Benchmarking lung cancer screening programmes with adaptive screening frequency against the optimal screening schedules derived from the ENGAGE framework: a comparative microsimulation study

Summary

Background

Lung cancer screening recommendations employ annual frequency for eligible individuals, despite evidence that it may not be universally optimal. The impact of imposing a structure on the screening frequency remains unknown. The ENGAGE framework, a validated framework that offers fully dynamic, analytically optimal, personalised lung cancer screening recommendations, could be used to assess the impact of screening structure on the effectiveness and efficiency of lung cancer screening.

Methods

In this comparative microsimulation study, we benchmarked alternative clinically relevant structured lung cancer screening programmes employing a fixed (annual or biennial) or adaptive (start with annual/biennial screening and then switch to biennial/annual at ages 60- or 65-years) screening frequency, against the ENGAGE framework. Individuals were eligible for screening according to the 2021 US Preventive Services Task Force recommendation on lung cancer screening. We assessed programmes' efficiency based on the number of screenings per death avoided (LDCT/DA) and the number of screenings per ever-screened individual (LDCT/ESI), and programmes’ effectiveness using quality-adjusted life years (QALY) gained from screening, lung cancer-specific mortality reduction (MR), and number of screen-detected lung cancer cases. We used validated natural history, smoking history generator, and risk prediction models to inform our analysis. Sensitivity analysis of key inputs was conducted.

Findings

ENGAGE was the best performing strategy. Among the structured policies, adaptive biennial-to-annual at age 65 was the best strategy requiring 24% less LDCT/DA and 60% less LDCT/ESI compared to TF2021, but yielded 105 more deaths per 100,000 screen-eligible individuals (10.2% vs. 11.8% MR for TF2021, p = 0.28). Fixed annual screening was the most effective strategy but the least efficient and was ranked as the fifth best strategy. All strategies yielded similar QALYs gained. Adherence levels did not affect the rankings.

Interpretation

Adaptive lung cancer screening strategies that start with biennial and switch to annual screening at a prespecified age perform well and warrant further consideration, especially in settings with limited availability of CT scanners and radiologists.

Funding

National Cancer Institute.

Research in context.

Evidence before this study

The US Preventive Services Task Force (USPSTF) recommends annual screening for lung cancer using low-dose CT (LDCT) for eligible individuals. Evidence supports that annual screening may not be the optimal screening frequency for all eligible individuals, with biennial and adaptive screening programmes emerging as potential alternatives. The impact of imposing a fixed screening frequency on the effectiveness and efficiency of lung cancer screening and how such programmes compare to adaptive strategies remains unknown. To identify the evidence preceding this study, we searched PubMed with the search string [((Lung Cancer{MeSH}[Title/Abstract]) AND (Screening{MeSH} [Title/Abstract])) AND (adaptive{MeSH}[Title/Abstract])], limited to English-language articles published in the past 10 years (up to June 2024). The abstracts and titles (n = 43) were screened to identify studies that evaluated the effectiveness of adaptive lung cancer screening. We identified 1 study that evaluated adaptive screening frequencies within the context of lung cancer screening, which found that adaptive screening schedules can reduce the associated harms of screening while maintaining its associated benefits.

Added value of this study

We used the ENGAGE framework, a validated framework offering analytically optimal, fully dynamic, personalised screening recommendations for ever-smoking individuals, to quantify the impact of imposing a fixed or an adaptive screening frequency to lung cancer screening. We assessed the performance of six clinically relevant screening strategies in terms of both effectiveness and efficiency and ranked them based on their proximity to the analytically optimal strategy derived from ENGAGE.

Implications of all the available evidence

Adaptive lung cancer screening programmes that start with biennial screening and switch to annual screening at a specific age have been found to perform well and improve the efficiency of screening as compared to annual screening. Policy makers should consider adaptive lung cancer screening programmes in future lung cancer screening guidelines development, especially in resource-constraint settings.

Introduction

According to the American Cancer Society, lung cancer afflicted more than 230,000 Americans and caused 130,000 deaths in 2022, thus remaining the leading cause of cancer-related death in the US.¹ Lung cancer screening using low-dose computed tomography (LDCT) has been shown to be effective in reducing lung cancer-specific mortality in two randomised clinical trials.^2,3

Existing lung cancer screening recommendations endorse annual screening examinations for all eligible individuals,4, 5, 6 partly for its practicality and ease of implementation. Studies have shown that for some individuals annual screening may be unnecessary, advocating for biennial lung cancer screening,7, 8, 9, 10 especially for individuals with a negative baseline screening exam.^8,10 Existing screening guidelines for breast cancer recommend annual screening for women aged 45–54 and biennial screening for women 55 years and older¹¹; onwards, we refer to strategies with different screening frequencies at specific ages as adaptive screening strategies. However, the effectiveness and efficiency of adaptive screening strategies for lung cancer remains unknown.

Regular screening frequency albeit being practical, is unlikely to be optimal for all eligible individuals, as it does not account for the dynamically varying lung cancer risk, screening history, life expectancy, and prior screening results of individuals. The impact of enforcing the annual screening frequency for all eligible persons on the effectiveness and efficiency of the overall lung cancer screening programme remains unknown given that the optimal screening schedule is elusive. Using principles from Operations Research, Toumazis et al.¹² developed the ENGAGE framework, which delivers personalised, analytically optimal screening schedules for ever-smoking individuals. ENGAGE assumes no fixed screening intervals, and thus delivers fully dynamic, age-specific screening schedules based on a formal assessment of the tradeoff between the expected benefits accrued from screening and the potential harms associated with LDCT screening. Although ENGAGE offers the optimal screening schedules for ever-smoking individuals, its dynamic screening recommendations makes its implementation challenging. Nevertheless, with ENGAGE in hand, one can benchmark the performance of alternative screening strategies that are more practical and easy to implement.¹³

The objective of this study is twofold: (i) to assess and compare the performance of adaptive lung cancer screening strategies against strategies with fixed screening frequencies; and (ii) to quantify the impact of enforcing a regular screening frequency on the effectiveness and efficiency of the overall screening programme.

Methods

Study design

We used a simulation model to assess the performance of lung cancer screening strategies that employ a fixed or an adaptive screening frequency for screen-eligible individuals per the 2021 USPSTF recommendation, and compared their effectiveness and efficiency against the ENGAGE derived screening strategy.

Structured lung cancer screening strategies

We assessed the effectiveness and efficiency of alternative screening policies assuming differing (fixed and adaptive) screening frequencies. In addition to the 2021 USPSTF recommendation on lung cancer screening (TF2021) that employs fixed annual screening, we considered the biennial screening strategy, along with four adaptive strategies that switch screening frequency from annual (A) to biennial (B), or vice-versa, at age 60 or 65 years, denoted by A60B, A65B, B60A, and B65A, respectively. We considered strategies that employ annual, biennial, or a combination of those two screening frequencies based on existing recommendations that endorse annual screening and recent evidence supporting that biennial screening may be optimal for some individuals.^8,12,14,15 For all strategies, individuals were eligible for screening if they satisfied the TF2021 eligibility criteria. We considered screening strategies that switch screening frequency at the ages of 60 and 65 years because those are at the middle of the age range for lung cancer screening (i.e., ages 50 to 80) according to the current USPSTF recommendation.

The ENGAGE framework

We benchmarked the effectiveness and efficiency of the aforementioned structured strategies against the analytically optimal strategy obtained from the ENGAGE framework. The ENGAGE framework¹² is based on a partially observable Markov decision process, a well-established stochastic optimisation framework used to optimise sequential medical decision making under uncertainty.16, 17, 18, 19, 20

The ENGAGE framework was informed by previously published and validated risk-prediction models and models developed within the Cancer Intervention and Surveillance Modeling Network (CISNET) Lung Working Group, which were used to inform the USPSTF recommendations on lung cancer screening.^21,22 Namely, ENGAGE utilises two risk-prediction models developed by Bach and colleagues²³ to inform the annual lung cancer incidence and death from causes other than lung cancer transitions. The transitions between the cancer and smoking states were estimated from a published and externally validated natural history model of lung cancer,^21,22,24 and a validated smoking history generator model developed within CISNET based on data from the US National Health Interview Survey,²⁵ respectively. Cancer state transitions were sex- and histology-specific whereas the smoking state transitions were age-, sex-, and birth cohort-specific. The ENGAGE framework has been shown to reproduce observed lung cancer incidence and mortality, demonstrating good validation.¹² ENGAGE selects the sequence of annual screening decisions (i.e., screen with LDCT or wait) that yields the maximum number of expected quality-adjusted life years (QALYs) gained from screening based on a formal assessment of the tradeoff between the health benefits associated with diagnosing lung cancer at an earlier stage via screening and the potential harms associated with LDCT (e.g., false positives). The ENGAGE-derived screening policies offer personalised, analytically-optimal, fully dynamic screening recommendations that outperform screening programmes which use a one-size-fits-all approach for eligible individuals, in terms of the effectiveness and the efficiency of the programme.¹² We have provided an overview of the ENGAGE model in Supplementary Methods (1) and an illustration of the underlying state-transition model in Supplementary Methods (2).

Study population

We evaluated the performance of each screening strategy on the same simulated population of US ever-smoking individuals from the 1950 birth cohort stratified by sex and smoking group. We simulated lung cancer-related events for 500,000 individuals using a published Markov state transition model¹² to evaluate the performance of alternative screening strategies. Individuals entered the model at age 50 years and were followed until death or age 100 years, whichever happened first. Analyses were stratified by sex, smoking status (current vs. former), and smoking intensity at age 50 years (light: <10 cigarettes/day, moderate: 10–19 cigarettes/day, and heavy: ≥20 cigarettes/day) which are henceforth collectively referred to as the smoking groups.

Outcomes

We assessed the efficiency of each strategy for each sex and smoking group using (i) the number of screenings per death avoided (LDCT/DA); and (ii) the number of screenings per ever-screened individuals (LDCT/ESI). Similarly, we assessed the effectiveness of each screening strategy using (i) QALYs gained from screening; (ii) lung cancer-specific mortality reduction (MR); (iii) screen-detected early-stage lung cancer cases (SDC); and (iv) total number of screen-detected lung cancer cases (early and advanced stages combined) (TSDC). We aggregated our subgroup results and derived population-based estimates by computing the weighted average of each measure across all sexes and smoking groups using the prevalence of each sex and smoking group within the US population as weights (Supplementary Table S1).¹²

For the purposes of this study, we extended the previously published ENGAGE framework to consider individuals’ discomfort while undergoing regular LDCT screening exams, in addition to the distress caused by indeterminate findings that was already incorporated in the model.¹² For our base-case analysis, we considered a disutility equal to three days (0.02) for each false-positive finding^26,27 and a disutility equal to eight hours (0.001) per LDCT screening.^28,29 Samples of the optimal screening schedules derived from the updated ENGAGE framework are provided in Supplementary Figures S1 and S2. We note that ENGAGE recommends lung cancer screening for individuals who are ineligible per the TF2021 criteria; that is, individuals aged over 80 years, people who quit smoking more than 15 years ago, and people with less than 20 pack-years of smoking history. Because the ENGAGE framework used as the benchmark offers personalised recommendations based on the smoking start and end ages of individuals, for practicality we conducted our analysis on the cohort of ever-smoking individuals who started smoking at age 18 years and either continue to smoke at age 50 or have quit smoking at age 45 years for the comparative analysis.

Statistics

Comparative analysis of screening strategies

We assessed the performance of each strategy and compared their effectiveness and efficiency using the frontier analysis, a well-known technique in health economics which allows evaluating two (possibly competing) criteria when comparing several alternatives (e.g., screening programmes of interest in this study).30, 31, 32 Given a pair of effectiveness and efficiency measures, the corresponding frontier is obtained by determining those screening programmes that are not dominated by any other screening programme based on the chosen criteria. For each sex and smoking group, we derived four classes of frontiers, one for each combination of effectiveness vs. efficiency measures, as follows: 1) MR vs. LDCT/DA; 2) MR vs. LDCT/ESI; 3) SDC vs. LDCT/DA; and 4) SDC vs. LDCT/ESI. We further derived the frontiers using the TSDC (vs. LDCT/DA and LDCT/ESI) metric for each strategy's effectiveness. We excluded QALYs from the comparative analysis because differences in QALYs gained from each strategy were clinically negligible.

We created a ranking system using the total distance of each screening strategy from the frontiers based on the rationale that strategies closer to the frontiers are more favorable. Specifically, for each sex and smoking group, we computed the distance of each policy from each of the four frontiers and calculated the weighted sum of the sex-specific distances to rank each policy for a given smoking group. We then obtained the overall rankings for each of the screening strategies by calculating their corresponding weighted sum of the distance across all frontiers and across all smoking groups. Also, we ranked the strategies assuming equal weights for each sex and smoking subgroup. We employed two well-known methods for computing the distances: the Manhattan and Euclidean distances (Supplementary Methods (3)).

Finally, for each strategy, we computed individual relative gain as compared to TF2021 for each effectiveness/efficiency measure, stratified by each smoking group and each sex. Through weighted sum of individual measures for all smoking groups and sexes, we computed the total relative gains or losses in effectiveness/efficiency (Supplementary Methods (4)). Similar analysis was repeated when strategies were compared against ENGAGE.

Sensitivity analysis

We examined the impact of imperfect adherence to lung cancer screening on the performance of the screening strategies and their overall ranking. In addition to the perfect adherence assumed for our base-case analysis, we considered three clinically meaningful values for the adherence rates: the reported adherence rate to lung cancer screening in current clinical practice (15%),^33,34 the adherence rate observed in other more established cancer screening programmes (70%),³⁵ and the midpoint of that range (45%).

We also examined the sensitivity of our findings to the disutility levels associated with false-positive findings and LDCT screening exams. ENGAGE is adaptive to the magnitude of these disutilities—that is, the optimal screening schedule will differ for different disutility values—which consequently will affect the benchmark for the structured strategies. We varied the magnitude of the negative effects associated with the false-positive findings and regular LDCT screening examinations ranging them from 1.5 days to 7 days (0.01–0.04; base-case value was 3 days) and from no disutility to one day (0–0.0027; base-case value was eight hours), respectively.

Further, we conducted probabilistic sensitivity analysis, for males and females separately, for individuals who have quit smoking or those who currently smoke. For the individuals who used to smoke with heavy intensity, we performed sensitivity analysis by simultaneously varying the health state transitions and disutility levels associated with the LDCT screening and false-positive findings. For all individuals who currently smoke, we additionally included varying the smoking intensity and status transitions. We excluded individuals who smoked with light and moderate intensity from the probabilistic sensitivity analysis as those are not eligible for screening per the 2021 USPSTF recommendation. For each analysis, we generated 250 combinations of values for the health state transitions, smoking intensity, and status transitions (only for current smokers), and the disutility values, by sampling from their respective Beta distribution with mean equal to the base-case values of each input parameter. For each combination of values, we simulated 100,000 individuals under each one of the screening strategies considered in our analysis and assessed their performance.

Ethics

Our study utilised simulated data generated by validated models. No identifiable information was used; therefore, no institutional review board (IRB) approval was needed.

Role of the funding source

Our funding source did not influence any stage of this study, particularly data collection, analysis and interpretation, and the writing of the manuscript. All the authors of this work had full access to the input data, were fully aware of the conducted analyses, and approve the content for publication.

Results

The fully dynamic screening strategy derived from ENGAGE outperforms all structured strategies (Table 1). At the population-level, despite being statistically significant, the differences in QALYs were clinically marginal (approximately 7.3 days per person between the strategies yielding the highest and lowest QALYs). ENGAGE achieved the highest MR (16.7% vs. 9.2%–11.8% from structured policies), the highest SDC (12,055 vs. 6340–9186 from the structured policies), while being the most efficient strategy based on the LDCT/DA measure (888 vs. 2640–3334).

Table 1.

Effectiveness and efficiency measures across the entire population using 0.02 and 0.001 as the base disutility values associated with false-positive results and LDCT discomfort, respectively.

Strategy	Effectiveness measures			Efficiency measures
	QALYs per person	MR	SDC	LDCT/DA	LDCT/ESI
ENGAGE	13.179	16.7%	12,055	888	9
TF2021	13.159	11.8%	9186	3334	16
A60B	13.158	9.2%	7006	3287	12
A65B	13.158	9.6%	7308	3324	13
B60A	13.171	10.7%	8525	2710	11
B65A	13.171	10.2%	8174	2682	10
Biennial	13.170	8.0%	6340	2640	8

Each measure is computed as the weighted average of the same measure across all sexes and smoking groups. QALYs: quality-adjusted life years; MR: mortality reduction; SDC: screen-detected early-stage lung cancer cases; LDCT/DA: number of screenings per death averted; LDCT/ESI: number or screenings per ever-screened individuals; TF2021: the 2021 US Preventive Services Task Force lung cancer screening recommendation; A60B (A65B): annual screenings until age of 60 (65) followed by biennial screenings thereafter; B60A (B65A): biennial screenings until age of 60 (65) followed by annual screenings thereafter.

The TF2021 achieved the highest values of MR and SDC among the structured strategies. However, it was the least efficient strategy requiring 3334 LDCT/DA (vs. 888–3324 for other strategies) and 16 LDCT/ESI (vs. 8–13 for other strategies). Strategies that start with biennial screening and then switch to annual screening yielded higher mortality reduction and were simultaneously more efficient than adaptive strategies that start with annual screening that then switch to biennial. Biennial screening was the most efficient screening strategy, but also the least effective. Similar trends were observed for all sex- and smoking-specific subgroup analyses (Supplementary Tables S2–S13).

Effectiveness-efficiency frontiers

Figs. 1 and 2 present the four classes of frontiers, for individuals who currently smoke and those who were smoking in the past, respectively, with heavy or moderate smoking intensity. Frontiers for the individuals with light smoking intensity are provided in Supplementary Figures S3 and S4 because very few individuals with light smoking intensity at age 50 become eligible in their lifetime.

Fig. 1.

Frontiers for current smokers are categorised into moderate smokers: (a) MR vs. LDCT/DA; (b) MR vs. LDCT/ESI; (c) SDC vs. LDCT/DA; (d) SDC vs. LDCT/ESI; and heavy smokers: (e)–(h) analogously. Strategies on each frontier are presented with bold text in the legend. Results are aggregated across the two sexes using the weights presented in Supplemental Table S1. The shaded region indicates the variability in the performance of the ENGAGE-derived screening strategy under different levels of the disutility values associated with false-positive results and regular LDCT exams. Note that, ENGAGE is the only screening strategy that is adaptive to the disutility levels whereas, the performance of the structured strategies is not affected by the disutility values, but their relative position to the frontier may change. MR: Mortality reduction; SDC: Screen-detected early-stage lung cancer cases; LDCT/ESI: Number of screenings per ever-screened individuals; LDCT/DA: Number of screenings per death avoided. TF2021: the 2021 US Preventive Services Task Force lung cancer screening recommendation; A60B (A65B): annual screenings until age of 60 (65) followed by biennial screenings thereafter; B60A (B65A): biennial screenings until age of 60 (65) followed by annual screenings thereafter. Note 1: For this analysis we only considered former smokers who start smoking at age 18 years.

Fig. 2.

Frontiers for former smokers are categorised into moderate smokers: (a) MR vs. LDCT/DA; (b) MR vs. LDCT/ESI; (c) SDC vs. LDCT/DA; (d) SDC vs. LDCT/ESI; and heavy smokers: (e)–(h) analogously. Strategies on each frontier are presented with bold text in the legend. Results are aggregated across the two sexes using the weights presented in Supplemental Table S1. The shaded region indicates the variability in the performance of the ENGAGE-derived screening strategy under different levels of the disutility values associated with false-positive results and regular LDCT exams. Note that, ENGAGE is the only screening strategy that is adaptive to the disutility levels whereas, the performance of the structured strategies is not affected by the disutility values, but their relative position to the frontier may change. MR: Mortality reduction; SDC: Screen-detected early-stage lung cancer cases; LDCT/ESI: Number of screenings per ever-screened individuals; LDCT/DA: Number of screenings per death avoided. TF2021: the 2021 US Preventive Services Task Force lung cancer screening recommendation; A60B (A65B): annual screenings until age of 60 (65) followed by biennial screenings thereafter; B60A (B65A): biennial screenings until age of 60 (65) followed by annual screenings thereafter. Note 1: For this analysis we only considered former smokers who start smoking at age 18 years and stop smoking at age 45 years. Note 2: Note that TF2021 and annual-to-biennial strategies are identical for former smokers considered in this study as those individuals become ineligible for screening after age 60 since they complete 15 years since smoking cessation. Similarly, biennial and biennial-to-annual screenings are equivalent for former smokers.

ENGAGE was the only screening strategy that remained on the frontier for all four combinations of effectiveness vs. efficiency measures across all smoking groups, except for the LDCT/ESI vs. MR frontier for individuals who used to smoke with moderate intensity (Figs. 1 and 2 and Supplementary Figures S3 and S4). The TF2021 strategy was on all the frontiers for individuals with (past or current) moderate smoking intensity. Biennial screening was only on the frontiers for those who currently smoke with heavy intensity that employ the LDCT/ESI metric for efficiency. Among the four adaptive strategies, only the B65A strategy was on the MR vs. LDCT/ESI frontier for current heavy smokers (Fig. 1f). All analyses involving TSDC are presented in Supplementary Figures S5 and S6.

Table 2 presents the ranking for the strategies based on the negative impact of adopting the structured screening frequency on the performance of the overall screening programme as quantified by their respective distance from the frontiers. B65A was the structured strategy closest to the frontier overall, albeit not being the best structured strategy in any of the smoking groups. Notably, TF2021 was ranked as the 5th best strategy, whereas A60B and A65B were the furthest from the frontiers. Rankings were consistent across both Euclidean and Manhattan distances (Table 2, Supplementary Table S14) and when the frontiers were derived using the arithmetic mean (not weighted average) (Supplementary Table S15). Sex-specific analyses yielded the same ranking as our base-case analysis (Supplementary Tables S16 and S17). Supplementary Table S18 presents the ranking of the strategies when TSDC was used instead of the SDC, whereas Supplemental Table S19 compares the alternative strategies when QALYs gained from screening are included in our analysis.

Table 2.

Structured strategies ranking based on the impact of imposing a structure on the screening frequency as quantified by the weighted sum of the Euclidean distances^b from the smoking group-specific frontiers, assuming perfect adherence.

Strategy	Distance from the frontiera						Overall distance	Overall ranking
	Former smoker			Current smoker
	Light	Moderate	Heavy	Light	Moderate	Heavy
ENGAGE	0	0.0004	0	0	0	0	0	1
B65A	0	0.2	22.3	16.3	2.9	8.0	11.4	2
B60A	0	0.2	22.3	18.1	1.4	10.0	11.7	3
Biennial	0	0.2	22.3	11.2	7.5	12.2	12.2	4
TF2021	0	0	29.3	21.8	0	19.8	15.5	5
A60B	0	0	29.3	15.6	10.0	24.9	17.4	6
A65B	0	0	29.3	17.6	8.8	28.6	17.9	7

TF2021: the 2021 US Preventive Services Task Force lung cancer screening recommendation; A60B (A65B): annual screenings until age of 60 (65) followed by biennial screenings thereafter; B60A (B65A): biennial screenings until age of 60 (65) followed by annual screenings thereafter.

^aThe total distances are calculated using the weighted sum over the two sexes.

^bDetails on how the Euclidean distance is calculated are provided in the Supplementary Methods.

Fig. 3 presents the total relative gains in effectiveness and efficiency for each strategy as compared to the TF2021 strategy using the weighted sum of individual relative gains over all smoking groups and sexes. Supplementary Figure S7 presents similar results when equal weights were imposed on the smoking groups and sexes. The figure shows that the relative gain in efficiency was higher for the biennial and adaptive biennial-to-annual strategies compared to TF2021 but with a relatively small loss in effectiveness. Similar analyses were performed by comparing the strategies against ENGAGE (Supplementary Figures S8 and S9).

Fig. 3.

Total relative (percentages of) gains in effectiveness and efficiency for each strategy compared to the 2021 US Preventive Services Task Force lung cancer screening recommendation using the weighted sum of individual relative gains over all smoking groups and sexes. Negative values indicate loss of effectiveness or efficiency. Supplementary Methods (2) further details the computations. TF2021: the 2021 US Preventive Services Task Force lung cancer screening recommendation; A60B (A65B): annual screenings until age of 60 (65) followed by biennial screenings thereafter; B60A (B65A): biennial screenings until age of 60 (65) followed by annual screenings thereafter.

Sensitivity analysis

Lower adherence rates did not alter the ranking of the structured strategies (Supplementary Table S20). Under imperfect adherence rates, ENGAGE was on the frontiers of all smoking groups. The impact of imposing a structure on the screening frequency diminished (that is, structured screening strategies were closer to the frontiers) as the adherence rates decreased. Under all adherence rates considered in our analysis, the TF2021 strategy was on the effectiveness-efficiency frontier for moderate smokers (current and former), but ranked consistently lower than the adaptive biennial-to-annual strategies.

The shaded region in each frontier (Figs. 1 and 2) represents the variability in the performance of the ENGAGE screening strategy for all possible combinations of the disutility values considered associated with false-positive results [0.01, 0.02 (base-case), 0.04] and the participant's discomfort from LDCT [0, 0.001 (base-case), 0.0027]. In general, the disutility levels did not affect the frontiers (i.e., the frontiers were comprised of the same strategies), but the distance of each structured strategy from the frontier may differ.

The results of the probabilistic sensitivity analysis are presented in Supplementary Figures S10–S17, stratified by sex and smoking group. Further, Supplementary Table S21 presents the uncertainty interval for each of the alternative effectiveness and efficiency measures associated with each strategy, stratified by sex, smoking status, and smoking intensity. The results were consistent with those obtained by using the base values for the input parameters.

Discussion

In this study, we assessed and compared the performance of structured lung cancer screening strategies, i.e., strategies that adopt a predetermined, regular (fixed or adaptive) screening frequency on the US population of screen-eligible individuals per the 2021 USPSTF criteria, against a fully dynamic, analytically optimal screening strategy derived from the ENGAGE framework. Using the ENGAGE strategy as the benchmark, we quantified the impact on the effectiveness and efficiency of the overall screening programme for each strategy due to adopting the structure on the screening frequency and created a ranking system to identify the structured strategy that yields outcomes closest to the ENGAGE-derived optimal screening strategy.

We showed that the effectiveness and efficiency of adaptive biennial-to-annual screening strategies are the closest to the fully dynamic, analytically optimal strategy derived from the ENGAGE framework. Interestingly, despite not being the best strategy in any of the measures used to assess the effectiveness and efficiency of the screening programmes, the adaptive biennial-to-annual strategies rank as the best structured strategy when all performance metrics were considered.

Implementation of a screening schedule with a single switch in the screening frequency at a predefined age is a practice that has been recommended for other cancer screening programmes (e.g., breast cancer screening),¹¹ and should not be a barrier for implementation since it retains a structure that is practical and easily implementable.

Our population-based results indicate that the TF2021 strategy is the most effective screening (based on lung cancer-specific mortality reduction and number of screen-detected lung cancer cases) among the structured policies given its aggressive screening regimen. However, annual screening of all eligible individuals impacts the efficiency of the screening programme, because the vast majority of these screening exams would be normal. Consequently, the TF2021 strategy is the least efficiency strategy considered in our analysis. The adaptive annual-to-biennial strategies are more efficient than the TF2021 strategy, but not as effective. Biennial screening is the most efficient strategy among structured policies, but the least effective. We showed that imposing a structure on lung cancer screening is affecting primarily the efficiency of the screening programme and secondarily its effectiveness.

Our findings provide important insights to decision makers regarding the effectiveness and efficiency of alternative easy to implement adaptive screening strategies. These insights could guide the development of future screening programmes that offer better balance between the effectiveness and efficiency of the screening programme. Findings from our study are particularly useful to guide lung cancer screening programmes in resource-constraint settings that are forced to consider the efficiency and cost-effectiveness of the screening programme in addition to its effectiveness.

Adherence to the lung cancer screening programme can exacerbate the impact of the structured screening frequency, although it does not affect their ranking. Through our sensitivity analysis, we found that the impact of imposing a structure on the screening frequency for low adherence rates, as the ones reported in the current clinical practice in the US, is small. However, as the adherence to the screening programme increases, the impact of enforcing a structure to screening frequency on the efficiency and the effectiveness of the overall screening programme worsens. This is important, especially considering that lung cancer screening uptake and adherence levels in the US have been trending upwards, exceeding the pre-COVID-19 pandemic levels.³⁶ Adherence to lung cancer screening is expected to continue to improve as the lung cancer screening programme matures and primary care providers and the public become more aware of the benefits of lung cancer screening. Therefore, as adherence rates improve, policy makers will have to re-evaluate the annual screening frequency, in order to recommend a screening programme that balances effectiveness and efficiency.

Our study has limitations. Although the ENGAGE framework provides the analytically optimal personalised screening strategies, it is not feasible to verify the optimality of these strategies in practice. The ENGAGE model utilises well validated models to inform the transitions between health and smoking states, but there is uncertainty around these estimates. Similarly, our estimates of the psychological effects of screening and indeterminate findings may differ from person to person and in time. Sensitivity analyses showed that our findings are robust to these changes, yet some caution is warranted when generalising the findings of this study. For example, the strategies derived from ENGAGE are personalised based on the sex, birth cohort, age to start smoking, age to stop smoking, smoking intensity, and screening history of individuals. Thus, for practicality, our analysis only considered individuals from the 1950 birth cohort, who started smoking at age 18, and quit smoking at age 45 years for people who quit smoking. Additionally, even though our analysis suggests that switching the screening frequency at age 65 is reasonable (as the mid-point of the USPSTF recommended screening period), a more thorough evaluation of alternative ages at which screening frequency switches is needed to identify the optimal switching age. Finally, the effectiveness and efficiency of the screening strategies are affected by several other confounding factors that are not included in our analysis, including race/ethnicity and access to healthcare, among others. For these reasons, generalising the conclusions of this study should be done with caution.

In conclusion, adaptive biennial-to-annual screening strategies significantly improves the efficiency of the screening programme at a moderate loss in effectiveness, offering better balance in the screening outcomes. Imposing a structure on the screening schedules impacts primarily the efficiency and secondarily the effectiveness of the overall screening programme. Adaptive biennial-to-annual screening warrants further consideration in future lung cancer screening guidelines development, especially in resource-constraint settings.

Contributors

Hemmati, Meza, Tammemägi, and Toumazis were responsible for conceptualization.

Hemmati, Ishizawa, and Toumazis were responsible for data curation.

Formal analysis was conducted by Hemmati, Ishizawa, and Toumazis.

Hemmati, Meza, Ostrin, Hanash, Antonoff, Schaefer, Tammemägi, and Toumazis were responsible for investigation.

Toumazis was responsible for Funding acquisition.

Hemmati, Meza, Tammemägi, and Toumazis were responsible for developing methodology.

Project administration was carried by Toumazis.

Ostrin, Hanash, Antonoff, Schaefer, and Toumazis provided resources.

Hemmati, Ishizawa, and Toumazis were responsible for developing software.

Supervision was provided by Toumazis.

Hemmati and Toumazis were responsible for data verification, validation, visualization, and writing the original draft.

All authors were responsible for reviewing and editing the draft.

All authors read and approved the final version of the manuscript.

Data sharing statement

All data and results are included in the manuscript and the supplemental materials.

Declaration of interests

Mehdi Hemmati has no conflict of interest to report. Sayaka Ishizawa has no conflict of interest to report. Edwin Ostrin has benefitted from “Early Detection Research Network Clinical Validation Center (NCI)” grant. Dr. Ostrin has presented on lung cancer screening in Astra Zeneca (April 2021) and Texas Association of Family Practitioners (Nov 23). He has received support for attending the 2020 Gene Systems (February, July 2023) ad GRAIL (December 2023). Dr. Ostrin has a patent, “intellectual property on a 4-protein blood biomarker panel for lung cancer early detection.” He has served in GRAIL Scientific Advisory Board (Dec 2023). Dr. Ostrin is also involved in continuing negotiation with 2020 Gene Systems (Gaithersburg, MD) for bringing blood biomarker panel for lung cancer to marker. Samir M. Hanash has received support from NCI Lung CVC. Mara Antonoff received payment or honoraria for lectures, presentations, speakers, bureaus, manuscript writing, or educational events held by Merck, Bristol Myers Squibb (BMS), Ethicon, and AstraZeneca. Andrew J. Schaefer has no conflict of interest to report. Martin C. Tammemägi has no conflict of interest to report. Iakovos Toumazis has received support from NIH/NCI (R37CA271187, U01CA253858, F32CA220961, and U01CA199284) and has served as an expert advisor to the American Cancer Society (ACS) Guideline Development Group for the update of the ACS lung cancer screening guideline.