Evaluating Lung Cancer Screening Uptake, Outcomes, and Costs in the United States: Challenges With Existing Data and Recommendations for Improvement

Abstract

The National Lung Screening Trial (NLST) reported substantial reduction in lung cancer mortality among high-risk individuals screened annually with low-dose helical computed tomography (LDCT). As a result, the US Preventive Services Task Force issued a B recommendation for annual LDCT in high-risk individuals, which requires private insurers to cover it without cost-sharing. The Medicare program also covers LDCT for high-risk beneficiaries without cost-sharing. However, the NLST findings may not be generalizable to the community setting because of differences in patients, providers, and practices participating in the NLST. Thus, examining uptake of LDCT screening in community practice is critical, as is evaluating the immediate and downstream outcomes of screening, including false-positive scans, follow-up examinations and adverse events, costs, stage of disease at diagnosis, and survival. This commentary presents an overview of the landscape of the data resources currently available to evaluate the uptake, outcomes, and costs of LDCT screening in the United States. We describe the strengths and limitations of existing data sources, including administrative databases, surveys, and registries. Thereafter, we provide recommendations for improving the data infrastructure pertaining to three overarching research areas: receipt of guideline-consistent screening and follow-up, weighing benefits and harms of screening, and costs of screening.

Lung cancer is the leading cause of cancer death in both men and women, accounting for about 154 000 cancer deaths annually in the United States (1). Surgical resection at an early stage of diagnosis offers the best chance for cure or long-term survival, making lung cancer particularly amenable for screening (2). In December 2013, based on the results from the National Lung Screening Trial (NLST), the US Preventive Services Task Force (USPSTF) concluded that in persons with high risk for lung cancer, the potential for annual low-dose helical computed tomography (LDCT) screening to reduce mortality outweighed its potential harms (3). The USPSTF B recommendation comprises adults aged 55 to 80 years who have a 30 pack-year smoking history and currently smoke or have quit within the past 15 years (3) and requires private insurers to cover LDCT screening for beneficiaries without patient cost-sharing. In 2014, the Centers for Medicare and Medicaid Services (CMS) added annual LDCT screening among eligible enrollees as an additional preventive service benefit under the Medicare program (4).

Understanding the generalizability of NLST results to community practice is a key health-care delivery and policy issue. Differences between the patients, providers, and practices who participated in the NLST and the general population have been highlighted in this regard (5,6). Notably, those who participated in the NLST were substantially younger, were less likely to be African American, were more likely to be former smokers, and had higher educational status than the general population of tobacco users in the United States (5). Given these differences, ensuring adherence to the annual screening schedule is a challenge, especially outside of a trial setting is a challenge. Furthermore, the NLST was conducted in a controlled setting: the participants’ eligibility was carefully ascertained, screening procedures followed a standard protocol, radiologic technologists were trained in image acquisition, radiologists were trained in image quality and standardized image interpretation, and many participating sites were well-recognized for their expertise in radiology and in the diagnosis and treatment of cancer (5). Conditions in community practice are often less ideal (7). Eligibility ascertainment may be less stringent in the community, and off-guideline use of LDCT may be common (8). Of particular concern in community settings are postsurgical morbidity and mortality. The NLST reported 1% mortality within 60 days among those undergoing resection for lung cancer (5). Estimates of postsurgical mortality among lung cancer patients in the community are much higher (2.3%–5.2%) (9–12). It is important, therefore, to examine how LDCT screening is being implemented in community practice and the immediate and downstream outcomes of screening, including false-positive scans, follow-up examinations and adverse events, unnecessary invasive workup for benign disease, costs, stage of disease at diagnosis, and survival. Box 1 lists some of the crucial research questions related to the use of LDCT screening in community practice.

Box 1.

Key research questions related to low-dose computed tomography (LDCT) screening in community practice

Receipt of guideline-consistent screening/follow-up

What proportion of eligible patients receive screening?

What proportion of noneligible patients receive screening?

What proportion of eligible patients receive shared decision-making (SDM) and smoking cessation guidance?

What is the quality of SDM and smoking cessation guidance?

What are barriers and facilitators to receipt of guideline-consistent screening?

What proportion of the screened receive follow-up examinations?

What proportion of the screened adhere to the annual screening schedule?

Weighing benefits and harms of screening

What is sensitivity of LDCT?

What is specificity of LDCT?

How prevalent are short-term adverse events due to screening?

How prevalent are longer-term adverse events due to screening?

How prevalent are invasive procedures ensuing from false-positive scans?

How prevalent is overdiagnosis due to screening?

What is the lung cancer stage distribution in screened and unscreened populations?

Is screening associated with reduced lung cancer mortality?

Is screening-associated smoking-cessation guidance effective?

Cost of screening

What is the medical cost of LDCT?

What is the cost-effectiveness of LDCT in community settings?

What is the nonmedical cost of LDCT?

According to the National Health Interview Survey (NHIS), the proportion of eligible individuals who reported LDCT screening in the past 12 months was 3.3% in 2010 and only marginally higher (3.9%) in 2015 (13). Given the slow uptake of LDCT screening, more observational data are required to identify determinants of LDCT screening at the patient, area, provider, and health system levels. Observational studies in community settings will also facilitate the development and evaluation of quality measures and interventions to promote screening practices that are consistent with scientific evidence. It is especially important to assess the performance characteristics of the Lung CT Screening Reporting and Data System (Lung-RADS)—the consensus classification of nodules issued by the American College of Radiology that is derived from several LDCT screening trials (14)—and fine-tune these criteria.

However, evidence on LDCT use in community practice in the United States remains sparse, reflecting limitations in existing data infrastructure to capture smoking history; eligibility for LDCT screening; receipt of screening and follow-up exams; adverse events and costs; cancer diagnoses, including stage; and treatment(s), recurrence, and survival.

In this commentary, we describe the strengths and limitations of existing data sources for monitoring the uptake, outcomes, and costs of LDCT screening, including administrative databases (health insurance claims and electronic health records), surveys, and registries. We also discuss implications for evaluating the comparative effectiveness of LDCT and then highlight opportunities for new types of data linkages and other approaches for improving the data infrastructure for evaluating LDCT screening.

Health Insurance Claims

Health insurance claims can be used to identify the receipt of LDCT scans and follow-up examinations and provide estimates of the immediate and downstream costs associated with screening exams. However, claims data are designed for payment and not to capture screening eligibility based on smoking history or absence of symptom. Although claims data contain some indirect indicators for smoking, including diagnosis codes for tobacco use disorder, records for counseling visits for smoking, and smoking-cessation prescription medication use, algorithms relying solely on these indicators are not sufficiently sensitive and will not yield all the necessary details for ascertaining screening eligibility (15).

Identifying screening LDCT scans in claims data also can be challenging. Some providers order LDCT scans as a substitute for diagnostic CT scans for symptom evaluation to reduce radiation exposure or to help patients avoid copays for diagnostic examinations (8). Although the CMS and most commercial insurance plans mandate that all LDCT screening examinations are accompanied by a billable shared decision-making (SDM) visit (4), it is unclear whether this requirement effectively deters incorrect coding.

Claims data, by design, include only subsets of the insured population of a geographic area. For instance, fee-for-service Medicare data are limited primarily to services rendered to beneficiaries aged 65 years and older whereas commercial claims mainly capture service utilization of a younger, working-age population enrolled in a specific plan. It is rarely possible to track enrollees when they switch from one commercial plan to another. Furthermore, because Medicare is the primary payer for Medicare beneficiaries, employment-based health insurance claims are effectively censored for many individuals when they enroll for Medicare. Only a few data resources allow continuous capture of receipt of LDCT and other services beyond Medicare enrollment. These include the state-based all-payer claims data, the Medicare Supplemental and Coordination of Benefits Database from MarketScan (for Medicare beneficiaries who register for supplemental plans) (16), and medical encounters data (for Medicare beneficiaries who enroll for Medicare Advantage within the same health system that had previously provided their employer-based coverage). Because the median age at diagnosis of lung cancer is 70 years (17), the fragmentation of insurance claims at age 65 years poses challenges for assessing issues such as adherence to or downstream outcomes of LDCT screening.

Finally, because these data can capture occurrence but not results of screening or diagnostic tests, they are of limited use in identifying positive scans or evaluating adherence to recommendations on LDCT reporting and management (eg, Lung-RADS category) and follow-up.

Cancer registry-insurance claims linkages, such as the Surveillance, Epidemiology and End Results-Medicare database (18), provide additional means to compare tumor stage, service utilization, costs, and outcomes between screened and unscreened groups. However, these data share limitations related to determination of LDCT screening eligibility, receipt of screening, LDCT results, and related outcomes.

Electronic Health Records (EHRs)

Compared with insurance claims, EHRs can provide a more comprehensive picture of the services delivered to all patients, including the uninsured, at a medical practice or a network of practices. Some EHR systems also can collect elements relevant to LDCT evaluations that are not captured in claims such as health status, problem lists, smoking history, details of SDM visits, and smoking cessation guidance.

The inherent and potential advantages of EHRs notwithstanding, these data also present many challenges. EHR vendors use disparate models for designing their EHR systems. Due to lack of interoperability between these data models, gaps in information may emerge if eligibility assessment, scan ordering, and receipt of scan happen at different practices. Often these gaps persist even after consolidation of medical practices or their mergers with hospitals or health systems. For this reason, EHR data from large, integrated delivery systems such as Kaiser Permanente or the Veterans Health Administration may be more comprehensive and better suited for evaluating LDCT screening than data from smaller practices or linked EHR claims data available from commercial insurers (19,20). However, even within these integrated delivery systems, many variables of interest (eg, exam indication, nodule characteristics, follow-up test results) may be available only as free text rather than in structured fields. As a result, extraction of lung cancer screening data from EHR systems can require either labor-intensive efforts by medical record abstractors or development of automated techniques such as natural language processing (8,20).

Irrespective of the source, the use of EHR data to assess LDCT can be constrained by their high cost, complexity, and limited availability to researchers external to the practice/organization generating the EHR.

Household Survey Data

Household surveys of population health and health care are widely used for research on cancer screening. These data are used to monitor national trends in LDCT uptake and identify disparities based on geographical and sociodemographic attributes. The principal source of information on the health of the civilian noninstitutionalized population of the United States is the NHIS conducted by the National Center for Health Statistics, a part of the Centers for Disease Control and Prevention (21). The annual NHIS includes questions on history of lung cancer, sociodemographic characteristics, smoking history (including pack-years and time since quitting), counseling for smoking cessation, occupation, health status, and health-care access. The response rate of the annual survey is approximately 70% of the eligible households (22). In addition, the NHIS Cancer Control Supplement is fielded periodically to assess knowledge, attitudes, and practices of cancer-related health behaviors and screening (23). Questions on receipt of chest x-ray and CT scan were included in NHIS Cancer Control Supplement 2010 and 2015, and national estimates of LDCT uptake from these years have been reported (13). These data are publicly available and can be freely downloaded from the National Center for Health Statistics website.

Unlike the NHIS, which is an in-person survey, another household survey administered by the Centers for Disease Control and Prevention, the Behavioral Health Risk Surveillance System (BRFSS), is a telephone survey designed to produce state-level estimates (24). The median response rate for BRFSS for all states, territories, and Washington, DC in 2016 was 47.0% and ranged from 30.7% to 65.0% (25). The first year the BRFSS includes an optional module for questions pertaining to lung cancer screening is 2018. These data will become publicly available for free download in 2020.

Another nationally representative cross-sectional survey is the Health Information National Trends Survey administered by the National Cancer Institute (NCI) to collect data about the use of cancer-related information by US adults aged 18 years and older (26). The Health Information National Trends Survey included a question on patient-provider discussion about testing to check for lung cancer in 2012, 2014, and 2017. The household response rates for these years ranged between 32% and 40% (17,27,28).

Large household surveys are not without limitations. Although most undertake cognitive testing to ensure that individuals understand survey questions (29), self-report of screening and smoking history can introduce bias. Firstly, questions concerning chest x-ray or CT scan in both NHIS and BRFSS do not refer explicitly to lung cancer screening. Instead, the respondents are asked whether they underwent the said examinations to “check” for lung cancer. Therefore, these surveys count as “screened” those individuals with symptoms who receive CT examinations to check for lung cancer (8). Additionally, individuals may not recall accurately whether they received a chest CT; for other organ sites, there is evidence that individuals tend to overreport receipt of screening (30–35). At the same time, surveys may underestimate smoking history because of social desirability in reporting health behaviors (36). Secondly, none of the existing surveys inquires about SDM for LDCT, provider recommendations for screening, postscreening follow-up examinations or invasive procedures, or costs attributable to LDCT screening. Finally, individuals who do not participate in surveys or respond only to parts of a survey may be systematically different from those who complete the full survey (37).

Lung Cancer Screening Registry

In view of the uncertainty around how best to manage individual lesions and the magnitude of the potential harms of LDCT screening in actual practice, the USPSTF encouraged the development of a registry to ensure that appropriate data on screening and subsequent follow-up are collected from screening programs (38). This USPSTF recommendation was followed by the CMS mandating participation in a CMS-approved registry as part of the national coverage determination for Medicare reimbursement for lung cancer screening (4). To date, the Lung Cancer Screening Registry (LCSR) sponsored by the American College of Radiology is the only CMS-approved lung cancer screening registry (39). The participating facilities (currently, a total of 2039) contribute the following data elements for all LDCT screening examinations regardless of insurance status: identifiers for the patient, interpreting radiologist, and ordering provider; make and model of the CT scanner; patient age, sex, height, and weight; detailed smoking history, including current smoking status, years since quitting if former smoker, and number of pack-years; documentation of SDM and smoking cessation guidance; date of screening examination, indication, effective radiation dose, resolution, and nodule characterization using Lung-RADS categories; and follow-up information through 12 months from the screening exam, including subsequent diagnostic procedures, tissue diagnosis, and lung cancer stage, if applicable (40).

An important objective of the LCSR is to ensure adequate measurement of the harms and benefits of lung cancer screening. Therefore, high-quality reporting of follow-up examinations and procedures is crucial for systematic audits of facilities and providers. That said, the completeness of capture of follow-up exams in the LCSR is uncertain, because nonradiologic diagnostic procedures are often done in facilities outside of where screening was performed. Although nonconformance to the reporting standards for eligibility, scan modalities, and results may lead to denial of a claim, inadequate reporting of the ensuing workup or noncompliance with Lung-RADS guidelines about follow-up may not preclude reimbursement for the scan.

By definition, the LCSR only includes information about individuals who receive LDCT; eligible individuals who do not receive screening are excluded. Therefore, the LCSR is of limited use in assessing the comparative effectiveness of LDCT. The LCSR is also not designed to collect information on the quality of SDM or smoking cessation guidance. Another limiting constraint of the LCSR is the stipulated period of 12 months postscan for reporting follow-up. Unless another screening scan occurs by the end of the 12-month period, the patient is likely to be lost to follow-up. There are also shortfalls in recording physical and psychological complications related to the scan or the resultant invasive procedures. The current formulation of the LCSR data entry tool allows for an adverse event to be captured in a structured field if it leads to death within 12 months of a scan. Less severe or long-term complications are likely to be missed.

Charting a Course of Action

We have thus far presented an overview of the landscape of the data resources currently available to evaluate the uptake, outcomes, and costs of LDCT screening in United States. In doing so, we highlighted some key merits and limitations of each data source. The general attributes of various datasets available for LDCT evaluations in community practice are presented in Table 1, andTable 2 summarizes the availability in these datasets of some of the important elements related to LDCT screening. Below, we present recommendations pertaining to the necessary data infrastructure for answering some of the key questions related to LDCT screening.

Table 1.

Attributes of common sources of data available for evaluation of LDCT screening in community practice^*

Data attributes	Commercial insurance and Medicare claims	EHR	Household surveys (NHIS/BRFSS)	ACR Lung Cancer Screening Registry
Population characteristics
Geographical representativeness	Medicare claims geographically representative for those 65+ years; commercial claims restricted to the insurance provider	Restricted to the practice/institution generating the EHR	Yes	Yes (data on screening recipients only), currently 2000+ sites across the US
Age representativeness	Medicare claims representative for those 65+ years; commercial claims restricted to the insurance provider and not representative for those 65+ years	Restricted to the practice/institution generating the EHR	Yes	Yes (data on screening recipients only)
Insurance representativeness	No	Yes	Yes	Yes
Provider representativeness	No	No	No	No
Sociodemographic information: age, sex, and race/ethnicity	Yes	Yes	Yes	Yes
Sociodemographic information: income and education	Aggregate information from the geographical area of residence may be available	Aggregate information from the geographical area of residence may be available	Yes	No
Health status	Claims-based information on comorbidities from services and diagnoses	Encounter- or problem-list-based information on comorbidities	Self-report	No
Data availability and cost
Availability for research	Yes	Usually proprietary	Yes	Usually proprietary
Timeliness	Time lag: 6 months–1 year for commercial claims, 2 years for Medicare claims	Usually current	1–2 years between survey and data release	Unknown
Cost of data	Usually expensive to acquire	Usually expensive to acquire	In public domain; free to download	Unknown

^*ACR = American College of Radiology; BRFSS = Behavioral Risk Factors Surveillance System; EHR = electronic health records; LDCT = low-dose computed tomography; NHIS = National Health Interview Survey.

Table 2.

Availability of LDCT-related attributes in common data sources^*

Data attributes	Commercial insurance and Medicare claims	EHR	Household surveys (NHIS/BRFSS)	ACR Lung Cancer Screening Registry
Information on eligibility
Pack-years in current smokers	No	Inconsistent	Yes	Yes
Pack-years in past smokers	No	Inconsistent	Yes	Yes
Time since quitting if past smoker	No	Inconsistent	Yes	Yes
Presence of symptoms	No	Inconsistent	No	Yes
Prior cancer history	No	Inconsistent	Yes	Yes
Clinical workflow
SDM	Yes, billing for SDM required for LDCT reimbursement; content unknown	Inconsistent	No	Yes, content unknown
Counseling for smoking cessation	Inconsistently reported and content unknown	Inconsistent	Yes, content unknown	Yes, content unknown
Outcomes data
Test result
Lung-RADS category	No	Inconsistent	No	Yes
Details of abnormality	No	Inconsistent	No	Yes
Diagnosis	Can identify treated patients	Inconsistent	Yes	Yes
Stage at diagnosis	No	Inconsistent	No	Yes
Adverse events	Yes	Yes	No	Only fatal adverse events occurring within 12 mo
Follow-up examinations	Yes	Yes	No	Short-term (12 mo after scan)
Cost of screening and treatment
Cost of initial scan	Yes	Yes	No	No
Cost of follow-up exams	Yes	Yes	No	No
Cost of adverse events	Yes	Yes	No	No
Cost of treatment	Available for screened and unscreened while covered by insurance	Yes	No	No
Recurrence			No	No
Survival	Available for Medicare claims; date of death not always available for commercial claims	Inconsistent	No	No
Data completeness	Commercial claims restricted to the primary payer; services covered by Medicare fee for service complete; Medicare Advantage claims unavailable	Data from integrated delivery systems more complete	Yes	Limited to 12 mo after scan

^*ACR = American College of Radiology; BRFSS = Behavioral Risk Factors Surveillance System; EHR = electronic health record; LDCT = low-dose computed tomography; Lung-RADS = Lung CT Screening Reporting and Data System; NHIS = National Health Interview Survey; SDM = shared decision making.

Receipt of Guideline-Consistent Screening and Follow-Up

The slow uptake of LDCT screening indicates that a large proportion of screening-eligible individuals may not have had an opportunity to have an SDM visit, may not have access to screening, or may not be recommended screening at all (41), because among those with an SDM visit, the majority do receive screening (42). Currently, the American Academy of Family Physicians considers that the evidence is insufficient to recommend for or against LDCT screening in persons at high risk for lung cancer based on age and smoking history (43). Concerns also have been raised about the quality of SDM being delivered (44). To elucidate physicians’ and patients’ knowledge and attitudes about lung cancer screening and other modifiable determinants of guideline-concordant screening, household surveys need to inquire about the contents of the SDM visit, including ascertainment of screening eligibility, discussions about harms and benefits of annual screening, referral or delivery of smoking cessation counseling, and barriers and facilitators to screening. Additionally, questions on the presence of symptoms possibly related to lung cancer can be used to distinguish diagnostic scans from lung cancer screening.

Introducing modifiers for details of smoking history in the International Classification of Disease coding system and a greater emphasis on documentation of smoking history are important first steps to facilitate the use of insurance claims and electronic health records for LDCT evaluations. Furthermore, linking administrative data with household surveys can facilitate evaluations of follow-up testing postscreening and adherence to the annual screening schedule. The NHIS-Medicare claims linkage, currently available for the years 1999 through 2013, is one such database that can be used to assess follow-up testing postscreening and adherence to the annual screening schedule (45). It would be useful to expand the linkage to include more recent years and other claim types.

The LCSR is another data source for evaluating follow-up examinations with the added advantage of the availability of scan results. However, the completeness of follow-up data in the LCSR needs to be examined.

Weighing Benefits and Harms of Screening

Comparisons of stage distribution and mortality between screened and unscreened groups and tracking these rates over time are necessary to evaluate the benefit of screening. These studies can be conducted within integrated delivery systems by enhancing the functionality of their EHR systems to report screening eligibility, receipt of screening, and scan results. For example, the NCI-funded Population-based Research to Optimize the Screening Process network is extracting lung cancer screening data from the EHRs of five different health-care delivery systems across the United States (46). Another study performed by the Veterans Health Administration—the Lung Cancer Screening Demonstration Project—used electronic tools to capture clinical data for screening coordinators (20).

The proportions of positive results were quite high in the LDCT arm of the NLST (approximately 24%–96% of which were false positive) and the Veterans Health Administration demonstration project (approximately 56%–98% of which false positive) (5,20). Further evaluation of the false positivity in community practice is required, especially the rates of invasive follow-up examinations and surgeries ensuing from false-positive scans and the complications thereof. Moreover, an estimated 20% of all lung cancers detected by LDCT in the NLST were likely to be indolent (47), raising concerns about overdiagnosis. In the absence of complete information on follow-up and diagnosis in the LCSR, these assessments can be facilitated by linking LCSR with administrative data and tumor registries. Valuable insights can be drawn from previous linkages between data from the Breast Cancer Surveillance Consortium screening registry data and claims data (48–51). Also, it would be useful to link the LCSR with tumor registry claims data linkages.

Furthermore, there is a dearth of studies on the potential psychological harms of LDCT screening, which include stress from anticipation of having a screening test; increased anxiety after receiving a positive or indeterminate screening result; feelings of guilt, shame, and anxiety in anticipation of possibly being diagnosed with lung cancer; psychological effects of workup of incidental findings and discovery of incidental disease; and psychological harms of nonbeneficial treatment due to screening (52). To address this deficiency, more questions on the psychological domains of screening can be included in household surveys.

Cost of Screening

Assuming eligibility for LDCT can be established, the linkage of administrative claims or screening registry data to tumor registries can facilitate examinations of costs of false-positive scans and overdiagnosis as well. Other potential sources of information on costs of LDCT are 1) linkage between household surveys that contain information about smoking and screening history and insurance claims (eg, the NHIS-Medicare linkage mentioned earlier), and 2) the Medical Expenditure Panel Survey (MEPS) (53). Each year, the MEPS recruits a sample of households who responded to the NHIS and conducts five interviews over a period of 2.5 years to collect information on service utilization and expenditures (53). With more LDCT-related questions in the NHIS and increasing prevalence of screening, the MEPS can be useful in examining costs of screening.

An important area of research is the value ascribed to the expenditures on LDCT screening. Initial analyses of the NLST data found annual screening to be cost-effective at $81 000 per quality-adjusted life-year gained (54). However, this estimate is based on several assumptions about the implementation of screening, which may not hold in clinical practice. Further economic inquiries are necessary using more robust data from the community, especially on downstream testing and adherence to the screening schedule.

Conclusion

The benefits of LDCT screening may be outweighed by the harms if it is inappropriately implemented in the community. A prerequisite to understanding and mitigating the harms of LDCT screening is a data infrastructure that comprehensively captures information about the screening continuum—including eligibility assessment, SDM, receipt of the scan, and the short- and long-term outcomes—through better tracking, reporting, and data sharing. In this commentary, we identified the merits and limitations of the available data for evaluating the implementation of LDCT screening in community practice and suggested improvements in the data infrastructure in view of the key unanswered questions related to LDCT screening.

Notes

Affiliations of authors: Surveillance and Health Services Research Program, Department of Intramural Research, American Cancer Society, Atlanta, GA (AR, KRY); Division of Cancer Control and Population Sciences, NCI, Bethesda, MD (VPDR); Thoracic Oncology Research Group, Division of Pulmonary, Critical Care, Allergy and Sleep Medicine, Medical University of South Carolina, Charleston, SC (GAS).

The authors have no conflicts of interest to disclose.