Clinical usefulness of polygenic risk scores in risk prediction models for lung cancer screening and lung nodule management - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Transl Oncol . 2026 Apr 15;68:102771. doi: 10.1016/j.tranon.2026.102771 Search in PMC Search in PubMed View in NLM Catalog Add to search Clinical usefulness of polygenic risk scores in risk prediction models for lung cancer screening and lung nodule management Véronique Boumtje Véronique Boumtje a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Véronique Boumtje a , Zhonglin Li Zhonglin Li a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Zhonglin Li a , Nathalie Gaudreault Nathalie Gaudreault a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Nathalie Gaudreault a , Victoria Saavedra Armero Victoria Saavedra Armero a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Victoria Saavedra Armero a , Dominique K Boudreau Dominique K Boudreau a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Dominique K Boudreau a , Sophie Plante Sophie Plante a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Sophie Plante a , Sabrina Biardel Sabrina Biardel a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Sabrina Biardel a , Aida Eslami Aida Eslami a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada b Department of Social and Preventive Medicine, Université Laval, Quebec City, Canada Find articles by Aida Eslami a, b , Simon Martel Simon Martel a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Simon Martel a , Sébastien Thériault Sébastien Thériault a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Sébastien Thériault a , Philippe Joubert Philippe Joubert a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada Find articles by Philippe Joubert a , Yohan Bossé Yohan Bossé a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada c Department of Molecular Medicine, Université Laval, Quebec City, Canada Find articles by Yohan Bossé a, c, ⁎ Author information Article notes Copyright and License information a Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada b Department of Social and Preventive Medicine, Université Laval, Quebec City, Canada c Department of Molecular Medicine, Université Laval, Quebec City, Canada ⁎ Corresponding author at: Institut universitaire de cardiologie et de pneumologie de Québec – Université Laval, Quebec City, Canada. [email protected] Received 2026 Feb 9; Revised 2026 Apr 2; Accepted 2026 Apr 10; Collection date 2026 Jun. © 2026 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). PMC Copyright notice PMCID: PMC13096907 PMID: 41990546 Highlights • Polygenic risk score utility in real-world lung cancer screening remains underexplored. • Polygenic risk score predicted incident lung cancer in both pre-screening populations and high-risk post-screening cohorts. • Polygenic risk score offers context-dependent clinical net benefit. Keywords: Polygenic susceptibility to lung cancer, Lung cancer screening, Lung nodule management Abstract Background Lung cancer mortality decreases as a result of low-dose computed tomography (LDCT) screening, but suffers from low uptake and high false-positive rates. The impact of integrating genetic risk using a polygenic risk score (PRS) to optimize lung cancer screening remains underexplored. Methods We developed a genome-wide PRS and evaluated its performance in pre- and post-screening contexts. Screening eligibility was assessed using two UK Biobank (UKB) subsets: UKB PLCO (n = 8957; PLCOm2012norace risk ≥2%) and UKB ScreeningCriteria (n = 74,024 meeting screening eligibility criteria). To evaluate nodule management, we used a cohort of 669 ever-smokers with PLCO ≥2% and Lung-RADS score ≥3, referred to as SYNERGIQC PLCO_LungRADS . Multivariable Cox models, time-dependent area under the curve (AUC), and decision curve analyses (DCA) evaluated association, discrimination, and clinical net benefit. Results In UKB PLCO , the PRS was associated with a hazard ratio (HR) of 1.18 per standard deviation. In UKB ScreeningCriteria , PRS showed HR of 1.34. Adding PRS to the PLCOm2012 model improved discrimination (AUC: 0.707 vs 0.696; P = 6.85e −27 [likelihood ratio test]) and correctly reclassified 9.2% of incident lung cancer cases. Six-year absolute risks stratified by PRS deciles indicate 3.1-fold increase in the top compared to the bottom decile. In SYNERGIQC PLCO_LungRADS , HR PRS was 1.22. In this context, DCA indicated a modest net benefit for decision thresholds between 10% and 30%. Conclusion PRS in lung cancer reveals context-dependent net benefit across studied populations. Although PRS adds limited value for determining screening eligibility, it may help reclassify borderline individuals and inform decisions regarding closer follow-up or invasive diagnostic procedures for high-risk screened groups. Introduction Lung cancer remains the leading cause of cancer-related mortality worldwide, with most cases diagnosed at an advanced stage when curative treatment options are limited [ 1 ]. Early detection through low-dose computed tomography (LDCT) screening reduces mortality by 20–24% in high-risk populations, as demonstrated by the national lung screening trial (NLST) and NELSON trial [ 2 , 3 ]. However, real-world implementation has been hindered by low uptake, high false-positive rates leading to unnecessary investigations and treatment of benign nodules, and overdiagnosis, which results in the detection of small, non-progressive cancers that would not cause clinical symptoms or death. The limitations of current risk targeting approaches are further highlighted by ever-smokers who fall just above or below clinical screening thresholds [ 4 ], as well as individuals who have never smoked but are at high risk of developing lung cancer due to inherited susceptibility [ 5 ]. Risk prediction models such as the prostate, lung, colorectal, and ovarian cancer screening trial model 2012 (PLCOm2012) [ 6 ], improve screening eligibility [ 7 ]. Similarly, the lung computed tomography screening reporting and data system (Lung-RADS) [ 8 ] provides the standard protocol for managing lung nodules detected during screening. Still, PLCOm2012 and Lung-RADS do not account for measured genetic factors. In the case of PLCOm2012, the model relies on family history as an indirect proxy for inherited susceptibility, which may explain inter-individual variability in lung cancer risk [ 9 ]. A personalized, integrative approach using genetic information and existing risk prediction models might address these limitations [ 10 ]. Polygenic risk scores (PRS), derived from genome-wide association studies (GWAS), offer a quantitative approach to estimate inherited risk by aggregating common germline variants [ 11 ]. For lung cancer, several susceptibility loci have been identified across diverse populations [ 12 ], and recent studies suggest that PRS adds independent predictive value beyond traditional clinical factors [ 11 , [13] , [14] , [15] , [16] ]. Yet, their utility in real-world screening contexts, especially when combined with validated risk prediction models, remains underexplored. In lung cancer screening programs, it remains unclear how individuals with low or high genetic risk, defined by the PRS could be integrated into the PLCOm2012 and Lung-RADS systems to optimize both the screening eligibility and the clinical workup of indeterminate nodules. For instance, individuals with high genetic risk may benefit from LDCT even with low or no smoking exposure, whereas those with low genetic risk may require greater smoking exposure to achieve a similar screening benefit. Similarly, after screening and given the same Lung-RADS score, individuals with a high PRS might warrant earlier follow-up or biopsy to detect fast-growing tumors, while those with a low PRS could potentially be reclassified for annual screening. In this study, we evaluate the statistical added value of integrating a genome-wide PRS into risk models for lung cancer screening and pulmonary nodules management. We leverage two prospective cohorts: UK Biobank (UKB) participants meeting screening eligibility criteria, and SYNERGIQC [ 17 ] cohort, derived from a lung cancer screening initiative in the province of Quebec. The first objective is to evaluate the PRS's ability to predict the risk of lung cancer in the pre-screening cohort. We assess whether integrating the PRS improves models’ risk discrimination and time-to-diagnosis prediction, with a specific focus on clinical decision thresholds, such as six-year lung cancer risk exceeding 2%. We further determine the PRS’s capacity to correctly reclassify individuals across the eligibility cut-off point by calculating the category-based net reclassification index (NRI) and the clinical net benefit. We hypothesize that the PRS will enhance screening eligibility. The second objective is to assess the PRS's utility for risk stratification in the post-screening population. We hypothesize that the PRS will improve risk stratification for individuals with indeterminate pulmonary nodules. Methods Study design We leveraged a prospective study including incident cases of lung cancer (UKB) and the provincial lung cancer screening program (SYNERGIQC). We constructed a pan-genomic PRS for lung cancer in UKB and SYNERGIQC. We evaluated the PRS clinical validity at two critical time points by modelling three decision-making contexts: UKB ScreeningCriteria and UKB PLCO as the pre-screening context and SYNERGIQC PLCO_LungRADS as the post-screening context. Study populations UK Biobank We used UKB, a prospective cohort with a range of health-related outcomes, as the screening context population [ 18 ]. From 502,166 available participants aged 37–73 years, we excluded those who were not of European ancestry (n = 92,783), those without genotype data (n = 485), and those with cancer at baseline (n = 191). Only ever-smokers aged 55–73 with ≥20 years of smoking history and ≤15 years since cessation were included. These criteria ensured that UKB participants were comparable in eligibility and genetic ancestry to participants in the lung cancer screening program in the province of Quebec. As a result, 74,024 participants met the screening eligibility criteria, henceforth referred as “UKB ScreeningCriteria ”. A high clinical risk subgroup (n = 8957) was further defined by a PLCOm2012norace risk score ≥2%, to match the Quebec lung cancer screening program. This subgroup is henceforth referred as “UKB PLCO ”. To calculate the PLCOm2012norace in UKB, we applied the equations used in Pasquinelli et al. [ 19 ] ( Appendix A ). In UKB, missing data were imputed using k-nearest neighbors (kNN; k = 10) algorithm [ 20 ]. Variables related to smoking had the highest proportion of missing value, but this was below 3% of the population. We converted educational levels to its equivalent in the cohort, while other variables, including family history, smoking status, and the presence of chronic obstructive pulmonary disease (COPD), remained unchanged. SYNERGIQC SYNERGIQC originates from the Quebec lung cancer screening program [ 17 ], a provincial multicenter pilot-study launched in 2021 for ever-smokers aged 55–74. The program initially enrolled participants, of whom 3542 were classified as high-risk based on a PLCOm2012norace risk score of ≥2%. This risk threshold was chosen according to the PanCan model [ 21 ], the precursor to the PLCOm2012 model validated in Canada [ 22 ]. From this high-risk group, participants were selected for the SYNERGIQC project, 2974 provided blood samples for research purposes to the AIRS biobank at the IUCPQ and CUSM sites. In SYNERGIQC, blood samples were collected during the first or second visit following enrolment in the screening program. As part of the Quebec lung cancer screening program, the Lung-RADS grading system (version 2022) was used to guide management decisions, such as close follow-up with LDCT or biopsy, consistent with previous reports [ 23 ]. In the current study, the subset of interest, referred to as “SYNERGIQC PLCO_LungRADS ”, was selected for genetic analysis. This subset modelled the context of indeterminate pulmonary nodules and included participants with at least one abnormal LDCT scan at screening. These participants were classified as Lung-RADS 3+ and were therefore considered to have an indeterminate pulmonary nodule (n = 669). SYNERGIQC PLCO_LungRADS data were collected locally, so all variables had no missing data. Outcome definition Incident lung cancer was defined by pathological confirmation or LDCT diagnosis in SYNERGIQC. For UKB, ICD9 and ICD10 related codes were used as described previously [ 24 ]. Supplementary Table 1 shows UKB data fields and codes for defining lung cancer cases. At the time of data analysis, the median follow-up was 13.7 years for UKB and 2.8 years for SYNERGIQC. Supplementary Table 2 presents the comparative overview of the SYNERGIQC and the UKB cohorts. Genotyping and imputation SYNERGIQC Genotyping was performed using the Illumina global screening array (GSA) version 3 BeadChip with the multi-disease (MD) drop-in panel, which adds approximately 50,000 disease-specific genetic markers. Quality control (QC) involved the exclusion of low-quality genetic variants with 10th percentile of Illumina GenCall score ≤ 0.1, a call rate < 0.97%, and a minor allele frequency (MAF) of <1%. Samples with missing genotypes were excluded, and Hardy-Weinberg equilibrium P < 1e-7 was used to evaluate genotype errors. Samples with sex discordance, duplication or non-European ancestry (via principal component analysis (PCA)) were also excluded. After QC, 669 samples and ∼511,000 variants on autosomal chromosomes remained for imputation. Supplementary Table 3 provides a summary of genotyping QCs for variants and samples. Ungenotyped single nucleotide polymorphisms (SNPs) were imputed via the TOPMed imputation server (minimac4, GRCh38). Variants with INFO <0.3 or MAF <0.001 were excluded. SNPs were harmonized to HapMap Phase 3 variants and lifted over to GRCh37 for matching with the reference panel used for variant selection. A total of 1,045,282 variants were selected for PRS calculation. UK Biobank Genotyping was performed on the Affymetrix UK BiLEVE Axiom and Axiom arrays. Imputation used the haplotype reference consortium (HRC) and UK10K+1000 Genomes reference panels. We excluded individuals with sex discordance, non-European ancestry, or samples with excess third-degree relatives (>10). Variants with INFO <0.3 or MAF <0.0001 were excluded. A total of 408,422 participants of European ancestry were included in the analysis. Polygenic risk score construction We constructed genome-wide PRS using summary statistics from a previously reported lung cancer GWAS including 29,266 cases and 56,450 controls of European ancestry [ 25 ]. The LDpred2-auto algorithm (R package bigsnpr v1.12.18), which does not require any validation set, was applied using linkage disequilibrium (LD) estimated from 408,422 UKB individuals. This method allows the target sample to also act as a reference panel for the LD matrix, helping to avoid overfitting, as long as the sample size is sufficient and representative of the GWAS study population. The PRS was calculated as a weighted sum of risk alleles, and standardized within each cohort. Supplementary Figure 1 illustrates the workflow. Risk modeling and statistical analyses Predictive models We developed multivariable Cox proportional hazards models (R package survival v3.8–3) to assess the predictive value of PRS, PLCOm2012norace, and their combination. Models were adjusted for age, sex, and the top ten genetic principal components (PCs). The study end date was October 25, 2022 for UKB and May 27, 2025 for SYNERGIQC. Discrimination Discriminatory performance was assessed using the area under the receiver operating characteristic curve (AUC) and Harrell’s concordance index (C-index), which measure a model's ability to distinguish between individuals who experienced the event earlier and those who experienced it later. Time-dependent AUCs were calculated at clinically relevant horizons (2 and 5 years) using inverse probability of censoring weighting [ 26 ], implemented via the R package timeROC v0.4. Risk stratification and cut-offs We applied the Youden index to derive optimal PRS thresholds for dichotomizing participants into high and low genetic risk groups [ 27 ]. These were compared to commonly used percentile-based thresholds (PRS80, PRS90). Kaplan-Meier diagnosis-free probability and log-rank tests evaluated time-to-diagnosis differences. Reclassification indices and net benefit To evaluate the clinical validity of the PRS models for lung cancer, we calculated the six-year absolute risk of lung cancer in UKB by combining the Cox proportional hazards model coefficients with the Efron [ 28 ] estimator for the baseline hazard function (R package riskRegression v12.21). We then compared the models with and without the PRS using the category-based net reclassification index only in UKB ScreeningCriteria . We quantified the clinical utility of PRS risk models applied to each study population with a decision curve analysis (R package rmda v1.6) across a range of lung cancer risk threshold probabilities (0–30%). In UKB, DCA was calculated from six-year absolute risk models, and the net benefits were compared between models with or without the PRS, either as a continuous variable or dichotomized using the Youden threshold. In SYNERGIQC PLCO_LungRADS , DCA was based on the Cox proportional hazards model, including the PRS Youden, to assess the net benefit of incorporating PRS for lung nodule malignancy status. Net benefit was calculated as the true-positive rate minus the false-positive rate based on the lung cancer risk threshold probabilities. It considers both the benefits of correctly identifying incident lung cancer (true positives) and the harms of intervening with individuals who do not yet have lung cancer (false positives). Classification accuracy and threshold optimization The performance of the lung cancer risk prediction models was evaluated by comparing their predicted values to the actual values for each cohort. The models were trained using 70% of the population and optimism-corrected discrimination estimates were derived using internal validation with 200 bootstrap resamples of the training set. Finally, the performance of the models on the remaining 30% of the population was evaluated. Confusion matrix was performed using the R package caret v7.0–1. Ethical approvals UK Biobank received approval from the British National Health Service, North West - Haydock Research Ethics Committee (16/NW/0274). The SYNERGIQC research project, was approved by the IUCPQ ethics committee (2022–3752). All participants of UKB and SYNERGIQC provided informed consent at the baseline assessment. Data availability The present analyses were performed as part of UK Biobank data application 25,205. The genome-wide PRS developed in this study is available in the PGS Catalog ( www.pgscatalog.org , score ID: PGS005397). ILCCO summary statistics [ 25 ] are available in the GWAS catalog under access number GCST004748. Results Study cohorts and baseline characteristics We analyzed 74,024 ever-smokers from the UK Biobank, representing a pre-screening population, of whom 8957 met the eligibility criteria for lung cancer screening according to the PLCOm2012norace model (≥ 2%) as well as 669 clinically high-risk post-screening participants from the SYNERGIQC cohort. Baseline characteristics are summarized in Supplementary Table 4 . In the UKB ScreeningCriteria cohort (n = 74,024), 3193 incident lung cancers (4.3%) were documented during a median follow-up of 13.7 years. In the UKB PLCO clinical selected subset (n = 8957), 1124 incident lung cancers (12.5%) were recorded. In both UKB subsets, lung cancer cases had higher smoking exposure and were more frequently afflicted by COPD ( Supplementary Table 4 ). A higher proportion of males was also observed among lung cancer cases, but this did not reach statistical significance in UKB PLCO . Age differed between the two groups in UKB subsets, but in opposite directions in UKB PLCO and UKB ScreeningCriteria , which we considered not clinically meaningful given the small mean differences between the groups. SYNERGIQC PLCO_LungRADS consisted of 669 high-risk post-screening ever-smokers, with a mean age of 65.1 ± 4.8 years and 47.8% male. At the time of data analysis, the median follow-up was 2.8 years (interquartile range 1.9–3.4), and 123 participants (18.4%) had a diagnosis of lung cancer. Of these 116 patients (94.3%) had pathological confirmation, while the remaining seven (5.7%) were diagnosed based solely on imaging, as obtaining a tissue sample for pathological analysis was not possible. SYNERGIQC PLCO_LungRADS participants diagnosed with lung cancer during the follow-up period were more likely to be male and were older than participants without a diagnosis ( Supplementary Table 4 ). However, participants with and without lung cancer did not differ in terms of smoking exposure or COPD. Association of PRS and PLCOm2012 with lung cancer incidence Genome-wide PRS was independently associated with lung cancer incidence in multivariable Cox proportional hazards models adjusted for age, sex, and PCs ( Fig. 1 ). In the UKB ScreeningCriteria cohort, both the PRS (HR per SD=1.34, 95% CI: 1.30–1.39; P = 9.66e −58 ) and PLCOm2012norace model (HR per SD=1.49, 95% CI:1.46–1.52; P = 2.32e −271 ) were significantly associated with lung cancer incidence. These associations remained true after mutual adjustment (PRS adjustedPLCOm2012norace HR per SD=1.29, 95%CI: 1.24–1.34; P = 9.41e −43 ; PLCOm2012norace adjustedPRS HR per SD=1.47, 95% CI: 1.44–1.50; P = 9.33e −315 ), indicating that each provides complementary predictive information. In UKB PLCO , each SD increase in the PRS was associated with an 18% increased risk of lung cancer incidence (HR per SD=1.18, 95% CI: 1.11–1.26; P = 1.13e −7 ). In SYNERGIQC PLCO_LungRADS , the effect size was similar, but estimated with less precision (HR per SD=1.22, 95% CI: 1.02–1.46; P = 3.02e −2 ). Fig. 1. Open in a new tab Association Between PRS and Lung Cancer Incidence in Cox Models. Forest plots of hazard ratios (HRs) for PRS and PLCOm2012norace (PLCO) in multivariable Cox proportional hazards models across cohorts. All models were adjusted for age, sex, and 10 first genetic principal components. P-value calculated treating the PRS as a continuous variable. Abbreviations: CI, confidence interval; HR, hazard ratio; PLCO, PLCOm2012norace. Discrimination of predictive performance Receiver operating characteristic (ROC) analyses demonstrated that the PRS modestly improved discrimination both pre- and post-screening ( Fig. 2 ). PRS alone in a model adjusted for age, sex, and PCs yielded AUCs of 0.629 in UKB ScreeningCriteria , 0.575 in UKB PLCO , and 0.654 in SYNERGIQC PLCO_LungRADS . In UKB ScreeningCriteria , integration with PLCOm2012norace increased AUC to 0.707 (likelihood ratio test P = 6.85e −27 vs. PLCOm2012norace AUC 0.696). Time-dependent ROC analysis ( Fig. 3 ) confirmed stable performance over clinically relevant time horizons. At 5 years in UKB ScreeningCriteria , the AUC increased to 0.603 (Youden index=0.155, 84th-85th percentile of the PRS), and in UKB PLCO , the AUC remained similar to 0.580 (Youden index=0.132 42nd-43rd percentile of the PRS); at 2 years in SYNERGIQC PLCO_LungRADS , the PRS Youden yielded an AUC of 0.579 (Youden index=0.147 45th-46th percentile of the PRS), suggesting an adaptable threshold based on cohort-specific distributions. Supplementary Table 5 presents the discriminative performance of the risk prediction models over time in UKB subsets and SYNERGIQC PLCO_LungRADS . Fig. 2. Open in a new tab Discrimination of Lung Cancer Risk Models in UK Biobank and SYNERGIQC PLCO_LungRADS . Receiver operating characteristic (ROC) curves comparing model (adjusted with age, sex, and first 10 ancestry principal components) performance across cohorts. Shown are AUCs for PRS-only models, PLCOm2012norace-only models, and the combined models. Abbreviations: PLCO, PLCOm2012norace. Fig. 3. Open in a new tab Time-Dependent AUC for PRS Prediction of Lung Cancer. Time-dependent ROC curves for PRS models evaluated at 2 years in SYNERGIQC PLCO_LungRADS and 5 years in UK Biobank. Risk stratification: PRS enables gradient discrimination Higher PRS values were consistently associated with increased lung cancer incidence ( Supplementary Table 6 ). In UKB ScreeningCriteria , joint stratification by PRS and PLCOm2012norace showed the poorest outcomes in dual high-risk groups and the best outcomes in dual low-risk groups (log-rank P < 1e −4 ; Supplementary Figure 2 ). In UKB PLCO , diagnosis-free probability was significantly lower in individuals with high PRS, with a consistent upward trend in incidence across all PRS quintiles (log-rank P < 0.001; Supplementary Figure 3 ). A similar trend was seen in SYNERGIQC PLCO_LungRADS (log-rank P PRS Youden =0.016 and P PRS>90% =0.028; Supplementary Figure 4 ) and incidence increased with PRS strata (2.1-fold higher in top decile vs. bottom 90%, P = 0.034). These results demonstrate the effectiveness of PRS, whether used alone or in conjunction with existing risk prediction models, to improve lung cancer risk stratification at different stages of screening. Classification accuracy and threshold optimization We compared the performance of different PRS thresholds (Youden, PRS80, PRS90) across cohorts, and in UKB ScreeningCriteria their combination with PLCOm2012norace ( Table 1 ). In UKB ScreeningCriteria , PLCOm2012norace alone balanced performance best (sensitivity 36.5%, specificity 87.1%; false positive rate (FPR) 12.9%). Adding PRS modestly increased sensitivity while keeping acceptable specificity: PRS80/90 + PLCOm2012norace reached ∼42% sensitivity with ∼83% specificity and FPR ∼17%. PRS Youden + PLCOm2012norace maximized sensitivity (44.5%) but lowered specificity (80.4%; FPR 19.6%). In UKB ScreeningCriteria , the positive predictive value (PPV) remained below 12%, which corresponds to the low prevalence of the disease. In UKB PLCO , PRS selected thresholds yielded high specificity (≥83%; FPR (4.8–16.9%), but low sensitivity (<21%), indicating that the PRS may be more useful as an exclusion tool rather than a primary screening test. In SYNERGIQC PLCO_LungRADS , PRS90 gave the best discrimination (sensitivity 38.2%; specificity 74.1%; FPR 25.9%). By contrast, the PRS Youden maximized sensitivity (70.6%) at the expense of low specificity (34.9%; FPR 65.1%), which may be advantageous when prioritizing early lung cancer detection in a high-risk post-screening population. These results suggest that the PRS adds context-dependent value in screening programs by maximizing detection in high-risk post-screening populations (SYNERGIQC PLCO_LungRADS ), excluding low-risk individuals in clinically selected populations (UKB PLCO ) and modestly improving triage in broader pre-screening populations (UKB ScreeningCriteria ). Table 1. Performance Metrics of PRS-Based Lung Cancer Risk Models Across Cohorts. Risk Prediction Model Sensitivity Specificity False Positive Rate Positive Predictive Value Negative Predictive Value Accuracy UKB ScreeningCriteria PRS Youden 29.8 80.0 20.0 6.3 96.2 77.9 PRS80 28.8 82.1 17.9 6.7 96.3 79.8 PRS90 25.6 83.4 16.6 6.5 96.2 81.0 PLCO 36.5 87.1 12.9 11.2 96.8 84.9 PRS80 + PLCO 41.9 82.4 17.6 9.6 97.0 80.7 PRS90 + PLCO 42.1 83.2 16.8 10.1 97.0 81.4 PRS Youden + PLCO 44.5 80.4 19.6 9.2 97.0 78.9 UKB PLCO PRS Youden 20.2 83.1 16.9 15.3 87.4 74.9 PRS quintile 9.4 93.6 6.4 18.0 87.3 82.5 PRS80 8.8 94.5 5.5 19.4 87.3 83.3 PRS90 6.8 95.2 4.8 17.5 87.1 83.6 SYNERGIQC PLCO_LungRADS PRS Youden 70.6 34.9 65.1 18.2 85.3 41.0 PRS quintile 58.8 42.8 57.2 17.4 83.5 45.5 PRS80 32.4 78.9 21.1 23.9 85.1 71.0 PRS90 38.2 74.1 25.9 23.2 85.4 68.0 Open in a new tab Sensitivity, specificity, and related metrics for PRS (e.g., Youden, PRS80, PRS90) alone or combined with PLCOm2012norace. PRS threshold (predefined PRS optimal threshold derived from Youden's index at 5 years in UKB subsets and 2 years in SYNERGIQC PLCO_LungRADS ). Bootstrap (200) validation used for unbiased estimates. Reclassification indices and net benefit Using a fully integrated risk model, we evaluated the six-year absolute risks stratified by PRS deciles in UKB ScreeningCriteria ( Fig. 4 ). Individuals in the 10th decile have a 3.1-fold higher risk (2.51%) than those in the first decile (0.81%). This demonstrates that the model successfully identified lung cancer events that mostly occur in the highest genetic risk tiers. Integrating the PRS into the full clinical model significantly improved the category-based net reclassification index (NRI: 0.032, 95% CI:0.012–0.056), primarily driven by a robust gain in case-detection (NRI+: 0.053, 95% CI: 0.033–0.077). The PRS correctly reclassified 9.2% of incident lung cancer cases (Pr (Up|Case):9.2%) that were otherwise classified as low-risk according to the PLCO criteria. While this increased sensitivity was accompanied by a modest increase in the reclassification of non-cases into the high-risk group (Pr (Up|Ctrl):4.6%), the overall net benefit remains positive, supporting the use of genetic susceptibility to capture high-risk individuals falling outside traditional screening thresholds in the pre-screening population. Fig. 4. Open in a new tab Predicted 6-year Lung Cancer Absolute Risk in UKB ScreeningCriteria Stratified by PRS Decile. Absolute risk was calculated using a multivariable Cox proportional hazards model adjusted for age, sex, 10 ancestry principal components, and the PLCO clinical risk score. The red line represents the mean population risk (1.5%) and the dashed green line the 2% threshold used in PLCOm2012 screening model. Error bars indicate 95% confidence intervals. Across clinically relevant lung cancer risk thresholds, PLCOm2012norace and PRS / PRS Youden yielded measurable net benefit ( Fig. 5 ). In UKB ScreeningCriteria ( Fig. 5 a), benefit was negligible, with the model’s curves approaching the "screen none" line beyond the ∼0.15 lung cancer risk threshold probability. When integrating PRS with PLCOm2012norace in UKB ScreeningCriteria ( Fig. 5 b), the PRS Youden + PLCOm2012norace and binary PLCOm2012norace-alone curves were essentially superimposed across thresholds ∼0.01–0.3, with a minimal incremental net benefit from adding PRS to PLCO in this setting. This suggests that PLCO already captures most actionable risk information in UKB ScreeningCriteria . In UKB PLCO ( Fig. 5 c), benefit was small and limited to a narrow range of thresholds (<∼0.10–0.20). In SYNERGIQC PLCO_LungRADS ( Fig. 5 d), the PRS Youden curve lay above both hypothetical "invasive procedures for all" and "invasive procedures for none" scenarios between ∼0.1 and ∼0.3 lung cancer risk threshold probability, indicating fewer unnecessary invasive procedures for a similar or higher number of detected cancers within this range. Collectively, these findings suggest that the PRS is most useful as a triage tool for managing lung nodules in high-risk post-screening groups, whereas its incremental value over a strong clinical model for selecting individuals for lung cancer screening appears limited in broader screening-eligible populations. Supplementary Figure 5 presents DCA of different PRS thresholds (Youden, PRS80, PRS90) in UKB subsets and SYNERGIQC PLCO_LungRADS . Fig. 5. Open in a new tab Decision Curve Analysis of Lung Cancer Risk Models. The x-axis is the risk threshold probability that changes from 0 to 1 (right truncated at 0.3) and the y-axis is the calculated net benefit for a given threshold probability. The colorful curves depict the net benefit of the risk model’s selection strategy for screening, whereas the dashed light gray and dotted gray lines display the net benefits in the alternative strategies of “intervention for all” (dashed light gray) versus “intervention for none” (dotted gray) in the data set. (a) presents risk models adjusted for age, sex and PC1–10, including PRS + PLCOm2012norace, PLCOm2012norace only or PRSYouden + PLCOm2012norace in UKB ScreeningCriteria . (b) presents risk models including PLCOm2012norace or PLCOm2012 binary (<2% threshold and equal or >2%) only, and PRS Youden + PLCOm2012 binary, age, sex and PC1–10 applied in UKB ScreeningCriteria . (c) presents risk models including PRS, age, sex and PC1–10 applied in the subset of participants with PLCOm2012norace risk equal or >2% from UKB PLCO . (d) presents risk models including PRS as a continuous variable, age, sex and PC1–10 applied in the SYNERGIQC PLCO_LungRADS cohort. Net benefit was calculated as the standardized difference between true positives and false positives, adjusted for the threshold probability. Discussion Our findings show that a genome-wide PRS adds independent and complementary information to established risk prediction models for lung cancer among ever-smokers eligible for LDCT screening. Across two well-characterized cohorts that differ in screening context and risk spectrum (UK Biobank and SYNERGIQC), the PRS remained a significant lung cancer predictor after adjusting for age, sex, genetic PCs, and PLCOm2012norace scores. Despite the consistent effect sizes observed (HRs ∼1.2–1.3 per SD) and modest but statistically significant gains in discrimination (ΔAUC = +0.011, P < 0.001), this shift represents a non-negligible refinement of individual risk trajectories. In UKB ScreeningCriteria, the model integrating genetic risk factor identified a high-risk stratum in the top PRS decile, with a mean 6-year absolute risk 3.1-fold increase compared to the bottom decile. Notably, this genetic adjustment moved the highest-risk group well above the 2.0% clinical threshold for LDCT eligibility. These results align with recent reports indicating that modern lung cancer PRS can stratify risk within smokers beyond traditional variables and that integrated models may help reclassification of borderline individuals into actionable screening categories [ 29 , 30 ]. Importantly, assessing the net benefit of the PRS using DCA allowed us to pinpoint three observations. First, in UKB ScreeningCriteria , adding the Youden PRS to a binary PLCOm2012norace did not materially improve net benefit at screening-relevant thresholds (∼1–4%), indicating that a strong clinical model already captures most actionable risk information. Second, in UKB PLCO , PRS alone offered incremental benefit confined to a narrow range of thresholds (<∼10–20%). Third, in SYNERGIQC PLCO_LungRADS , a Youden-dichotomized PRS yielded modest but tangible net benefit across ∼10–30% risk thresholds, implying fewer unnecessary invasive additional examinations for similar or greater case detection. Interpreting DCA in screening is well established and emphasizes net benefit as the decision-relevant quantity across varying preferences and harms [ 31 , 32 ]. The differential PRS performance between the pre- and post-screening cohorts is biologically and design consistent. The absence of DCA improvement when PRS is added to PLCOm2012norace in UKB ScreeningCriteria suggests that the PRS-integrated approach maintains a clinical utility by potentially capturing high-risk individuals who might fall below the threshold of traditional criteria models such as PLCOm2012 or Bach [ 33 , 34 ]. These models rely heavily on smoking duration and intensity and they are not correlated with genetic risk, either biologically or statistically. In UKB PLCO with low event rate and the participant profile filtered based on PLCOm2012, the net benefit of a binary PRS is modestly noticeable at a limited range of thresholds; at such prevalence, PRS primarily acts as an exclusion tool for low-risk participants not identified by the clinical risk model, with minimal additional benefit over "screen none”. In contrast, SYNERGIQC PLCO_LungRADS is enriched with higher baseline risk and LDCT abnormalities (Lung-RADS ≥3) already identified during screening. In this context, additional genetic information may help inform decisions for patients with similar Lung-RADS score but in different genetic risk groups. The integrated model could justify up-classifying the management to an immediate biopsy or PET/CT scan for patients in the top PRS decile. Conversely, clinicians might feel more confident opting for short-term surveillance rather than invasive procedures for patients with a low PRS, especially those with poor performance status and significant comorbidities. For screening programs using risk-model thresholds of 2% 6-year range [ 23 , 35 ], our results suggest that the PRS should be used to manage indeterminate lung nodules rather than as a first-line tool to select participants for lung cancer screening. Guidelines for managing patients with pulmonary nodules support the use of predictive models to classify patients into high- or low-risk groups [ 36 ]. However, it has not been clearly demonstrated that current prediction models, such as the Brock model used for initial risk assessment, offer better results than clinical judgment, and are not universally accepted [ 37 ]. Future research should focus on developing and validating algorithms that integrate artificial intelligence-based nodule interpretation, the Brock model score, and biomarkers such as PRS. This is clinically relevant, given the ongoing concerns about overdiagnosis and false positives in LDCT programs [ 38 ], as well as the need for more effective management of indeterminate pulmonary nodules. Beyond average effects, PRS gradients in our data identified subgroups with substantially different incidence and timing of diagnosis. Consistent with recent large-scale analyses, individuals in the highest PRS strata experience elevated risk and reach the absolute risk thresholds relevant for screening or lung nodule investigation, whereas those in lower PRS strata accumulate risk more slowly, implications that support personalized screening and management of lung lesions [ 39 ]. Taken together, these observations reinforce that germline susceptibility reflects elements of intrinsic risk that are not fully captured by current risk prediction models and provide insight into how to better monitor individuals identified as high-risk before and after screening. A recent study suggests that risk prediction models for smokers and nonsmokers are essential for lung cancer screening [ 40 ]. However, there are controversies regarding the direction of the PRS effect on smoking across ethnic groups [ 13 , 16 , 41 ]. UK Biobank studies found an increased HR for lung cancer with both smoking status and genetic risk [ 18 ]. Zhu et al. [ 14 ]. reported that the interaction between smoking and PRS regarding lung cancer incidence was significant. This interaction was not observed in the China Kadoorie Biobank, suggesting smoking-related lung cancer risk differences by ethnicity. Further studies are needed to clarify the interaction between smoking and the PRS, and to determine whether the PRS is more informative in ever or never-smokers. Strengths of this work include evaluation across two independent cohorts with longitudinal follow-up; a genome-wide PRS derived with state-of-the-art methodology (LDpred2) and robust LD resources; and a multi-metric assessment spanning discrimination, time-to-event performance, and decision analysis. The usefulness of PRS prediction models to enhance screening eligibility and manage pulmonary nodules is shown in three population contexts. Finally, using >70,000 UK Biobank participants bolsters generalizability within European ancestry populations. Some limitations merit emphasis. First, our analyses were restricted to individuals of European ancestry. The portability of PRS across ancestries remains imperfect due to differences in LD, allele frequencies, and GWAS-detected genetic architecture; multi-ancestry development and recalibration will be essential for equitable deployment [ 42 , 43 ]. Future research must prioritize multiethnic biobanks in order to develop PRSs that will benefit global populations in research and clinical settings, offering an opportunity to promote health equity. Second, the relatively short median follow-up in SYNERGIQC PLCO_LungRADS (∼2.8 years) may underestimate long-term risk separation and may have missed slow growing/low grade lung cancers that warrant mid to long-term clinical follow-up. Third, we focused on PLCOm2012norace; other validated models (e.g., Bach, LLPv2) and their programmatic thresholds are used internationally, and head-to-head comparisons with PRS augmentation warrant further studies [ 44 , 45 ]. Fourth, although our DCA suggests that unnecessary invasive procedures could be reduced while achieving comparable or improved benefits, realized reductions will require validation and prospective evaluation. Finally, we did not incorporate blood-based biomarkers (e.g., ctDNA, small RNAs), which are rapidly evolving and may further enhance early detection and screening programs when combined with imaging and genetic risk factors [ 46 , 47 ]. Collection of a blood sample prior to LDCT, followed by genotyping, would represent an ideal approach for implementing PRS in real-world settings, enabling its use both for refining lung cancer risk prediction in the pre-screening setting and for optimizing the management of indeterminate lung nodules in the post-screening setting. The convergence of risk model-based screening pathways already in practice, inexpensive and scalable genotyping, and maturing PRS methodology support the potential future integration of PRS into screening workflows [ 48 ]. Health-economic modeling suggests that risk-stratified strategies that include PRS can be cost-effective under plausible assumptions, though real-world data and randomized designs are needed to establish value, safety, and acceptability [ 49 , 50 ]. Conclusion PRS-guided decision making offers context-dependent clinical net benefit depending on the studied populations. In pre-screening populations with strong clinical models, genetic predisposition can reclassify individuals above or below a screening threshold. In high-risk post-screening cohorts, a model with PRS alone can yield modest net benefit across a range of lung cancer risk thresholds. With attention to ancestry equity, benchmarking, and prospective evaluation of PRS-based processes, these results support targeted research on PRS aimed at testing the hypothesis that it could reduce unnecessary diagnostic procedures for indeterminate nodules. Funding This work was supported by the Consortium Québécois sur la Découverte du Médicament (CQDM), SYNERGIQC program in partnerships with AstraZeneca, Pfizer, Biomark and the IUCPQ Foundation through a generous donation from Mr. Normand Lord. CRediT authorship contribution statement Véronique Boumtje: Writing – original draft, Visualization, Methodology, Formal analysis, Data curation. Zhonglin Li: Methodology, Data curation. Nathalie Gaudreault: Writing – review & editing, Resources, Project administration, Methodology. Victoria Saavedra Armero: Writing – review & editing, Resources, Methodology. Dominique K. Boudreau: Writing – review & editing, Resources, Methodology. Sophie Plante: Writing – review & editing, Data curation. Sabrina Biardel: Writing – review & editing, Data curation. Aida Eslami: Writing – review & editing, Formal analysis. Simon Martel: Writing – review & editing, Resources, Investigation. Sébastien Thériault: Writing – review & editing, Supervision, Investigation, Conceptualization. Philippe Joubert: Writing – review & editing, Investigation, Funding acquisition. Yohan Bossé: Writing – review & editing, Supervision, Funding acquisition, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgments The authors would like to thank the team at the IUCPQ-UL and CUSM AIRS Biobank site ( www.biobanque.ca ) for their valuable assistance as well as their collaborators, including Dr Nicole Ezer at CUSM, Dr Kim Boutet at CISSS Côte-Nord, Dr Nicole Bouchard at CIUSSS de l’Estrie-CHUS, Dr Nathalie Daaboul at CIUSSS de la Montérégie-Centre, and Dr Laila Samy at CISSS-Laval, for their help with patient recruitment and sample collection. This research has been conducted using the UK Biobank Resource under Application Number 25205. Véronique Boumtje is the recipient of the Vanier Canada Graduate Scholarships awarded by the Canadian Institutes of Health Research. Philippe Joubert holds a Junior 2 Clinical Research Scholar award from the FRQS and holds a Canada Research Chair in Identification of predictive and diagnostic biomarkers to improve lung cancer management. Footnotes Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.tranon.2026.102771 . Appendix. Supplementary materials mmc1.docx (1.3MB, docx) References 1. Bray F., Laversanne M., Sung H., et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J. Clin. 2024;74(3):229–263. doi: 10.3322/caac.21834. 2024-05-01. [ DOI ] [ PubMed ] [ Google Scholar ] 2. De Koning H.J., Van Der Aalst C.M., De Jong P.A., et al. Reduced lung-cancer mortality with volume CT screening in a randomized trial. N. Engl. J. Med. 2020;382(6):503–513. doi: 10.1056/nejmoa1911793. 2020-02-06. [ DOI ] [ PubMed ] [ Google Scholar ] 3. The National Lung Screening Trial Research Team Reduced lung-cancer mortality with low-dose computed tomographic screening. N. Engl. J. Med. 2011;365(5):395–409. doi: 10.1056/nejmoa1102873. 2011-08-04. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Jonas D.E., Reuland D.S., Reddy S.M., et al. Screening for lung cancer with low-dose computed tomography: updated evidence report and systematic review for the US Preventive Services Task Force. JAMA. 2021;325(10):971–987. doi: 10.1001/jama.2021.0377. March 9, 2021. [ DOI ] [ PubMed ] [ Google Scholar ] 5. Lam S., Bai C., Baldwin D.R., et al. Current and future perspectives on computed tomography screening for Lung Cancer: a roadmap from 2023 to 2027 from the International Association for the Study of Lung Cancer. J. Thorac. Oncol. 2024;19(1):36–51. doi: 10.1016/j.jtho.2023.07.019. 2024-01-01. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Tammemägi M.C., Katki H.A., Hocking W.G., et al. Selection criteria for lung-cancer screening. N. Eng. J. Med. 2013;368(8):728–736. doi: 10.1056/NEJMoa1211776. February 21, 2013. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Tammemägi M.C., Church T.R., Hocking W.G., et al. Evaluation of the lung cancer risks at which to screen ever- and never-smokers: screening rules applied to the PLCO and NLST cohorts. PLoS. Med. 2014;11(12):1549–1676. doi: 10.1371/journal.pmed.1001764. Published 2014. Dec 2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. LungRADSv2022 2025. https://www.acr.org/-/media/ACR/Files/RADS/Lung-RADS/Lung-RADS-2022.pdf Accessed November 18. 9. Lebrett M.B., Smith M.J., Crosbie E.J., et al. Validation of lung cancer polygenic risk scores in a high-risk case-control cohort. Genet. Med. 2023;25(8) doi: 10.1016/j.gim.2023.100882. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Li L., Pang S., Starnecker F., Mueller-Myhsok B., Schunkert H. Integration of a polygenic score into guideline-recommended prediction of cardiovascular disease. Eur. Heart. J. 2024;45(20):1843–1852. doi: 10.1093/eurheartj/ehae048. 2024-05-27. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Dai J., Lv J., Zhu M., et al. Identification of risk loci and a polygenic risk score for lung cancer: a large-scale prospective cohort study in Chinese populations. Lancet Respir. Med. 2019;7(10):881–891. doi: 10.1016/S2213-2600(19)30144-4. 2019-10-01. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Bosse Y., Amos C.I. A decade of GWAS results in lung cancer. Cancer Epidemiol. Biomark. Prev. Apr 2018;27(4):363–379. doi: 10.1158/1055-9965.EPI-16-0794. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Boumtje V., Manikpurage H.D., Li Z., et al. Polygenic inheritance and its interplay with smoking history in predicting lung cancer diagnosis: a French-canadian case-control cohort. EBioMedicine. 2024;106 doi: 10.1016/j.ebiom.2024.105234. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Zhu M., Lv J., Huang Y., et al. Ethnic differences of genetic risk and smoking in lung cancer: two prospective cohort studies. Int. J. Epidemiol. 2023;52(6):1815–1825. doi: 10.1093/ije/dyad118. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Bossé Y., Martel S. Germline variants invited to lung cancer screening. Lancet Respir. Med. 2019;7(10):832–833. doi: 10.1016/S2213-2600(19)30188-2. [ DOI ] [ PubMed ] [ Google Scholar ] 16. Zhang P., Chen P.-L., Li Z.-H., et al. Association of smoking and polygenic risk with the incidence of lung cancer: a prospective cohort study. Br. J. Cancer. 2022;126(11):1637–1646. doi: 10.1038/s41416-022-01736-3. 2022-06-01. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Martel S., Lizotte H., Boily G., et al. EP.04A.04 real-world data of LDCT lung cancer screening implementation in a public healthcare system: quebec three 3-year experience. J. Thorac. Oncol. 2024;19(10):S460–S461. doi: 10.1016/j.jtho.2024.09.843. 2024-10-01. [ DOI ] [ Google Scholar ] 18. Sudlow C., Gallacher J., Allen N., et al. UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS. Med. 2015;12(3):1549–1676. doi: 10.1371/journal.pmed.1001779. Published 2015 Mar 31. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Pasquinelli M.M., Tammemägi M.C., Kovitz K.L., et al. Risk prediction model versus United States Preventive Services Task Force lung cancer screening eligibility criteria: reducing race disparities. J. Thorac. Oncol. 2020;15(11):1738–1747. doi: 10.1016/j.jtho.2020.08.006. 4 August 2020. [ DOI ] [ PubMed ] [ Google Scholar ] 20. Zhang Z. Introduction to machine learning: k-nearest neighbors. Ann. Transl. Med. Jun 2016;4(11):218. doi: 10.21037/atm.2016.03.37. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Tammemagi M.C., Schmidt H., Martel S., et al. Participant selection for lung cancer screening by risk modelling (the Pan-Canadian Early Detection of Lung Cancer [PanCan] study): a single-arm, prospective study. Lancet Oncol. 2017;18(11) doi: 10.1016/S1470-2045(17)30597-1. [ DOI ] [ PubMed ] [ Google Scholar ] 22. Jantzen R., Ezer N., Camilleri-Broët S., Tammemägi M.C., Broët P. Evaluation of the accuracy of the PLCOm20126-year lung cancer risk prediction model among smokers in the CARTaGENE population-based cohort. CMAJ. Open. 2023;11(2):E314–E322. doi: 10.9778/cmajo.20210335. 2023-03-01. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Darling G.E., Tammemägi M.C., Schmidt H., et al. Organized lung cancer screening pilot: informing a province-wide program in Ontario, Canada. Ann. Thorac. Surg. 2021;111(6):1805–1811. doi: 10.1016/j.athoracsur.2020.07.051. 111. [ DOI ] [ PubMed ] [ Google Scholar ] 24. Larsson S.C., Carter P., Kar S., et al. Smoking, alcohol consumption, and cancer: a mendelian randomisation study in UK Biobank and international genetic consortia participants. PLoS. Med. 2020;17(7):1549–1676. doi: 10.1371/journal.pmed.1003178. 2020-07-23. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. McKay J.D., Hung R.J., Han Y., et al. Large-scale association analysis identifies new lung cancer susceptibility loci and heterogeneity in genetic susceptibility across histological subtypes. Nat. Genet. Jul 2017;49(7):1126–1132. doi: 10.1038/ng.3892. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Blanche P., Dartigues J.-F., Jacqmin-Gadda H. Estimating and comparing time-dependent areas under receiver operating characteristic curves for censored event times with competing risks. Stat. Med. 2013;32(30):5381–5397. doi: 10.1002/sim.5958. [ DOI ] [ PubMed ] [ Google Scholar ] 27. Martínez-Camblor P., Pardo-Fernández J.C. The Youden Index in the generalized receiver operating characteristic curve context. Int. J. Biostat. 2019;15(1) doi: 10.1515/ijb-2018-0060. [ DOI ] [ PubMed ] [ Google Scholar ] 28. Hertz-Picciotto I., Rockhill B. Validity and efficiency of approximation methods for tied survival times in Cox regression. Biometrics. Sep 1997;53(3):1151–1156. doi: 10.2307/2533573. [ DOI ] [ PubMed ] [ Google Scholar ] 29. Ma Z., Zhu Z., Pang G., et al. Development and validation of a lung cancer polygenic risk score incorporating susceptibility variants for risk factors. Int. J. Cancer. 2025;156(5):953–963. doi: 10.1002/ijc.35210. [ DOI ] [ PubMed ] [ Google Scholar ] 30. Hu J., Ye Y., Zhou G., Zhao H. Using clinical and genetic risk factors for risk prediction of 8 cancers in the UK Biobank. JNCI. Cancer Spectr. 2024;8(2) doi: 10.1093/jncics/pkae008. April 2024. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Vickers A.J., Elkin E.B. Decision curve analysis: a novel method for evaluating prediction models. Med. Decis. Mak. 2006;26(6):565–574. doi: 10.1177/0272989x06295361. 2006-11-01. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Vickers A.J., Calster B.V., Steyerberg E.W. A simple, step-by-step guide to interpreting decision curve analysis. Diagn. Progn. Res. 2019;3(1) doi: 10.1186/s41512-019-0064-7. 2019-12-01. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Bach P.B., Kattan M.W., Thornquist M.D., et al. Variations in lung cancer risk among smokers. JNCI: J. Natl. Cancer Inst. 2003;95(6):470–478. doi: 10.1093/jnci/95.6.470. [ DOI ] [ PubMed ] [ Google Scholar ] 34. Ten Haaf K., Jeon J., Tammemägi M.C., et al. Risk prediction models for selection of lung cancer screening candidates: a retrospective validation study. PLoS. Med. Apr 2017;14(4) doi: 10.1371/journal.pmed.1002277. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Toumazis I., Cao P., De Nijs K., et al. Risk model-based lung cancer screening: a cost-effectiveness analysis. Ann. Intern. Med. 2023;176(3):320–332. doi: 10.7326/M22-2216. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Callister M.E.J., Baldwin D.R., Akram A.R., et al. British Thoracic Society guidelines for the investigation and management of pulmonary nodules: accredited by NICE. Thorax. 2015;70(Suppl 2) doi: 10.1136/thoraxjnl-2015-207168. ii1-ii54. [ DOI ] [ PubMed ] [ Google Scholar ] 37. Loverdos K., Fotiadis A., Kontogianni C., Iliopoulou M., Gaga M. Lung nodules: a comprehensive review on current approach and management. Ann. Thorac. Med. Oct-Dec 2019;14(4):226–238. doi: 10.4103/atm.ATM_110_19. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Black W.C., Chiles C., Church T.R., et al. Lung cancer incidence and mortality with extended follow-up in the national Lung screening trial. J. Thorac. Oncol. 2019;14(10):1732–1742. doi: 10.1016/j.jtho.2019.05.044. 2019-10-01. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Hung R.J., Warkentin M.T., Brhane Y., et al. Assessing lung cancer absolute risk trajectory based on a polygenic risk model. Cancer Res. Mar 15 2021;81(6):1607–1615. doi: 10.1158/0008-5472.CAN-20-1237. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Hsin-Hung C., En-Kuei T., Yun-Ju W., Fu-Zong W. Impact of annual trend volume of low-dose computed tomography for lung cancer screening on overdiagnosis, overmanagement, and gender disparities. Cancer ImAging. 2024;24(1):73. doi: 10.1186/s40644-024-00716-5. Jun 12. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 41. Duncan M.S., Diaz-Zabala H., Jaworski J., et al. Interaction between continuous pack-years smoked and polygenic risk score on lung cancer risk: prospective results from the Framingham heart study. Cancer Epidemiol. Biomarkers Prev. Apr 3 2024;33(4):500–508. doi: 10.1158/1055-9965.Epi-23-0571. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Xiang R., Kelemen M., Xu Y., et al. Recent advances in polygenic scores: translation, equitability, methods and FAIR tools. Genome Med. 2024;16(1) doi: 10.1186/s13073-024-01304-9. 2024-02-19. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Duncan L., Shen H., Gelaye B., et al. Analysis of polygenic risk score usage and performance in diverse human populations. Nat. Commun. 2019;10(1):3328. doi: 10.1038/s41467-019-11112-0. 2019-07-25. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Gabe R., Crosbie P.A.J., Vulkan D., et al. Prospective evaluation of lung cancer screening eligibility criteria and lung cancer detection in the Yorkshire lung screening trial. J. Thorac. Oncol. 2025;20(4):425–436. doi: 10.1016/j.jtho.2024.12.016. [ DOI ] [ PubMed ] [ Google Scholar ] 45. Niu C., Lyu Q., Carothers C.D., et al. Medical multimodal multitask foundation model for lung cancer screening. Nat. Commun. 2025;16(1):1523. doi: 10.1038/s41467-025-56822-w. 2025-02-11. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Lam W.K.J., Bai J., Ma M.-J.L., Cheung Y.T.T., Jiang P. Circulating tumour DNA analysis for early detection of lung cancer: a systematic review. Ann. Transl. Med. 2024;12(4):64. doi: 10.21037/atm-23-1572. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Ren F., Fei Q., Qiu K., Zhang Y., Zhang H., Sun L. Liquid biopsy techniques and lung cancer: diagnosis, monitoring and evaluation. J. Exp. Clin. Cancer Res. 2024;43(1) doi: 10.1186/s13046-024-03026-7. 2024-04-01. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. The All of Us Research Program Genomics Investigators Genomic data in the All of Us Research Program. Nature. 2024;627(8003):340–346. doi: 10.1038/s41586-023-06957-x. 2024-03-14. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 49. Zhao Z., Gu S., Yang Y., et al. A cost-effectiveness analysis of lung cancer screening with low-dose computed tomography and a polygenic risk score. BMC. Cancer. 2024;24(1) doi: 10.1186/s12885-023-11800-7. 2024-01-13. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Wang X., Zhang Z., Ding Y., et al. Impact of individual level uncertainty of lung cancer polygenic risk score (PRS) on risk stratification. Genome Med. 2024;16(1):22. doi: 10.1186/s13073-024-01298-4. Published 2024 Feb 5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials mmc1.docx (1.3MB, docx) Data Availability Statement The present analyses were performed as part of UK Biobank data application 25,205. The genome-wide PRS developed in this study is available in the PGS Catalog ( www.pgscatalog.org , score ID: PGS005397). ILCCO summary statistics [ 25 ] are available in the GWAS catalog under access number GCST004748. Articles from Translational Oncology are provided here courtesy of Neoplasia Press ACTIONS View on publisher site PDF (1.9 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top