|Home | About | Journals | Submit | Contact Us | Français|
A randomized trial has demonstrated that lung cancer screening reduces mortality. Identifying participant and program characteristics that influence the cost-effectiveness of screening will help translate trial results into benefits at the population level.
Six U.S. cohorts (males and females aged 50, 60, or 70) were simulated in an existing patient-level lung cancer model. Smoking histories reflected observed U.S. patterns. We simulated lifetime histories of 500,000 identical individuals per cohort in each scenario. Costs per quality-adjusted life-year gained ($/QALY) were estimated for each program: CT screening; stand-alone smoking cessation therapies (4–30% 1-year abstinence); and combined programs.
Annual screening of current and former smokers aged 50–74 cost between $126,000–$169,000/QALY (minimum 20 pack-years of smoking) or $110,000–$166,000/QALY (40 pack-year minimum), compared to no screening and assuming background quit rates. Screening was beneficial but had a higher cost per QALY when the model included radiation-induced lung cancers. If screen participation doubled background quit rates, the cost of annual screening (at age 50, 20 pack-year minimum) was below $75,000/QALY. If screen participation halved background quit rates, benefits from screening were nearly erased. If screening had no effect on quit rates, annual screening cost more but provided fewer QALYs than annual cessation therapies. Annual combined screening/cessation therapy programs at age 50 cost $130,500–$159,700/QALY, compared to annual stand-alone cessation.
The cost-effectiveness of CT screening will likely be strongly linked to achievable smoking cessation rates. Trials and further modeling should explore the consequences of relationships between smoking behaviors and screen participation.
The National Lung Screening Trial (NLST) released results earlier than planned upon demonstrating that screening with CT (as compared to screening with chest X-ray) reduced lung cancer mortality by 20%.1, 2 As noted by NLST investigators,1 the mortality reduction observed in the self-selected, volunteer population (n=53,456) under controlled trial settings will not eliminate all uncertainties surrounding the effectiveness or value of screening in the general population.3 Modeling can complement randomized trials by simulating screening in populations with characteristics different than trial participants under different scenarios, and comparing screening to other lung cancer control interventions.
Because of the strong causal link between smoking and lung cancer, mortality reductions possible with screening will depend in part on whether screen participation alters a smoker's likelihood of quitting. CT screening has been described as a potential `teachable moment' for motivating continuing smokers to quit.4–7 Alternatively, if no lung cancer is detected, smokers could believe they will not develop cancer and have been given license to continue to smoke. Which of these effects, if either, will be observed in the NLST is not yet clear. European studies have reported that trial participants have higher quit rates than the general population, but small or no differences in quit or relapse rates between participants in the control and screened arms.8, 9
For this analysis, we used an existing microsimulation model that previously predicted mortality reductions between 15% (15 years of follow-up) and 28% (6 years of follow-up) for 5 annual screens in smokers with 20 pack-years of exposure,10 compared to 20% reduction (6 years of follow-up) for 3 annual screens in smokers with 30 pack-years of exposure as reported by the NLST.1 Our model simulated cohorts of individuals representative of the U.S. population and not specifically the NLST cohort. Treatment costs and measures of health-related quality of life, as well as risks of secondary lung cancers due to radiation exposure from CT examinations, were incorporated. The mortality reduction reported from the NLST was not used as an input or calibration endpoint for this analysis.
The purpose of our study was to estimate the cost-effectiveness of CT screening for lung cancer in the U.S. population and to identify characteristics of lung cancer screening programs (i.e., screen frequency and adherence, eligibility, follow-up program, effects on cessation, and costs) with the largest influences on the cost-effectiveness of screening. We compared smoking cessation therapies to screening and to programs that combined smoking cessation and screening.
This study used publicly-available de-identified human subject data and the single cohort Lung Cancer Policy Model (LCPM), previously used to predict the long-term effectiveness of screening.10, 11(see Supplemental Digital Content 1, which provides additional methods and model details, Supplemental Digital Content 2, which provides a schematic of the model, Supplemental Digital Content 3, which provides model inputs specific to this analysis, Supplemental Digital Content 4, which provides costs used for this analysis, and www.cisnet.cancer.gov/profiles for details on the LCPM, including references for model inputs, calibration and validation, and prior applications).
The LCPM is a patient-level microsimulation model of lung cancer development, progression, detection, treatment, and survival. Lung cancer natural history parameters were previously estimated via calibration against tumor registry data on age-specific lung cancer incidence rates, distributions of size, stage and cell types of incident lung cancers, and lung cancer-specific survival. Lung cancers have varied progression rates and can be 5 major histologies. Model validation included reproducing endpoints from 2 screening studies (rates of positive screens, stage and cell type distributions) and 2 cohort studies (mortality).
The model simulates symptomatic, incidental, and (for screening scenarios) screen detection of benign and malignant pulmonary nodules. Sensitivity of screening CT exams is a function of the diameter and location (central versus peripheral) of a pulmonary nodule. CT has a sensitivity of 0.63 for peripheral nodules of 1mm in diameter, 0.77 for peripheral nodules of 4mm in diameter, and 1.0 for peripheral nodules of 8mm in diameter or larger. CT was assumed to have lower sensitivity for central nodules (75% of that for peripheral nodules of the same size) to account for obstruction by the aorta, etc. Pulmonary nodules (benign and malignant) detected on imaging exams are triaged by size. In the base case, nodules <4mm are not followed, nodules 4–8mm are followed by serial high-resolution CT (at months 1, 3, 6, 9, 12, and 24) and nodules ≥8mm are biopsied with bronchoscopy, fine-needle biopsy, or video-assisted thoracoscopy (VATS). In sensitivity analyses, we simulated fewer high-resolution CT examinations (4–6mm nodules followed up at months 9 and 24, and 6–8mm at months 6, 12, and 24).
Survival after diagnosis is modeled explicitly as a function of treatment and underlying disease characteristics (not using a stage-shift as in previous studies12). Staging and treatment of non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC) followed National Comprehensive Cancer Network consensus guidelines in place in 2000. Invasive exams and surgical resection are associated with operative mortality in the model.
Smoking histories (ages of starting and stopping smoking and an average dose) representative of 6 cohorts of U.S. white males and females aged 50, 60, and 70 years in 1990 were derived from survey data and used to generate six fixed cohorts of 500,000 individuals.11, 13 Competing mortality risks were stratified by smoking status, age, and sex. No lung cancer screening was recommended during the 1990s; thus the no-intervention scenario corresponds to observed incidence.
Current smokers faced an annual background rate of smoking cessation of 3%,14 uncorrelated with pack-years. Except where noted, screen participation was uncorrelated with the probability of cessation.
We investigated the effects of a smoking cessation therapy consisting of bupropion and nicotine replacement therapy (NRT) prescribed to current smokers. Omitting the distinction between therapy uptake and efficacy, 1-year abstinence rates evaluated were 4%, 8%, 16% or 30% (versus 3% background), reflecting estimates that vary widely depending on the population and intervention.15–19 Unless offered cessation therapy at the next program visit, individuals with elevated cessation rates were assumed to revert to the background 3% after 1 year.
In the model, features of the screening program (CT sensitivity, frequency of screening exams, eligibility for screening) are translated into estimates of the effectiveness of screening. In other words, the effectiveness of screening (i.e., the relative reduction in mortality) is a model output, generated as a non-linear function of the benefits and harms of screening. Screening effectiveness cannot be directly varied to identify thresholds for screening to be cost-effective.
Screening leads to detection of asymptomatic, prevalent lung cancers as well as benign pulmonary nodules. The proportion of screening exams with positive results (both false positive [benign] and true positive [cancer]) varies with the age of the cohort, the number of screens and the definition of a positive test. On a baseline (prevalent) screen with small (<4mm) nodules categorized as `not suspicious for lung cancer,' the positivity rate is approximately 20%.13 The model predicts that screening leads to an excess of lung cancer cases; the magnitude of the excess varies with numbers of screens and length of follow-up.10, 11 Harms from screening include operative mortality, costs of follow-up examinations (for both cancer and ultimately benign nodules), and an increased risk of subsequent lung cancers (with all attendant costs and outcomes) arising from radiation exposure from screening and follow up CT examinations (see Supplemental Digital Content 1, which provides details of radiation risk component). No disutilities from screening-related anxiety were considered.
Except where noted, all eligible individuals were assumed to participate in screening. In a sensitivity analysis, 70% of eligible individuals adhered to their screening schedule.
Except where noted, screening had no effect on the probability of smoking cessation.8 The direction and magnitude of any independent effect of screening on smoking cessation are unknown, so we postulated a wide range of pessimistic to optimistic scenarios in sensitivity analyses: each instance of screening participation could reduce (to 1.5%; pessimistic scenario) or increase (to 4% or 6%; optimistic scenarios) an individual's probability of quitting smoking over the subsequent year.
Diagnosis, staging, and treatment costs were derived from Medicare reimbursements. Costs of diagnostic tests (including screening CTs) were estimated by 2006 Medicare payments (per CPT code in the Physician Fee Schedule). Details of treatment costs by phase of care are available elsewhere.20 SEER-Medicare linked data were used to create patient-level monthly costs, grouped into baseline (pre-diagnosis), initial (30 days to 7 months after diagnosis), continuing, and terminal (final month of life) phases of care and stratified by stage, histology (NSCLC vs. SCLC), and treatment (surgery, radiation, chemotherapy). The initial phase excluded the first 30 days post-diagnosis, which encompassed costs of dissimilar staging algorithms. Baseline (non-lung cancer) medical costs were estimated as costs accrued in a 12-month period prior to presentation with lung cancer (excluding 3 months immediately preceding diagnosis to subtract costs of treating undiagnosed lung cancer). This analysis used constant costs (averaged over 1992–2003, in constant 2006$). We replaced continuing-phase treatment costs with baseline costs for long-term (>3 or 5 years) survivors in a sensitivity analysis.
Patient-time costs for diagnostic tests and treatments were estimated using age-specific wages of an average worker in 2006 (U.S. Bureau of Labor Statistics). Caregiver time (e.g., accompanying patient to chemotherapy) was assumed equal to patient time for diagnostics, surveillance, and initial phases of care, and represented by the cost of hospice care for the terminal phase. Costs of pharmaceuticals (e.g., anti-emetics, smoking cessation therapies) were estimated by average wholesale prices (Red Book, ThompsonReuters).
The cost of a smoking cessation intervention (see below) was estimated at $300, based on a combination of 1 month of 24-hour nicotine transdermal patches (Watson labs, 7 or 14 mg; $188) plus 30 days of 300mg bupropion HCl (Watson labs, 150mg tablets; $116). Separately, the cost of cessation was increased 6-fold to represent a longer treatment duration and/or additional costs required to achieve the same cessation rates.
Baseline quality of life (QOL) weights for non-lung cancer states, stratified by age and gender, were U.S.-specific standardized values derived from survey participants' self-reported health as measured by the EQ-5D.21 To avoid a possible increase in health-related QOL from treatment for lung cancer, we multiplied the baseline weights by weights for lung cancer states. Studies of QOL in lung cancer patients provide cell type, stage, and treatment-specific QOL scores derived from patients using a variety of assessment instruments22–24 and transformed into utilities.22 In the base case, patients undergoing resection (lobectomy) experienced a 3-month disutility and then a return to baseline.25
Costs and quality-adjusted life-years (QALYs) were discounted at 3% annually.26 Incremental cost-effectiveness ratios were calculated for non-dominated strategies in 2006 U.S. $/QALY.26 A strategy was dominated if it cost more but provided fewer benefits than a comparator or if it had a higher incremental cost-effectiveness ratio than a more effective strategy (weakly dominated).
Ratios were first calculated for CT screening versus no intervention. For the base case and the scenarios that incorporated risks of radiation-induced lung cancer, individuals with ≥20 pack-years were eligible. Scenarios with more-restrictive eligibility or independent effects of screen participation on cessation but no radiation risks were also evaluated.
Alternative program characteristics (CT sensitivity, screening adherence, diagnostic test costs, operative mortality, follow-up of nodules <4mm) and assumptions regarding QOL (a return to baseline QOL at 1, 6, 12, or 24 months after lobectomy, or never27) were simulated in 50 year-old males.
Cessation alone and a combined cessation/screening were simulated in 50 year-olds. The combined program consisted of screening CT for current and former smokers, plus bupropion and NRT prescribed to current smokers at the screening examination.
We simulated lifetime histories of the six fixed cohorts of 500,000 individuals described above in each scenario. The number of 1st-order trials is scalable and 2nd-order uncertainty is not considered in these analyses. Thus, p-values for comparisons between strategies are uninformative and not reported.
Per person screened (≥20 pack-years), the incremental lifetime healthcare costs for a single screen (vs. no screen) varied from $1,778–$3,637 and provided between 0.009– 0.022 additional QALYs ($144,000–$207,000/QALY). Corresponding gains in undiscounted life-years ranged from 0.018–0.045 (7–16 days). One-time screening provided reductions in lung cancer-specific mortality of between 5.02%–7.52% compared to no screening.
In all cohorts single screening was excluded by weak dominance. Compared to no screening, annual screening of persons with at least 20 pack-years of smoking history reduced lung cancer-specific mortality by 17.98%–25.16% at 10 years at a cost of $126,000–$169,000/QALY (Table 1). Including risks of radiation-induced lung cancer yielded smaller mortality reductions and higher costs ($139,000–$203,000/QALY, Table 1, follow-up CT dose of 58mGy). Alternate estimates for organ doses for follow-up CTs (both 10mGy and 90mGy were used) yielded a range of $133,000–$247,000/QALY (not shown). Follow-up algorithms with fewer CT examinations yielded ranges of $121,000–$160,000/QALY without radiation risks, and $127,000–$160,000/QALY with radiation risks (Table 1).
Restricting screening to individuals with ≥40 pack-years, current smokers, or current smokers and recent quitters yielded ratios from $110,000–$166,000/QALY (Table 2). Single screens were weakly dominated.
In cohorts of 50 year-olds, if screen participation itself were associated with a doubling (to 6%) of the background cessation rate, annual screening (perfect adherence) was associated with a cost (vs. no screening) of $73,000/QALY (men) and $40,000/QALY (women), while a 33% increase (to 4%) of background cessation resulted in cost-effectiveness ratios of $105,000/QALY (men) and $97,000/QALY (women). In the pessimistic scenario in which screen participation was associated with a 50% reduction (to 1.5%) in cessation, the cost-effectiveness ratios were $880,000/QALY (men) and >$1 million/QALY (women; Table 3).
Treatment costs that return to baseline at 36 or 60 months after resection (vs. base case 100 months) yielded ratios of $134,000/QALY or $135,000/QALY (women) and $145,000/QALY or $146,000/QALY (men; Table 3).
In a cohort of 50 year-old men, imperfect (70%) adherence to annual screening was associated with a cost (vs. no screening) of $180,000/QALY (Table 4). Perfect sensitivity of CT for peripheral pulmonary nodules yielded a ratio of $141,000/QALY (Table 4), comparable to the base case: most of the additional nodules detected were <4mm and therefore not followed. Neither following nodules <4mm nor a reduction in the cost of screening CT was associated with a ratio of <$100,000/QALY. Reductions in operative mortality for lobectomy and invasive staging examinations reduced the ratio to $141,000/QALY. Prolonged delays before a return to baseline health after resection yielded greater costs per QALY.
One-time smoking cessation therapy costing $300 cost between $11,400/QALY (30% cessation, men) and $69,300/QALY (4% cessation, women), compared to no intervention (Table 5). Annual smoking cessation therapy programs offered additional benefits at costs from $12,500–$69,400/QALY. An annual combination strategy yielded more benefits than annual cessation therapy alone, but at an incremental cost from $130,500–$159,700/QALY (see Supplemental Digital Content 5.doc, which is a figure legend for Supplemental Digital Content 6.tif, which is a plot of costs versus effects).
Cessation therapy costing $1,800 and yielding 4% abstinence was weakly dominated by annual screening in 50-year olds. Results for abstinence rates of 8% or higher followed a pattern similar to the $300 cessation therapy: ratios <$60,400/QALY for annual cessation therapy and >$100,000/QALY for annual combination programs (not shown).
The NLST recently provided evidence that individuals randomized to 3 annual CT screens had 20% lower lung cancer-specific mortality than individuals randomized to 3 annual chest X-ray screens.1 The NELSON trial (n=15,822) is powered to show a mortality reduction of 25% from CT screening, compared to usual care.28 It was not possible to simulate these trials or to calibrate to the NLST results, because we did not have individual-level demographics and smoking histories (pack-years decomposed into dose and duration over time) necessary to predict lung cancer risks. We therefore simulated cohorts of individuals representative of the U.S. population. Further, we did not include a module for chest x-ray screening and could not simulate the NLST's follow-up patterns for individuals with mm-sized nodules (the study design did not specify a protocol). Our analysis, completed prior to release of the NLST results, predicted a range of 17.76% to 25.16% reduction in lung cancer-specific mortality at 10 years, depending on the number of screens (4 to 10), the cohort, and whether radiation risks are included (see Table 1), compared to no intervention. Other modeling studies of the effectiveness of CT screening completed prior to the NLST predicted benefits ranging from no benefit29 to reductions in lung cancer-specific mortality of 8.0% to 45.6%, depending on model assumptions and screening program characteristics (e.g., eligibility criteria, screening modality and frequency, patient management, and length of follow-up).10, 12, 30–34
We found that unless screen participation increases smoking cessation, lung cancer screening was considerably more expensive than other U.S. screening programs. Colorectal cancer screening — widely viewed to be cost-effective – has cost-effectiveness ratios (versus no screening) in the range of $13,000–$32,000/LY (2006$).35 Breast cancer screening of women aged ≥40 with mammography has a cost-effectiveness ratio compared to no screening of $47,700/QALY (2006$).36
We predicted low costs per QALY for cessation therapy, consistent with estimates that range from cost-saving to $17,000/QALY (2006$).37–39 Much of the gain in life expectancy after cessation is due to decreases in deaths from causes other than lung cancer.40
Our base-case estimates of $126,000–$169,000/QALY for annual screening of 50–70 year-old ever-smokers is comparable to a prior estimate of $143,000/QALY (2006$) for annual CT screening of current smokers.12 Our microsimulation approach includes a natural history model calibrated to tumor registry data rather than a stage shift as the mechanism for screening effectiveness, so did not require estimates of screening biases as inputs. Our approach simulates symptom-related and incidental detection of lung cancers in addition to screen detection and permitted evaluation of strategies employing cessation therapy and programs that employed different follow-up algorithms and eligibility. We showed that the influences of eligibility, screening frequency, adherence, frequency of follow-up CT exams, and accumulated radiation risks on cost-effectiveness were smaller in magnitude than influences of cessation. Our model suggests that results from screening trials should be interpreted with consideration given to the specifics of any cessation component.
Our `no-intervention' scenarios were fit to observed incidence rates and employed contemporaneous inputs but are historical, and omit advances in staging and targeted therapies. Our assumption that patients undergo guideline care was common to all scenarios but overestimates the percentage of patients who undergo lobectomy. Our results may thus overestimate the gain in QALYs attributable to screening. We allow for non-operative candidates (7.7%), but an additional 16–20% (depending on age) of early-stage patients in SEER do not undergo surgery for unspecified reasons. Our analysis was limited to whites due to insufficient individual-level data in other populations on smoking histories and cancer outcomes necessary for model development.
We assumed a societal perspective, which dictated inclusion of all costs without regard to who pays them. We used Medicare fee schedules which were in part designed to approximate the resource use costs of all medical interventions, including the initial screening examination which is not typically reimbursed.26 Use of health-related quality of life indexes other than the EQ-5D for the general population would have yielded different QALY totals but because trends by age are similar across standard indexes,41 the incremental change in QALYs and therefore costs per QALY between scenarios would be similar to those we estimated.
Imperfect uptake of cessation therapies would likely translate to abstinence rates closer to the lower end (4%) of the range we evaluated than the maximum (30%). We did not vary screening's effect on cessation according to the test result, and did not consider perceptions of lung cancer risk, which may influence participation and cessation.7, 42 Our predicted mortality reduction from annual screening in the absence of radiation-induced lung cancers (>16% in all cohorts) exceeded a threshold (1%–4%) postulated to outweigh risks of radiation-induced lung cancers (50–52 year-old male current smokers).43 When risks of radiation-induced lung cancers were included in the model, estimated mortality reductions at 15 years were within 1% (on an absolute scale) of the base case estimates in male cohorts and within 3% in female cohorts. Reducing the frequency of follow-up CT examinations reduced the magnitude of the effect of radiation risk on 15-year mortality reductions (within 1% in all cohorts) from screening but did not alter the conclusions: the risk of radiation-induced lung cancers was outweighed by the reduction in deaths from lung cancer attributed to screening. Fewer follow-up CT examinations reduced radiation risks, but also delayed detection of some small, growing cancers: mortality reductions were lower compared to the base case, and the cost for all 6 cohorts remained over $120,000/QALY.
Changes to several model inputs could result in higher costs per QALY than estimated in this study: an increased cost of the screening CT exam or other treatments (for instance, to reflect retail prices rather than opportunity costs and the societal perspective we adopted); addition of PET-CT in the staging algorithm (although costs of PET-CT would be offset by some reductions in excisional biopsies); and addition of targeted therapies for lung cancer such as erlotinib (although the cost per QALY would offset by increased survival for patients with sensitizing mutations).
In conclusion, results from a microsimulation model suggest that the cost-effectiveness of CT screening programs will be strongly influenced by smoking cessation rates among screen participants. The specific eligibility criteria will be more influential on the cost-effectiveness of screening than other program characteristics such as the screening test cost or radiation risks. Unless screen participation increases the probability of cessation, screening with helical CT may cost over $100,000/QALY compared to no intervention, and be more expensive and provide fewer QALYs than an annual cessation intervention. A combined screening/cessation program would offer benefits to both current and former smokers, but would cost over $100,000/QALY. Understanding behaviors surrounding smoking and screen participation6, 7 will be critical for translating trial results into population benefits.
Rob Boer, PhD, David Levy, PhD, Karen M. Kuntz, Sc.D., David C. Christiani, M.D., M.P.H., CISNET Lung investigators, Michael Gilmore, MBA.
Funded by American Cancer Society (RSG 2008A060554), National Cancer Institute (R01CA97337 [Gazelle], R00CA126147 [McMahon]). The National Electrical Manufacturers Association funded two authors (GSG, ACT) for an unrelated project. GSG is a consultant to GE Healthcare.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.