Cost-effectiveness of lung cancer screening with low-dose computed tomography in heavy smokers: a microsimulation modelling study

Background: Lung cancer screening with low-dose computed tomography (LDCT) reduces lung cancer mortality. The aim of this study was to evaluate the cost-effectiveness of lung cancer screening with LDCT in a high-risk population. Methods: The study used an adapted micro-simulation model in a cohort of Dutch heavy smokers for a lifetime horizon from a health insurance perspective. The main outcomes included average cost-effectiveness ratio (ACER), incremental cost-effectiveness ratio (ICER) and lung cancer mortality reduction. The comparator was no screening. Scenarios with different screening intervals and starting and stopping ages were evaluated for 100,000 male heavy smokers and 100,000 female heavy smokers. A cost-effectiveness threshold of 60 k€ per life year gained (LYG) was assumed acceptable. Results: The evaluated screening scenarios yielded ACERs ranging from 17.7 to 32.4 k€/LYG for men and from 17.8 to 32.1 k€/LYG for women. The lung cancer mortality reduction ranged from 9.3% to 16.8% for men and from 7.8% to 13.7% for women. The optimal screening scenario was annual screening from 55 to 80 years for men and biennial screening from 50 to 80 years for women, with an ICER of 51.6 and 45.8 k€ per LYG compared with its previous efficient alternative, respectively. Compared with no screening, the optimal screening scenario yielded an ICER of 27.6 k€/LYG for men and 21.1 k€/LYG for women. The mortality reduction of lung cancer was 15.9% for men and 10.6% for women. Conclusions: Lung cancer LDCT screening is cost-effective in a high-risk population. The optimal screening scenario is dependent on sex.


INTRODUCTION
Lung cancer is the most commonly diagnosed cancer and the leading cause of cancer death in the Netherlands 1 .The life expectancy is estimated to be 13 years shorter for heavy smokers than for non-smokers 2 .Screening for lung cancer by low-dose computed tomography (LDCT) of the chest has been shown to prevent premature death by detection of developing cancers at an early stage 3 .The National Lung Screening Trial (NLST) found that three annual screenings with LDCT in (ex-)smokers aged 55 to 74 years reduced lung cancer mortality by 20% six years after baseline compared with three annual screenings with chest radiography 4 .The Dutch-Belgian Randomized Lung Cancer Screening Trial (NELSON) confirmed the benefits of LDCT screening for lung cancer, showing that lung-cancer mortality reduced by 24% for men and 33% for women in the LDCT screening group as compared with the no-screening group at 10 years of follow-up 5 .The European position statement on lung cancer screening recommended that Europe should prepare for the implementation of LDCT screening 3 .However, there is still debate on the optimal screening strategy 3 .Given that the Multicentric Italian Lung Detection trial found no difference in mortality comparing annual and biennial screening intervals at 5-year follow-up 6 , the benefits and harms of biennial screening need to be further investigated, and the optimal screening strategy requires further investigation.
A model-based economic evaluation of different lung cancer screening strategies could provide a reference for policy-makers to facilitate selecting an optimal strategy for lung cancer screening.Previous modeling studies that estimated the cost-effectiveness in Europe concluded that lung cancer screening can be cost-effective [7][8][9] .However, these studies yielded inconclusive results on the average cost-effectiveness ratio (ACER) relative to no screening, ranging from 16.8 to 48.4 k€ per life year gained (LYG) [7][8][9] .In addition, these studies were limited to studying only a single (annual) screening interval or they excluded possible harms from screening (such as false positives and radiation risk) [7][8][9] , or did not consider the cost of immunotherapy 8,9 .The aim of this study was therefore to assess the cost-effectiveness of various LDCT lung cancer screening strategies in a high-risk population, overcoming the limitations of previous studies.

Micro-simulation model
The micro-simulation model SiMRiSc was used and adapted for the purpose of lung cancer LDCT screening.This model has previously successfully been used to investigate the cost-effectiveness of breast cancer screening programs.The structure of the model and its underlying assumptions have been extensively described 10,11 .The basic principle of the model was that lung cancers were detected at an earlier stage when screening was implemented compared with no screening.Consequently, participants with lung cancer have a longer survival owing to the smaller tumour size and lower probability of positive lymph nodes and metastasis at detection.Thus, the life expectancy of the population in a screening setting is higher than that in a no-screening setting.The microsimulation model simulates the life history of each individual in the considered population from 20 years old until death in the presence and in the absence of LDCT screening.Several modules are incorporated in the model.The model allows for the simulation of various screening intervals.Remarkably, the sensitivity of LDCT was a function of tumor size instead of a fixed value independent of tumour size.The sensitivity was 0% for tumours of size less than 3 mm, 100% for tumours of size larger than 5 mm and a continuous function for tumours of size between 3 and 5 mm.Furthermore, the model included a module for radiation-induced tumour risk based on the model in the BEIR VII report 12 .A detailed description of the model is presented in Supplementary.

Simulated population
The model simulated two cohorts of 100,000 male and 100,000 female heavy smokers in the Netherlands from 20 years old until death.All simulations were repeated 10 times to assure that the standard error of the simulated outcomes was always less than 5% of point estimate.The average value of the 10 simulated results was derived and presented.A heavy smoker was defined as a current smoker who smokes at least 20 cigarettes per day according to the Dutch Central Bureau of Statistics 13 .

Parameters of the model
In the simulation, every individual was assumed to die at a predetermined natural death age, which was sampled from the cumulative mortality distribution 14 (Table S1).Lifetime risk for developing lung cancer and the mean age (and spread) at the time of lung cancer diagnosis was derived from the estimated lung cancer incidence for male and female heavy smokers.The incidence of lung cancer among heavy smokers was based on the lung cancer incidence in the general population 15 and the population attributable fraction for lung cancer due to tobacco smoking 16 (Table S2).In the 'tumor growth module', the volume doubling time of lung cancers was based on the publication of Henschke et al 17 .The self-detection size in the 'self-detection module' was based on the article of Rami-Porta et al 18 .The lung cancer survival parameters were derived from the literature on survival by stage of lung cancer (Table S3 and Table S4) 19 .The function that described the sensitivity of LDCT as a function of tumour diameter 20 and the specificity of LDCT was based on the published data 21 .The 'tumour induction module' consisted of the average dose per LDCT scan 22 and the risk of lung cancer induction from ionising radiation 12 .We used a health insurance perspective.Costs related to screening, diagnosis and treatment of lung cancer were considered and valued in euro [23][24][25] .Details of cost are presented in Supplementary.Discounting of 4% for costs and 1.5% for health effects (LYG) was applied according to Dutch guidelines 26 .To allow for international comparison, we also applied a discount rate of 3% for both costs and effects 27 .Values of all input parameters were independently taken from the literature and are summarised in Table 1.

Lung cancer survival
Table S3 and Table S4, Figure S1 19

Validation of the model
The population used for the validation of the model consisted of Dutch heavy smokers aged 50 to 75 years, similar to the population of the NELSON study28.The model was validated by comparing the simulated outcomes (number of screen-detected lung cancers and interval lung cancers per 1,000 screened individuals, and size distribution of screen-detected lung cancers) with the observed data from the first and second screening rounds of the NELSON lung cancer screening trial28, as shown in Table S5 and Table S6.

Screening scenarios
The evaluated screening scenarios combined different key characteristics of LDCT screening strategies: screening interval and start and stop age of screening.Annual and biennial screening intervals were considered.The considered values for the screening start age were 50, 55 and 60 years, and for the screening stop age were 75, 80 and 85 years, which covers all of the current recommendations regarding screening age.Overall, 18 scenarios were modelled for men and women separately.Perfect attendance of screening was assumed for all the base-case scenarios.

Outcomes and cost-effectiveness
The primary outcomes of the model assessed for each scenario were: ACER, incremental cost-effectiveness ratio (ICER) and lung cancer mortality reduction.The ACER was estimated as the ratio of the difference in costs to the difference in health effects of the investigated screening scenario compared with no screening.The secondary outcomes were LYG, number of lung cancer deaths averted, interval lung cancers, false positives, radiation-induced lung cancers and additional costs relative to no screening.The LYG was the difference in death age of the simulated population between a screening and no-screening setting.
The efficient scenarios were selected based on the cost per LYG and per averted lung cancer death in male and female heavy smokers.Scenarios that were more costly and less effective (fewer LYG or less lung cancer deaths prevented) than other scenarios or a combination of other scenarios were ruled out.The remaining screening scenarios were considered efficient and constituted the efficient frontier.For each efficient screening scenario, the ICER was estimated as the ratio of incremental costs to incremental health effects (LYG or averted lung cancer deaths) of a screening strategy relative to the previous efficient scenario.The Dutch National Health Care Institute uses a threshold of 80 k€ per quality-adjusted life year (QALY) gained for high-burden diseases 29 .Given that the cost per LYG is usually lower than the cost per QALY gained 30 and the mean utility score was 0.74 for lung cancer survivors based on a Dutch investigation 31 , a conservative estimation of the cost-effectiveness threshold of 60 k€/LYG was used in this study.The scenario with the highest ICER below the threshold was considered optimal.

Sensitivity analysis
One-way sensitivity analyses were carried out to explore parameter uncertainty of the most cost-effective scenario at the assumed threshold.We varied the baseline values of the input parameters by an increase or decrease of two standard deviations for the base case analysis.Cost values were increased or decreased by 50% of the value of the base case analysis.Imperfect attendance was evaluated by assuming 50% attendance rates.The values of the input parameters in the sensitivity analyses are presented in Table S7.Tornado plots were constructed to visualise the impact of parameter uncertainty on the ACER.

RESULTS
The LYG across all the screening scenarios compared with no screening ranged from 4991 to 8641 for men and from 4854 to 8741 for women (Table 2).The ACERs ranged from 17.7 to 32.4 k€/LYG for men and from 17.8 to 32.1 k€/LYG for women compared with no screening (Table S8).The lung cancer mortality reduction ranged from 9.3% to 16.8% for men and from 7.8% to 13.7% for women (Table 2).The scenario representing the screening scenario of the NELSON study (A-50-75) had a cost of 31.4 k€/LYG for men and 30.9 k€/LYG for women, with 14.9% and 12.4% mortality reduction respectively, compared with no screening.Of the evaluated scenarios, six were judged to be efficient for men and seven were efficient for women based on the cost per LYG (Figure 1).The efficient frontier consisted of a mix of annual and biennial scenarios.The estimated ICER ranged from 17.7 to 188.0 k€/LYG for men and from 17.8 to 191.1 k€/LYG for women compared with its previous efficient scenario (Table 3).The outcomes of all scenarios in which a discount rate of 3% for both costs and effects was applied are presented in Table S9.
Assuming a cost-effectiveness threshold of 60 k€/LYG as acceptable for the Dutch healthcare system, the optimal screening strategy was annual screening from 55 to 80 years old (A-55-80) for male heavy smokers, yielding an ICER of 51.6 k€/LYG compared with its previous efficient scenario.The optimal screening strategy for female heavy smokers was biennial screening from 50 to 80 years old (B-50-80), with an ICER of 45.8 k€/LYG compared with its previous efficient scenario.Compared with no screening, the optimal screening scenario yielded a cost of 27.6 k€/LYG and 15.9% mortality reduction of lung cancer for men, and a cost of 21.1 k€/LYG and 10.6% mortality reduction of lung cancer for women (Table 3).

Sensitivity analysis
The most influential factors on the ACER were lifetime risk of lung cancer and screening cost.An increase in the lifetime risk by 2 standard deviations in the optimal screening scenarios resulted in a 33% and 28% decrease of the base ACER, for male and female heavy smokers, respectively.The ACER changed by more than 40% after a 50% variation of screening cost.Detailed results of sensitivity analyses are available in Figure S2 and Figure S3.

DISCUSSION
A previously published simulation model was applied and validated for the purpose of the analysis presented here.The model reproduced the observed data of the first and second screening rounds of the NELSON study to a high accuracy.The simulations indicated that lung cancer screening with LDCT is cost-effective in a high-risk population.At a cost-effectiveness threshold of 60 k€/LYG, the most promising scenario was annual screening from the age of 55 to 80 years for male heavy smokers and biennial screening from the age of 50 to 80 years for female heavy smokers, yielding a cost of 27.6 k€/LYG and 21.1 k€/LYG relative to no screening, respectively.
The model-based cost-effectiveness analyses on lung cancer screening with LDCT in other European countries applied a discount rate of 3% for both costs and effects.When the same discounting rate was applied in our study, the ACER ranged from 23.3 to 51.9 k€/LYG for men and 23.3 to 50.2 k€/LYG for women.Our results are comparable with the results of model-based cost-effectiveness analyses on lung cancer screening with LDCT conducted in Switzerland but higher than those of analyses conducted in Germany.The cost-effectiveness analysis conducted in Switzerland indicated that annual or biennial screening for the high-risk population with various smoking history may be cost-effective at a cost of 25.6 to 48.4 k€/LYG compared with no screening 7 .The simulation analysis in conducted in Germany showed that annual screening may be cost-effective at a cost of 19.3 k€/LYG and 30.3 k€/QALY in heavy smokers aged 55 to 75 years compared with standard clinical care 9 .The lower treatment cost applied in that study could contribute to the difference in ACER in the present study.
Several studies evaluated the cost-effectiveness of lung cancer screening with LDCT outside Europe, especially in the United States, Canada and Australia.The US Preventive Services Task Force recommended annual screening for a population aged 55 to 80 years with 30 or more pack-years after balancing the benefits and harms of 576 scenarios 32 .
A cost-effectiveness analysis in a Canadian population indicated that annual screening for individuals aged 55 to 80 years required 43.0 to 50.0 k$/LYG compared with no screening 33 .However, in a simulation analysis of an Australian population using NLST criteria, lung cancer screening with LDCT was not likely to be cost-effective owing to the high cost of 138.0 kAU$/LYG (approximately 104.7 k$/LYG) 34 .
Our analysis indicated different optimal scenarios for men and women and lung cancer screening was more cost-effective in women than in men (an ACER of 21.1 k€/ LYG for women vs 27.6 k€/LYG for men), which is in line with the previous findings 35,36 .
The optimal screening scenario for women indicated that screening should be started at an earlier age than for men and a biennial screening should be adopted.This is in line with the previous studies, showing that women diagnosed with lung cancer were significantly younger than men 37,38 .In addition, given that women are more vulnerable to radiation-induced tumours than men 22 , biennial screening is warranted.
Remarkably, the scenario of the NELSON study (A-50-75) was not included in the efficient frontiers.However, this scenario was very close to the efficient frontier based on the cost per LYG, which is consistent with a previous study conducted in Germany 8 .
The selection criteria of the NELSON population were made based on a statistical power analysis for mortality reduction 39 , and cost-effectiveness was not the main aim, which might explain the efficiency difference compared with other scenarios.
Our model has some limitations.First, not all the values of the input parameters were derived from the Dutch setting, which might have an impact on the effectiveness of screening.However, we incorporated this uncertainty by applying (normal and lognormal) distributions on the input parameters and evaluated the impact using a sensitivity analysis.In addition, because the external validation of the model to the observed data from a lung cancer screening program in the Dutch population showed good results, the evaluation of cost-effectiveness is considered valid.Second, we could not evaluate the ICER of lung cancer screening in populations with different smoking histories in terms of pack-years due to a lack of data.However, we evaluated the ICER by varying the lung cancer lifetime risk in the sensitivity analyses, indicating that lung cancer screening in a more high-risk population will be more cost-effective, as expected.Third, a single-treatment cost was used for stage I to stage III tumours owing to the lack of data for each histological stage, which might lead to an overestimation of the cost per LYG because of the shift to an earlier stage (lower treatment cost) from screening.Fourth, our analysis focused only on quantitative outcomes.The quality of life of patients with lung cancer and the disutility associated with LDCT screening were not incorporated in our model.However, the NELSON study indicated that the impact of LDCT screening itself on the quality of life was negligible in the long term 40 .Fifth, assuming that all tumours are spherical is a limitation.In the NELSON study more than half of the malignant nodules were polygonal and irregular 41 .A spiculated, non-spherical growth might yield a smaller self-detection size owing to the larger probability of being symptomatic, which implies an overestimation of the cost-effectiveness.However, from the sensitivity analyses it follows that the influence of self-detection size on costeffectiveness is only minor.Therefore, we estimate that this limitation will not change the major outcomes of our simulations.
In conclusion, the results from a microsimulation model show that lung cancer screening with LDCT is cost-effective in a high-risk population.At a cost-effectiveness threshold of 60 k€/LYG, the optimal screening scenario for male heavy smokers is annual screening from 55 to 80 years old, yielding a cost of 27.6 k€/LYG and 15.9% mortality reduction of lung cancer relative to no screening.The optimal screening scenario for female heavy smokers is biennial screening from 50 to 80 years old, yielding a cost of 21.1 k€/LYG and 10.6% mortality reduction of lung cancer relative to no screening.

Model description
Several modules are incorporated in the model.

Cost
The cost of a LDCT scan (€ 172) was obtained from the Dutch Healthcare Authority 2 .The costs of diagnosis (€ 1,908 per patient) and treatment of lung cancer were extracted from a study on the costs of lung cancer in the Netherlands 3 , indexed to January 2020 4 .The treatment cost per patient was k€ 37.9 for stage I-III lung cancer and k€ 27.6 for stage IV lung cancer without including immunotherapy cost 3 .The amount of patients treated with immunotherapy and the related costs were estimated for this study based on experts' opinion.Approximately 58% of patients with stage IV lung cancer would be eligible for immunotherapy treatment according to the experts.The expected cost of pembrolizumab treatment course, which dominates the market in the Netherlands for lung cancer treatment by immune checkpoint inhibitors, was estimated as k€ 40 -60 per patient per year in 2019 5 .We used the average value (k€ 50) as an estimation of immunotherapy cost for this study.

Lung cancer incidence in heavy smokers
The lung cancer incidence module was used in the model to assign tumours to participants in the population at a certain age, according to the lung cancer incidence probability.In the model, a normal distributed lung cancer incidence as a function of age was assumed.The probability density function was a Gaussian function normalized to the lifetime risk of developing lung cancer and had the form of: Chapter 7 Cost-effectiveness of lung cancer screening in heavy smokers

Lung cancer incidence in heavy smokers
The lung cancer incidence module was used in the model to assign tumours to participants in the population at a certain age, according to the lung cancer incidence probability.In the model, a normal distributed lung cancer incidence as a function of age was assumed.The probability density function was a Gaussian function normalized to the lifetime risk of developing lung cancer and had the form of: Where p is the risk of acquiring lung cancer during that year, α is the participant's age, f is the lifetime risk, µ is the mean age and σ is the spread.
The number of new lung cancer cases by age in men and women was obtained from the Dutch cancer registry 7 .The latest data on average number of lung cancer cases in 5 years from 2012-2016 were used.We used the population attributable fraction of lung cancer for tobacco smoking 8 to estimate the number of lung cancer cases attributable to smoking and to non-smoking factors.The number of smokers in 2010 were obtained from Dutch statistics 9 (assuming a 10 year latency between smoking and lung cancer).By using the risk estimate of heavy smoking and lung cancer 10 , the lung cancer incidence was estimated in male and female heavy smokers.The estimated lung cancer incidence used in the model is presented in Table S2.
Table S2 Where p is the risk of acquiring lung cancer during that year, α is the participant's age, f is the lifetime risk, µ is the mean age and σ is the spread.
The number of new lung cancer cases by age in men and women was obtained from the Dutch cancer registry 7 .The latest data on average number of lung cancer cases in 5 years from 2012-2016 were used.We used the population attributable fraction of lung cancer for tobacco smoking 8 to estimate the number of lung cancer cases attributable to smoking and to non-smoking factors.The number of smokers in 2010 were obtained from Dutch statistics 9 (assuming a 10 year latency between smoking and lung cancer).By using the risk estimate of heavy smoking and lung cancer 10 , the lung cancer incidence was estimated in male and female heavy smokers.The estimated lung cancer incidence used in the model is presented in Table S2.

Lung cancer survival
The lung cancer survival in the model was a function of tumor size and years after diagnosis with considering the presence of lymph nodes and metastasis.The survival of lung cancer was reported by IASLC in the 8 th of the TNM classification for lung cancer.It has been shown that the correlation between tumor size and survival can be very well described by 12 : where F is the survival chance after diagnosis and D is the tumor diameter in mm.Q and Z are constants and are found through curve fitting.If applied to the IASLC data, fits of the equation can be made to obtain Q and Z parameters at different times after diagnosis.Q and Z can then be described into a continuous function of time after tumor diagnosis by fitting a power function for Q and a logarithmic function for Z: .(/) = 0/ + 1(/) = 2 ln / + 6 where a, b, c and d are constants and t is given in years.
For each of the four NM-stages in Table S3, the survival of lung cancer was modeled as a function of tumor size.The fitting to survival data yielded the following values for the constants a to d listed in Table S4.The survival chance was visualized in Figure S1.
where F is the survival chance after diagnosis and D is the tumor diameter in mm.Q and Z are constants and are found through curve fitting.If applied to the IASLC data, fits of the equation can be made to obtain Q and Z parameters at different times after diagnosis.Q and Z can then be described into a continuous function of time after tumor diagnosis by fitting a power function for Q and a logarithmic function for Z:

Lung cancer survival
The lung cancer survival in the model was a function of tumor size and years after diagnosis with considering the presence of lymph nodes and metastasis.The survival of lung cancer was reported by IASLC in the 8 th of the TNM classification for lung cancer.It has been shown that the correlation between tumor size and survival can be very well described by 12 : where F is the survival chance after diagnosis and D is the tumor diameter in mm.Q and Z are constants and are found through curve fitting.If applied to the IASLC data, fits of the equation can be made to obtain Q and Z parameters at different times after diagnosis.Q and Z can then be described into a continuous function of time after tumor diagnosis by fitting a power function for Q and a logarithmic function for Z: .(/) = 0/ + 1(/) = 2 ln / + 6 where a, b, c and d are constants and t is given in years.
For each of the four NM-stages in Table S3, the survival of lung cancer was modeled as a function of tumor size.The fitting to survival data yielded the following values for the constants a to d listed in Table S4.The survival chance was visualized in Figure S1.
where a, b, c and d are constants and t is given in years.
For each of the four NM-stages in Table S3, the survival of lung cancer was modeled as a function of tumor size.The fitting to survival data yielded the following values for the constants a to d listed in Table S4.The survival chance was visualized in Figure S1.
Table S4.The value of input parameters for lung cancer survival in the model * estimated from IASLC publication (T descriptor) 13 .# estimated from IASLC publication (N descriptor) 14 .

Parameter
$ estimated from IASLC publication (TNM classification) 1 .The survival of metastatic lung cancer was a function of time after diagnosis and was chosen to be independent of tumor size (c = d = 0).

Validation
The number of simulated screen-detected lung cancers and interval lung cancers was compared to the observed incidence separately for the first and second screening rounds.The simulated results were considered valid when the simulated point estimates fell within the empirical confidence interval.
In the validation of the model, more lung cancers were detected in the first round and less lung cancers were detected in the second round compared to the observed NELSON data.This could be explained by the fact that in the NELSON study nodules with a diameter less than 4.6 mm were considered to be negative, while in the simulation model of this study, cancers larger than 3 mm could be detected by LDCT.Therefore the malignant nodules with a diameter between 3 mm and 4.6 mm were counted in the second round in the NELSON study but in the first round in the model of this study.This might lead to a slight underestimation of the ACER because lung cancers with a smaller size usually have a better survival and lower cost.The validation results are summarized in Table S5 and Table S6.

Cost-effectiveness with the Dutch discount rate
The data used for the cost-effectiveness analysis of lung cancer screening with the Dutch discount rate are provided.The following tables of the original research article are based on these data, including "Table 2", "Table 3" and "Table S8" 1 .The data of 18 scenarios for men and women are presented in the excel file "Output_Main results_ Men.xlsx" and "Output_Main results_Women.xlsx",respectively.The name of each worksheet tab indicates the performed scenario.For example, "A-50-75" signifies annual screening from 50 to 75 years old, and "B-50-75" signifies biennial screening from 50 to 75 years old.In each scenario, the following data are included for screening and no screening: the size of the simulated population, the number of lung cancers in the screened population, the number of participants with a screen-detected lung cancer, the number of participants that died from lung cancer, the number of life years, the number of interval lung cancers, the number of false positive results and the total costs.In addition, a summary table of the output of each scenario is provided, which includes the mortality reduction compared to no screening, the average cost-effectiveness ratio (ACER) per averted lung cancer death, the ACER per life years gained (LYG) and the number of radiation-induced lung cancers.The cost and the discounted LYG are the result after discounting by 4% for cost and 1.5% for LYG.

Cost-effectiveness with the international discount rate
The data used for the cost-effectiveness analysis of lung cancer screening with the international discount rate are provided.A discount rate of 3% for both cost and LYG was applied.The "Table S9" of the original research article is based on these data 1 .The data of 18 scenarios for men and women are presented in the excel file "Output_Main results_Men_International discount.xlsx","Output_Main results_Women_International discount.xlsx",respectively.For the description of the content of the two excel files we refer to "1.2.1 Cost-effectiveness with the Dutch discount rate".

Sensitivity analysis
The data used for the sensitivity analysis of the original research article are provided.The scenarios for the sensitivity analysis were described in the original research article 1 .The "Table S9", "Figure S2" and "Figure S3" of the original research article are based on these data 1 .The data for men and women are presented in the excel files "Output_Sensitivity analysis_Men_A-55-80.xlsx" and "Output_Sensitivity analysis_Women_B-50-80.xlsx", respectively.The name of each worksheet tab indicates the variable that was varied in the sensitivity analysis.For the content of the worksheet tabs we refer to "1.2.1 Costeffectiveness with the Dutch discount rate".

Model validation
The data used for the model validation of the research article are provided in the excel file "Output_Validation".The data of the number of screen-detected lung cancers, the number of interval lung cancers and the size distribution of the screen-detected tumours in the first and second screening rounds are presented in 3 worksheet tabs with the corresponding names.The excel file contains the data for men and women separately, as well as the combined data.The "Table S5" and "Table S6" of the original research article are based on these data 1 .

EXPERIMENTAL DESIGN, MATERIALS AND METHODS
The related research article was designed to evaluate the cost-effectiveness of lung cancer screening with LDCT in a high-risk population using a micro-simulation model 1 .
The design, materials and methods are clearly described in the research article.Briefly, the micro-simulation model SiMRiSc was used.This model has previously successfully been used to evaluate the cost-effectiveness of breast cancer screening programs.It was adapted for the purpose of lung cancer LDCT screening.The model was validated by comparing the simulated outcomes to the observed data from the Dutch-Belgian Randomized Lung Cancer Screening (NELSON) trial.The evaluated screening scenarios combined different key characteristics of LDCT screening strategies: screening interval, and start and stop age of screening.The evaluated outcomes included the average cost-effectiveness ratio, incremental cost-effectiveness ratio, lung cancer mortality reduction, life years gained, number of lung cancer deaths averted, interval lung cancers, false positives, radiation-induced lung cancers and additional costs relative to no screening.One-way sensitivity analyses were performed to explore the parameters uncertainty of the most cost-effective scenarios.The technical details of the model were described in the supplementary of the original research article 1 .

Figure 1 .
Figure 1.The cost-effectiveness in cost per life years gained (top) and cost per lung cancer death averted (bottom) of all evaluated scenarios for men (left) and women (right).Annual screening intervals are shown in red, and biennial screening intervals in blue.The scenarios that constitute an efficient frontier are labeled (A-annual, B-biennial -screening start age -screening stop age)

Figure S1 .
Figure S1.The survival chance in a color scale between 0 (red) and 1 (blue) as a function of time since diagnosis (in years) and tumor diameter (in mm) at the moment of diagnosis.(A): for tumors of any size with negative lymph nodes and metastasis (T 1-4 N 0 M 0 ); (B) for tumors of any size with positive lymph nodes and negative metastasis (T 1-4 N 1-3 M 0 ); (C) for metastatic tumors (M 1a-b ) of any size (T 1-4 N 0-3 M 1a-b ); (D) for metastatic tumors (M 1c ) of any size (T 1-4 N 0-3 M 1c ).

Table 1 .
Input parameters of the SiMRiSc model

Table 2 .
Outcomes of the scenarios per 100,000 male and per 100,000 female heavy smokers Scenario a LC = lung cancer, LYG = life year gained.The lifetime number of lung cancer deaths without screening was estimated at 7,308 per 100,000 male heavy smokers and 7,679 per 100,000 female heavy smokers.Costs were discounted by 4% annually, and LYG were discounted by 1.5% annually.ascreening interval (A-annual, B-biennial) -screening start age -screening start age.

Table 3 .
Cost-effectiveness of screening scenarios on the efficient frontier for 100,000 male and 100,000 female heavy smokers Scenario

a Discounted additional cost vs no screening, (in million €) ACER vs no screening (in k€/LYG) ACER vs no screening (in k€/one averted lung cancer death) ICER vs the previous efficient scenario (in k€/LYG or k€/one lung cancer death averted) Men Women Men Women Men Women Men Women
ACER = average cost-effectiveness ratio; ICER = incremental cost-effectiveness ratio; k€ = 1000 euro; NE: not efficient.LYG = life years gained.Costs were discounted by 4% annually, and LYG was discounted by 1.5% annually.ascreening interval (A-annual, B-biennial) -screening start age -screening stop age b optimal strategy for men at the cost-effectiveness threshold of 60 k€/LYG.c optimal strategy for women at the cost-effectiveness threshold of 60 k€/LYG The 'life expectancy module' determined the natural death age of the individuals in the simulated population.The 'cancer incidence module' determined the cancer incidence as a function of age.The 'tumor growth module' was based on an exponential lung tumor growth as a function of time.All lung cancers are assumed to be spherical allowing calculation of tumor volume based on diameter.Tumor volume doubling times (VDT) were sampled randomly from a log-normal distribution for each individual with cancer.The 'self-detection tumor size module' determined the diameter of tumors detected without screening.For individuals with a lung cancer, a diameter of self-detection was sampled from a lognormal distribution.The 'tumor survival module' determined the survival chance as a function of the TNM stage at diagnosis1, where we used diameter as a proxy for T stage.Sensitivity and specificity of LDCT screening were included where sensitivity was a function of tumor size.Finally, the 'tumor induction module' determined the risk of radiation-induced lung cancers.

Table S1 .
All-cause cumulative mortality of heavy smokers in the Netherlands Source: Dutch statistics6.

.
Cumulative incidence of lung cancer in heavy smokers for men and women in the Netherlands

Table S2 .
Cumulative incidence of lung cancer in heavy smokers for men and women in the Netherlands

Table S3 .
Probability of positive lymph nodes and metastasis in each T categoryThe lung cancer survival in the model was a function of tumor size and years after diagnosis with considering the presence of lymph nodes and metastasis.The survival of lung cancer was reported by IASLC in the 8 th of the TNM classification for lung cancer.It has been shown that the correlation between tumor size and survival can be very well described by12 :

Probability of positive lymph nodes and metastasis by tumor sizeTable S3 .
Probability of positive lymph nodes and metastasis in each T category

Probability of positive lymph nodes and metastasis by tumor sizeTable S3 .
Probability of positive lymph nodes and metastasis in each T category

Table S5 .
Validation of screen-detected lung cancers and interval lung cancers from the simulation

Table S6 .
Validation of size distribution of screen-detected lung cancers from the simulation

Table S7 .
The values of the input parameters in the sensitivity analysis The low and high values were derived from a decrease or increase of two standard deviations of the value in the base case analysis.Cost values were decreased or increased by 50% of the value in the base case analysis.Attendance rate was decreased by 50% of the value in the base case analysis. *

Table S8 .
The cost per life years gained and cost per lung cancer death averted of all evaluated scenarios *screening interval (A-annual, B-biennial) -start screening age -stop screening age.ACER, average cost-effectiveness ratio.k€: 1000 euro.Costs were discounted by 4% annually, and LYG were discounted by 1.5% annually.

Table S9 .
Outcomes of the scenarios per 100,000 male and per 100,000 female heavy smokers (a discount rate of 3% for both costs and LYG was applied)