Equity gaps in pulmonary nodule risk prediction for never-smokers: a real-world evidence-informed narrative review and policy analysis

Abstract

Background and Objective

Current lung cancer screening and pulmonary nodule management pathways were built largely around age, smoking exposure, and computed tomography (CT) morphology. These frameworks may under-recognize selected never-smoker subgroups because their risk scores often fall below conventional thresholds, especially East Asian women with molecularly enriched lung adenocarcinoma risk. This review evaluates whether real-world cohort evidence supports an equity concern in risk prediction and management for never-smokers with pulmonary nodules.

Methods

We conducted targeted narrative evidence mapping of guideline documents, prediction model studies, external validation studies, East Asian never-smoker screening cohorts, molecular epidemiology studies, and diagnostic-delay outcome studies. Priority was given to peer-reviewed primary studies, prospective cohorts, guidelines, and official society statements. Commercial calculators, news reports, legal commentaries, and non-peer-reviewed secondary sources were excluded.

Key Content and Findings

Smoking-centered screening eligibility remains the dominant entry point into low-dose computed tomography (LDCT) screening, while never-smokers are excluded. Cancer Screening Program in Urban China (CanSPUC)-derived Chinese risk models identified locally relevant predictors, including chronic respiratory disease, family history, menopause, education, tuberculosis history, and other non-tobacco factors. The Taiwan Lung Cancer Screening in Never-Smoker Trial (TALENT) prospective cohort showed clinically meaningful LDCT-detectable invasive lung cancer in selected never-smokers. Hospital-based persistent nodule cohorts suggest attenuation of screening-derived prediction models such as Brock when applied outside their original development setting. Model-based evidence from the National Lung Screening Trial (NLST) and International Association for the Study of Lung Cancer (IASLC) data indicates that diagnostic delays can reduce survival estimates for fast-growing early non-small cell lung cancer (NSCLC), and Veterans Health Administration (VHA) data show worse outcomes when surgery for clinical stage I NSCLC is delayed beyond 12 weeks. Molecular epidemiology further demonstrates that East Asian never-smoker lung adenocarcinoma is enriched for actionable drivers, particularly EGFR sensitizing mutations.

Conclusions

The available real-world evidence does not justify indiscriminate LDCT screening for all never-smokers. It supports recalibrated, equity-aware risk pathways for selected never-smoker subgroups. Reforms should include subgroup calibration of nodule models, incorporation of non-tobacco and molecular risk proxies, closed-loop incidental nodule tracking, and shared decision-making that addresses false-negative consequences in populations outside the training distribution of traditional models.

Introduction

Risk prediction models have become an essential part of pulmonary nodule management. The Fleischner Society recommendations, the Mayo Clinic model, the Brock/PanCan model, and the PLCOm2012 (Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial model 2012) screening risk model all attempt to transform heterogeneous clinical and imaging information into reproducible decisions about surveillance, positron emission tomography-computed tomography (PET-CT), biopsy, or surgery (1-5). These tools have real clinical value. They reduce unnecessary invasive procedures, protect patients from anxiety and procedure-related morbidity, and help health systems allocate finite imaging and specialist resources.

However, the evidentiary success of smoking-centered frameworks creates a blind spot. The most influential screening trials and prediction models were developed in populations enriched for older current or former smokers in Western settings. As a result, age and smoking exposure became dominant entry criteria for screening and important variables in risk estimation. When these tools are applied to populations outside the original training distribution, especially never-smokers and East Asian women, apparent statistical objectivity may conceal limited generalizability to these populations.

This issue is not merely theoretical. Lung cancer in never-smokers accounts for approximately 25% of the global lung cancer burden and is increasingly recognized as a distinct clinical entity (6). In East Asia, more than 50% of lung cancers in women arise in never-smokers, with adenocarcinoma predominating (7). Real-world low-dose computed tomography (LDCT) screening cohorts from China and Korea report pulmonary nodule prevalence of approximately 2.4–16.2% and lung cancer detection rates of approximately 0.45–0.85% among never-smoking women (8,9). Therefore, a population labeled low risk by smoking-centered tools may overlap with a population that stands to benefit substantially from early diagnosis and molecularly informed treatment. In East Asian populations, lung adenocarcinoma among never-smokers often has a molecular profile enriched for actionable driver alterations, especially EGFR sensitizing mutations (7,10).

This review focuses on women because East Asian female never-smokers carry the highest population-level burden of never-smoker lung cancer, their adenocarcinomas are enriched for actionable EGFR mutations, and both Cancer Screening Program in Urban China (CanSPUC) and Taiwan Lung Cancer Screening in Never-Smoker Trial (TALENT) have identified sex-specific risk structures (e.g., reproductive, hormonal, and environmental factors) that smoking-based models fail to capture (7,10-13).

The purpose of this review is not to argue that all never-smokers should undergo LDCT screening, nor to dismiss the harms of overdiagnosis. Rather, we argue that real-world evidence supports a narrower and more actionable claim: selected never-smoker subgroups may be systematically under-triaged by current screening and nodule-management pathways, and these pathways should be recalibrated using equity-aware, population-specific evidence. The review includes explicit evidence mapping strategy. A conceptual framework illustrating potential mechanisms of risk misclassification and diagnostic delay in selected never-smoker populations is shown in Figure 1. We present this article in accordance with the Narrative Review reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1511/rc).

Figure 1.

Conceptual framework of potential mechanisms contributing to risk misclassification and diagnostic delay in selected never-smoker populations. Arrows represent hypothesized relationships, not validated causal links. The framework integrates evidence from multiple populations and illustrates potential relationships among smoking-centered screening criteria, potential limitations in model transportability, diagnostic delay, and the molecularly enriched adenocarcinoma phenotype observed in East Asian never-smokers. It is intended to guide future research, not to support universal screening. PLCOm2012, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial model 2012; USPSTF, U.S. Preventive Services Task Force.

Methods

We used a combination of evidence-identification approaches. First, a structured PubMed search was performed to identify studies addressing pulmonary nodules, screening, and risk prediction in never-smokers; the full search strategy is provided in Table S1. Second, landmark model development studies (3-5) and diagnostic or treatment delay studies (14,15) were identified through targeted supplementary searches focusing on specific evidence domains without applying the never-smoker restriction, combined with backward and forward citation tracking. Third, relevant guideline repositories and major thoracic society statements were searched directly. This approach was designed to support targeted narrative evidence mapping and is summarized in Table 1. Searches were updated through January 31, 2026.

Table 1. The search strategy summary.

Items	Specification
Date of search	January 31, 2026
Databases and other sources searched	PubMed (principal structured search; complete search strategy provided in Table S1)
	Guideline repositories (GIN, NICE, USPSTF) and official society statements (ATS, Fleischner Society, IASLC) were searched
	Targeted supplementary searches and backward/forward citation tracking were performed for landmark risk prediction model studies (Mayo, Brock, and PLCOm2012) and diagnostic and treatment delay studies (14,15)
Search terms used	Core search dimensions: (I) disease and population: pulmonary nodule, lung cancer, never-smokers/non-smokers; (II) research methods: risk prediction, risk models, lung cancer screening, low-dose CT; (III) equity themes: health equity, health disparities, equity gaps; (IV) clinical outcomes: diagnostic delay, treatment delay, survival prognosis; (V) molecular characteristics: EGFR mutations, driver mutations; (VI) study types: real-world studies, cohort studies, external validation studies
	Combined MeSH subject headings and free text terms using Boolean operators AND/OR (complete PubMed search strategy is provided in Table S1)
Timeframe	No start date restriction to January 31, 2026
Inclusion and exclusion criteria	Inclusion criteria: (I) peer-reviewed original research: prospective cohort studies, retrospective real-world cohort studies, risk model development and external validation studies; (II) high-quality secondary research: systematic reviews, meta-analyses; (III) official documents: national/international clinical guidelines, formal statements from professional societies; (IV) English-language publications; (V) studies containing verifiable core numerical results
	Exclusion criteria: (I) commercial risk calculators, news reports, legal commentaries, non-peer-reviewed preprints; (II) case reports, single-center small-sample studies (n<100), pure expert commentaries; (III) conference abstracts (not formally published in full text); (IV) literature with unavailable full text
Selection process	Literature screening: independently completed by 5 researchers (Shangqing Xu, Ning Wei, Xiaoyan Han, Yan Sun, Huiling Li) through title-abstract screening and full-text review. Disagreements were resolved through discussion with the corresponding author Minjie Ma to reach consensus
	Evidence verification: all core numerical conclusions were traced back to original literature for verification, forming an independent evidence verification log to ensure data accuracy
	Quality assessment: priority was given to large-sample cohorts (n>10,000), multicenter studies, and content recommended by official guidelines
Additional considerations	Key numerical conclusions must cite original studies rather than secondary citations, and both training and validation set results should be reported
	Clearly distinguish between the intended design purpose of models and their actual application scenarios to avoid overinterpreting research conclusions
	Gray literature and unpublished internal data were not included
	Conclusions that could not be verified in peer-reviewed literature were either deleted or explicitly labeled with their evidence level

ATS, American Thoracic Society; CT, computed tomography; GIN, Guidelines International Network; IASLC, International Association for the Study of Lung Cancer; NICE, National Institute for Health and Care Excellence; USPSTF, U.S. Preventive Services Task Force.

Eligible sources included peer-reviewed primary cohort studies, prospective screening studies, model-development or validation studies, systematic reviews or official statements, and national or international guidelines. For key numerical claims, we preferred the original publication over news reports, calculators, conference summaries, or legal websites. If a number could not be verified in a peer-reviewed source, it was either removed or explicitly downgraded. The review did not involve new patient-level data and therefore did not require institutional ethics approval.

Table 1 summarizes the search strategy used in this review, and Table 2 summarizes the evidence mapping strategy. The aim was to support a policy and clinical-practice argument, not to calculate pooled estimates. Accordingly, no meta-analysis was performed.

Table 2. Evidence mapping strategy and source hierarchy.

Domain	Preferred source type	Key examples used	How evidence was used
Guidelines and eligibility	Official guidelines and society documents	Fleischner 2017; USPSTF 2021; Lung-RADS v2022	Define current standard pathways and identify where never-smokers are structurally excluded
Prediction models	Original model studies and validation studies	Mayo; Brock/PanCan; PLCOm2012; PLCO risk-threshold evaluation	Clarify model purpose, development setting, and subgroup limitations
East Asian never-smoker cohorts	Prospective or large real-world cohorts	TALENT; CanSPUC-derived Chinese models	Quantify non-tobacco risk factors and selected never-smoker LDCT yield
Clinical consequences	Large databases and outcome modeling	NLST/IASLC delay model; VHA stage I NSCLC cohort	Support the clinical importance of delayed diagnosis and treatment
Molecular epidemiology	Clinical molecular cohorts and reviews	East Asian never-smoker lung adenocarcinoma data	Explain why traditional low-risk labels can conflict with actionable molecular risk

CanSPUC, Cancer Screening Program in Urban China; IASLC, International Association for the Study of Lung Cancer; LDCT, low-dose computed tomography; Lung-RADS, Lung CT Screening Reporting and Data System; NLST, National Lung Screening Trial; NSCLC, non-small cell lung cancer; PLCO, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; PLCOm2012, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial model 2012; TALENT, Taiwan Lung Cancer Screening for Never Smoker Trial; USPSTF, U.S. Preventive Services Task Force; VHA, Veterans Health Administration.

Smoking-centered screening eligibility and structural exclusion

The current U.S. Preventive Services Task Force (USPSTF) recommendation applies annual LDCT screening to adults aged 50 to 80 years who have at least a 20 pack-year smoking history and currently smoke or quit within the past 15 years (2). This recommendation is evidence-based for its target population and is not, by itself, an inequitable policy. The problem arises when its categorical structure is interpreted as implying that never-smokers are uniformly low-risk or that risk factors outside smoking are clinically secondary.

Because never-smokers generally fail smoking-based eligibility criteria, their early lung cancers are often detected incidentally rather than through organized screening. After an incidental nodule is found, these individuals are then triaged by nodule algorithms and malignancy models whose development populations may not reflect their risk phenotype. This creates a two-stage disadvantage: exclusion from screening entry followed by potential risk misclassification during nodule assessment.

PLCOm2012 illustrates this boundary. It is more sophisticated than categorical pack-year criteria and performs well for selecting many ever-smokers, yet evaluation of PLCO (Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial) risk thresholds reported that none of 65,711 PLCO never-smokers reached a PLCOm2012 risk threshold of at least 0.0151 (16). This finding demonstrates structural exclusion of never-smokers at the population level, as smoking-weighted risk scores for this group never reach the actionable threshold.

CanSPUC evidence: non-smoking Chinese risk is not smoking-free risk

Two CanSPUC-derived analyses are especially relevant because they provide population-based Chinese evidence rather than extrapolation from Western smoker-enriched cohorts. A study focused on nonsmoking Chinese women included 151,834 participants, randomly split into training and validation sets of 75,917 each. It identified age, chronic respiratory disease history, first-degree family history of lung cancer, menopause, and benign breast disease history as predictors. The training-set areas under the curve (AUCs) for 1-, 3-, and 5-year risk were 0.762, 0.718, and 0.703; validation-set AUCs were more moderate at 0.646, 0.658, and 0.650, respectively. The original study did not report 95% confidence intervals for these AUC estimates (11).

A separate CanSPUC-based non-smoker model included 214,764 eligible subjects and identified age, sex, education, family history, tuberculosis history, and absence of hyperlipidemia as independent predictors. Its training-set AUCs were 0.753, 0.752, and 0.755 for 1-, 3-, and 5-year risk; validation-set AUCs were 0.668, 0.678, and 0.685. Confidence intervals were not reported in the original publication (12).

These two CanSPUC-derived cohorts overlap; the female-specific cohort is a subset of the larger non-smoker cohort. They should not be regarded as independent replications, but are presented to illustrate sex-specific and general non-smoker risk structures, respectively. These studies should be interpreted carefully. Their discrimination is moderate, not definitive, and neither study proves that all Chinese never-smokers warrant screening. Their importance is different: they demonstrate that clinically relevant risk structure exists among non-smokers and that this structure is not reducible to zero pack-years. For a nodule management system, this means that family history, prior pulmonary disease, reproductive or hormonal factors, and population-specific environmental exposures should not be treated as peripheral variables. These AUC values reflect discrimination only; calibration data stratified by smoking status are not available for these cohorts, though the original models showed acceptable calibration in their overall development samples.

TALENT: prospective LDCT evidence in selected never-smokers

TALENT provides one of the strongest prospective data points for selected East Asian never-smokers. TALENT was a nationwide, multicenter prospective cohort study conducted at 17 tertiary centers. Eligible participants were aged 55 to 75 years, had never smoked or had fewer than 10 pack-years and had stopped for more than 15 years, meaning that the cohort was composed predominantly, but not exclusively, of never-smokers, and had at least one risk factor such as family history of lung cancer, passive smoke exposure, prior pulmonary tuberculosis or chronic obstructive pulmonary disease, or substantial cooking-related exposure (13).

The published TALENT report included 12,011 participants and reported a clinically meaningful burden of LDCT-detectable invasive lung cancer at baseline. Earlier conference reporting and the final article indicate an overall lung cancer detection rate of approximately 2.6% and an invasive cancer detection rate of approximately 2.1% in this predominantly never-smoker cohort, with most screen-detected cancers at stage 0–I (13,17). Of note, the published TALENT report did not provide detection rates stratified by never-smoker versus light former-smoker status, so the precise contribution of former smokers cannot be quantified; this is discussed as a limitation. The key implication is not mass screening for all never-smokers. Rather, TALENT challenges the assumption that never-smokers with enriched risk factors are uniformly too low-risk to benefit from structured early detection.

TALENT also requires caution. The authors and commentators noted the possibility of overdiagnosis, particularly when LDCT identifies indolent lesions in populations with high prevalence of small subsolid nodules. For policy purposes, this is precisely why a risk-enriched and population-specific approach is preferable to universal screening.

Overdiagnosis remains a serious concern. Gao and Welch analyzed data from 12 million Taiwanese women and found that LDCT screening promotion was associated with a 6-fold increase in early-stage lung cancer incidence (2.3 to 14.4 per 100,000), no change in late-stage incidence, and largely stable mortality—suggesting that an estimated 7,000–12,000 women were overdiagnosed (18). This ecological evidence contrasts with the TALENT prospective cohort, which enrolled a risk-enriched population. The two studies are complementary: together they reinforce that indiscriminate screening of never-smokers is not justified, while risk-enriched pathways warrant continued investigation with rigorous overdiagnosis monitoring.

Transferability of nodule models to external populations

The Mayo and Brock models remain useful decision aids, but their limitations must be transparent. The Mayo model was developed from clinical solitary pulmonary nodules and includes age, smoking, prior extrathoracic cancer, nodule diameter, spiculation, and upper-lobe location (3). The Brock/PanCan model, developed in a screening context, incorporates age, sex, family history, emphysema, nodule size, upper-lobe location, nodule count, spiculation, and nodule type (4). In its original setting, Brock showed high discrimination. The question is whether that performance transfers reliably to hospital-based incidental nodule populations and never-smoker subgroups.

Recent real-world evidence suggests caution. In a hospital-based study of persistent pulmonary nodules (130 retrospective and 301 prospective participants, mixed smoking status, US hospital-based setting), the Brock model had an AUC of approximately 0.679, and some confirmed cancers had Brock risk scores below the commonly used 10% threshold (19). Given that this cohort was not restricted to never-smokers or East Asian populations, these findings illustrate challenges in model transportability in an external setting but do not directly demonstrate poor performance in East Asian never-smokers. Dedicated prospective external validation studies with adequate never-smoker sample sizes and direct calibration assessment are needed before drawing subgroup-specific conclusions.

Low model scores should not automatically reassure clinicians when applied outside the original training distribution. This consideration is particularly relevant for selected never-smoker populations, including East Asian women with persistent nodules or non-tobacco-related risk factors, for whom dedicated validation and calibration studies remain needed. The AUC values reported for risk models in screening populations and for malignancy prediction in persistent nodules reflect different prediction tasks and should not be compared directly; they are presented here only to illustrate the range of model performance in different clinical settings.

Diagnostic delay: why false negatives matter

Risk prediction is not only a statistical exercise; it changes time. A low-risk label can lengthen surveillance intervals, reduce urgency, and increase the probability of loss to follow-up. In slow-growing benign or indolent nodules, this may be appropriate. In fast-growing early non-small cell lung cancer (NSCLC), delay can be consequential.

A radiology study modeled the prognostic effect of diagnostic delays using the National Lung Screening Trial (NLST) and International Association for the Study of Lung Cancer (IASLC) data. For a 20-mm tumor with a 60-day volume-doubling time in the NLST database, estimated 10-year lung cancer-specific survival decreased from 83.4% to 76.5%, 66.8%, and 32.3% after 90, 180, and 365 days of delay, respectively. IASLC data showed similar decreases in 5-year overall survival from 81.2% to 73.4%, 62.4%, and 23.3% (14). These are model-estimated figures, not randomized trial observations, but they quantify the plausible biological cost of prolonged observation for aggressive tumors.

Treatment delay data point in the same direction. A Veterans Health Administration (VHA) retrospective cohort of 9,904 patients with clinical stage I NSCLC found that surgical delay beyond 12 weeks from radiographic diagnosis was associated with increased recurrence risk and worse overall survival; each week beyond 12 weeks was associated with a small but statistically significant increase in recurrence hazard (hazard ratio =1.004; 95% confidence interval, 1.001–1.006; P=0.002) (15). These findings do not imply that every indeterminate nodule should be rushed to surgery. They do mean that surveillance systems should be robust enough to identify growth, persistence, and evolving high-risk features without letting low-risk labels become passive neglect.

Molecular epidemiology: the biological basis of the equity gap

The fairness concern becomes more compelling when molecular epidemiology is added. East Asian never-smoker lung adenocarcinoma is enriched for oncogenic drivers. A review of the East Asian experience reported that EGFR sensitizing mutations constitute approximately 60–78% of never-smoker adenocarcinomas in this region (7). Other molecular cohorts have shown that a high proportion of East Asian never-smoker lung adenocarcinomas harbor known oncogenic alterations in a limited set of genes, including EGFR, HER2, ALK, and KRAS (20).

This molecular profile matters because it inverts the intuition embedded in smoking-centered prediction. Smoking remains the strongest population-level risk factor for lung cancer overall. Yet within lung adenocarcinoma, absence of smoking in East Asian women is not simply absence of risk; it often signals a different etiologic and molecular pathway. These patients may be younger, may have fewer smoking-related comorbidities, and may have tumors that are molecularly targetable if diagnosed in time.

Therefore, the equity gap is not that every never-smoker has high absolute risk. It is that current models can assign low probability to a subgroup whose cancers, when present, are enriched for actionable biology and can suffer clinically meaningful harm from delayed diagnosis.

Equity-aware policy and clinical reforms

Based on the evidence summarized in Table 3, the following reforms to promote equity merit consideration. First, subgroup calibration should become routine. Any nodule risk model used in a multiethnic health system should report performance by smoking status, sex, ancestry or region when appropriate, nodule type, and clinical setting. Minimum reporting should include discrimination, calibration-in-the-large, calibration slope, false-negative rate, and decision-curve analysis by subgroup. Where subgroup performance is poor, risk thresholds should be recalibrated rather than imported unchanged.

Table 3. Verified real-world evidence supporting an equity-aware pathway for selected never-smokers.

Evidence domain	Cohort/source	Verified quantitative finding	Interpretation for this review	Caution
Chinese non-smoking women	CanSPUC-derived JMIR cohort	n=151,834; training AUCs 0.762, 0.718, 0.703; validation AUCs 0.646, 0.658, 0.650 for 1-, 3-, 5-year risk	Non-tobacco variables can stratify lung cancer risk among nonsmoking Chinese women	Moderate external performance; not a stand-alone screening mandate
Chinese non-smokers overall	CanSPUC-derived Frontiers cohort	n=214,764; training AUCs 0.753–0.755; validation AUCs 0.668–0.685	Population-specific models can identify risk beyond smoking	Variables and calibration may require local validation
Asian never-smoker LDCT screening	TALENT prospective cohort	12,011 selected never-smokers; overall detection around 2.6%; invasive detection around 2.1%	Selected never-smokers can have meaningful screen-detectable disease burden	Overdiagnosis possible; avoid universal screening claim
Model transferability	Persistent nodule hospital cohort	Brock AUC about 0.679; sybil about 0.678; logistic regression 0.729	Screening-derived models attenuate in hospital-based persistent nodule settings	Recent study; requires independent replication
Diagnostic delay	NLST/IASLC model	20-mm tumor, VDT 60 days: 10-year LC-specific survival estimated to decrease from 83.4% to 32.3% after 365-day delay	Long observation windows can harm fast-growing early NSCLC	Model-based estimates; not all nodules are fast-growing cancers
Treatment delay	VHA stage I NSCLC cohort	n=9,904; delay >12 weeks associated with recurrence and worse survival; HR for recurrence per week beyond 12 weeks 1.004 (95% CI, 1.001–1.006; P=0.002)	Timely evaluation matters after suspicious radiographic diagnosis	Veteran population mostly male; not a nodule-screening cohort
Molecular enrichment	East Asian never-smoker LUAD literature	EGFR sensitizing mutations roughly 60–78% in East Asian never-smoker adenocarcinoma	Traditional low-smoking risk can coexist with actionable molecular risk	Molecular prevalence does not prove screening benefit alone
Structural exclusion	PLCO threshold evaluation	0 of 65,711 PLCO never-smokers reached PLCOm2012 risk ≥0.0151	Smoking-weighted eligibility can mechanically exclude never-smokers	Model was designed primarily for screening selection among ever-smokers

AUC, area under the curve; CanSPUC, Cancer Screening Program in Urban China; CI, confidence interval; HR, hazard ratio; IASLC, International Association for the Study of Lung Cancer; JMIR, Journal of Medical Internet Research; LC, local control; LDCT, low-dose computed tomography; LUAD, lung adenocarcinoma; NLST, National Lung Screening Trial; NSCLC, non-small cell lung cancer; PLCO, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; PLCOm2012, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial model 2012; TALENT, Taiwan Lung Cancer Screening for Never Smoker Trial; VDT, volume doubling time; VHA, Veterans Health Administration.

Second, screening eligibility should evolve from pack-years alone to risk-enriched pathways. The evidence does not support indiscriminate LDCT for all never-smokers. It supports models that include family history, chronic lung disease, prior tuberculosis, environmental and indoor exposure proxies, sex-specific variables, and potentially molecular or polygenic risk markers when validated. CanSPUC and TALENT provide a foundation for this direction, but health-economic evaluation and overdiagnosis monitoring remain essential.

Third, incidental nodule care should be closed loop. Lung nodule management often fails not only because of model error but because follow-up responsibility is fragmented. Natural language processing of radiology reports, nodule registries, automated reminders, and explicit assignment of follow-up ownership can reduce passive loss to follow-up. The American Thoracic Society has identified disparities in lung nodule evaluation as a research priority and emphasized the need for interventions addressing timely guideline-concordant follow-up (21).

Fourth, shared decision-making should include model uncertainty. When a never-smoker with enriched risk factors receives a low model score, clinicians should explain that the score was generated from variables and populations that may not fully represent that patient. The conversation should balance overdiagnosis and false-positive harms against the false-negative consequence of delayed recognition, especially for persistent subsolid nodules, family history, and Asian never-smoker adenocarcinoma phenotypes.

Fifth, liquid biopsy and AI should be developed as adjuncts, not shortcuts. Circulating tumor DNA, methylation assays, extracellular vesicle markers, and radiomics may eventually refine early nodule risk. However, early-stage tumors can shed little ctDNA, and AI systems can reproduce dataset bias. Future AI models should therefore be evaluated using external, multiethnic, never-smoker-enriched cohorts and fairness constraints before being used to lower or raise surveillance intensity.

Strengths and limitations of this review

This review has several strengths. It verifies the main quantitative claims using peer-reviewed real-world cohorts, prospective screening data, official guidelines, and major outcome databases. It distinguishes between population screening, nodule risk prediction, diagnostic delay, and molecular epidemiology rather than merging them into a single claim. It also avoids the misleading conclusion that never-smokers as a whole should be screened indiscriminately.

The limitations are equally important. This is not a systematic review, and no pooled estimate was calculated. The review does not report patient-level data from our institution. Some datasets, such as TALENT and CanSPUC, are highly relevant to Chinese and East Asian settings but may not generalize globally. Diagnostic-delay survival estimates are model-based and depend on assumptions about tumor volume-doubling time. Finally, the molecular enrichment of EGFR among East Asian never-smokers does not by itself prove screening benefit; it strengthens the biological plausibility of a targeted risk pathway.

Additionally, the TALENT cohort included a small proportion of light former smokers, and detection rates were not reported separately for true never-smokers; therefore, a minor contribution from former smokers cannot be excluded. Direct calibration data for risk models in never-smoker subgroups are scarce; the equity concern described here is therefore primarily a threshold problem rather than statistically proven underestimation, and future studies should report calibration stratified by smoking status. Furthermore, our evidence base is predominantly East Asian; the proposed equity-aware reforms should be regarded as a conceptual framework informed by these populations, and their applicability to other never-smoker groups requires local validation. Our search was restricted to English-language databases (primarily PubMed) and did not include Chinese-language databases such as CNKI or Wanfang. Given the East Asian focus of this review, this risks publication bias and may underrepresent Chinese real-world evidence—a methodological equity gap that parallels the clinical equity gap we describe. Future systematic reviews should integrate Chinese-language databases.

Conclusions

Current pulmonary nodule risk prediction frameworks have prevented substantial harm from overdiagnosis and unnecessary invasive procedures. Their value should be preserved. The real-world evidence synthesized here indicates, however, that selected never-smoker subgroups, especially East Asian women with non-tobacco risk factors and molecularly enriched lung adenocarcinoma biology, may not be adequately represented by smoking-centered screening eligibility or traditional nodule risk models.

The practical response is not to abandon probability-based medicine, but to make it more precise and more equitable. Future pathways should recalibrate model thresholds in never-smokers, incorporate population-specific non-tobacco predictors, maintain closed-loop incidental nodule tracking, and support shared decision-making that explicitly communicates model uncertainty. In this sense, equity-aware nodule management is not a departure from evidence-based medicine; it is the next step in making evidence-based medicine valid for the populations to whom it is applied.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.