO lung cancer screening study are an underrepresentation of Black race and Hispanic ethnicity, limited availability of occupational and environmental exposure data, and the fact that the trial population represents a higher socioeconomic status than the general population based on educational level, which may limit the ability to apply these results to the general population.
The strengths of the PLCO Lung Cancer Screening Trial include the large sample size, inclusion of women and never-smokers, excellent long-term follow-up of participants, and availability of the biorepository. In summary, the results of this study showed that PLCO has maintained excellent participant compliance and continues to robustly detect lung cancers at later screening rounds and that these screen-detected cancers at later rounds have a similar favorable stage distribution as those detected at baseline. A determination of whether early detection with CXR will translate into a mortality benefit of screening awaits the final PLCO results, which based on design assumptions are anticipated in 2015. In addition, the National Lung Screening Trial, a randomized trial comparing low-dose CT with CXR, is ongoing in the United States (33), and results from the PLCO trial will be critical to interpretation of that study.
This study was supported by contracts from the Division of Cancer Prevention, National Cancer Institute (NCI), National Institutes of Health to the screening centers, Westat Inc, and Information Management Services.
D. A. Lynch receives research support from Siemens Inc for research into positron emission tomography-computed tomography of pulmonary nodules. The sponsors had no role in the study design, collection of data, analysis and interpretation of the data, decision to submit the manuscript for publication, and the writing of the manuscript.
The National Cancer Institute (NCI) participated in the trial design and provided oversight during the conduct of the study. We thank the investigators of the screening centers: Lombardi Cancer Research Center, Georgetown University, Washington, DC; Henry Ford Health System, Detroit, MI; Huntsman Cancer Institute, University of Utah School of Medicine, Salt Lake City, UT; Marshfield Clinic Research Foundation, Marshfield, WI; Pacific Health Research Institute, Honolulu, HI; University of Alabama at Birmingham School of Medicine, Birmingham, AL; University of Colorado Health Sciences Center, Denver, CO; University of Minnesota School of Public Health, Minneapolis, MN; Pittsburgh Cancer Institute/Magee Women’s Hospital, University of Pittsburgh, Pittsburgh, PA; Washington University School of Medicine, St Louis, MO; the staff of the PLCO Screening Centers; Ms Barbara O’Brien and staff of Westat Inc, Rockville, MD, for trial management; Pamela Marcus, PhD, Epidemiologist, Division of Cancer Prevention, National Cancer Institute, for her expertise in analysis of prior lung screening trials; Anthony B. Miller, MD, Professor Emeritus, Department of Public Health Services, University of Toronto, Ontario, Canada, for review of the manuscript and many helpful suggestions; and Marie Fleisner, Marshfield Clinic, for editorial assistance. Most importantly, we acknowledge the study participants who made this trial possible.