A cost-effectiveness analysis of lung cancer screening with low-dose computed tomography and a polygenic risk score

China, with 1/3 proportion of smoking population across the world has substantial cancer burden while lung cancer remains the leading cause of cancer-related death. The effectiveness for mortality reduction of lung cancer screening programs has been well confirmed by several trials (e.g. National Lung Screening Trail) and the challenge for lung cancer screening now seemed to be the high false-positive rate of Low-Dose Computed Tomography (LDCT). To make the existing cancer screening programs more efficient targeting, polygenic risk scores (PRSs) are introduced. PRS have the potential to identify individuals at risk of different type of cancers, optimizing treatment, and predicting survival outcomes. We searched PubMed, EMBASE, and Web of Science between January 1, 2000, and July 30, 2023, with no language restrictions, using the terms “China” or “Chinese”, “lung cancer”, “polygenic risk score” or “PRS” or “genetic test” and “cost-effectiveness”, to identify published economic evaluations on PRS-based strategy for lung cancer screening in China. We found no previous studies describing the cost-effectiveness of PRS-based lung screening in China. Only one previous study evaluated the effect of PRS-based screening based on modelling using UK metrics.

The comparative cost-effectiveness analysis used a Markov state-transition model to assess the potential effect and costs of the screening strategies incorporating PRS or not. We found that the screening strategy incorporating PRS was estimated to be cost-effective compared with non-screening, with an ICUR of CNY 156,691.93 to CNY 221,741.84 (initial start age range across 50-74 years) per QALY gained. The strategy that screened using LDCT alone from 70-74 years annually could obtain an ICER of CNY 80,880.85 per QALY gained, which was the most cost-effective strategy. The introduction of PRS as an extra eligible criteria was associated with making strategies cost-saving but also lose the capability of gaining more LYs compared with LDCT screening alone.

Our findings suggest that lung cancer screening programs incorporating PRS of existing performance would hardly be cost-effective using the willingness-to-pay threshold of 1 time GDP per capita, and the optimal screening strategy for lung cancer still remains to be LDCT screening alone for now, suggesting that we should be more conservative in considering LDCT screening with PRS for lung cancer.

In this modelling study, the Markov model on lung cancer screening that developed by our previous work was used and adapted for the purpose of assessing the potential impact of LDCT screening with and without a PRS from a societal perspective. Important assumptions and the overall structure of the model have been thoroughly described before and in supplementary material [19, 20]. Per China guideline for the screening and early detection of lung cancer (2021,Beijing) [21] recommended, 3 hypothetical cohorts of 10,000 current and former smokers aged 50–74 years old were simulated until death or age 79 years (mean life expectancy in China),named non-screening cohort, LDCT screening cohort and LDCT&PRS screening cohort. Unlike the normal LDCT screening modality, individuals who enter the cohort of LDCT&PRS were assumed to have received PRS assessment and were included to the top 5% high risk based on PRS. All the simulated individuals from two screening cohorts undergo annual screening until the simulation ended. We further superimposed screening and diagnostic follow-up interventions onto the natural history model for lung cancer and obtained population-level outcomes. Data sources, main outcomes, and the full research design are shown in Fig. 1. The model was run with a cycle length of 1 year and a discount rate of 5% was applied to both costs and effectiveness. The model construction and all the simulations were conducted using Treeage Pro, version 2021 (Treeage Software). The study was performed according to the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) and was approved by the ethics committee of the Jiangsu Province Hospital of Chinese Medicine; informed consent was not applicable because this was a modeling study.

For this modelling analysis, we used China age-stratified data for lung cancer incidence and integrated the effect of smoking rate to model incidence rates for the initial probability of lung cancer for those in the cohorts of non-screening or LDCT screening alone [22‐24]. According to the 3 PRS-defined quantiles (ie, the top 5%,5%-95%, and the bottom 5%), we then calculated the relative risk(RR) of the PRS for lung cancer based on the published estimates of the standardized rates of lung cancer events of the three groups of heavy smokers with diverse genetic risk in China Kadoorie Biobank (CKB) cohort [17]. The proportion of clinical stage for lung cancer detected by LDCT was derived from screening results of the Wenling Lung Cancer Screening Program, which was initiated in 2018 to conduct annual LDCT screening for local populations at high risk of lung cancer with follow-up for 3 years. A total of 20130 asymptomatic individuals were screened by the program by the end of December, 2022, and 287 patients were diagnosed with lung cancer; details of the proportions by cancer stage are presented in Table 1. Annual screening followed the same screening protocol as in the Cancer Screening Program in Urban China, which determined positive findings by morphologic features and the size of the nodule [25]. As for those diagnosed by normal clinical pathways, the probability that diagnosed clinically is detailed by stage in Table 1 based on a hospital-based multi-center lung cancer retrospective clinical epidemiological survey in China(LuCCRES) [26]. The probability of health to all-cause death was estimated as all-cause mortality for smokers by age [24, 27]. The probability of lung cancer-specific death was derived from a study by Zhang et al. [28] and was adjusted for smoking status [29, 30]. The probability that a cancerous state progressed to a more advanced state or to a maintenance state is detailed by cancer stage in Table 1 according to Haaf’s work [31]. The sensitivity and specificity for LDCT were based on a study that enrolled 9,522 person-times over five screening rounds from 2014 to 2018 in Sichuan, China [32]. Perfect attendance to screening was assumed for base-case analysis and the uptake rates by different screening modality were incorporated in scenario analysis [33].

A total estimated cost for the lung cancer screening program consisted of two parts, the direct screening cost and the indirect screening cost. Screening related cost data were surveyed by the work team of a local lung cancer screening program for the expenses for public advertising, screening invitation management, staff salary and depreciation of screening machinery. For the indirect screening cost, we conducted a survey to estimate the expenses for transportation and wage for missed work for the participants.

We estimated the treatment cost of lung cancer by stage based on the database of local medical insurance bureau, which including 4,947 patients and 107,248 relevant records. Given the potential diversity in treatment cost across the nation, we adapt the treatment cost by stage using published metrics form China Health Statistics Yearbook 2020 [38]. The cost of maintenance by stage was calculated using the standard follow-up process and the unit price of each test per the price list of medical services in public medical institutions. All the costs in this study are expressed in CNY and are discounted to the price level of 2022 at a discount rate of 5%.

For quality-of-life adjustment, we used the utility values for lung cancer state by stage based on a EQ-5D-3L survey from 2586 lung cancer patients in 8 provinces and 12 cities in China through the Cancer Screening Program in Urban China(CanSPUC). In addition to, we derived the utility value of CIS stage from a global systematic review by Sturza et al. [36]. The utility value for the maintenance state of each stage was derived from a domestic thesis in 2016 [37].

We compared 15 alternative strategies as shown in Table 2. The first 5 strategies involved non-screening for all the heavy smokers as blank control. The remaining 10 strategies were defined by combinations of risk stratification approaches (smoking pack-years or PRS) and initial screening age from 50 to 70 years by 5-year age bands. We describe these strategies in Table 2.

In this study, primary outcomes included life years (LYs), quality adjusted life years (QALYs), and costs of different strategies. Given the #0 Non-screening strategy as reference, a strategy was deemed cost-effective if the incremental cost-effectiveness ratio (ICER), namely the difference between the overall costs of the two strategies divided by the difference between the total QALYs gained, was lower than the cost-effectiveness threshold of 1–3 times Gross Domestic Production (GDP) per capita per QALY gained (CNY 85,698–257,094) [39].

The robustness of the outcomes to uncertainties in the parameter estimates was examined through a series of univariate sensitivity analyses. The cost of screening, treatment cost as well as maintenance cost and consumer price index (CPI) rate were set to vary by 30% compared to base case values. The discount rate was set to range from 0 to 8%. The RR of the PRS for lung cancer was set to range from 2.64 to 5.99. The sensitivity and specificity of LDCT test were set to range from (0.632, 0.648) to (0.948, 0.972). Furthermore, Probability sensitivity analysis (PSA) was also performed with 10,000 iterations to assess the joint uncertainties in the values of input parameters. Input parameters were randomly drawn from beta, lognormal or gamma distribution (see Table 1). As for the scenario analysis, we evaluated the health benefits and harms associated with a lung cancer screening program that incorporated the uptake rate of different screening modalities among Chinese high-risk population for lung cancer.

In the absence of screening, the total number of lung cancer death per 100,000 heavy smokers aged between 50–79 years were estimated to range from 4,434 to 10,586. The introduction of a screening program led to a decrease of lung cancer deaths, with the reduction rate of lung cancer death ranging from 0.31% to 15.80% across a diverse set of screening strategies. About 95% false-positive cases could be averted by incorporating PRS in the screening program in relative to LDCT screening alone. The LYs and QALYs across all the screening strategies compared with non-screening ranged from 60.26 to 134.93 and from 59.83 to 134.27, respectively. To be specific, screening strategies using PRS as extra eligible criteria obtained lower LY and QALY gained than LDCT screening alone (see Table 3). Compared to non-screening, the #1LDCT strategies cost between CNY 104,998.56 and CNY 176,565.66 per LY gained. The #2 PRS&LDCT strategies cost between CNY 191,110.06 and CNY 260,918.20 per LY gained. When adjusting to QALYs, the #1LDCT strategies would cost between CNY 808,80.85 and CNY 150,050.15 per QALY gained. The #2 PRS&LDCT strategies would cost between CNY 156,691.93 and CNY 221,741.84 per QALY gained. All showed an ICER below 3 times GDP per capita (CNY257,094) per QALY gained. Assuming a cost-effectiveness threshold of 1time GDP per capita (CNY 85,698) per QALY gained for the Chinese healthcare system, only annual LDCT screening with the start age of 65–74 and 70–74 years old were cost-effective, yielding an ICER of CNY 85,332.16 and CNY 80,880.85 per QALY gained compared with non-screening. Table 3 provides the outcomes of the model simulation.

Results of sensitivity analyses are shown in Fig. 2 and Fig. 3. The most influential factors on the ICER were specificity and sensitivity of LDCT, as well as discount rate. The results were robust to the changes of the important values from base-case analysis with no variation exceeding 3 times GDP per capita (CNY257,094) per QALY gained, but also generally exceeding 1 times GDP per capita (CNY85,698) (Fig. 3). Notably, the #1LDCT screening strategy compared with the #0 Non-screening strategy with a start age older than 55 years had better than 90% likelihood of being cost-effective when the willingness-to-pay threshold was 3 times GDP per capita (CNY257,094). Meanwhile, the probability of #2 PRS&LDCT screening strategy to be cost-effective ranged from 33.77%-79.68%, varying from different start age. While the per capita GDP (CNY 85,698) serves as the threshold for absolutely cost-effective, the acceptability at willingness-to-pay threshold ranged from 1.44% to 34.18% for #1LDCT screening strategy and from 0.26% to 2.54% for #2 PRS&LDCT screening strategy (Table 4).

The tornado diagram illustrates the change in the incremental cost-effectiveness ratio (ICER), which was defined as the cost of the PRS&LDCT screening strategy minus the cost of the LDCT screening divided by the difference of the quality-adjusted life-year of the two strategies when important input parameters were varied for both strategies (1 strategy at a time) by 10% ~ 30% higher or lower than their base-case values (shown in Sect. 2.5 Sensitivity analysis and scenario analysis). The vertical axis (dotted dark line) on the left shows the estimated ICER for the base-case analysis, and the vertical axis on the right showed the willingness-to-pay. The column with black color in the tornado diagram showed when the input parameters decrease, their impact for the results. Similarly, the column with grey color showed when the input parameters increase, their impact for the results. 

Abbreviations: LDCT, low-dose computed tomography; PRS, polygenic risk score; LC, lung cancer; CIS, carcinoma in situ; CPI, consumer price index.

The dashed circle is the 95% confidence interval, which indicates the robustness of the model operation. The dashed lines are displayed as the cost-effectiveness threshold of 1 times GDP per capita (CNY85,698) and 3 times GDP per capita (CNY257,094) per QALY gained, respectively. The dots above the dashed line are cost-effective.

Abbreviations: LDCT, low-dose computed tomography; PRS, polygenic risk score; WTP, willingness-to-pay threshold. In a previous study, a discrete choice experiment was used to create scenarios on several different possible modalities for the implementation of lung cancer screening in Chinese context [38]. The uptake rate varied from different screening modalities by mixed-logit model. The uptake rate of screening by blood test would be decreased by 0.08 compared with the baseline, i.e. LDCT screening. The compliance rate of LDCT screening in CanSPUC from 2013 to 2018 remained 34.41%, 37.25%, and 48.21% in urban areas of Shanxi, Henan, and Zhejiang Provinces, respectively [5‐7]. However, we found a substantial improvement (91%) on the compliance rate of LDCT in Wenling lung cancer screening program than those reported by CanSPUC. As CanSUPC was a national cancer screening program targeting five cancer types (lung cancer, female breast cancer, liver cancer, upper gastrointestinal cancer, and colorectal cancer) using a combined screening modality. Given the effect on the compliance rate for the combined screening modality of five cancer type might varied from separate screening for each cancer type, we hence used the compliance rate of LDCT screening from the Wenling lung cancer screening program in this study. The compliance rate of PRS test was then estimated as 83.72% for scenario analysis. When we analysed the impact of the compliance rate of LDCT and PRS test, we observed similar patterns as those obtained from our base case analysis with a perfect attendance, despite some differences on the absolute effects due to discrepancies in the compliance rate of the two cohorts (Supplementary Table S6).