Risk prediction for lung cancer screening: a systematic review and meta-regression

Risk prediction for lung cancer screening: a systematic review and meta-regression | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Risk prediction for lung cancer screening: a systematic review and meta-regression Ramin Rezaeianzadeh , Crystal leung , Soo Jeong Kim , Kayly Choy , Kate Johnson , Miranda Kirby , Stephen Lam , Benjamin Smith , Mohsen Sadatsafavi doi: https://doi.org/10.1101/2025.09.10.25335529 Ramin Rezaeianzadeh 1 Respiratory Evaluation Sciences Program, Collaboration for Outcomes Research and Evaluation, Faculty of Pharmaceutical Sciences, University of British Columbia , Vancouver, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: raminrzn{at}student.ubc.ca Crystal leung 1 Respiratory Evaluation Sciences Program, Collaboration for Outcomes Research and Evaluation, Faculty of Pharmaceutical Sciences, University of British Columbia , Vancouver, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Soo Jeong Kim 1 Respiratory Evaluation Sciences Program, Collaboration for Outcomes Research and Evaluation, Faculty of Pharmaceutical Sciences, University of British Columbia , Vancouver, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kayly Choy 1 Respiratory Evaluation Sciences Program, Collaboration for Outcomes Research and Evaluation, Faculty of Pharmaceutical Sciences, University of British Columbia , Vancouver, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kate Johnson 2 Collaboration for Outcomes Research and Evaluation, Faculty of Pharmaceutical Sciences, University of British Columbia, Vancouver , Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Miranda Kirby 5 Department of Physics, Toronto Metropolitan University , Toronto, ON, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Stephen Lam 4 Division of Respirology, Department of Medicine , BC Cancer, Vancouver, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Benjamin Smith 3 Department of Medicine, McGill University , Montreal, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Mohsen Sadatsafavi 1 Respiratory Evaluation Sciences Program, Collaboration for Outcomes Research and Evaluation, Faculty of Pharmaceutical Sciences, University of British Columbia , Vancouver, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Background Lung cancer (LC) is the leading cause of cancer mortality, often diagnosed at advanced stages. Screening reduces mortality in high-risk individuals, but its efficiency can improve with pre- and post-screening risk stratification. With recent LC screening guideline updates in Europe and the US, numerous novel risk prediction models have emerged since the last systematic review of such models. We reviewed risk-based models for selecting candidates for CT screening, and post-CT stratification. Methods We systematically reviewed Embase and MEDLINE (2020–2024), identifying studies proposing new LC risk models for screening selection or nodule classification. Data extraction included study design, population, model type, risk horizon, and internal/external validation metrics. In addition, we performed an exploratory meta-regression of AUCs to assess whether sample size, model class, validation type, and biomarker use were associated with discrimination. Results Of 1987 records, 68 were included: 41 models were for screening selection (20 without biomarkers, 21 with), and 27 for nodule classification. Regression-based models predominated, though machine learning and deep learning approaches were increasingly common. Discrimination ranged from moderate (AUC≈0.70) to excellent (>0.90), with biomarker and imaging-enhanced models often outperforming traditional ones. Model calibration was inconsistently reported, and fewer than half underwent external validation. Meta-regression suggested that, among pre-screening models, larger sample sizes were modestly associated with higher AUC. Conclusion 75 models had been identified prior to 2020, we found 68 models since. This reflects growing interest in personalized LC screening. While many demonstrate strong discrimination, inconsistent calibration and limited external validation hinder clinical adoption. Future efforts should prioritize improving existing models rather than developing new ones, transparent evaluation, cost-effectiveness analysis, and real-world implementation. Introduction Lung cancer remains one of the most common, and one of the deadliest, malignancies worldwide, accounting for an estimated 11.6 percent of all new cancer diagnoses and 18.4 percent of cancer related deaths in 2018. 1 Diagnosis at an advanced stage drives dismal outcomes, with a five year survival of only about 3 percent, whereas detection at an early stage improves five year survival to approximately 62 percent. 2 Most lung cancer cases are diagnosed at later stages. Almost half (44.8%) diagnosed lung cancer cases from 2017 to 2021 were already spread to distant parts of the body. 34 Screening for lung cancer offers the possibility to shift the diagnosis stage towards earlier, more curable stages of its progression. 5 Large randomized trials have established low-dose computed tomography (LDCT) as the preferred screening tool for individuals at high risk. 6–13 In particular, the U.S. National Lung Screening Trial (NLST) demonstrated a 20 percent relative reduction in lung cancer mortality with LDCT versus chest radiography. 14 Accordingly, major guideline bodies now recommend annual LDCT screening for individuals exceeding certain smoking levels. In 2021, the U.S. Preventive Services Task Force (USPSTF) expanded its recommendations for screening to cover adults aged 50–80 years with a ≥ 20 pack-year smoking history who currently smoke or quit within the past 15 years. 15 Parallel efforts in Europe and elsewhere have followed: after trials such as NELSON 11 and MILD 8 , countries including Croatia 16 (nationwide since 2020), the Czech Republic (2022) 17 , Poland 18 (2021–22), and the United Kingdom 18 (2023) have adopted LDCT screening programs. In 2022, the European Commission urged all 27 EU member states to implement a “stepwise” approach to lung cancer screening. 19 Canada’s Task Force on Preventive Health Care similarly recommends screening adults aged 55–74 years with a ≥ 30 pack-year history for up to three consecutive years. 20 Despite these advances, current eligibility criteria still include many who will never develop lung cancer and exclude a substantial proportion of those who ultimately die from it. 21 , 22 To improve the efficiency of screening, risk-based strategies have been proposed that incorporate additional clinical, demographic, and radiologic factors. Such risk-based strategies are generally applied for two purposes: firstly, to identify individuals at highest baseline risk who stand to benefit most from LDCT screening, and secondly, to stratify malignancy risk in persons with pulmonary nodules detected on LDCT. There are no universally accepted risk scoring algorithms, and developing new algorithms is an active area of research. A 2020 systematic review identified 75 novel risk prediction models published until 2020. 23 Perhaps due to recent policy updates (e.g., the 2021 USPSTF guidelines), evolving regulatory frameworks in Europe and Asia, and the emergence of machine-learning– driven approaches, several new models have been proposed since the publication of the aforementioned systematic review. In light of the above, we have conducted an updated appraisal. The purpose of this work was to identify and describe novel risk prediction models developed across diverse jurisdictions since 2020, with a focus on appraising statistical methodology, incorporation of novel biomarkers, and various statistical strategies for predicting the risk of lung cancer among high-risk individuals. Methods We conducted a systematic literature search to identify relevant studies published between January 2020 and November 2024. This review updates a previous systematic review on risk-based lung cancer screening conducted by Toumazis et al. published in 2020. 23 The search strategy, developed in consultation with an in-house librarian, combined both controlled vocabulary and free-text terms, including “lung cancer screening”, “risk prediction”, “risk-based screening”, and “risk assessment”, ( Supplementary Table 1 ). We developed and documented our search strategy prior to execution, to ensure comprehensive retrieval of literature, across the databases used. Studies were eligible for inclusion if they met the following criteria: (1) the study proposed a novel lung cancer prediction model, (2) the full-text article was obtainable, and (3) the manuscript was written in English. Articles were excluded if any of the following applied: (1) the record was a review article, editorial, letter, or commentary, (2) the study was not published in a peer-reviewed journal, (3) the study was not conducted in a screening setting (such as studies involving only full-dose CT scans or invasive tissue collection for model training), or (4) the study examined lung cancer risk in populations with an existing underlying condition. View this table: View inline View popup Table S1: Search strategy The search was conducted in the Ovid Embase and Medline databases. We screened results in two stages. In the first stage, titles and abstracts were independently reviewed by two reviewers (RR and CL). In cases of disagreement, a third reviewer (MS) was consulted to resolve conflicts, ensuring a rigorous and unbiased study selection process. In the subsequent stage, potentially eligible studies underwent a full-text review using the same process for resolving discrepancies. Data extraction was carried out by one reviewer (RR) using an adapted version of a previously published extraction form 24 , and the extracted data were subsequently reviewed by two additional reviewers (CL and KC) for accuracy and consistency. Information extracted from each study included details regarding: (1) the risk factors incorporated into each lung cancer prediction model, (2) the study design, (3) the model type, (4) the targeted population, (5) the prediction type, and (6) key performance metrics. The performance metrics included measures of accuracy (sensitivity, specificity, predictive values at reported threshold values), calibration measures assessing the agreement between predicted probabilities and observed outcomes, and discrimination measures such as the area under the receiver operating characteristic (ROC) curve. Performance measures were only interpreted if they were from an external validation (assessed in a sample not used for training the model) or from an internal validation (based on the same data used to fit the model) after optimism correction (e.g., via bootstrapping or cross validation). We considered uncorrected estimates of model performance from internal validation uninterpretable. The data were summarized in tabulated form with accompanying narrative summaries to enable direct comparisons of predictive performance across models. We categorized models by the screening step they address, that is risk-based lung cancer screening models used to determine screening eligibility, without biomarkers and with biomarker data. Or post-screening malignancy risk prediction models for screen-detected pulmonary nodules. Meta-regression We conducted an exploratory meta-regression to explore heterogeneity among the included studies on estimates of discrimination. We extracted study-level data on area under the curve (AUC, or c-statistic) values with their standard errors (SE). For each AUC estimate, we recorded the following variables: sample size, prediction model category (regression [baseline]; machine-learning approaches; neural networks; all else were categorized as other), validation type (internal [baseline]; external), and, in pre–screening studies, whether biomarkers were included in the model. If a study reported multiple AUCs for different subgroups (e.g., in different validation samples), we included all of them as long as the samples did not overlap. To account for clustering of AUCs within each study, we conducted meta-regressions with robust variance estimation using the robumeta package. Given heterogeneity in model specifications and the time points at which AUC was measured, we did not report pooled AUC estimates by screening stage. All analyses were performed in R (version 4.4.2). Results Of the 1,987 records identified through database searching (1,344 from Embase and 643 from MEDLINE), 379 duplicates (one manually and 378 by Covidence) were removed, leaving 1,608 unique records for title and abstract screening. Of these, 1,389 were excluded, and the remaining 217 articles underwent full-text retrieval (all were successfully retrieved). Full-text assessment led to exclusion of 148 studies. 68 studies met inclusion criteria and were incorporated into the review (Figure 1). These studies reported on 41 new models on risk prediction models before screening, of which 21 included a biomarker. 27 new models were developed to predict the risk of malignancy in identified lung nodules after screening. Download figure Open in new tab Table 1 provides the list of new risk prediction models for lung cancer risk for screening selection without incorporating a biomarker. The left-most column provides a shorthand label for each model to facilitate referral to the table in the text below. View this table: View inline View popup Table 1. Risk-based lung cancer screening models not incorporating biomarkers. These models were developed and validated across diverse populations, including Asian ever-smokers, general populations in the UK and China, female non-smokers. Samples also varied and included clinical trials and health-system registries; prediction methodology involved both traditional regression and machine-learning approaches. Of the 20 risk models, six followed prospective designs (e.g. Shanghai-LCM, Guo 2021, OWL), ten had retrospective designs (e.g. Levi, CanPredict, Guo 2023), two used combined sets of data (ALIGNED, UCL-I), one used a case-control design (LLP-V3), and one analyzed trial data (Survival Stacking Ensemble). Regression-based approaches predominated in model development: eleven studies fitted Cox proportional-hazards models (e.g.: Shanghai-LCM, CanPredict, Jung, Park), two used logistic regression (LLPv3, ALIGNED) and one linear regression (Kent & Medway Lung Cancer Risk tool). Machine-learning methods were applied in four studies, ensemble trees (Levi), a mixed ensemble (UCL-I), machine learning (Chandran) and a neural network (Yeh), and one study built stacked Cox ensembles on trial data (Survival Stacking Ensemble). Optimism-corrected internal values of discrimination were reported for almost all models and spanned from moderate to excellent: the neural-network model by Yeh reached the highest c-statistic (0.902), while and the regression-based CanPredict had similar values (0.897 in women and 0.904 in men); most Cox or logistic models clustered between 0.75 and 0.83, for example Shanghai-LCM 0.78, LLPv3 0.81, UCL-I 0.81, Yeo 0.83 and Park 0.82, while Ma, NCC-LCm and the survival-stacking ensemble lay in the mid-0.70s. By contrast, quantitative metrics related to internal calibration appeared only in five studies. Calibration slopes were close to unity in Shanghai-LCM (0.98), NCC-LCm (0.97), LLPv3 (0.90) and Park (E/O 1.00); UCL-I reported an E/O ratio of 1.0 but a slope of 1.35, indicating some mis-calibration, whereas Shanghai-LCM’s E/O ratio of 0.70 suggested under-prediction The remaining models stated that calibration was inspected graphically. Overall, discrimination was generally acceptable to strong across models and formal internal calibration assessment was often lacking. External discrimination metrics were available for ten models. Discrimination (C-statistic/AUC), and were generally lower than corresponding values from internal validation. They ranged from 0.70 for the stacked Cox ensemble tested in NLST to 0.88 for CanPredict in men and 0.87 in women; intermediate values were 0.70 for Shanghai-LCM, 0.67–0.75 for NCC-LCm, 0.74 for ALIGNED, 0.76–0.77 for Ma, 0.78 for UCL-I on PLCO, 0.82 for Park, and 0.72–0.81 for Chandran across two health-care datasets. Quantitative metrics of calibration appeared in only three studies: NCC-LCm reported slopes 0.95–0.97 and E/O ratios 0.95–1.04, UCL-I an E/O ratio of 1.00, and Park an E/O of 0.99; other models offered graphical assessments or none. Overall, regression-based models achieved moderate to excellent discrimination (C-statistics 0.71–0.90) with generally good internal calibration when evaluated. Machine-learning approaches performed comparably, with some models (Yeh neural network, CanPredict) surpassing regression-based methods in AUC. Nevertheless, external validation was unevenly applied yet usually confirmed reasonably preserved discrimination. ThresholdLbased sensitivity and specificity varied according to chosen cut-points. Among the 21 screening models that incorporated biomarkers ( Table 2 ), nine used prospective designs, six case-control, two in retrospective, two used pooled data, and one analysis of trial data. Logistic regression was the most common modelling approach (8 models); four used time-to-event Cox-based methods, six studies applied machine-learning algorithms such as XGBoost, Random Forest or multi-algorithm ensembles; and one employed a flexible Royston–Parmar survival model. As expected, logistic regression predominated in case-control and nested case-control designs, whereas machine-learning and survival models were mainly used in prospective designs. View this table: View inline View popup Table 2. Risk-based lung cancer screening models incorporating biomarker data. Internal discrimination varied by approach. MachineLlearning models frequently achieved the highest values, Siqi Zhang’s XGBoost and Random Forest reported training AUCs of 0.99–1.000, though overfitting remains a concern. RegressionLbased models typically achieved AUCs or C-statistics between 0.80 and 0.86 (e.g: Gould’s MES AUC 0.86; Fahrmann’s four-protein panel AUC 0.85). Polygenic risk scores alone performed modestly (AUCs ∼0.58–0.62) but improved when combined with clinical variables (e.g., Wei CKB model AUC 0.71). Calibration metrics were less consistently reported: Nguyen et al. (HUNT lung – SNP Model) reported average O/E ratios of 0.87 (discovery) and 0.96 (validation), whereas most others provided only graphical calibration plots or Hosmer–Lemeshow tests. External validation was reported for six models, and again were lower than internal validation values (MetaPRS, Hunt Lung SNP, LunaCAM-S, TNSF-SQ, Jacobsen et al., and Yao et al.). Discrimination varied from a C-statistics of 0.58 for the MetaPRS score to an AUC 0.916 for the HUNT Lung-SNP model, Yao metabolomic panel AUC 0.945, LunaCAM-S AUC 0.901, TNSF-SQ AUC 0.714, and Jacobsen AUC 0.74. Two models exceeded an AUC of 0.90 (HUNT Lung-SNP, Yao et al., LunaCAM-S). Calibration was seldom quantifiedin external validation; Nguyen reported an O/E ratio of 0.96. Heterogeneity in datasets, metrics, and reporting limits direct comparison across models. Overall, Internal discrimination covered a broad span, from an AUC 0.591 for a 19-variant polygenic score to 1.00 for a random-forest model in UK Biobank, with most models lying between 0.80 and 0.88. For the six models that reported external validation, C-index/AUC values ranged from 0.58 (MetaPRS) to 0.95 (metabolomic panel by Yao), and three exceeded 0.90. Calibration was rarely quantified: Nguyen reported an observed-to-expected ratio of 0.96, while most other studies provided only graphical checks. Reported operating points showed wide variation, sensitivity spanned 15.3–29.7% for TNSF-SQ (specificity ≥ 94.7 %) to 90.6 % for a ctDNA methylation panel (specificity 61.9 %), with the metabolomic model achieving 81.0 % sensitivity at 97.9 % specificity. Table 3 reports on the 27 models in that predicted malignancy risk in screen-detected pulmonary nodules. Among these, analyses of lung-cancer screening trials dominated, with fourteen having analyzed trial data, namely LCi; Brock & LCP-CNN; S-RRL/B-RRL; Lu; qADD++; Rundo; Sybil; Venkadesh 1; Venkadesh 2; Kaiwen Xu; Trajanovski; Ebrahimpour; Paul; and CXR-LC. Seven models stem used retrospective designs (Huijie, Kaeum, Chang, Yao, cLung-RADS, Jieke Liu and Gao). Two had combined data from multiple sources (Dahai Liu and INTEGRAL-Radiomics), and three used prospective designs (LCP-CNN, Xiao and Wu). View this table: View inline View popup Table 3. Malignancy risk prediction models for screen-detected pulmonary nodules. Analytically, conventional Cox or (regularised) logistic regression were used in 10 models; LCi, Kaeum, Chang, Yao, Lu, Rundo, Jieke Liu, Xiao, Wu and Kaiwen Xu. 16 studies used machine-learning approaches, nine of them convolutional neural networks (Brock/LCP-CNN, the Huijie-CNN, CXR-LC, Venkadesh 1, DL-based cLung-RADS, Dahai Liu, the prospective LCP-CNN cohort, Sybil and Venkadesh 2). Seven machine-learning models relied on reinforcement learning, gradient boosting, support-vector machines or ensemble trees: S-RRL/B-RRL (radiomics RL), INTEGRAL-Radiomics (XGBoost/Random Forest), qADD++ (linear SVM), Ebrahimpour (gradient boosting), Gao (feed-forward NN), Trajanovski (projection NN) and Paul (radiomic/tabular NN). Internal discrimination across models ranged from moderate to high: the LCi Cox model achieved AUCs of 93.7% at one year and 84.0% at five years, whereas the CXR-LC CNN yielded AUCs of 0.73 for six-year incidence and 0.76 for twelve-year incidence. Calibration also varied: the LCi model exhibited calibration slopes below unity (1-year slope = 0.80), suggesting slight risk overestimation, while the CXR-LC model’s O/E ratio was 1.01 (95% CI 0.87–1.14). The INTEGRAL-Radiomics model showed an E/O ratio of 1.02 with a calibration-in-the-large (CITL) of 0.69, alongside a cross-validated AUC of 0.93. In contrast, the Yao logistic model demonstrated perfect internal calibration (E/O = 1.00, CITL = 0.00, calibration slope = 1.00) together with an AUC of 0.91. Overall, several models combined strong discrimination (AUCs ≥ 0.91) with acceptable calibration, though calibration slopes and E/O ratios indicate varying degrees of risk estimation accuracy. External discrimination AUCs varied across models: the LCi model achieved 0.90 at 1-year, 0.81 at 3-years and 0.80 at 5-years; the CXR-LC model showed 6-year incidence AUCs of 0.66; Venkadesh1 yielded 0.93; Dahai Liu reported 0.88; LCP-CNN reached 0.835 at VUMC and 0.92 at OUH; Sybil recorded 0.81 at MGH and 0.80 at CGMH; Venkadesh2 achieved 0.97 using current + prior CT and 0.89 for the latest CT; Jieke Liu’s radiomic signature and nomogram both gave 0.98; Ebrahimpour obtained 0.76; Gao reported 0.91; Xiao 0.81; and Wu 0.75. In aggregate, models incorporating quantitative imaging, whether via machine learning or deep-learning feature extraction, achieved very good to excellent discrimination, frequently surpassing AUC of 0.90 internally and maintaining AUC ≥ 0.80 upon external testing. Regression-based models based on large trial data also demonstrated robust discrimination (AUC ∼0.85–0.90) with reliable calibration metrics. Reported sensitivities (61–96%) and specificities (61–98%) reflected differing clinical priorities. The typical decline in discrimination and calibration on external validation underscores the problem of model fitting and the need for local recalibration. The relative stability of the performance of deep-learning algorithms across diverse cohorts is promising for broader clinical deployment, contingent upon rigorous validation in real-world settings. Meta-regression analysis of discriminatory performance Table 4 presents the results of meta-regression of discriminatory performance for pre-screening models (24 clusters; 81 AUC values). Larger sample sizes were associated with slightly higher AUC (p = .030). Machine-learning models showed a non-significant increase of 0.037 in AUC compared to regression models (p = 0.307); external validation was associated with a non-significant change of –0.012 (p = 0.439), and inclusion of biomarkers conferred a non-significant difference of +0.033 (p = 0.337). Regression estimates are presented in supplementary Table S2 . View this table: View inline View popup Table S2. RVE Hierarchical Effects Model (Pre-screening) View this table: View inline View popup Download powerpoint Table 4: Characteristics of studies at the pre-screening stage (including meta-regression P-values) Table 5 presents the post–screening meta-regression results (19 clusters; 62 effect sizes). Here, the association between c-statistic and sample size was not statistically significant (p = 0.759). We observed the general trends in the direction of other associations, but none reached statistical significance: machine-learning (other than neural networks) v. regression (β = +0.035; p = 0.476), neural networks v. regression (β = +0.047; p = 0.415), external v. internal validation (β = –0.001; p = 0.957). Regression estimates are presented in supplementary Table S3 . View this table: View inline View popup Download powerpoint Table S3. RVE Hierarchical Effects Model (Post-screening) View this table: View inline View popup Download powerpoint Table 5: Characteristics of studies at the post-screening stage (including meta-regression P-values) Discussion This systematic review identified 68 new risk prediction models for lung cancer screening developed since 2020. Across all categories, discrimination metrics (C-statistics or AUC) generally ranged from moderate (≈0.70) to excellent (>0.90), with post-screening machine-learning and deep-learning approaches achieving the highest values. Calibration, when formally evaluated, was typically good for regression-based models but underreported for many complex algorithms. External validation was inconsistently applied. Available results suggest attenuation and discrimination, which is a typical manifestation of the model being overfitted in their original sample, as well as drift in calibration. The latter can also be a manifestation of overfitting, as well as differences across populations and sampling methods (e.g. trial-based data for development, and electronic health records for validation). Screen-selection models without biomarkers were trained on routine demographic and smoking variables; most used classical Cox or logistic regression, few used non-image ML methods (e.g., gradient-boosted trees, random forests), and only one employed neural networks. In models incorporating biomarkers, inputs typically included laboratory, protein, genetic, or multi-omic markers; most used regression, some used ML, and CNNs were not used. For nodule-malignancy models, only a small subset supplement CT features with blood biomarkers, whereas the majority combine imaging with basic clinical data; approaches span regression, non-CNN ML, and CT-based CNNs. Our meta-regression showed in pre–screening models, larger sample sizes were associated with modest but statistically significant improvements in discrimination, while the use of machine-learning algorithms, external validation, and biomarker inclusion showed favorable point estimates but did not achieve statistical significance. In post–screening models, although the baseline discrimination was substantially higher, no model feature, including sample size or advanced algorithmic approaches, emerged as a significant predictor of performance. For time-to-event data, AUCs are time-dependent and were reported inconsistently across studies, harmonization and pooling were not feasible, thus limiting our meta-regression. These findings suggest that while sample size plays a role in model discrimination early in the screening process, other factors such as model complexity and validation strategy may have more nuanced effects. The effect size s in absolute terms were small (albeit this is typical for the c-statistic 93 ), and the confidence bounds were wide. These findings can be used in hypothesis-generation. They may point toward patterns worth testing in future research or larger datasets, but are not confirmatory evidence of an effect. These models are developed across diverse populations: in ever-smokers and never-smokers, across Asian and Western cohorts, and in both community screenings and clinical trial settings. Prospective cohort designs enhanced the generalizability and statistical power of several models. However, fewer than half of the models in each category underwent robust external validation, and many relied on retrospective or case–control designs that are prone to selection bias. Calibration reporting was uneven, with many machine-learning and deep-learning studies omitting calibration entirely. With 68 new models since 2020, a substantial increase, efforts should shift from developing additional algorithms to validating those already available. Performing head-to-head comparisons of regression, machine-learning, and deep-learning approaches within the same cohorts can elucidate not only differences in accuracy, but also ease of use and acceptance by individuals, providers, and communities. In addition, future studies can assess cost-effectiveness by examining how biomarkers and imaging algorithms influence resource use, program costs, and patient outcomes. Finally, carrying out implementation studies to explore how these tools integrate into screening settings, and developing strategies for sharing individualized risk with patients. These steps will help move lung cancer risk models from development into practice. Our review was limited by heterogeneity in model reporting: differing follow-up horizons, risk thresholds, and performance metrics complicated direct comparisons. Publication bias may favor models with higher apparent performance. Finally, rapidly evolving machine-learning methods underscore the need for continual evidence synthesis. Conclusions A wide spectrum of lung cancer risk models now exists, from simple equations to sophisticated biomarker- and image-based algorithms, and with methodology varying from conventional risk equations to sophisticated machine learning algorithms. While many tools seem to demonstrate strong discrimination, the relative paucity of rigorous external validation and formal calibration reporting hampers clinical adoption. Future efforts should emphasize transparent, standardized evaluation across populations, economic evaluations, and implementation studies to realize the potential of personalized risk prediction in lung cancer screening and nodule management. Data Availability All data produced in the present work are contained in the manuscript Supplementary Material Download figure Open in new tab Figure S1 Forest plot of study-level pooled AUCs (95 % CI) by model type and screening stage Footnotes Authors’ contribution: Sources of support: This study was funded by a Project Grant from The Canadian Institutes of Health Research. References 1. ↵ Bray F , Ferlay J , Soerjomataram I , Siegel RL , Torre LA , Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries . CA: A Cancer Journal for Clinicians . 2018 ; 68 ( 6 ): 394 – 424 . doi: 10.3322/caac.21492 OpenUrl CrossRef PubMed 2. ↵ cancer CCS/ S canadienne du . Survival statistics for non–small cell lung cancer. Canadian Cancer Society . August 2023 . Accessed April 8, 2025. https://cancer.ca/en/cancer-information/cancer-types/lung/prognosis-and-survival/non-small-cell-lung-cancer-survival-statistics 3. CDC . U.S. Cancer Statistics Lung Cancer Stat Bite. United States Cancer Statistics . June 5, 2024 . Accessed May 22, 2025. https://www.cdc.gov/united-states-cancer-statistics/publications/lung-cancer-stat-bite.html 4. Nations JA , Brown DW , Shao S , Shriver CD , Zhu K . Comparative Trends in the Distribution of Lung Cancer Stage at Diagnosis in the Department of Defense Cancer Registry and the Surveillance, Epidemiology, and End Results data, 1989-2012 . Mil Med . 2020 ; 185 ( 11 -12): e2044 - e2048 . doi: 10.1093/milmed/usaa218 OpenUrl CrossRef PubMed 5. ↵ National Lung Screening Trial Research Team , Aberle DR , Adams AM , et al. Reduced lung-cancer mortality with low-dose computed tomographic screening . N Engl J Med . 2011 ; 365 ( 5 ): 395 – 409 . doi: 10.1056/NEJMoa1102873 OpenUrl CrossRef PubMed Web of Science 6. The National Lung Screening Trial : Overview and Study Design1 . Radiology . 2011 ; 258 ( 1 ): 243 – 253 . doi: 10.1148/radiol.10091808 OpenUrl CrossRef PubMed Web of Science 7. Wille MMW , Dirksen A , Ashraf H , et al. Results of the Randomized Danish Lung Cancer Screening Trial with Focus on High-Risk Profiling . Am J Respir Crit Care Med . 2016 ; 193 ( 5 ): 542 – 551 . doi: 10.1164/rccm.201505-1040OC OpenUrl CrossRef PubMed 8. ↵ Pastorino U , Rossi M , Rosato V , et al. Annual or biennial CT screening versus observation in heavy smokers: 5-year results of the MILD trial . Eur J Cancer Prev . 2012 ; 21 ( 3 ): 308 – 315 . doi: 10.1097/CEJ.0b013e328351e1b6 OpenUrl CrossRef PubMed 9. Paci E , Puliti D , Lopes Pegna A , et al. Mortality, survival and incidence rates in the ITALUNG randomised lung cancer screening trial . Thorax . 2017 ; 72 ( 9 ): 825 – 831 . doi: 10.1136/thoraxjnl-2016-209825 OpenUrl Abstract / FREE Full Text 10. Infante M , Cavuto S , Lutman FR , et al. Long-Term Follow-up Results of the DANTE Trial, a Randomized Study of Lung Cancer Screening with Spiral Computed Tomography . Am J Respir Crit Care Med . 2015 ; 191 ( 10 ): 1166 – 1175 . doi: 10.1164/rccm.201408-1475OC OpenUrl CrossRef PubMed 11. ↵ Horeweg N , van der Aalst CM , Vliegenthart R , et al. Volumetric computed tomography screening for lung cancer: three rounds of the NELSON trial . Eur Respir J . 2013 ; 42 ( 6 ): 1659 – 1667 . doi: 10.1183/09031936.00197712 OpenUrl Abstract / FREE Full Text 12. Gohagan JK , Marcus PM , Fagerstrom RM , et al. Final results of the Lung Screening Study, a randomized feasibility study of spiral CT versus chest X-ray screening for lung cancer . Lung Cancer . 2005 ; 47 ( 1 ): 9 – 15 . doi: 10.1016/j.lungcan.2004.06.007 OpenUrl CrossRef PubMed Web of Science 13. Blanchon T , Bréchot JM , Grenier PA , et al. Baseline results of the Depiscan study: a French randomized pilot trial of lung cancer screening comparing low dose CT scan (LDCT) and chest X-ray (CXR) . Lung Cancer . 2007 ; 58 ( 1 ): 50 – 58 . doi: 10.1016/j.lungcan.2007.05.009 OpenUrl CrossRef PubMed Web of Science 14. ↵ null null . Reduced Lung-Cancer Mortality with Low-Dose Computed Tomographic Screening . New England Journal of Medicine . 2011 ; 365 ( 5 ): 395 – 409 . doi: 10.1056/NEJMoa1102873 OpenUrl CrossRef PubMed Web of Science 15. ↵ Darden ME , Hoagland A. Lung Cancer Screening and USPSTF Recommendations . JAMA Network Open . 2025 ; 8 ( 2 ): e2458916 . doi: 10.1001/jamanetworkopen.2024.58916 OpenUrl CrossRef 16. ↵ Arnold H. Paving the way for lung cancer screening: learning from Central and Eastern Europe. The Lung Cancer Policy Network . October 2, 2024 . Accessed April 21, 2025. https://www.lungcancerpolicynetwork.com/paving-the-way-for-lung-cancer-screening-learning-from-central-and-eastern-europe/ 17. ↵ Borilova S , Dusek L , Jakubikova L , et al. Lung Cancer in the Czech Republic . Journal of Thoracic Oncology . 2023 ; 18 ( 3 ): 271 – 277 . doi: 10.1016/j.jtho.2022.11.023 OpenUrl CrossRef 18. ↵ Wait S , Alvarez-Rosete A , Osama T , et al. Implementing Lung Cancer Screening in Europe: Taking a Systems Approach . JTO Clin Res Rep . 2022 ; 3 ( 5 ): 100329 . doi: 10.1016/j.jtocrr.2022.100329 OpenUrl CrossRef PubMed 19. ↵ Ward B , Vašáková MK , Cordeiro CR , et al. Important steps towards a big change for lung health: a joint approach by the European Respiratory Society, the European Society of Radiology and their partners to facilitate implementation of the European Union’s new recommendations on lung cancer screening . ERJ Open Research . 2023 ; 9 ( 3 ). doi: 10.1183/23120541.00026-2023 OpenUrl Abstract / FREE Full Text 20. ↵ Care CTF on PH . Recommendations on screening for lung cancer . CMAJ . 2016 ; 188 ( 6 ): 425 – 432 . doi: 10.1503/cmaj.151421 OpenUrl FREE Full Text 21. ↵ Pinsky PF , Berg CD . Applying the National Lung Screening Trial eligibility criteria to the US population: what percent of the population and of incident lung cancers would be covered? J Med Screen . 2012 ; 19 ( 3 ): 154 – 156 . doi: 10.1258/jms.2012.012010 OpenUrl CrossRef PubMed 22. ↵ Gracie K , Kennedy MPT , Esterbrook G , et al. The proportion of lung cancer patients attending UK lung cancer clinics who would have been eligible for low-dose CT screening . Eur Respir J . 2019 ; 54 ( 2 ): 1802221 . doi: 10.1183/13993003.02221-2018 OpenUrl Abstract / FREE Full Text 23. ↵ Toumazis I , Bastani M , Han SS , Plevritis SK . Risk-Based lung cancer screening: A systematic review . Lung Cancer . 2020 ; 147 : 154 – 186 . doi: 10.1016/j.lungcan.2020.07.007 OpenUrl CrossRef PubMed 24. ↵ Fernandez-Felix BM , López-Alcalde J , Roqué M , Muriel A , Zamora J . CHARMS and PROBAST at your fingertips: a template for data extraction and risk of bias assessment in systematic reviews of predictive models . BMC Medical Research Methodology . 2023 ; 23 ( 1 ): 44 . doi: 10.1186/s12874-023-01849-0 OpenUrl CrossRef PubMed 25. Yang JJ , Wen W , Zahed H , et al. Lung Cancer Risk Prediction Models for Asian Ever-Smokers . J Thorac Oncol . 2024 ; 19 ( 3 ): 451 – 464 . doi: 10.1016/j.jtho.2023.11.002 OpenUrl CrossRef PubMed 26. Levi M , Lazebnik T , Kushnir S , Yosef N , Shlomi D . Machine learning computational model to predict lung cancer using electronic medical records . Cancer Epidemiol . 2024 ; 92 : 102631 . doi: 10.1016/j.canep.2024.102631 OpenUrl CrossRef 27. Guo LW , Lyu ZY , Meng QC , et al. A risk prediction model for selecting high-risk population for computed tomography lung cancer screening in China . Lung Cancer . 2022 ; 163 : 27 – 34 . doi: 10.1016/j.lungcan.2021.11.015 OpenUrl CrossRef PubMed 28. Pan Z , Zhang R , Shen S , et al. OWL: an optimized and independently validated machine learning prediction model for lung cancer screening based on the UK Biobank, PLCO, and NLST populations . EBioMedicine . 2023 ; 88 : 104443 . doi: 10.1016/j.ebiom.2023.104443 OpenUrl CrossRef 29. Liao W , Coupland CAC , Burchardt J , et al. Predicting the future risk of lung cancer: development, and internal and external validation of the CanPredict (lung) model in 19·67 million people and evaluation of model performance against seven other risk prediction models . The Lancet Respiratory Medicine . 2023 ; 11 ( 8 ): 685 – 697 . doi: 10.1016/S2213-2600(23)00050-4 OpenUrl CrossRef PubMed 30. Guo L , Meng Q , Zheng L , et al. Lung Cancer Risk Prediction Nomogram in Nonsmoking Chinese Women: Retrospective Cross-sectional Cohort Study . JMIR Public Health Surveill . 2023 ; 9 : e41640 . doi: 10.2196/41640 OpenUrl CrossRef 31. Howell D , Buttery R , Badrinath P , et al. Developing a risk prediction tool for lung cancer in Kent and Medway , England: cohort study using linked data. BJC Rep . 2023 ; 1 ( 1 ): 16 . doi: 10.1038/s44276-023-00019-5 OpenUrl CrossRef PubMed 32. Chandran U , Reps J , Yang R , Vachani A , Maldonado F , Kalsekar I . Machine Learning and Real-World Data to Predict Lung Cancer Risk in Routine Care . Cancer Epidemiol Biomarkers Prev . 2023 ; 32 ( 3 ): 337 – 343 . doi: 10.1158/1055-9965.EPI-22-0873 OpenUrl CrossRef PubMed 33. Ma Z , Lv J , Zhu M , et al. Lung cancer risk score for ever and never smokers in China . Cancer Commun (Lond ) . 2023 ; 43 ( 8 ): 877 – 895 . doi: 10.1002/cac2.12463 OpenUrl CrossRef PubMed 34. ↵ Wang F , Tan F , Shen S , et al. Risk-stratified Approach for Never- and Ever-Smokers in Lung Cancer Screening: A Prospective Cohort Study in China . Am J Respir Crit Care Med . 2023 ; 207 ( 1 ): 77 – 88 . doi: 10.1164/rccm.202204-0727OC OpenUrl CrossRef PubMed 35. Yeo Y , Shin DW , Han K , et al. Individual 5-Year Lung Cancer Risk Prediction Model in Korea Using a Nationwide Representative Database . Cancers (Basel ) . 2021 ; 13 ( 14 ): 3496 . doi: 10.3390/cancers13143496 OpenUrl CrossRef PubMed 36. Field JK , Vulkan D , Davies MPA , Duffy SW , Gabe R . Liverpool Lung Project lung cancer risk stratification model: calibration and prospective validation . Thorax . 2021 ; 76 ( 2 ): 161 – 168 . doi: 10.1136/thoraxjnl-2020-215158 OpenUrl Abstract / FREE Full Text 37. Jani BD , Sullivan MK , Hanlon P , et al. Personalised lung cancer risk stratification and lung cancer screening: do general practice electronic medical records have a role? Br J Cancer . 2023 ; 129 ( 12 ): 1968 – 1977 . doi: 10.1038/s41416-023-02467-9 OpenUrl CrossRef PubMed 38. Chien LH , Chen TY , Chen CH , et al. Recalibrating Risk Prediction Models by Synthesizing Data Sources: Adapting the Lung Cancer PLCO Model for Taiwan . Cancer Epidemiol Biomarkers Prev . 2022 ; 31 ( 12 ): 2208 – 2218 . doi: 10.1158/1055-9965.EPI-22-0281 OpenUrl CrossRef PubMed 39. Jung AW , Holm PC , Gaurav K , et al. Multi-cancer risk stratification based on national health data: a retrospective modelling and validation study . The Lancet Digital Health . 2024 ; 6 ( 6 ): e396 – e406 . doi: 10.1016/S2589-7500(24)00062-1 OpenUrl CrossRef 40. Callender T , Imrie F , Cebere B , et al. Assessing eligibility for lung cancer screening using parsimonious ensemble machine learning models: A development and validation study . PLOS Medicine . 2023 ; 20 ( 10 ): e1004287 . doi: 10.1371/journal.pmed.1004287 OpenUrl CrossRef 41. Yeh MCH , Wang YH , Yang HC , Bai KJ , Wang HH , Li YCJ . Artificial Intelligence-Based Prediction of Lung Cancer Risk Using Nonimaging Electronic Medical Records: Deep Learning Approach . J Med Internet Res . 2021 ; 23 ( 8 ): e26256 . doi: 10.2196/26256 OpenUrl CrossRef PubMed 42. Guo LW , Lyu ZY , Meng QC , et al. Construction and Validation of a Lung Cancer Risk Prediction Model for Non-Smokers in China . Front Oncol . 2022 ; 11 : 766939 . doi: 10.3389/fonc.2021.766939 OpenUrl CrossRef 43. Alonso E , Calle X , Gurrutxaga I , Beristain A . Survival Stacking Ensemble Model for Lung Cancer Risk Prediction . Stud Health Technol Inform . 2024 ; 321 : 155 – 159 . doi: 10.3233/SHTI241083 OpenUrl CrossRef PubMed 44. Park B , Kim Y , Lee J , Lee N , Jang SH . Risk-based prediction model for selecting eligible population for lung cancer screening among ever smokers in Korea . Transl Lung Cancer Res . 2021 ; 10 ( 12 ): 4390 – 4402 . doi: 10.21037/tlcr-21-566 OpenUrl CrossRef PubMed 45. Gould MK , Huang BZ , Tammemagi MC , Kinar Y , Shiff R . Machine Learning for Early Lung Cancer Identification Using Routine Clinical and Laboratory Data . Am J Respir Crit Care Med . 2021 ; 204 ( 4 ): 445 – 453 . doi: 10.1164/rccm.202007-2791OC OpenUrl CrossRef PubMed 46. Fahrmann JF , Marsh T , Irajizad E , et al. Blood-Based Biomarker Panel for Personalized Lung Cancer Risk Assessment . J Clin Oncol . 2022 ; 40 ( 8 ): 876 – 883 . doi: 10.1200/JCO.21.01460 OpenUrl CrossRef PubMed 47. Zhang S , Yang L , Xu W , et al. Predicting the risk of lung cancer using machine learning: A large study based on UK Biobank . Medicine (Baltimore ) . 2024 ; 103 ( 16 ): e37879 . doi: 10.1097/MD.0000000000037879 OpenUrl CrossRef PubMed 48. 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 OpenUrl CrossRef PubMed 49. Cortés-Ibáñez FO , Johnson T , Mascalchi M , Katzke V , Delorme S , Kaaks R . Serum-based biomarkers associated with lung cancer risk and cause-specific mortality in the German randomized Lung Cancer Screening Intervention (LUSI) trial . Transl Lung Cancer Res . 2023 ; 12 ( 12 ): 2460 – 2475 . doi: 10.21037/tlcr-23-548 OpenUrl CrossRef PubMed 50. Horsfall LJ , Clarke CS , Nazareth I , Ambler G . The value of blood-based measures of liver function and urate in lung cancer risk prediction: A cohort study and health economic analysis . Cancer Epidemiol . 2023 ; 84 : 102354 . doi: 10.1016/j.canep.2023.102354 OpenUrl CrossRef 51. Nguyen OTD , Fotopoulos I , Nøst TH , et al. The HUNT lung-SNP model: genetic variants plus clinical variables improve lung cancer risk assessment over clinical models . J Cancer Res Clin Oncol . 2024 ; 150 ( 8 ): 389 . doi: 10.1007/s00432-024-05909-w OpenUrl CrossRef PubMed 52. Chen A , Wu E , Huang R , et al. Development of Lung Cancer Risk Prediction Machine Learning Models for Equitable Learning Health System: Retrospective Study . JMIR AI . 2024 ; 3 : e56590 . doi: 10.2196/56590 OpenUrl CrossRef 53. Wang Z , Xie K , Zhu G , et al. Early detection and stratification of lung cancer aided by a cost-effective assay targeting circulating tumor DNA (ctDNA) methylation . Respir Res . 2023 ; 24 ( 1 ): 163 . doi: 10.1186/s12931-023-02449-8 OpenUrl CrossRef PubMed 54. Yu H , Raut JR , Schöttker B , Holleczek B , Zhang Y , Brenner H . Individual and joint contributions of genetic and methylation risk scores for enhancing lung cancer risk stratification: data from a population-based cohort in Germany . Clinical Epigenetics . 2020 ; 12 ( 1 ): 89 . doi: 10.1186/s13148-020-00872-y OpenUrl CrossRef PubMed 55. Tang Y , You D , Yi H , Yang S , Zhao Y . IPRS: Leveraging Gene-Environment Interaction to Reconstruct Polygenic Risk Score . Front Genet . 2022 ; 13 : 801397 . doi: 10.3389/fgene.2022.801397 OpenUrl CrossRef 56. Chien LH , Chen CH , Chen TY , et al. Predicting Lung Cancer Occurrence in Never-Smoking Females in Asia: TNSF-SQ, a Prediction Model . Cancer Epidemiol Biomarkers Prev . 2020 ; 29 ( 2 ): 452 – 459 . doi: 10.1158/1055-9965.EPI-19-1221 OpenUrl Abstract / FREE Full Text 57. Hung RJ , Warkentin MT , Brhane Y , et al. Assessing Lung Cancer Absolute Risk Trajectory Based on a Polygenic Risk Model . Cancer Res . 2021 ; 81 ( 6 ): 1607 – 1615 . doi: 10.1158/0008-5472.CAN-20-1237 OpenUrl Abstract / FREE Full Text 58. Bhardwaj M , Schöttker B , Holleczek B , Benner A , Schrotz-King P , Brenner H . Potential of Inflammatory Protein Signatures for Enhanced Selection of People for Lung Cancer Screening . Cancers (Basel ) . 2022 ; 14 ( 9 ): 2146 . doi: 10.3390/cancers14092146 OpenUrl CrossRef PubMed 59. Zhang S , Yang F , Wang L , Si S , Zhang J , Xue F . Personalized prediction for multiple chronic diseases by developing the multi-task Cox learning model . PLoS Comput Biol . 2023 ; 19 ( 9 ): e1011396 . doi: 10.1371/journal.pcbi.1011396 OpenUrl CrossRef PubMed 60. Jia G , Lu Y , Wen W , et al. Evaluating the Utility of Polygenic Risk Scores in Identifying High-Risk Individuals for Eight Common Cancers . JNCI Cancer Spectr . 2020 ; 4 ( 3 ): pkaa021 . doi: 10.1093/jncics/pkaa021 OpenUrl CrossRef 61. Jacobsen KK , Kobylecki CJ , Skov-Jeppesen SM , Bojesen SE . Development and validation of a simple general population lung cancer risk model including AHRR-methylation . Lung Cancer . 2023 ; 181 : 107229 . doi: 10.1016/j.lungcan.2023.107229 OpenUrl CrossRef 62. Lian J , Vardhanabhuti V . Metabolic biomarkers using nuclear magnetic resonance metabolomics assay for the prediction of aging-related disease risk and mortality: a prospective, longitudinal, observational, cohort study based on the UK Biobank . Geroscience . 2024 ; 46 ( 2 ): 1515 – 1526 . doi: 10.1007/s11357-023-00918-y OpenUrl CrossRef PubMed 63. Wei X , Sun D , Gao J , et al. Development and evaluation of a polygenic risk score for lung cancer in never-smoking women: A large-scale prospective Chinese cohort study . Int J Cancer . 2024 ; 154 ( 5 ): 807 – 815 . doi: 10.1002/ijc.34765 OpenUrl CrossRef PubMed 64. Yao Y , Wang X , Guan J , et al. Metabolomic differentiation of benign vs malignant pulmonary nodules with high specificity via high-resolution mass spectrometry analysis of patient sera . Nat Commun . 2023 ; 14 ( 1 ): 2339 . doi: 10.1038/s41467-023-37875-1 OpenUrl CrossRef PubMed 65. Wang N , Yao C , Luo C , et al. Integrated plasma and exosome long noncoding RNA profiling is promising for diagnosing non-small cell lung cancer . Clinical Chemistry and Laboratory Medicine (CCLM ) . 2023 ; 61 ( 12 ): 2216 – 2228 . doi: 10.1515/cclm-2023-0291 OpenUrl CrossRef 66. Schreuder A , Jacobs C , Lessmann N , et al. Combining pulmonary and cardiac computed tomography biomarkers for disease-specific risk modelling in lung cancer screening . Eur Respir J . 2021 ; 58 ( 3 ): 2003386 . doi: 10.1183/13993003.03386-2020 OpenUrl Abstract / FREE Full Text 67. Chetan MR , Dowson N , Price NW , Ather S , Nicolson A , Gleeson FV . Developing an understanding of artificial intelligence lung nodule risk prediction using insights from the Brock model . Eur Radiol . 2022 ; 32 ( 8 ): 5330 – 5338 . doi: 10.1007/s00330-022-08635-4 OpenUrl CrossRef PubMed 68. Yuan H , Gao Z , He X , et al. Application of logistic regression and convolutional neural network in prediction and diagnosis of high-risk populations of lung cancer . Eur J Cancer Prev . 2022 ; 31 ( 2 ): 145 – 151 . doi: 10.1097/CEJ.0000000000000684 OpenUrl CrossRef PubMed 69. Wang Y , Zhou C , Ying L , et al. Leveraging Serial Low-Dose CT Scans in Radiomics-based Reinforcement Learning to Improve Early Diagnosis of Lung Cancer at Baseline Screening . Radiol Cardiothorac Imaging . 2024 ; 6 ( 3 ): e230196 . doi: 10.1148/ryct.230196 OpenUrl CrossRef 70. Lu MT , Raghu VK , Mayrhofer T , Aerts HJWL , Hoffmann U. Deep Learning Using Chest Radiographs to Identify High-Risk Smokers for Lung Cancer Screening Computed Tomography: Development and Validation of a Prediction Model . Ann Intern Med . 2020 ; 173 ( 9 ): 704 – 713 . doi: 10.7326/M20-1868 OpenUrl CrossRef PubMed 71. Choi K , Park JS , Kwon YS , et al. Development of lung cancer risk prediction models based on F-18 FDG PET images . Ann Nucl Med . 2023 ; 37 ( 10 ): 572 – 582 . doi: 10.1007/s12149-023-01858-5 OpenUrl CrossRef PubMed 72. Chang HT , Wang PH , Chen WF , Lin CJ . Risk Assessment of Early Lung Cancer with LDCT and Health Examinations . Int J Environ Res Public Health . 2022 ; 19 ( 8 ): 4633 . doi: 10.3390/ijerph19084633 OpenUrl CrossRef PubMed 73. Venkadesh KV , Setio AAA , Schreuder A , et al. Deep Learning for Malignancy Risk Estimation of Pulmonary Nodules Detected at Low-Dose Screening CT . Radiology . 2021 ; 300 ( 2 ): 438 – 447 . doi: 10.1148/radiol.2021204433 OpenUrl CrossRef PubMed 74. Warkentin MT , Al-Sawaihey H , Lam S , et al. Radiomics analysis to predict pulmonary nodule malignancy using machine learning approaches . Thorax . 2024 ; 79 ( 4 ): 307 – 315 . doi: 10.1136/thorax-2023-220226 OpenUrl Abstract / FREE Full Text 75. Yao B , Huang X , Wu F , et al. A Novel Model Using Serum Thymidine Kinase 1 and Low-dose Computed Tomography Parameters to Predict Three-year Lung Cancer Risk in People with Pulmonary Nodules: A Retrospective Study . J Cancer . 2024 ; 15 ( 3 ): 737 – 746 . doi: 10.7150/jca.90428 OpenUrl CrossRef PubMed 76. Meng Q , Li B , Gao P , et al. Development and Validation of a Risk Stratification Model of Pulmonary Ground-Glass Nodules Based on Complementary Lung-RADS 1.1 and Deep Learning Scores . Front Public Health . 2022 ; 10 : 891306 . doi: 10.3389/fpubh.2022.891306 OpenUrl CrossRef 77. Liu D , Sun X , Liu A , et al. Predictive value of a novel Asian lung cancer screening nomogram based on artificial intelligence and epidemiological characteristics . Thorac Cancer . 2021 ; 12 ( 23 ): 3130 – 3140 . doi: 10.1111/1759-7714.14140 OpenUrl CrossRef PubMed 78. Massion PP , Antic S , Ather S , et al. Assessing the Accuracy of a Deep Learning Method to Risk Stratify Indeterminate Pulmonary Nodules . Am J Respir Crit Care Med . 2020 ; 202 ( 2 ): 241 – 249 . doi: 10.1164/rccm.201903-0505OC OpenUrl CrossRef PubMed 79. Lu H , Kim J , Qi J , et al. Multi-Window CT Based Radiological Traits for Improving Early Detection in Lung Cancer Screening . Cancer Manag Res . 2020 ; 12 : 12225 – 12238 . doi: 10.2147/CMAR.S246609 OpenUrl CrossRef PubMed 80. Zhou C , Chan HP , Chughtai A , Hadjiiski LM , Kazerooni EA , Wei J . Pathologic categorization of lung nodules: Radiomic descriptors of CT attenuation distribution patterns of solid and subsolid nodules in low-dose CT . Eur J Radiol . 2020 ; 129 : 109106 . doi: 10.1016/j.ejrad.2020.109106 OpenUrl CrossRef 81. Rundo L , Ledda RE , di Noia C , et al. A Low-Dose CT-Based Radiomic Model to Improve Characterization and Screening Recall Intervals of Indeterminate Prevalent Pulmonary Nodules . Diagnostics (Basel ) . 2021 ; 11 ( 9 ): 1610 . doi: 10.3390/diagnostics11091610 OpenUrl CrossRef PubMed 82. Mikhael PG , Wohlwend J , Yala A , et al. Sybil: A Validated Deep Learning Model to Predict Future Lung Cancer Risk From a Single Low-Dose Chest Computed Tomography . JCO . 2023 ; 41 ( 12 ): 2191 – 2200 . doi: 10.1200/JCO.22.01345 OpenUrl CrossRef PubMed 83. Venkadesh KV , Aleef TA , Scholten ET , et al. Prior CT Improves Deep Learning for Malignancy Risk Estimation of Screening-detected Pulmonary Nodules . Radiology . 2023 ; 308 ( 2 ): e223308 . doi: 10.1148/radiol.223308 OpenUrl CrossRef PubMed 84. Liu J , Xu H , Qing H , et al. Comparison of Radiomic Models Based on Low-Dose and Standard-Dose CT for Prediction of Adenocarcinomas and Benign Lesions in Solid Pulmonary Nodules . Front Oncol . 2021 ; 10 : 634298 . doi: 10.3389/fonc.2020.634298 OpenUrl CrossRef 85. Ebrahimpour L , Després P , Manem VSK . Differential Radiomics-Based Signature Predicts Lung Cancer Risk Accounting for Imaging Parameters in NLST Cohort . Cancer Med . 2024 ; 13 ( 20 ): e70359 . doi: 10.1002/cam4.70359 OpenUrl CrossRef 86. Gao R , Tang Y , Khan MS , et al. Cancer Risk Estimation Combining Lung Screening CT with Clinical Data Elements . Radiol Artif Intell . 2021 ; 3 ( 6 ): e210032 . doi: 10.1148/ryai.2021210032 OpenUrl CrossRef 87. Xiao H , Shi Z , Zou Y , et al. One-off low-dose CT screening of positive nodules in lung cancer: A prospective community-based cohort study . Lung Cancer . 2023 ; 177 : 1 – 10 . doi: 10.1016/j.lungcan.2023.01.005 OpenUrl CrossRef PubMed 88. Wu Z , Tan F , Xie Y , et al. A strategy to reduce the false-positive rate after low-dose computed tomography in lung cancer screening: A multicenter prospective cohort study . Cancer Med . 2023 ; 12 ( 13 ): 14781 – 14793 . doi: 10.1002/cam4.6106 OpenUrl CrossRef PubMed 89. Xu K , Khan MS , Li TZ , et al. AI Body Composition in Lung Cancer Screening: Added Value Beyond Lung Cancer Detection . Radiology . 2023 ; 308 ( 1 ): e222937 . doi: 10.1148/radiol.222937 OpenUrl CrossRef PubMed 90. Trajanovski S , Mavroeidis D , Swisher CL , et al. Towards radiologist-level cancer risk assessment in CT lung screening using deep learning . Comput Med Imaging Graph . 2021 ; 90 : 101883 . doi: 10.1016/j.compmedimag.2021.101883 OpenUrl CrossRef PubMed 91. Paul R , Schabath MB , Gillies R , Hall LO , Goldgof DB . Hybrid models for lung nodule malignancy prediction utilizing convolutional neural network ensembles and clinical data . J Med Imaging (Bellingham ) . 2020 ; 7 ( 2 ): 024502 . doi: 10.1117/1.JMI.7.2.024502 OpenUrl CrossRef PubMed 92. Robbins HA , Cheung LC , Chaturvedi AK , Baldwin DR , Berg CD , Katki HA . Management of Lung Cancer Screening Results Based on Individual Prediction of Current and Future Lung Cancer Risks . Journal of Thoracic Oncology . 2022 ; 17 ( 2 ): 252 – 263 . doi: 10.1016/j.jtho.2021.10.001 OpenUrl CrossRef PubMed 93. ↵ Baker SG , Schuit E , Steyerberg EW , et al. How to interpret a small increase in AUC with an additional risk prediction marker: Decision analysis comes through . Stat Med . 2014 ; 33 ( 22 ): 3946 – 3959 . doi: 10.1002/sim.6195 OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted September 12, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Risk prediction for lung cancer screening: a systematic review and meta-regression Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Risk prediction for lung cancer screening: a systematic review and meta-regression Ramin Rezaeianzadeh , Crystal leung , Soo Jeong Kim , Kayly Choy , Kate Johnson , Miranda Kirby , Stephen Lam , Benjamin Smith , Mohsen Sadatsafavi medRxiv 2025.09.10.25335529; doi: https://doi.org/10.1101/2025.09.10.25335529 Share This Article: Copy Citation Tools Risk prediction for lung cancer screening: a systematic review and meta-regression Ramin Rezaeianzadeh , Crystal leung , Soo Jeong Kim , Kayly Choy , Kate Johnson , Miranda Kirby , Stephen Lam , Benjamin Smith , Mohsen Sadatsafavi medRxiv 2025.09.10.25335529; doi: https://doi.org/10.1101/2025.09.10.25335529 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Epidemiology Subject Areas All Articles Addiction Medicine (570) Allergy and Immunology (864) Anesthesia (301) Cardiovascular Medicine (4443) Dentistry and Oral Medicine (444) Dermatology (383) Emergency Medicine (609) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1513) Epidemiology (15234) Forensic Medicine (30) Gastroenterology (1127) Genetic and Genomic Medicine (6610) Geriatric Medicine (669) Health Economics (999) Health Informatics (4544) Health Policy (1370) Health Systems and Quality Improvement (1613) Hematology (543) HIV/AIDS (1266) Infectious Diseases (except HIV/AIDS) (15925) Intensive Care and Critical Care Medicine (1104) Medical Education (623) Medical Ethics (147) Nephrology (668) Neurology (6610) Nursing (346) Nutrition (999) Obstetrics and Gynecology (1147) Occupational and Environmental Health (957) Oncology (3339) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (665) Pediatrics (1693) Pharmacology and Therapeutics (693) Primary Care Research (714) Psychiatry and Clinical Psychology (5454) Public and Global Health (9242) Radiology and Imaging (2203) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1197) Rheumatology (596) Sexual and Reproductive Health (715) Sports Medicine (530) Surgery (713) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a020e58bfab68e2e',t:'MTc3OTg0MTY3Ng=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();