Public Preferences for Risk-Stratified Lung Cancer Screening in Ireland: Protocol using a Citizens’ Jury and Stated Preference Methods

Challenges with eligibility criteria and harms

LCS is inherently risk-based: screening very low-risk populations provides little or no mortality benefit and may increase harms^8–10. Potential harms include false positives, radiation exposure, and invasive follow-up procedures that may cause anxiety and morbidity^11,12. Overdiagnosis is an additional concern, leading to the detection and treatment of cancers that would never have become symptomatic, with consequences for patients and health systems^13–15.

Most existing programmes, including those informed by NLST and NELSON, have relied on age and smoking history as eligibility criteria¹⁶. These are easy to apply but risk both over-inclusion (screening individuals unlikely to benefit) and under-inclusion (excluding some at higher risk due to additional factors)^17,18.

Role of risk prediction models

To address these limitations, multivariable risk prediction models have been developed, incorporating additional demographic and clinical factors such as family history and respiratory disease^19,20. More than 20 models exist, with PLCOm2012 the most extensively validated and widely applied^21–25.

Evidence from large cohorts shows that such models outperform categorical criteria, improving sensitivity and positive predictive value while reducing overdiagnosis^24,26–28. Cost-effectiveness analyses suggest that risk model–based selection prevents more deaths, increases life-years gained, and reduces the number needed to screen compared with age and smoking-based criteria^23,29,30. These features increase the likelihood of programme adoption in resource-constrained health systems. In the UK, both the Liverpool Lung Project and PLCOm2012 models have informed eligibility in the UK Lung Cancer Screening Trial and the NHS Targeted Lung Health Check programme^31,32.

However, the performance of risk models varies between populations, influenced by baseline risk and demographic characteristics^33,34. Thresholds applied in the US or UK populations may not translate directly to Ireland, where smoking patterns, healthcare access, occupational exposures, and population health profiles differ. Furthermore, Ireland's younger demographic structure and rural geography present distinct implementation challenges that require locally informed solutions.

Population-specific thresholds and communication with low-risk groups

Determining appropriate risk thresholds is crucial. Lower thresholds increase sensitivity but at the cost of more false positives, while higher thresholds reduce unnecessary investigations but risk missing potentially curable cancers³³. Critically, there is limited guidance on how best to communicate with individuals assessed as low risk, who should avoid screening but remain vigilant for symptoms, particularly as risk status can change over time with continued smoking or ageing³⁵.

In Ireland, a pilot trial is applying NHS-based thresholds^36,37: offering screening to individuals aged 55–74 with a 6-year PLCOm2012 risk ≥1.51% or a 5-year Liverpool Lung Project (LLPv2) risk ≥2.5%. Whilst this pilot will provide valuable operational insights, further research is required to determine whether these imported cut-offs are clinically appropriate, cost-effective, and acceptable to the Irish population. Additionally, the Irish primary care context—characterised by fee-for-service general practice and variable integration with secondary care—necessitates careful consideration of implementation pathways that differ from the NHS model.

The importance of patient values and shared decision-making

Preference studies indicate a wide variation in how individuals balance the potential benefits and harms of LCS^38,39. This underlines the importance of shared decision-making (SDM), which supports informed, preference-sensitive choices^40,41. The European Respiratory Society and European Society of Radiology highlight SDM as a central component of effective screening, recommending co-designed decision aids and clinician training⁴².

Whilst clinical guidelines endorse SDM^43–45, and evidence suggests it may increase informed uptake without increasing decisional conflict^46,47, implementation remains limited in practice. Barriers include time constraints in consultations, gaps in clinician knowledge about risk communication, and patients’ expressed preference for direct recommendations rather than deliberative discussions⁴⁸. Compounding these challenges, clinicians often overestimate screening benefits and underestimate harms, risking systemically biased consultations⁴⁹. Structured supports – including evidence-based decision aids, communication training, and patient navigation services are therefore essential⁵⁰.

Embedding SDM within LCS programmes is particularly relevant to primary care, where most screening eligibility discussions and follow-up management are likely to occur. In Ireland, where general practitioners serve as gatekeepers to specialist services and enjoy high levels of patient trust, primary care represents both the logical setting for screening discussions and a key determinant of programme success. However, without purpose-designed tools and clear communication frameworks, general practitioners may face significant implementation burdens.

Rationale for this study

Despite compelling evidence for risk-stratified LCS and growing international implementation experience, critical uncertainties remain before a screening programme can be responsibly introduced in Ireland. These uncertainties operate at three distinct levels.

First, at the population level, it is unclear whether international risk thresholds—such as those derived from PLCOm2012 in US cohorts—are appropriate for the Irish population, which differs in smoking epidemiology, cancer incidence patterns, and healthcare system characteristics. Calibration studies are needed, but threshold selection cannot be determined by clinical criteria alone.

Second, at the programme design level, evidence gaps exist regarding optimal approaches to: (a) communicating with individuals categorised as low risk who are ineligible for screening; (b) supporting risk re-assessment as individuals age or their risk factors evolve; and (c) integrating screening pathways within Ireland's mixed public-private primary care system. International models may not be directly transferable.

Third, at the individual decision-making level, there is limited understanding of how Irish patients and the public weigh competing screening attributes (e.g., false positive risks versus mortality reduction), what risk thresholds they consider acceptable for participation, and what forms of decision support align with their preferences and values.

Importantly, no published studies have systematically explored public preferences for risk-stratified LCS design within the Irish context, nor integrated deliberative, preference elicitation, and individual threshold methods to inform screening policy. Addressing these evidence gaps through participatory research is essential to ensure that any future Irish screening programme is not only clinically effective and cost-efficient but also acceptable, equitable, and trusted by the communities it aims to serve.

Aims and objectives

This study addresses these evidence gaps through deliberative and quantitative approaches to inform Irish LCS policy.

In the first phase, a Citizens’ Jury (CJ) will explore public values, ethical considerations and preferences regarding risk-stratified versus population-based screening approaches, providing a contextual understanding of how the Irish public conceptualises screening acceptability, equity, and trust.

The second phase will use a Discrete Choice Experiment (DCE) to quantify preferences for key screening programme attributes. The selected attributes will be informed by the outcomes of the CJ deliberations, and may include eligibility criteria, recruitment strategies, screening locations, communication approaches, and follow-up pathways.

The third phase will apply a Threshold Technique (TT) to identify individuals’ personal risk cut-offs for screening, enabling direct comparison with PLCOm2012 benchmarks and imported UK thresholds.

The integration of these methods will support evidence-based, patient-centred recommendations to guide the design of a future Irish LCS programme. Outputs will include: (a) acceptable risk thresholds for screening eligibility calibrated to Irish public preferences; (b) weighted preference parameters for programme design optimisation; (c) co-designed frameworks for shared decision-making; and (d) key inputs for cost-effectiveness modelling and population impact assessment.

Study design

This programme of work adopts a sequential, exploratory design to inform the development of a risk-stratified LCS programme in Ireland. Three interconnected studies will be undertaken:

Reporting will follow established guidance: ISPOR^51–53 and CHERRIES⁵⁴ for the DCE, TT, and GRIPP2⁵⁵ for patient and public involvement.

Phase 1: Citizens’ Jury

Rationale and design

Citizens’ juries provide a structured forum for lay deliberation on ethically complex health policy questions, combining balanced evidence with facilitated discussion to surface informed public values^56–58. For LCS, CJs can clarify the acceptability of risk thresholds, perceived fairness of eligibility criteria, and preferences for communicating benefits and harms. CJ findings will directly inform DCE attribute development and TT scenario framing.

Participants and Recruitment

Two juries of 10–12 participants each (total n=20–24) will be convened, consistent with published practice for effective deliberation and manageable group dynamics^59–61. Purposive sampling will ensure representation across age, sex, ethnicity, socioeconomic status (proxied by General Medical Services eligibility), smoking history, and geography (urban/rural), guided by Central Statistics Office data. Participants aged 40–74 years will be recruited, reflecting the screening-eligible population (55–74 years in established programmes) plus those approaching eligibility. This age range ensures preference data are policy-relevant whilst maintaining recruitment feasibility. Exclusion criteria are current employment in oncology or respiratory medicine, or a personal history of lung cancer. Experience of other cancers will not preclude participation. Recruitment will be facilitated through general practice networks, community organisations, and targeted social media, with quota monitoring. We will pursue snowball sampling for underrepresented groups. Participants will be reimbursed for time and expenses in line with national public involvement guidance.

Sample size justification

Deliberative studies do not employ statistical power calculations. The proposed size balances inclusivity with the need for high-quality deliberation, aligns with precedent in cancer screening CJs^62,63 and is adequate to reach thematic saturation within and across juries while preserving deliberative validity.

Materials and procedure

One jury will be held in person (Dublin) and one online to increase accessibility. Each jury will run over two days. Day one will deliver a balanced evidence presentation on LCS efficacy, harms (false positives, overdiagnosis, radiation exposure), international eligibility approaches (population-based vs risk-stratified), equity considerations, and communication strategies. Day two will comprise facilitated deliberations on risk thresholds, fairness and transparency of stratification, communication preferences, and low-risk management. Independent facilitators will follow a standardised guide. Pre- and post-jury questionnaires will assess attitudes and knowledge change. Sessions will be audio-recorded, transcribed verbatim, and pseudonymised.

Evidence Presentation

Evidence packs will be prepared in plain language using established guidance on risk communication (e.g., absolute risks, natural frequencies, icon arrays). Content will cover the benefits and harms of LCS, eligibility approaches (age/smoking vs. risk-based), and ethical considerations. Materials will be co-developed by the research team and PPI representatives, then reviewed by an independent advisory panel (GP, respiratory clinician, and two PPI representatives) for accuracy, neutrality, and accessibility. Materials will be piloted with three lay reviewers who do not participate in the juries. During the jury, expert witnesses representing diverse perspectives (e.g., oncology, general practice, ethics, public health) will present short, structured inputs with time for questions.

Facilitation

Jury sessions will be moderated by independent facilitators with backgrounds in deliberative engagement and no involvement in the study design or analysis, to minimise bias. Facilitators will receive a briefing on lung cancer and deliberative techniques, but will not provide substantive content. Their role will be to encourage inclusive participation, manage time, and ensure respectful dialogue.

Deliberation process

The aim is to surface areas of agreement, disagreement, and conditional positions rather than forced consensus. Strong majority views (>75% agreement) will be noted as such. Where views are divided, both positions will be reported with supporting rationales. Anonymous voting may be employed for specific propositions whilst preserving space for minority views.

Outputs and validation

At the end of each session, facilitators will provide a real-time summary of the discussion, which participants will review and amend (member checking). Draft recommendations will be circulated to participants for final comment. These validated outputs will then inform attribute selection for the DCE and framing of TT scenarios.

Data analysis

Analysis will follow codebook thematic analysis⁶⁴ supported by NVivo. An initial coding frame, informed by the deliberation guide and research questions, will be iteratively refined through team consensus. NVivo will support data management. Two analysts will double-code a subset of transcripts to enhance dependability; discrepancies will be resolved by discussion. Analyst reflexivity will be supported through research diaries documenting interpretive decisions and regular team debriefs. Results will be reported according to COREQ and CJCheck (participants, context, facilitators, information presented, decision processes, outputs).

Phase 2a: Discrete Choice Experiment

Rationale

The DCE will estimate the relative importance of screening programme attributes and quantify trade-offs the public is willing to make between benefits and harms, complementing the CJ by providing parameterisable preference weights suitable for modelling and policy design.

Participants and Recruitment

A nationally representative sample of adults aged 40–74 years will be recruited via a market research company (e.g., REDC). Quotas will target representation by age, sex, region, socioeconomic indicators, ethnicity, and smoking status. Eligibility requires residence in Ireland and the capacity to consent in English.

Sample size justification

The minimum sample size for main effects was estimated at 75 using the Johnson–Orme (2003) heuristic (c = 3, t = 10, a = 2)⁶⁵. As the pooled multinomial logit model will include up to three degrees of freedom of covariate interactions—either smoking (3 levels, 2 df) or age-sex (2 df) and potentially smoking with age or sex (3 df)—this figure was scaled to reflect the increased parameter count. Models with 2–3 df of interactions entail roughly three- to four-times as many coefficients as a main-effects specification; thus, the Johnson–Orme floor was multiplied by this parameter ratio (≈3–4×) to maintain comparable precision per coefficient^66,67. A further modest information-dilution factor (1.5×) was applied to account for the lower information content of between-person covariates^53,68. This yields a target range of 340–450 completers, incorporating expected non-completion (400 invitations @ 15% = 340; 500 @ 10% = 450). A working estimate of ≈400 completers is therefore cited throughout this document. As no universal DCE sample-size rule exists, the adequacy of this estimate will be verified empirically through pilot testing^53,68.

Attribute development

Attributes will be developed through four stages:

1.  Initial candidate pool: Generated from citizens' jury deliberations, review of LCS preference studies, and stakeholder interviews with general practitioners (n=3–5) and screening coordinators.

2.  Prioritisation workshop: Patient and public involvement panel (n=6–8) will rate attributes for relevance, comprehensibility, and distinctiveness.

3.  Refinement: Research team consensus on 5–7 attributes balancing policy relevance, preference heterogeneity, and cognitive burden.

4.  Level specification: Informed by Irish pilot data, NHS Targeted Lung Health Checks, and PLCOm2012 thresholds.

Anticipated attributes (subject to citizens' jury and patient involvement input):

•  Eligibility criteria (age plus smoking history versus risk score threshold)

•  Risk communication approach (categorical [high/medium/low] versus absolute percentage)

•  Screening location (hospital versus community clinic versus mobile unit)

•  Invitation approach (general practitioner-initiated versus population mailout versus opportunistic)

•  Follow-up pathway (nurse-led versus general practitioner-led versus integrated multidisciplinary clinic)

•  False positive risk

•  Mortality reduction

Levels will span plausible policy-relevant ranges. The full attribute table with rationales will be provided in supplementary materials per ISPOR guidelines

Attributes and levels will be finalised iteratively through CJ findings, literature, and PPI input^69–72. Development will be transparently reported per ISPOR recommendations and qualitative reporting guidance for formative work, including criteria for attribute inclusion/exclusion, piloting feedback, and final wording. Levels will be policy-relevant and plausible to avoid anchoring or dominance.

Pilot testing

Attributes and survey flow will be piloted in two stages: (i) cognitive interviews (n≈3) with PPI members to assess comprehension, plausibility, and wording; and (ii) an online pre-test (n≈3) to assess completion time, task burden, and response patterns. Feedback will inform revisions before full deployment.

Experimental design and survey flow

A D-efficient fractional factorial design will be generated and blocked into versions to minimise respondent burden. Each participant will complete 8–12 choice tasks presenting two unlabelled alternatives (final decision to be specified after piloting). Task order will be randomised.

The survey will include an introduction, plain-language explanations, worked examples, and comprehension checks. Internal validity will be assessed through one fixed-choice task with a dominant alternative (to identify inattentive responding) and one repeated choice task at different survey positions (test-retest reliability). Participants failing the dominance test will be excluded from primary analysis but retained for sensitivity assessment. No opt-out option will be included to focus on trade-offs between feasible programme configurations.

Data analysis

Responses will be analysed using a pooled multinomial logit (MNL) model under the random utility framework. Each respondent completes ten choice tasks, allowing estimation of average preferences for each attribute level across the sample. Attributes will be effects-coded, and utility will be expressed as a linear combination of attribute levels. The model will include up to three degrees of freedom of covariate interactions—either smoking (three levels), age–sex, or potentially smoking with age or sex—to capture observed (systematic) heterogeneity. In this specification, preferences vary predictably according to respondent characteristics, which is appropriate since future predictions will rely on known covariates rather than repeated choices from the same individuals.

Model performance and parsimony will be assessed using typical information criteria (log-likelihood, AIC, and BIC) when comparing alternative covariate specifications or interaction structures. Statistical significance will be set at p<0.05. All analyses will be conducted using R.

To further examine unexplained preference variation, latent class and mixed logit models will be estimated as exploratory extensions. The latent class model will identify whether respondents cluster into a small number of discrete preference profiles based on their choice patterns, with demographic and other respondent characteristics examined post hoc for association with class membership. The mixed logit model will test whether allowing coefficients to vary randomly across individuals materially improves model fit. These additional models will be undertaken with awareness of their greater complexity and potential instability if the effective sample size is limited, and findings will therefore be interpreted descriptively rather than inferentially.

Phase 2b: Threshold Technique

Rationale

The threshold technique elicits the tipping point at which individuals accept or decline screening, based on the minimum acceptable benefit. Unlike the DCE, which captures relative preferences across multiple attributes simultaneously, the TT provides precise individual-level risk-benefit thresholds comparable to international benchmarks such as PLCOm2012 cut-offs.

Participants and recruitment

The TT will be administered to the same sample completing the DCE, ensuring within-sample complementarity and enabling direct comparison of preference structures across methods. Sampling, recruitment, and eligibility criteria are described in Phase 2a.

Design

We will use an adaptive staircase design to determine each participant’s personal risk cut-offs for LCS. This will be achieved through a series of binary choice iterations conducted via the following protocol:

1.  Starting risk: Participants will be randomly assigned to one of four starting values: 0.75%, 1.5%, 3%, or 6% six-year lung cancer risk. This randomisation will enable assessment of anchoring effects without explicit priming.

2.  Step rules: Initial steps of +- 1 percentage point. After the first reversal (i.e. change from accept to decline, or vice-versa), step size halves to +- 0.5% points, then to +- 0.25% points after the second reversal.

3.  Convergence criterion: Threshold is defined when the participant reverses direction twice within a 0.5% range and confirms the position on a third presentation at that level.

4.  Maximum iterations: 10 iterations per participant to limit burden. Participants not achieving convergence will be classified as non-convergent and analysed separately.

At each iteration, participants will see:

•  Their hypothetical personal 6-year lung cancer risk at the specified level (e.g. “Your risk is 2%”)

•  Number needed to screen to prevent one lung cancer death at that risk level

•  Expected number of people (per 100 screened) experiencing a false positive result requiring follow-up imaging or procedures

•  Visual representation (Cates plot showing 100 people at this specified risk level)

To maximise comprehension, scenarios will use plain language, absolute risk figures, and visual decision aids. Cates plots will be the primary format; however, alternative formats such as icon arrays and bar charts will be pilot-tested with PPI members to ensure accessibility across literacy levels. Preferred formats will be retained for the main survey. Number needed to screen, and number needed to harm at each risk level will be derived from published PLCOm2012 validation studies and UK pilot data, with all assumptions documented in supplementary materials.

Following this information presentation, the participant will be asked at each iteration: "Given this risk level, would you want to participate in lung cancer screening?" [Yes/Unsure/No]. Responses of "Unsure" will prompt the midpoint between accept/reject values.

The threshold technique will be embedded within the same online survey platform as the discrete choice experiment. After completing discrete choice tasks, participants will proceed directly to threshold scenarios. In each iteration, participants will indicate whether they would accept screening at the presented risk level. If screening is accepted, the subsequent scenario presents a lower risk level; if declined, the subsequent scenario presents a higher risk level. The staircase continues until convergence is achieved or 10 iterations are completed.

Data analysis

Results will be summarised as medians with 95% confidence intervals for the Minimum Acceptable Benefit (MAB) — the minimum absolute mortality reduction at which participants choose to accept screening, determined by the elicited threshold risk level within the adaptive staircase and its corresponding modelled screening outcomes. Descriptive analysis will show the distribution of thresholds visualised via histograms with a comparison with the PLCOm2012 1.5% benchmark. Summary statistics will be stratified by key demographic and clinical characteristics.

Each participant’s elicited threshold risk (%) for accepting lung cancer screening will be treated as a quasi-continuous outcome. The primary analysis will use linear regression to examine associations between threshold risk and age, sex, smoking status, socioeconomic status, and family history of lung cancer. Regression coefficients will represent mean differences in threshold (%) with 95% confidence intervals. Model assumptions will be checked using residual diagnostics, and results will also be summarised as medians with interquartile ranges.

Sensitivity analyses will test for anchoring and convergence effects. A secondary model will include anchoring group (0.75%, 1.5%, 3%, 6%), time-to-converge, and their interaction to assess whether anchoring effects vary by the speed of convergence.

To aid interpretation of the results from the DCE and TT elements, we will conduct an exploratory analysis of the relationship between TT-derived risk thresholds and DCE decisions at the participant level. Optional open-ended questions will explore potential explanations for the participants' choices.

Research context

The interdisciplinary research team includes patients, academics (public health, epidemiology, psychology, social sciences), healthcare professionals, and policymakers, ensuring a well-rounded approach to recruitment strategy development. The study will take place in Ireland, where healthcare is delivered through a mixed public-private system overseen by the Health Service Executive (HSE). The government is transitioning toward universal healthcare via the Sláintecare Implementation Strategy. Publicly funded cancer services are managed by the National Cancer Control Programme (NCCP), while population-based screening programmes, such as breast, cervical, and bowel screening, are run by the National Screening Service (NSS).

Patient and Public Involvement (PPI)

PPI is embedded throughout, with contributions from the PRICAN PPI network and the Irish Lung Cancer Community. Members informed study design, attribute selection, and visual aid development, and will contribute to interpretation and dissemination. Activities will be reported using GRIPP2⁵⁵.

Ethics and informed consent

Ethical approval granted by the RCSI REC (Ethics Review Code: 202510014). All participants will provide informed consent and receive a Participant Information Leaflet outlining study procedures, risks, benefits, and their right to withdraw. No minors are involved, and consent will be obtained directly from all participants.

Strengths and limitations

The CJ provides legitimacy and transparency in identifying attributes for subsequent quantitative work, addressing recognised gaps in DCE development. The DCE will produce population-level utility estimates, while the TT will generate precise individual tipping points, enabling comparison with internationally applied thresholds such as PLCOm2012. The study embeds PPI throughout, in line with GRIPP2 guidance, ensuring that outputs are meaningful and implementable. Primary care relevance is central, with outputs designed to support GP-led SDM.

Limitations should also be noted. CJ samples are necessarily small and may not capture the full spectrum of perspectives, particularly among marginalised groups. Stated preference methods are vulnerable to hypothetical bias, meaning that expressed choices may not fully predict behaviour. The absence of an opt-out in the DCE could reduce realism, although it allows more precise estimation of policy-relevant trade-offs. Online survey administration and English-only materials may exclude some groups, despite purposive sampling and quota strategies. Finally, while the study informs programme design, its scope does not extend to health-economic modelling or pilot implementation, which will be required before policy adoption.

Comparison with existing literature

International evidence supports the superiority of risk-model-based eligibility over categorical criteria^24,26–28, but optimal thresholds vary across populations³³. UK experience with the Liverpool Lung Project and NHS Targeted Lung Health Checks demonstrates feasibility, yet transferability to Ireland requires calibration. Our TT explicitly addresses this evidence gap.

Studies have consistently shown heterogeneity in how individuals balance the benefits and harms of LCS^38,39, reinforcing the importance of SDM. However, uptake of SDM remains limited in practice, constrained by consultation time, clinician knowledge, and misperceptions of benefit and harm. By co-designing decision aids and training content grounded in empirical preference data, this study responds directly to those barriers.

Methodologically, the combined use of CJ, DCE, and TT is novel in LCS research. It aligns with best practice for transparent attribute selection⁵¹ and adds precision through estimation of the minimum acceptable benefits derived from each participant’s elicited risk threshold. This approach addresses a rarely examined but crucial element for aligning screening programme design with public tolerance for harm.

Policy and practice implications

The findings of this study will provide population-specific thresholds for risk-based eligibility, allowing comparison with international standards and supporting calibration of models such as PLCOm2012 for use in the Irish context. They will also inform the development of SDM tools that communicate absolute benefits and harms in clear and accessible formats, suitable for integration into GP–patient discussions in primary care. Importantly, the study will generate equity-related insights by identifying which screening attributes are valued most by underserved or higher-risk groups, thereby responding to calls for screening programmes that reduce, rather than exacerbate, inequalities⁷³. In addition, the work will provide guidance on how best to communicate with individuals who fall below the eligibility threshold, ensuring that these groups remain engaged, vigilant for symptoms, and aware that their risk status may change over time.

Taken together, these outputs directly align with national priorities, including the NCCP Early Diagnosis Plan 2022–2025⁷⁴ and the National Cancer Strategy 2017–2026⁷⁵, while also complementing European initiatives such as SOLACE⁷⁶ and the EU Cancer Mission^77,78. The study therefore offers timely evidence to support both national decision-making and broader European efforts to expand equitable access to lung cancer screening.