Positive Screens are More Likely in a National Lung Cancer Screening Registry than the National Lung Screening Trial

INTRODUCTION

In 2011, the National Lung Screening Trial (NLST) reported a 20% reduction in lung cancer mortality with annual LDCT screening relative to chest radiography.¹ Evidence supporting LDCT screening was extended by the Dutch-Belgian Lung-Cancer Screening Trial (NELSON), published in 2020, which reported a lung cancer mortality reduction of 24% in men and 33% in women with LDCT screening compared to no screening.² The United States Preventive Services Task Force (USPSTF) recommends annual lung cancer screening (LCS) with LDCT for individuals at high risk for lung cancer. These recommendations were introduced in 2013, shortly following publication of the NLST, and were updated in 2021 to expand the LCS eligible population.³ Annual LCS is covered by private and commercial insurers without cost-sharing under the 2010 Patient Protection and Affordable Care Act and is also a covered preventive health benefit by the Centers for Medicare & Medicaid Services’ (CMS) Medicare program.^4,5

Despite two large randomized controlled trials (NLST and NELSON) demonstrating the efficacy of LCS in reducing mortality and national policies facilitating insurance coverage, uptake of LCS lags.⁶ A report from the American Lung Association reports a national screening rate of 5.8%, and while other data suggest rates of approximately 20% for a few U.S. states, LCS is considerably underutilized relative to other screening tests.^7–10 Reasons for the slow uptake are multifactorial, including concerns about the false-positive rate of LDCT, the relatively high incidence of incidental findings unrelated to lung cancer, access barriers to LDCT facilities and downstream screening workup, and uncertainly regarding if and how the benefits demonstrated in clinical trials will translate to community screening.^11–15 The current body of LCS literature lacks a robust understanding of real-world LCS results, particularly regarding Lung-RADS results and how they compare to the NLST.

Our purpose was to compare Lung-RADS results between the NLST and community screening in a representative national screening population using data from the American College of Radiology (ACR) LCS Registry (LCSR). Under the 2015 LCS Medicare coverage decision, data entry into a CMS approved registry was required for LCS coverage, and the ACR’s LCSR was the only registry receiving CMS approval.¹⁶ While entry into the LCSR is no longer required per the 2021 LCS Medicare coverage decision, the former CMS decision required a minimum reporting of data elements, including radiation dose, smoking history, and Lung-RADS results.^5,16 We aimed to compare Lung-RADS scores between the LCSR and NLST and evaluate associations between Lung-RADS results and patient characteristics.

METHODS

RESULTS

Figure 1 shows cohort construction. For the LCSR, there was a final sample size of 686,011 participants receiving a T0 LDCT. We excluded 129,772 participants because they did not meet NLST eligibility criteria and an additional 24,004 LDCTs because they had a clinical diagnosis or signs/symptoms of lung cancer prior to screening. Of these, 11,031 participants were excluded because the exam indication was other than LCS, and 2,343 because the results were incomplete or missing. The final NLST sample included 26,432 participants. We excluded 22 participants not meeting NLST eligibility criteria, 263 who had no valid screening exam, one for whom the T0 screening occurred before randomization, and four because the screening results were incomplete or missing.

Table 1 shows the study population characteristics. Supplementary Table 1 gives study population characteristics by year for the LCSR population. Relative to the NLST, the LCSR population was older (mean age (SD): 64.0 (5.4) versus 61.4 (5.0) years), p<0.001) and included more females (47.9% versus 40.9%, p<0.001). The LCSR included more individuals who were currently smoking than formerly smoking (currently versus formerly smoking: 61.5% versus 37.8% for LCSR; 48.1% versus 51.9% for NLST, p<0.001). Pack years was lower in the LCSR than the NLST (mean (SD): 47.3 (17.9) versus 54.8 (19.9), p<0.001). BMI differed between both populations (p<0.001), with the LCSR having a higher proportion of individuals who were underweight (3.8% versus 1.2%, LCSR versus NLST respectively), as well as higher proportion of individuals with obese BMI (35.8% versus 28.9%m, LCSR versus NLST respectively). When stratified into groups of currently and formerly smoking, the LCSR continued to have a higher proportion of underweight and obese individuals (Supplementary Tables 2a–b).

Table 2 reports the crude results comparing the Lung-RADS distribution for the LCSR and NLST. Supplementary Table 3 gives the year-by-year breakdown of Lung-RADS scores for the LCSR population. The Lung-RADS distribution differed between the LCSR and NLST populations (p<0.001). Relative to the NLST, the LCSR had an overall lower proportion of Lung-RADS 1 examinations (55.6% NLST versus 40.8% LCSR) and a higher proportion of Lung-RADS 2 examinations (30.7% NLST versus 41.8% LCSR). There was an overall higher proportion of “positive” screening examinations (defined as Lung-RADS categories 3, 4A, 4B, or 4X) in the LCSR group (119,269/686,011, 17.4%) versus (3,610/26,432, 13.7%). This was predominantly driven by the higher proportion of Lung-RADS 3 examinations in the LCSR group (Lung-RADS 3 – 10.2% LCSR versus 6.4% NLST). The overall proportions of Lung-RADS 4A and 4B were marginally lower in the LCSR population compared to the NLST, whereas the proportion of Lung-RADS 4X examinations was minimally higher. In Table 2, for LDCTs in the NLST classified as falling into a cross-category (i.e., Lung-RADS 3/4A, 3/4A/4B, 4A/4B), the highest Lung-RADS category was assumed. In a sensitively analysis, which assumed the lowest Lung-RADS category for any such cross-category exams, there was no substantial change to these results (Supplementary Table 4).

Table 3 shows results of a multinomial logistic regression model evaluating the outcome of Lung-RADS score (with Lung-RADS 1 as the reference value). After controlling for age, gender, smoking status, pack-years, and BMI, the LCSR population were more likely to have higher Lung-RADS scores. Namely, the LCSR population had a higher likelihood of Lung-RADS scores 2, 3, 4A, 4B, and 4X compared to the NLST (adjusted odds ratio (aOR) and 95%CI >1 for Lung-RADS scores 2, 3, 4A, 4B, 4X relative to Lung-RADS 1). Additionally, increasing age was associated with higher Lung-RADS. Compared to females, males were more likely to have Lung-RADS 4B results (aOR, 95%CI: 1.05, 1.01–1.08) and less likely to have Lung-RADS 2 results (aOR, 95%CI: 0.91, 0.89–0.91). Currently smoking status was associated with a higher Lung-RADS score compared to formerly smoking status, and increasing pack-years was also associated with a higher Lung-RADS score. Compared to a normal BMI, those with BMI in the obese and overweight categories were less likely to have higher Lung-RADS scores (i.e., likelihood of Lung-RADS scores 2, 3, 4A, 4B, 4X was overall lower for obese and overweight individuals). For those with BMI in the underweight category, they were less likely to have a Lung-RADS 2 examination (aOR, 95%CI: 0.88, 0.85–0.91), though there was no significant difference between underweight and normal weight individuals and the other Lung-RADS scores of 3, 4A, 4B, or 4X.

DISCUSSION

In this study, comparing LCS results in a large national population from the ACR’s LCSR with the NLST, the clinically screened population had higher Lung-RADS scores, even after adjustment for known confounders, including smoking and age. The higher Lung-RADS scores in the clinical population are consistent with other smaller studies and have several implications.¹⁸

First, the higher Lung-RADS scores may reflect higher lung cancer rates in clinical LCS than research trials. This been suggested by prior smaller studies.^18,19 Our analysis implies that reasons for a potentially higher lung cancer rate would extend beyond older age and higher current smoking rates in the clinical cohort. One possibility is differences in exposures, such as tobacco or other environmental exposures. Though pack-years was reportedly lower for the clinical cohort than the NLST, smoking history may not be accurately captured or potentially under-reported in the LCSR. Capturing smoking history in the electronic health record is a significant challenge.²⁰ For example, in one study, measures of smoking intensity and years smoked were missing or inaccurate in >50% of participants; further, pack-years data were accurate in only 20% of patients.²¹ Relevant to smoking history, it should be noted also that clinical practice differs from a controlled clinical trial setting. For the NLST, smoking history was an enrollment criteria for randomization, and therefore may have been more rigorously captured than what is readily feasible in a clinical setting.²² It is also possible that higher comorbidity rates in the clinical cohort could contribute to a higher likelihood of lung cancer.^23,24 For instance, the presence of emphysema is linked to a higher likelihood of lung cancer, even after controlling for pack-years.²⁵ While detailed comorbidity data in the LCSR precluded specific investigation in this study, from an implementation perspective, a potentially higher lung cancer rate in clinical LCS underscores the requisite for programs to ensure appropriate follow up of positive screens and appropriate resources to facilitate care of LCS-detected cancers. This is critical in ensuring that the benefits of early lung cancer detection observed in trials translate to clinical screening.^1,2

Another explanation for the higher Lung-RADS scores in the clinical setting could be higher false positive rates. While use of Lung-RADS reduces false positive rates, false positives pose a challenge for LCS, contributing to unnecessary downstream procedures, unnecessary costs, poorer LCS adherence, and patient worry.^3,12,26–30 These concerns are potentially heightened given evidence of higher rates of invasive procedures and procedural-related complications in real-world LCS versus the NLST.²⁹ Other evidence demonstrates that screened individuals undergoing imaging for false positive LCS exams are less likely to adhere to annual screening.²⁸ Potential reasons for false positive examinations in clinical screening are multifactorial, spanning differences in practice settings, radiologic interpretations, and patient populations. The highly controlled setting of a clinical trial is challenging to replicate in real-world practice. The NLST was conducted at 33 academic centers with screening interpretation performed primarily by cardiothoracic radiologists receiving dedicated training on image acquisition/quality and interpretation. In clinical settings, heterogeneity in practice types and interpreting radiologists likely contribute to differences in LCS results. For example, Henderson, et al. reported higher rates of positive screens (defined as Lung-RADS 3, 4A, 4B, or 4X) in community screening settings (where interpretating radiologists were likely to be general radiologists) versus academic settings (where radiologists were subspecialty cardiothoracic radiologists).³¹ Though limiting LDCT interpretation to subspecialty-trained cardiothoracic radiologists may be impractical towards facilitating LCS access to the nearly 14.5 million LCS-eligible individuals in the U.S., the updated CMS LCS decision, which eliminates requirements for radiologists to interpret a minimum number of chest CTs and document continuing medical education, could exacerbate challenges associated with false positives.^5,32 Along the same lines, the updated CMS decision also omitted the previous requirement for LCS programs to submit data to a centralized registry.⁵ Although these policies were put forth with intent to broaden LCS access, quality assurance and outcomes monitoring are key components of high-quality screening.³³ For example, in the case of breast cancer screening, the Mammography Quality and Standards Act (MQSA) mandates minimal requirements for radiologists and data monitoring of screening results, including pathologic correlation and periodic auditing.³⁴ The elimination of centralized data reporting and minimum requirements for interpretating radiologists further heightens the need and responsibility for individual screening programs to perform internal quality assurance and data monitoring. This is crucial to ensure that false positives and unnecessary harms are reduced.

The higher Lung-RADS scores in the clinical population also draws attention to the need for continued research to improve pulmonary nodule classification. While LDCT has high specificity for nodule detection, discrimination between malignant and benign nodules is often not straightforward. For instance, in NLST, 96.4% of positive LDCTs were false positive (resulting in no cancer diagnosis).¹ While Lung-RADS, which was created to standardize LDCT reporting, reduces false positives, there is continued room for improvement.¹² As LDCT expands, Lung-RADS has evolved to incorporate new evidence, with the most recent iteration released in 2022 incorporating new guidance to reduce false positives from infectious/inflammatory nodules, certain airway-associated findings, and benign intrapulmonary lymph nodes.³⁵ As LCS grows, it is prudent for Lung-RADS to continue to incorporate new evidence from real-world screening, so as to ensure that patients with malignant nodules are expeditiously recognized while those with benign nodules are averted from unnecessary follow up and/or harm. As research advances, it is also possible that future paradigms for LCS could expand beyond Lung-RADS to incorporate emerging technologies, such as blood biomarkers or radiomics, which may aid in the distinguishing benign from malignant nodules.^36,37

Aside from the higher Lung-RADS scores in the LCSR, patient characteristics between the NLST and clinical LCS populations differed, with the latter being older, more likely to be currently smoking, and more likely to have an obese or underweight BMI. While these differences did not explain the higher Lung-RADS scores in the clinical cohort, they have ramifications for real-world screening. The older age, higher current smoking rates, and differences in BMI are in keeping with other evidence suggesting that NLST participants were overall healthier than the clinically eligible LCS population.^24,38 For real-world LCS implementation, this underscores the need for shared decision making (SDM) conversations discussing individual patient risks versus benefits of screening. A key component of SDM is patient health status, including consideration of medical comorbidities and life expectancy.³⁹ While the LCSR lacks detailed life expectancy and comorbidity data, the higher obesity rates in the clinical LCSR cohort could imply higher comorbidities and potentially lower life expectancy. Robust data links obesity to early mortality and comorbidities across multiple organ systems, and the co-occurrence of smoking and elevated BMI has particularly negative health consequences.^40,41 Regarding the higher proportion of underweight individuals in the clinical cohort, this may imply higher rates of emphysema in the clinical cohort, as emphysema is associated with lower BMI.⁴² Taken together, the observed differences in patient characteristics may imply that the risks of LCS are higher in clinical settings that in research trials. This is supported by evidence demonstrating higher rates of complications following LCS-related invasive procedures in clinical screening, with older patients carrying relatively higher risk. ^26,29 Given the higher Lung-RADS scores in the clinically screened population, which translates to a higher likelihood of downstream workup following screening, our results reiterate the necessity to prioritize screening in individuals in whom net benefit outweighs risks.

Our study has limitations. The study does not include lung cancer outcomes, as cancer rates could not be reliably assessed from the LCSR dataset, which may be due to underreporting in the LCSR, a limitation discussed in previously published work.⁴³ As such, it is not possible to precisely determine if the positive screens observed in the LCSR were due to higher rates of lung cancer or higher rates of false positive exams. Lung-RADS was introduced after the NLST, hence the NLST did not use Lung-RADS. We retrospectively re-classified the NLST LDCTs into Lung-RADS version 1.1 scores using available nodule descriptors. An updated Lung-RADS was issued in 2022, and whether the conclusions drawn here would remain similar using Lung-RADS 2022 is unknown.⁴⁴ Additionally, the timeline of LDCT acquisition in both data sets differs. LDCTs in the NLST occurred between 2002–2007, whereas the ACR LCSR data was from 2015–18. While multidetector CT was used in the NLST, LDCTs from the LCSR would likely be acquired on more contemporaneous scanners (i.e., more detector row elements, faster scan acquisition).⁴⁵ It is also likely that image reconstruction differed between the two datasets, with more contemporary exams from the LCSR likely reconstructing at thinner axial slices, which would increase sensitivity of nodule detection.⁴⁶ Thinner axial slices, in addition to other contemporary reconstruction strategies in clinical practice, such as maximum intensity projection (MIP), likely contribute to the higher proportion of Lung-RADS 2 scores observed in the LCSR versus the NLST.⁴⁶ Importantly, there are likely unmeasured confounders not captured in this study (i.e., comorbidities, practice and/or radiologist type) which may contribute to differences in Lung-RADS distributions between the two populations. Additionally, this study focused on baseline screens, and whether similar patterns would be observed at repeat screening is unknown, though is a focus of future research.

CONCLUSIONS

We compared real-world LCS results from the LCSR to the NLST. We report the key finding that Lung-RADS scores were higher in the clinically screened population than the NLST, even after adjusting for confounders including age and smoking. This finding has implications for real-world LCS, as it would imply higher rates of follow-up testing downstream of LCS and potentially higher rates of lung cancer than the NLST. Future work should focus on determining long-term outcomes in the clinical LCS population, including lung cancer rates and mortality, as well as on improvements in nodule classification that enable accurate identification of malignant nodules while reducing false positive exams.

Supplementary Material

Data Statement:

IFG declares that they had full access to all of the data in this study and take complete responsibility for the integrity of the data and the accuracy of the data analysis.