Model design and structure
Our economic evaluation followed a cohort of former and current smokers aged between 50 and 74 years with a smoking history of at least 25 pack-years and without a history of LC. The time horizon of the analysis was 30 years. The concept of the economic model assumes that LC cases develop at a given rate in the simulated cohort, and may be discovered by different screening strategies, at a different rate depending on the screening strategy (that is, some cases may remain detectable, but undiscovered). By base case, the evaluation compared two scenarios: one that assumed annual or biannual screening for LC with LDCT against another that assumed no organised screening using LDCT for LC (meaning that patients can still be diagnosed with LC by showing symptoms and being referred to undergo chest X-rays). LC screening with LDCT was terminated in the model once the mean age of the cohort reached 75 years.
The estimated size of the target population – 50–74 years of age, current or former heavy smokers – was set at 644 400 people, an estimate based on census data and a national public health survey from Hungary. LDCT screening uptake was assumed to be 50%, whereas the probability of responding to screening invitation in the “no LDCT screening” strategy was assumed to be 42,80%, an estimate based on participation rate in LC screening programs between 2018 and 2020.
As our economic evaluation was a cost-utility analysis, we aggregated our estimates on outcomes by combining utility values assigned to health states with the time spent in the given health state to calculate QALYs, while costs were expressed in HUF over the time horizon of 30 years. Using a cost-utility analysis framework enables to estimate the economic impact of introducing LDCT screening for LC through the effect on multiple outcomes. Outcomes and costs were both discounted by an annual rate of 3.7%. Estimates on the survival of the cohort (such as changes in life expectancy, disease-specific mortality) were obtained from the mortality.org database and a published analyses of local retrospective data on lung cancer patients (6).
To match the characteristics of disease progression as well as to ensure comparability with earlier analysis, we designed a Markov cohort state-transition model with a cycle length of 28 days to conduct a cost-effectiveness analysis that assigns estimates of utility values and costs to health states. The model structure consists of the following health states: alive (no diagnosis of cancer), localised SCLC, advanced SCLC, localised NSCLC and advanced NSCLC, as well as death as a terminal, absorbing health state (final health state of patients, as they can only enter this health state). Localised disease was defined as being Tis/T1-T3A, N0, M0 according to the TNM system. Advanced disease was defined as any other stage that does not match the definition of localised disease.
The model also contains a health state to capture false positive and indeterminate cases of LC in the LDCT arm. The impact of LDCT and discovery of LC on costs and outcomes are both captured by the respective tunnel health states (that is, patients can only spend one cycle, and cannot remain in this health state and moves to the next state). Microsoft Excel for Microsoft 365 was used to program the economic model (Redmond, WA).
The simulated cohort enters the simulation in the alive (and cancer-free) health state and moves toward the discovery of LC under the “no organised screening” strategy, followed by health states describing the type of LC. When simulating the “screening for LC with LDCT” strategy, patients in the simulated cohort may also enter the health states describing the type of LC via LDCT screening (that is, their disease is discovered by LDCT in an asymptomatic stage), although they can also be classified as false positives or indeterminate cases and return to the alive (no diagnosis of cancer) health state. Patients diagnosed with LC may stay alive or die of the disease at different rates depending on the particular health state. Death from other causes may occur in any health state of the model, according to the initial distribution of the cohort by age and gender. Figure 1. provides an overview of the structure on the economic model used for this evaluation.
Model Inputs
The clinical input parameters of the economic evaluation were sourced from the HUNCHEST-II study for the effectiveness of screening, while data on the epidemiology and clinical features of LC (incidence, diagnosis, histology-specific survival), as well as utility values for each health state were collected from secondary data sources identified by targeted literature review. Estimates of resource use and unit costs were retrieved from expert interviews and from the analysis of cost databases and retrospective claims data of the National Health Insurance Fund Manager, respectively. Input values for parameters related to the local policy context of economic evaluations such as the discount rate, time horizon of the analysis, and the cost-effectiveness threshold were adapted from the local guidance on economic evaluations (7). Components of direct costs were diagnostics, treatment (including surgery, medication, radiation therapy respective follow-up), and hospice costs funded by the public payer. Table 3 provides a summary of the input values used in the analysis.
Scenarios and sensitivity analyses
To observe the structural- and parameter uncertainties in the economic evaluation, we conducted both deterministic one-way sensitivity analyses and probabilistic sensitivity analyses, respectively. For the deterministic one-way sensitivity analyses, we evaluated the impact of each included variable on the cost-effectiveness results over a range of possible input values while holding all other variables constant and visualised the results on a tornado chart. Probabilistic sensitivity analyses using a second-order Monte-Carlo simulation for 1 000 trials to generate random sets of input parameter values were carried out to assess the impact of parameter uncertainty.