Cost‐effectiveness of risk‐based low‐dose computed tomography screening for lung cancer in Switzerland

Throughout Europe, computed tomography (CT) screening for lung cancer is in a phase of clinical implementation or reimbursement evaluation. To efficiently select individuals for screening, the use of lung cancer risk models has been suggested, but their incremental (cost‐)effectiveness relative to eligibility based on pack‐year criteria has not been thoroughly evaluated for a European setting. We evaluate the cost‐effectiveness of pack‐year and risk‐based screening (PLCOm2012 model‐based) strategies for Switzerland, which aided in informing the recommendations of the Swiss Cancer Screening Committee (CSC). We use the MISCAN (MIcrosimulation SCreening ANalysis)‐Lung model to estimate benefits and harms of screening among individuals born 1940 to 1979 in Switzerland. We evaluate 1512 strategies, differing in the age ranges employed for screening, the screening interval and the strictness of the smoking requirements. We estimate risk‐based strategies to be more cost‐effective than pack‐year‐based screening strategies. The most efficient strategy compliant with CSC recommendations is biennial screening for ever‐smokers aged 55 to 80 with a 1.6% PLCOm2012 risk. Relative to no screening this strategy is estimated to reduce lung cancer mortality by 11.0%, with estimated costs per Quality‐Adjusted Life‐Year (QALY) gained of €19 341, and a €1.990 billion 15‐year budget impact. Biennial screening ages 55 to 80 for those with 20 pack‐years shows a lower mortality reduction (10.5%) and higher cost per QALY gained (€20 869). Despite model uncertainties, our estimates suggest there may be cost‐effective screening policies for Switzerland. Risk‐based biennial screening ages 55 to 80 for those with ≥1.6% PLCOm2012 risk conforms to CSC recommendations and is estimated to be more efficient than pack‐year‐based alternatives.


| INTRODUCTION
Lung cancer (LC) is the leading cause of cancer-related mortality in Europe. 1 Clinical LC diagnosis typically occurs in a metastasized stage; 5-year LC survival is only 11%. 2 To facilitate diagnosis at an earlier cancer stage, individuals at high risk of LC may benefit from low-dose computed tomography (CT) screening, which has shown LC mortality reductions of 20% in the US National Lung Screening Trial (NLST) and 24% in the Dutch-Belgian lung-cancer screening trial (Nederlands-Leuvens Longkanker Screenings Onderzoek [NELSON]). 3,46][7][8] The benefits, harms and costs, may vary by the strategy employed, urging careful selection of the screening strategy. 9,10The United States Preventive Services Taskforce (USPSTF) recommends annual CT screening for individuals aged 50-80 with at least 20 pack-years smoked (PYs) and maximally 15 years since smoking cessation. 11Screening with these criteria is estimated to be cost-effective, but not the most efficient strategy in terms of costs per quality-adjusted life year (QALY) gained. 9,10 make LC screening more effective, it may be beneficial to invite individuals based on LC risk, rather than categorical criteria such as PYs smoked.Such a strategy would invite all individuals above a certain model-based LC risk for CT screening. 12One such model is the PLCOm2012 model of 6-year LC incidence risk, which uses smoking history and intensity, age, race, education, body mass index (BMI), presence of chronic obstructive pulmonary disease (COPD) and personal/family cancer history. 13Screening based on the PLCOm2012 model has been shown to improve efficiency of screening in the NLST population, 14 a finding recently supported by interim results from the International Lung Screening Trial (ILST). 15e incremental benefits and harms of risk-based LC screening are not known for the European setting, despite individual risk prediction models being recommended for selection into LC screening. 16e UK-based Targeted Lung Health Check (TLHC) program employs the PLCOm2012 model for selection into screening, with favourable interim results. 17,18However, it is unknown whether the chosen 1.51% risk threshold, combined with the targeted age range of 55 to 74 years, represents the optimal strategy for other European countries.Moreover, there are no known estimates of the incremental harms and benefits of risk-based screening in the European setting, relative to PY-based criteria.[21][22][23][24] Recently, the Swiss Cancer Screening Committee (CSC) issued a recommendation in favour of LC screening. 25Pending a reimbursement decision, the committee suggests biennial screening focusing on younger populations (eg, 55-80 years rather than 60-85 years) with moderate smoking histories (eg, smokers from 20 PYs and including ex-smokers), without a specific recommendation for a risk-or PY-threshold.
In this study, we present a microsimulation-based costeffectiveness and budget impact analysis of risk-based screening, from a public payer perspective.This study builds on our analyses for the CSC-commissioned Health Technology Assessment report. 25,26Here, we present the cost-effectiveness of CT screening, and the set of most efficient screening strategies.Compliance to CSC recommendations was considered to assess implementation feasibility.

| METHODS
We performed a microsimulation-based cost-effectiveness analysis of LC screening.We include the CHEERS 2022 checklist in Data S1A. 27

| MISCAN model
We used the MISCAN-Lung natural history model, as calibrated to individual-level data from the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO) and the NLST. 28The model has informed USPSTF 2013 and 2021 screening recommendations. 29,30,31 For each simulated individual, a smoking history was established, with probabilities of smoking initiation and cessation specific to birth cohort and sex.Methods and data underlying the smoking histories are reported in Data S1B.The age of death from other causes than LC is established, accounting for individual smoking behaviour. 32Mortality rates by smoking exposure are adapted from the literature and validated against published estimates for Switzerland, as reported in Data S1B. 33,34 the individual develops LC, the onset age is generated based on the smoking history.At onset a cancer histology is drawn from a distribution consistent with Swiss LC incidence.Over time, the cancer may progress while remaining preclinical, or be detected clinically.When detected, a stage-and histology-specific time until death from LC is drawn from the country-specific survival distribution.Model parameters have been described previously 28 and are included in Data S1B.
CT screening rounds were simulated at the strategy-specific interval, with individual eligibility by age determined by the smoking history.LC may be detected with a sensitivity specific to the LC histology and stage at the time of screening.Relative to our previous study, 9 CT sensitivity is incremented by 5 percentage points for stage 1A to 2 to reflect advances in screening since the NLST. 3,35A simulation of the NELSON protocol, as shown in the methodological supplement, finds a 5-percentage point increase to best replicate published NELSON lung cancer mortality rate ratios.This increment is subject to sensitivity analysis to account for uncertainty surrounding the sensitivity improvement.A screen-detected LC has a stage-specific probability of the lung cancer death being prevented.If unsuccessful, the age of death is applied from the life history without screening.

| Population
We studied birth cohorts 1940 to 1979, extending our previous study of cohorts 1935 to 1965.Cohort sizes by year and sex reflect the population composition of December 31, 2020. 36Swiss smoking data by 5-year cohort and 5-year age group determined rates of smoking initiation and cessation and cigarettes per day, as described in Data S1B. 37Cohort mortality tables are adjusted for LC mortality and smoking-related mortality.The use of cohort life tables represents a change relative to our previous study, for which only period lifetables were available. 9The LC histology distribution and LC survival were adjusted to incidence data from the Swiss National Institute of Cancer Epidemiology and Registration (NICER). 38For 10 million simulated individuals, LC outcomes were generated and recorded up to age 100.

| Evaluated outcomes
For each simulated individual, we recorded life years (LYs), QALYs lived and LC outcomes.To determine cost-effectiveness of a screening strategy, we evaluated the population gain in LYs and QALYs, relative to the scenario without screening.Strategies that gained the most QALYs for their level of cost constitute the cost-effective frontier.Secondary outcomes, such as follow-up procedures, secondary CT scans and lung biopsies were recorded per rates from the NELSON trial (Table S1).Costs, LYs, and QALYs were discounted at 3% relative to 2023, the presumed start of screening.Cost-effectiveness of a strategy on the efficiency frontier is given by the incremental cost-effectiveness ratio (ICER): the incremental cost per QALY gained relative to the next-cheapest strategy on the frontier.To set a cost-effectiveness threshold, we maintained a Willingness-to-Pay (WTP) of €38 000, equal to the most recent EU gross domestic product (GDP) per capita. 39Additionally, we considered whether strategies meet CSC suggestions of screening moderate smokers aged 55 to 80.

| Screening strategies
We simulated 1512 strategies, varying by starting age, stopping age, screening interval and eligibility requirement (Table 1).The CSC recommends biennial screening for Switzerland, citing capacity concerns with respect to annual screening. 25We therefore took biennial screening as the base-case of feasible strategies, considering triennial and annual screening as sensitivity analyses.As eligibility requirements, we considered PY-based strategies, employed in the NLST 4 and advised by the USPSTF, 11 and smoking duration-based strategies, employed in the NELSON trial. 3Additionally, we simulated screening eligibility based on PLCOm2012 risk levels, 13 as used in the ILST. 15A B L E 1 Characteristics of the evaluated screening scenarios.We used the reduced-form PLCOm2012 model, which considers sex, smoking duration, cigarettes per day (CPD) and years since smoking cessation, with performance very close to the complete model. 40The reduced-form model assumes reference values for covariates included in the complete model, which means that real-world screening may include more individuals, or include them at an earlier age.Our basecase assumed perfect screening attendance, with lower attendance simulations included as a sensitivity analysis.

| Costs and health utilities
Cost and utility values are given in Table 2.A public payer perspective was employed to align the cost-effectiveness analysis with a policy maker perspective.We included risk-assessment and invitation costs.
Costs of LC care from the University Hospital Zurich were used from our previous study, adjusted for inflation and increased use of novel LC therapies. 9LC-attributable costs for 1112 patients were included, T A B L E 2 Cost and QALY input.values from a meta-analysis of LC health utilities. 41Terminal-phase utilities are applied for the final 6 months, otherwise stage-specific utilities are applied in tandem with the costing phase (initial or continuing care).Population-level utilities were taken from a study of Swiss individuals. 42The lower value of the age-specific and lung cancer health state utility is applied.LDCT screening cost was estimated at €420.

| Sensitivity analyses
We repeated our analysis with discount rates of 0.0%, 1.5%, 3.0% (the base-case), 4.5% and 6.0%.The cost and utility values were subject to a univariate sensitivity analysis, as well as a multivariate probabilistic sensitivity analysis (PSA), for which statistical distributions are reported in Table 2.
We studied the effect of the assumed CT sensitivity for earlystage LC.The CT sensitivity by stage and histology per estimates from the NLST and the PLCO was considered, 28 as well as a 5 (the basecase), 10 and 15 percentage point increase in CT sensitivity for stages 1A to 2.
Additionally, we considered a scenario where screening limited to those with a minimum 5-year remaining life expectancy.This scenario reflects the potential impact of shared decision-making preceding entrance into a screening programme.This strategy has been shown to reduce overdiagnosis projections in population-level screening. 10 also repeat our analysis with varying attendance rates.We consider 100%, 75% and 50% attendance rates.We assumed nonattendance to be caused partly by never-attendance and partly by incidental non-attendance.For an attendance level p, it is assumed 1À√p of eligibles never attend, while √p of eligibles attend √p of their scans for an overall attendance rate of p.
Finally, we test our adjustment of cessation rates to fit Swiss LC incidence, relative to the alternative of adjusting background risk.The methodological Data S1B specifies this analysis, which finds screening to be 4.5% less cost effective (per the RISK16 ACER) when decreasing baseline risk relative to our assumption of increased cessation.

| RESULTS
We report primary results in the main text, with additional Tables and Figures reported in Data S1C.would depart from the CSC-suggested age range and screening coverage.RISK11 may therefore be the most feasible.If we maximize life-years gained (LYG), RISK11 remains on the frontier, with a €23 138 ICER relative to RISK10 (Table S2).

| Screening Interval
Current programmes of LC screening advise annual screening. 4,6wever, the CSC recommend biennial screening in light of capacity concerns. 25Figure 1 shows the efficiency frontiers of annual, biennial and triennial strategies (the complete set of strategies are shown in Figure S1).We find that for strategies with expenses similar to strategy RISK11, the incremental benefit of annual screening is marginal.
The closest annual strategy to RISK11 in projected screening volume is RISK4a (Data S3), screening ages 60 to 80 from 3.0% risk.We estimate it would yield 9.7% more QALYs for 12.7% additional costs and 20.5% more screens.Both TLHC-like (annual screening ages 55-75  S4) is found to be dominated by the most efficient biennial strategies.

| Budget Impact
We calculated the budget impact of screening cohorts 1940 to 1979 with the RISK11 strategy, which is CSC-compliant and estimated to be cost-effective.Figure 2  Swiss CT volume. 43

| Sensitivity analyses
Table S6 and Figure S6 show the changes in costs and QALYs for the efficient biennial strategies for 50% and 75% screening attendance.
RISK11 has a 3.7% higher ACER (€20 056) at 50% attendance, suggesting screening with imperfect attendance is less efficient, but still cost-effective.
Table S5 shows the change in outcomes when screening is limited to those with a minimum 5-year life expectancy, per their individually generated other-cause mortality age.For RISK11, this reduces projected overdiagnosis from 4.9% to 0.9%, with 7.3% fewer screens and 0.3% fewer QALYs gained.
We evaluate the sensitivity of our results to cost and utility inputs.The ACER of RISK11 is evaluated at the bounds of the 95% confidence interval (95% CI) of input-specific distributions per

| DISCUSSION
We present the cost-effectiveness of LC screening in Switzerland.
Our estimates from the MISCAN-Lung model find screening to be a cost-effective measure, consistent with previous European and US estimates.Relative to annual screening, biennial and triennial screening are expected to reduce the total QALY benefit.However, less frequent screening reduces the required CT capacity, and is still estimated to be cost-effective.In 2019, 1.18 million CT scans were conducted in Switzerland. 43Biennial screening ages 55 to 80 from 1.6% PLCOm2012 risk would require an estimated 172 000 additional scans in the first year, a 15% increase.A TLHC-like strategy of annual screening ages 55 to 75 from 1.51% risk would require 290 000 scans (+25%).The USPSTF2021 strategy of annually screening ages 50 to 80 from 20 PYs is estimated to require 530 000 scans (+45%).Even with imperfect attendance, the CT volume for annual screening may be difficult to achieve, warranting deference to biennial screening.
We find screening to be more cost-effective than our previous analysis of older cohorts. 9We attribute the difference to the increased life expectancy of the newer cohorts, yielding more QALYs per life saved.Our analysis also includes higher CT sensitivity estimates, which favour screening effectiveness.
We find the cost per QALY gained of screening those with 1.6% PLCOm2012 risk to be robust to changes in input parameters.Our PSA showed a 95% CI of the ACER of €10 545 to €28 609.This suggests that screening with this strategy is cost-effective at our assumed cost-effectiveness threshold of €38 000, even for unfavourable parameter combinations.However, the assumed independence of cost input distributions means that unfavourable cost scenarios across inputs may have a larger effect than estimated here.
The cost-effectiveness of screening is sensitive to the CT cost, and terminal LC care costs.The optimal strategy will therefore depend on CT affordability.Terminal care costs for LC are also of interest for the cost-effectiveness of screening.5][46][47][48] This may improve the cost-effectiveness of screening, since these costs are partly supplanted by surgical costs for the early-detected cancers.However, if targeted therapies are implemented for earlier-stage cancers, this stage-shift effect may diminish. 49Of the quality of life inputs, screening cost-effectiveness was most sensitive to early stage LC utility.
Screening efficiency may be improved when participation depends on remaining life expectancy.Although an idealized scenario, in practice screening may be reserved for those without excess morbidities prohibitive of benefiting from screening.The benefits of screening should therefore, in practice, be between the base scenario in which every eligible individual is screened, and the scenario in which only those with a minimum 5-year life expectancy are screened.
Consequently, the base-case overdiagnosis projection of 4.9% of screen-detected cases, represents a pessimistic scenario.
Our study results are limited by the validity of the MISCAN-Lung model as applied to this particular context.Structural assumptions on the natural history of lung cancer (such as the preclinical sojourn time length) and the effectiveness of screening (eg, use of a stage-shift or cure model) are known to influence the estimated benefits and harms. 50Comparative modelling studies 10,24 that aggregate various model specifications may give a more robust estimate of the effectiveness of lung cancer screening.Future research may also focus on more elaborate recalibration of the smoking dose-response model to novel epidemiological contexts, which may improve the projected lung cancer burden for a particular setting.Real-world lung cancer screening effectiveness will also depend on the success of encouraging (repeat) attendance.There is further debate about the assumptions regarding quality of life of lung cancer patients, and potential impacts on quality of life from indeterminate or false positive findings. 51e cost-effectiveness of LC screening may increase further with novel strategies of screening.The 4-IN-the-LUNG-RUN trial, 52 currently underway in five European countries, will investigate whether individuals with a negative baseline scan may benefit equally from a biennial screening as they would from an annual scan.Our analysis includes annual and biennial strategies, but does not consider personalized intervals.4-IN-THE-LUNG-RUN results may inform whether baseline-dependent risk stratification may improve screening efficiency.Screening has also been shown to be associated with smoking cessation, 53 which our analysis does not assume to occur in excess of the cessation rate without screening.

| CONCLUSION
We present the first comparative cost-effectiveness analysis of riskbased and PY-based screening for a European country.Incorporating recommendations from the CSC, we project the optimal strategy for Switzerland would be biennial screening of smokers and ex-smokers with 1.6% PLCOm2012 risk between the ages of 55 and 80.
42te: Cost values for the treatment and detection of lung cancer (LC) and Probabilistic Sensitivity Analysis (PSA) distributions used to test robustness of the results to these values.In the univariate PSA, values are taken from the given distribution for each input separately, ceteris paribus.In the multivariate case, all values are taken from their given distribution, assuming zero covariance between the cost inputs.Norm utility values are taken from Perneger et al.42SE values are calculated from the SD reported in the publication, with n-values for each age category supplied through correspondence with the authors.Costs are converted to € from CHF values per the September 1, 2022 exchange rate of 1.020 €/CHF.Outcomes per 100 000 individuals alive in 2023 for strategies on the efficiency frontier.Initial care (first 3 months after diagnosis), terminal care (final 6 months of life) and continuing care (up to a maximum of 5 years).For individuals with LC, we use quality of life a N(a, b) refers to a normal distribution with mean a and SD b. b 54T A B L E 3

Table 3
Simulated undiscounted outcomes of the most efficient biennial screening strategies, as well as a few inefficient strategies of interest.Outcomes are given per 100 000 individuals alive at the assumed start of screening of 2023.Strategies are sorted by their position on the efficiency frontier, equivalent to a sorting by the number of Quality Adjusted Life Years (QALYs) gained.NS refers to the non-screening scenario.PY1 to PY3 report scenarios in which a screening strategy is employed with eligibility based on a minimum of pack-years (PY) smoked and a maximum of years since smoking cessation (cess).RISK1 to RISK19 report strategies which base CT (Computed Tomography) screening eligibility on the reduced-form PLCOm2012 risk model.Stratified costs in millions of EUR and Incremental Cost Effectiveness Ratios (ICERs) for strategies on the efficiency frontier.Costs per 100 000 individuals alive in 2023 for each strategy on the efficiency frontier, stratified by primary cost category.The Incremental Cost Effectiveness Ratio (ICER) of each strategy is also given, which reports the incremental cost per QALY (Quality Adjusted Life Year) gained of implementing a strategy, relative to the strategy preceding it on the efficiency frontier.Cost values are given as millions of EUR per 100 000 individuals, except for the ICERs which are given nominally.Costs, LYs and QALYs are discounted by 3% annually starting from 2023.The number of computed tomography (CT) scans per 100 000 individuals are not discounted.
summarizes the most efficient biennial screening strategies.Without screening, we project 7011 lifetime LC cases per 100 000 individuals alive in 2023, associated with 4757 LC deaths.Screening strategies on the efficient frontier reduce LC mortality by 4.4% for strategy PY1 (biennial screening ages 60-75 those with >40 pack-years), to 17.6% for RISK19 (annual screening ages 55-85 those with >1.0%PLCOm2012 risk).T A B L E 3 (Continued) Abbreviations: LYG, life-years gained; LY, life-years; NNS, number of unique individuals needed to screen to achieve the given outcome; SD, screen-detected.aUSPTF-2021, and TLHC strategies, not part of the efficiency frontier.CSC1 (Cancer Screening Committee) to CSC4 report pack year-based strategies that follow suggestions in the recent CSC recommendations.25 b Total false positive results from all CT screening events, per rates from the NELSON trial.cBiopsies pertaining to false positive screening outcomes.dOverdiagnosis is reported as the percentage of simulated screen-detected cancers that were not associated with a lung-cancer death in the non-screening scenario over a lifetime horizon.TOMONAGA ET AL.T A B L E 4 Abbreviation: LY, life-years.aEligibility requirement, based on PLCOm2012 13 risk or pack-year (PY) based with a maximum number of years of smoking cessation (cess).3.1 | Effective screening strategies risk, maintains the CSC-suggested age range, and has a similar population coverage as the 20PY eligibility criterion (17.5% for RISK11, 15.1-18.6%for CSC1-CSC4).However, RISK11 is estimated to cost 9.1% less and require 11% fewer CT scans, yielding only 2% fewer QALYs.Furthermore, RISK11 is on the efficiency frontier, with an ACER of €19 341 (7.9% less than CSC2) and a €29 852 ICER (relative to strategy RISK10).With a €38 000 WTP, screening may be expanded up to strategy RISK15, screening ages 50-80 those with 1.3% risk.Although cost-effective per the estimated ICER, RISK15 QALYs gained vs Incremental Costs (per 100 000 individuals alive in 2023) vs No Screening by strategy.Incremental costs and QALYs (Quality-Adjusted Life Years) relative to no screening for the efficiency frontiers of biennial and annual screening strategies (ie, the selection of strategies that realize the highest number of QALYs at a given level of cost), as well as selected strategies of interest.Strategies include screening those with 20 pack-years (PYs) between ages 55 and 80, the Targeted Lung Health Check (TLHC) strategy and the United States Preventive Services Task Force (USPSTF) 2021 recommended strategy.Outcomes are scaled to 100 000 individuals alive at the presumed start of screening of 2023.Both QALYs and costs are discounted at a rate of 3%.The strategies constituting the efficiency frontier are reported in Table3.RISK11 represents a strategy of biennially screening of smoking individuals with 1.6% PLCOm2012 risk between the ages of 55 and 80. RISK12 represents a strategy of biennial screening of individuals with 1.7% PLCOm2012 risk between the ages of 50 and 80. Diagonal lines report the QALYs at each cost level required to meet a given willingness-to-pay (WTP) threshold.

Table 2 .
Figures S2 and S3 report the results.We find the cost-