Ethics Statement

This study was approved by the Research Ethics Board of Sunnybrook Health Sciences Centre on behalf of the Institute for Clinical Evaluative Sciences (ICES). Individual consent for access to de-identified data was not required.

MISCAN-Lung

The MIcrosimulation SCreening ANalysis (MISCAN) Lung model was used for this analysis. Other versions of the MISCAN model have been used to investigate the cost-effectiveness of screening programs for breast, colorectal, cervical, and prostate cancers [19–22]. The MISCAN-Lung model used in these analyses was previously calibrated to individual-level data from the NLST and the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial, from which information on the preclinical duration of lung cancer and the effectiveness of CT screening were derived [23,24]. MISCAN-Lung was one of the models used to inform the USPSTF on their recommendations for lung cancer screening [13]. The structure of the model and its underlying assumptions have been described previously and are detailed in S1 Text; the characteristics of the investigated population are described in the following section and in S2 and S3 Texts [23,24].

In brief, MISCAN-Lung simulates life histories for each individual in the considered population from birth until death in the presence and absence of screening. For each individual, a smoking history is generated based on data on the investigated population. A person’s smoking history influences the probability of developing preclinical lung cancer as well as the probability of dying from other causes. The model considers four histological types of lung cancer: adenocarcinoma/large cell carcinoma/bronchioloalveolar carcinoma, squamous cell carcinoma, other non-small cell carcinoma, and small cell carcinoma. Once preclinical lung cancer has developed, it is assumed to progress through stages IA to IV. During each stage, the cancer may be detected due to symptoms, after which the person is assumed to undergo treatment with associated treatment costs. Lung cancer survival after clinical diagnosis is dependent on the histology and stage of the cancer and the person’s gender. If the screening component is activated, preclinical lung cancers may be detected by screening (at the expense of screening-related costs), which may alter a person’s life history: detection by screening allows treatment at an earlier stage, which may cure the individual, allowing him or her to resume a normal (lung cancer-free) life history. The probability that an individual is cured due to early detection differs by the stage at detection. Screening may also result in serious harms, such as overdiagnosis (the detection of a disease that would never have been detected if screening had not occurred), which may lead to unnecessary (invasive) follow-up procedures, treatments, and anxiety. The effects of screening are derived through utilizing information on the preclinical duration of lung cancer, the screen-detectability of lung cancer, and relevant information on the examined population (such as smoking behavior and other-cause mortality corrected for smoking history) to model the life histories of individuals in the presence and absence of screening [25]. Through comparing the life histories in the presence of screening with the corresponding life histories in the absence of screening, MISCAN-Lung can estimate the effectiveness and costs of screening scenarios.

Simulated Population

Three different birth cohorts were investigated: 1940–1949 (ages 66–75 in 2015), 1950–1959 (ages 56–65 in 2015), and 1960–1969 (ages 46–55 in 2015). These cohorts represent approximately 5.2 million individuals in 2016 in the age range for which the USPSTF currently recommends lung cancer screening [12,26]. Birth tables for each cohort were derived from information from Statistics Canada and the Ontario Ministry of Finance [26,27]. Ontario-specific data on smoking behavior were used to model smoking initiation and cessation probabilities and the average number of cigarettes smoked per day (divided into five categories) by age and gender for each cohort [28–32]. Life tables by birth year and gender were extracted from the Canadian Human Mortality Database and adjusted for smoking behavior and lung cancer mortality, as shown graphically for persons born in 1955 in S2 Text Figs C–F [32]. Further information on the methods used to model smoking behavior and adjustment of the life tables for smoking behavior is provided in S2 Text. The age- and gender-specific lung cancer incidence, histology proportions, and stage proportions estimated by MISCAN-Lung were compared to observed data from the Ontario Cancer Registry from 2007–2009, in which screening did not occur, and are detailed in S3 Text [33].

Screening Scenarios

In total, 576 potential screening scenarios were evaluated. The evaluated scenarios considered different combinations of the following characteristics: age to start screening; age to stop screening; screening interval; and screening eligibility regarding cumulative smoking exposure, years since smoking cessation (for former smokers, defined as individuals who have quit smoking permanently), and whether or not former smokers were excluded from further screening after they reach a maximum number of years since cessation (Table 1). Two types of cumulative smoking criteria, of which one is used at a time in an evaluated scenario, were distinguished: the first type was based on the cumulative number of pack-years, used in the NLST (“NLST-like”) [1]; the second type of cumulative smoking criteria was based on the criteria used in the Dutch-Belgian lung cancer screening trial (NELSON), which evaluated a person’s smoking duration and average number of cigarettes per day separately (“NELSON-like”) [34]. Of the 576 screening scenarios, 216 screening scenarios considered “NLST-like” criteria (including the criteria currently recommended by the USPSTF), whereas 360 screening scenarios considered “NELSON-like” criteria. Perfect attendance to screening was assumed for the base-case investigation. Estimations regarding screen-related procedures such as false-positive results (defined as receiving a positive screening test result when lung cancer is not found after diagnostic work-up), follow-up CT examinations, screen-related biopsies/bronchoscopies, and non-lung cancer-related surgeries were derived from individual-level data from the CT-arm of the NLST and are described in S4 Text. Limited information on morbidity and mortality has been reported from randomized controlled trials. In a subgroup analysis of NELSON participants, 1% of participants were found to have incidental findings that required additional work-up procedures or treatment [35]. However, no information on morbidity or mortality was reported. In the NLST, 0.06% of the positive screening tests in the low-dose CT group that did not result in a diagnosis of lung cancer were associated with a major complication after an invasive procedure [1]. Overall, six individuals with a positive screening test in the low-dose CT group that did not result in a diagnosis of lung cancer died within 60 d after an invasive diagnostic procedure (0.04%), but it was unknown whether these deaths were caused by complications of the diagnostic procedures [1]. Thus, given the low occurrence of invasive procedures along with a low frequency of major complications, the occurrence of morbidity or death related to screen-related follow-up procedures is expected to be minor.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of the lung cancer screening scenarios evaluated by the MISCAN-Lung model.

https://doi.org/10.1371/journal.pmed.1002225.t001

Costs

The analyses were conducted from a third-party health care payer perspective. Fully allocated costs for lung cancer treatment were estimated by stage, age, and gender from the date of diagnosis until the date of death or last known date of follow-up (by person-month) through data from the Ontario Health Insurance Plan (OHIP), the Canadian Institute for Health Information (CIHI), the Ontario Drug Benefit Plan database, the Ontario Chronic Care database, the Ontario Home Care database, and the Ontario New Drug Funding Program. These datasets were linked using unique encoded identifiers and were analyzed at the ICES. Controls without a lung cancer diagnosis from the Registered Persons Database (a roster of all OHIP beneficiaries) were matched to 12,713 staged cases of lung cancer from the Ontario Cancer Registry (10 controls matched per case), based on age, sex, median household income, and census tract on the date of diagnosis of the case. Fully allocated costs were estimated similarly for controls. The fully allocated costs of controls were subtracted from the fully allocated costs for cases in order to obtain the attributable costs of lung cancer care by phase of care (initial, continuing, and terminal care) [36]. By incorporating the fully allocated costs of lung cancer care by phase of care, it is taken into account that individuals whose lung cancer death is averted will on average incur higher costs over their remaining lifetime. In MISCAN-Lung, the attributable costs for stage I were assumed for the modeled stages IA and IB, whereas the attributable costs of stage III were assumed for the modeled stages IIIA and IIIB.

Each person’s eligibility for lung cancer screening was assumed to be free of misclassification error. Therefore, upon entering the eligible age range for the considered screening scenarios, each ever-smoking individual was assumed to receive an invitation for a lung cancer risk assessment. It was assumed that half of all ever-smokers would accept this invitation; half of the individuals who participated in the risk assessment were assumed to request a consultation with a primary care physician about their risk. Costs for screening-related events were determined using 2013 data from OHIP and CIHI. The costs for screening invitations and fixed costs related to the screening program, such as costs for the screening registry, program infrastructure, communications, and advertising, were derived from those incurred in the recent establishment of the colorectal screening program administered by Cancer Care Ontario. The costs for risk assessments were estimated assuming that screening program staff trained in health communication would administer the assessments. Fixed costs were counted up to the year in which the last individuals in the cohorts are eligible for screening (2045 for screening scenarios that end at age 75, 2050 for screening scenarios that end at age 80).

All costs were expressed in Canadian dollars (using May 2013 levels as a base) and were adjusted to reflect the May 2015 prices for health care services using the Ontario Consumer Price Index derived from Statistics Canada [37]. A lifetime time horizon for the costs and effects of screening was applied to each simulated person. Annual discount rates of 3% were applied to both costs and effects, using 2015 as the reference year [38]. The estimated attributable costs of lung cancer care by phase of care and the estimated costs related to the screening program are presented in Tables 2 and 3, respectively. To reflect the uncertainty in these cost estimates, sensitivity analyses were performed, which varied the costs by 50%, as described in a later section of this manuscript. Although cost-effectiveness thresholds have been proposed in the past, there is no official cost-effectiveness threshold employed in the Canadian health care system [37]. Therefore, a cost-effectiveness threshold of $50,000 Canadian dollars ($41,114 in May 1, 2015, US dollars) per life-year gained was chosen, similar to previous Canadian cost-effectiveness studies [39,40].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Attributable costs (in Canadian dollars) estimates used in the MISCAN-Lung model.

Costs were estimated per phase of lung cancer care per person-year of treatment by gender, age, and stage of disease for Ontario, Canada.

https://doi.org/10.1371/journal.pmed.1002225.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Costs (in Canadian dollars) of screening-related events and fixed program costs used in the MISCAN-Lung model.

https://doi.org/10.1371/journal.pmed.1002225.t003

Benefits, Harms, and Cost-Effectiveness of Screening Scenarios

For each screening scenario, the number of lung cancer deaths prevented, life-years gained, proportion of individuals ever screened, number of CT examinations, screen-related biopsies/bronchoscopies, false-positive screens, non-lung cancer-related surgeries, overdiagnoses, and costs were compared with a situation in which screening does not occur, from 2015 onward. Screening scenarios that were more costly and less effective (i.e., fewer life-years gained) than other scenarios were ruled out as non-efficient by simple dominance. Scenarios that were more costly and less effective than a combination of other scenarios were also ruled out as non-efficient by extended dominance. The remaining screening scenarios constitute the frontier of efficient screening scenarios, i.e., the efficient frontier. For each efficient screening scenario, the ICER was determined, calculated as the incremental net costs per incremental life-year gained compared to the previous efficient screening scenario.

Sensitivity Analyses

A number of sensitivity analyses were performed to investigate which groups of cost estimates and attendance assumptions have the greatest influence on the cost-effectiveness estimates by varying the costs for CT examinations by 50% compared with the base-case analyses, varying the attributable costs of lung cancer care by phase of care by 50% compared with the base-case analyses, and imperfect attendance rates for screening: low attendance (33% overall compliance rate), average attendance (55% overall compliance rate), and high attendance (64% overall compliance rate). Sensitivity analyses were performed for all 576 scenarios to investigate the effects of variations in assumptions on the composition of the efficient frontier.

Screening Scenarios on the Efficient Frontier

The net discounted costs and life-years gained (from 2015 onwards) for each scenario were used to determine the screening scenarios on the efficient frontier, i.e., the scenarios that provide the highest number of life-years gained for their costs, in the base-case analysis, as shown in Figs 1 and 2. The scenarios that are on the efficient frontier are described in Table 4 and shown in Fig 2. A complete overview of the net discounted costs and life-years gained of all investigated screening scenarios is presented in S1 Appendix Figs A–H. All outcomes were reported per 100,000 individuals alive in 2015.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The cost-effectiveness of all 576 investigated lung cancer screening scenarios in the base-case analysis.

Results are presented per 100,000 individuals alive in 2015 and are discounted by 3% annually. Scenarios on the efficient frontier are described in Table 4.

https://doi.org/10.1371/journal.pmed.1002225.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The incremental cost-effectiveness of the lung cancer screening scenarios on the efficient frontier.

Results are presented per 100,000 individuals alive in 2015 and are discounted by 3% annually. Scenarios on the efficient frontier are described in Table 4.

https://doi.org/10.1371/journal.pmed.1002225.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Cost-effectiveness estimates for lung cancer screening scenarios on the efficient frontier.

https://doi.org/10.1371/journal.pmed.1002225.t004

All scenarios on the efficient frontier consist of annual screening (Table 4), while biennial screening is dominated. Assuming a cost-effectiveness threshold of $50,000 Canadian dollars ($41,114 May 1, 2015, US dollars) per life-year gained as acceptable for the Canadian health care system, Scenario #2 was considered optimal: current and former smokers (who quit ≤ 10 y ago) who smoked ≥40 pack-years would be screened annually between ages 55–75, yielding an ICER of $41,136 Canadian dollars ($33,825 May 1, 2015, US dollars) per life-year gained. If 50% lower or higher attributable costs were assumed, the ICER of this scenario was estimated to be $38,240 ($31,444 May 1, 2015, US dollars) or $48,525 ($39,901 May 1, 2015, US dollars), respectively. If 50% lower or higher costs for CT examinations were assumed, the ICER of this scenario was estimated to be $28,630 ($23,542 May 1, 2015, US dollars) or $73,507 ($60,443 May 1, 2015, US dollars), respectively.

In addition, the benefits and harms of all scenarios on the efficient frontier were examined (Table 5). Scenario #2 would reduce lung cancer mortality in the overall population (which includes non-eligible individuals) by 9.05%, preventing 251 lung cancer deaths and gaining 2,531 life-years (undiscounted) over the lifetime of the program (i.e., on average, 10.08 life-years would be gained for each lung cancer death prevented). However, in Scenario #2, 9.56% of the overall population would receive at least one screen, requiring 91,692 CT screens and follow-up examinations. This scenario would lead to 14,729 false-positive screens and 163 surgeries for potentially benign disease (in persons in whom lung cancer is not detected) and 350 biopsies/bronchoscopies (in persons in whom lung cancer is not detected). Ultimately, 12.53% of all screen-detected cancers would be overdiagnosed, leading to 82 overdiagnosed cases. Based on the estimated number of individuals in the examined cohorts in 2016, Scenario #2 is estimated to screen 499,261 individuals at least once, require 4,788,523 CT examinations, and prevent 13,108 lung cancer deaths, while 4,282 cases of lung cancer would be overdiagnosed [26,27]. The average annual non-discounted costs compared to no screening would be approximately $1,400,000 Canadian dollars ($1,151,178 May 1, 2015, US dollars) over the considered time period; however, the annual costs are higher in the first years compared to later years, due to diminishing numbers of individuals meeting the eligibility criteria. For example, the average non-discounted costs compared to no screening are approximately $5,000,000 Canadian dollars ($4,111,350 May 1, 2015, US dollars) for 2015–2020 compared with approximately $1,600,000 Canadian dollars ($1,315,632 May 1, 2015, US dollars) in 2030–2035.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Overview of selected benefits and harms (per 100,000 individuals alive at the start of 2015) of the screening scenarios on the efficient frontier (effect estimates are not discounted).

https://doi.org/10.1371/journal.pmed.1002225.t005

Effects of Screening Scenario Characteristics on Cost-Effectiveness

Scenarios with older starting ages have lower costs compared with scenarios that start at younger ages but also yield a smaller number of life-years gained (S1 Appendix Fig A). Raising the age to stop screening from 75 to 80 increases the costs and the number of life-years gained, but differences are modest (S1 Appendix Fig B). A comparison of scenarios by smoking eligibility criteria indicates that there is little difference between using NLST-like or NELSON-like smoking eligibility criteria (S1 Appendix Fig C).

S1 Appendix Figs D–G demonstrate the importance of cumulative smoking criteria on cost-effectiveness. Each increase in the cumulative smoking requirement for enrollment substantially decreases the costs while modestly decreasing the number of life-years gained in both NLST-like (S1 Appendix Fig D) and NELSON-like (S1 Appendix Fig E) screening scenarios. In other words, scenarios that apply stringent cumulative smoking eligibility criteria are closer to the efficient frontier than those that apply less restrictive cumulative smoking eligibility criteria. In general, scenarios that require only 20 pack-years are dominated by scenarios that apply more stringent (higher) pack-year criteria.

Increasing the maximum number of years since smoking cessation (S1 Appendix Fig F) and not excluding individuals from further screening once they reach the maximum number of years since cessation (S1 Appendix Fig G) both increase the costs and the numbers of life-years gained. However, the effects of these criteria are less pronounced than those related to cumulative smoking requirements.

S1 Appendix Fig H shows the effects of annual screening compared with biennial screening. Although biennial screening scenarios have substantially lower costs compared with annual screening scenarios, the life-years gained are also substantially lower. S1 Appendix Fig H demonstrates that annual screening scenarios dominate biennial screening scenarios.

Sensitivity Analyses

Altering assumptions about attendance rates, CT examination costs, and attributable costs impacted the scenarios on the efficient frontier in the base-case analysis to varying degrees. Fig 3 provides an overview of the scenarios on the efficient frontier in the base-case analysis along with the discounted life-years gained and costs for these scenarios in the sensitivity analyses.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The incremental cost-effectiveness of the lung cancer screening scenarios on the efficient frontier and their corresponding cost-effectiveness throughout different sensitivity analyses.

Results are presented per 100,000 individuals alive in 2015 and are discounted by 3% annually. The relative ranking of the scenarios is consistent across sensitivity analyses (i.e., if a scenario is more costly and gains more life-years than another scenario in the base-case analysis, this is also the case in all sensitivity analyses).

https://doi.org/10.1371/journal.pmed.1002225.g003

Altering assumptions also impacted the composition of the efficient frontier, as shown in S5 Text. When the attendance rates were varied, lower attendance rates were observed to shift scenarios with less restrictive criteria, especially with regards to smoking behavior, on the efficient frontier (S5 Text Tables A–C). This may be due to the fixed costs of the screening scenarios, which are independent of the number of screened individuals; at lower levels of participation these costs have a greater influence on the cost-effectiveness than the costs of CT examinations.

When the attributable costs were varied, it was observed that halving the attributable costs had little effect on the scenarios on the efficient frontier (S5 Text Table D). When the attributable costs were doubled, it was observed that scenarios with less restrictive criteria, especially with regards to smoking cessation, were shifted on the efficient frontier (S5 Text Table E).

When the costs of CT examinations were varied, it was observed that halving the costs of CT examinations also shifted scenarios with less restrictive criteria, in particular with regards to smoking cessation, on the efficient frontier (S5 Text Table F). Doubling the costs of CT examinations had the greatest effect of all sensitivity analyses; the scenarios with the most restrictive criteria with regards to age and smoking were shifted on the efficient frontier and the least costly scenarios on this efficient frontier favored biennial screening (S5 Text Table G).

Scenario #2 was on the efficient frontier across all sensitivity analyses, with the exception of assuming the lowest attendance rates (S5 Text Table H). In contrast, although Scenario #5 closely resembles the eligibility criteria that were used in the NLST, it was not on the efficient frontier in any of the sensitivity analyses.