Plasma proteomic profiles associated with low-dose computed tomography findings and incident lung cancer: a prospective pilot study

Introduction

Lung cancer remains one of the most significant global health challenges, has a high incidence rate and is a leading cause of cancer-related mortality worldwide. In 2024 alone, an estimated 2.6 million new cases of lung cancer and 1.9 million deaths occurred worldwide accounting for 19.1% of all cancer deaths globally. Despite advances in therapeutic interventions, the 5-year survival rate for lung cancer patients remains dismally low at approximately 19%, primarily due to the late stage at which the disease is often diagnosed. Early detection is critical for improving survival rates, as it enables timely intervention in localized disease (1).

Low-dose computed tomography (LDCT) has emerged as a promising tool for the early detection of lung cancer in high-risk populations such as long-term smokers. The National Lung Screening Trial (NLST) demonstrated that LDCT screening could reduce lung cancer mortality by approximately 20% compared to chest radiography (2). Similar results were reported by the NELSON trial, in which LDCT screening led to a 26% reduction in lung cancer mortality. Instead of using lesion diameter evaluation, the NELSON study used volumetric assessment resulting in markedly lower rates of false positives (3). However, LDCT screening is not without its challenges such as false-positive findings, which can lead to unnecessary invasive procedures and patient anxiety (3,4). Additionally, LDCT’s effectiveness can be compromised by factors such as patient adherence, radiation exposure, and variability in interpretation. These challenges highlight the need for complementary approaches to enhance the accuracy and feasibility of lung cancer screening (5,6).

Blood-based proteomic profiling has gained attention as a potential adjunct to LDCT in lung cancer screening (7,8). Proteomics, the large-scale study of proteins, offers a window into the dynamic molecular landscape of cancer, capturing changes in protein expression and modification that may reflect early tumor development. Recent studies have demonstrated the utility of proteomic signatures in distinguishing lung cancer patients from healthy controls, with some panels showing promise in detecting early-stage disease. Many of the identified predictive proteins relate to systemic inflammation suggesting that this plays an important role in lung carcinogenesis (9,10). Furthermore, combining proteomic data with radiographic modalities like LDCT could improve diagnostic accuracy, potentially reduce false positives and enhance risk stratification (10,11).

In this study, we utilized material collected in a prospective LDCT lung cancer screening trial (12). Our hypothesis was that blood proteomic profile analysis could improve risk prediction for lung cancer and stratification of LDCT findings. To create novel predictive models for stratifying LDCT findings, and cancer detection, we applied a proximity extension assay (PEA) (Olink 96 Immuno-Oncology panel) to analyze baseline plasma samples in a high risk population for lung cancer parallel with computed tomography (CT) screening results. The study was not designed to replace LDCT, to serve as a post-LDCT nodule management, or to evaluate clinical utility. We present this article in accordance with the STROBE reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0646/rc).

Methods

LDCT cohort

The present study is an exploratory biomarker substudy of the prospective LDCT-SC-FI study, in which 200 participants with a marked smoking history who also met the NELSON study eligibility criteria for lung cancer screening were recruited at Oulu University Hospital. The design and smoking-cessation outcomes of the parent study are described in our previous publication (12). Blood-based biomarker analysis was prespecified as an exploratory outcome of the LDCT-SC-FI study, with blood samples prospectively collected for this purpose (Appendix 1). The present study evaluates plasma proteomic biomarkers within this prespecified framework. The study recruitment occurred from November 2022 to April 2023, and the second round of LDCT examinations was finished by April 2024. The final clinical data update was carried out in August 2024. Briefly, the inclusion criteria included an age of 50–74 years, active smoking status, and a smoking history of ≥15 cigarettes/day for ≥25 years or smoked ≥10 cigarettes/day for ≥30 years. The most important exclusion criteria included previous malignancy and chest CT performed within one year of the study entry.

Participants in the study were randomized 1:1 to smoking cessation with smartphone application or by written materials. LDCT scans and blood samples were collected at study entry and 1-year follow-up in both study arms. The primary end-point of the study was smoking cessation at 3 and 6 months after inclusion while secondary end-points included reduction in smoking, performance of LDCT screening, and cost evaluation. The sample size calculations were based on the primary end-point of the study (smoking cessation). In brief, the study was powered (80%) with 155 subjects to detect 15% difference with 90% confidence (75% vs. 90%) in the number of active smokers. With the expected dropout rate, the sample size was adjusted to 200.

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Northern Ostrobothnia Hospital District, the regional ethics committee responsible for Oulu University Hospital at the time of the study approval (EETTKM 21/2022). All the participants signed an informed consent before any study procedures. The study subjects were not compensated for their participation. Of note, LDCT lung cancer screening is not among the publicly funded cancer screenings in Finland. The study was registered on 18 November 2022 in the ClinicalTrials.gov database under registration number NCT05630950.

LDCT screening

Two hundred study participants were screened for lung cancer with LDCT at baseline and at 1 year. Nodule detection and evaluation algorithm followed the NELSON study (13). LDCT findings were at baseline categorized by nodule volume: Negative and NODCAT 1–4. For NODCAT 3 lesions, a 3-month follow-up CT was performed to assess nodule volume doubling time and categorize the growth into growth categories (GROWCATs) A, B and C. At 1 year, LDCT scans were performed on 183 participants (91.5% of the initial study population). Any detected nodules were categorized again by nodule volume and by nodule volume doubling time. All the radiographic images were assessed by a thoracic radiologist with computer assisted lesion assessment including volume definition.

• NELSON criteria: variable is based on the nodule evaluation algorithm of NELSON lung cancer screening study and groups findings into three categories: (I) negative; (II) indeterminate; (III) positive. Indeterminate is described as NODCAT 3 (at baseline), new NODCAT 2–3 or GROWCAT B (at follow-up). Positive findings include NODCAT 4 (at baseline), new NODCAT 4 or GROWCAT C (at follow-up). Classification of participants in three categories was based on the LDCT finding with the highest category on either screening round.
• New lung cancer: this variable indicates whether a new lung cancer diagnosis occurred between baseline and the last available follow-up (minimum follow-up of 16 months after baseline).

Plasma samples

Of the 200 study participants, 193 individuals had peripheral blood samples available from the study entry. Blood samples were drawn in 10 milliliter EDTA blood collection tubes and processed by standardized protocols. After centrifugation at 1,811 g for 10 minutes, plasma was separated and aliquoted into cryotubes and stored at −80 ℃.

Plasma protein analysis platform

One hundred ninety-three pseudonymized plasma samples (40 μL) were analyzed with PEA using Olink target 96 Immuno-Oncology panel (BioXpedia; Aarhus, Denmark). The samples were randomized on three plates and normalization was done according to Olink instructions using inter-plate controls. Olink platform uses a relative protein concentration unit on a log₂ scale called normalized protein expression (NPX) which is derived from qPCR cycle threshold. Due to the exploratory nature of the study we decided to use all the protein data provided, protein expression levels below the limit of detection (LOD) were used and missing values imputed.

Statistical analysis

All statistical analyses were performed using R (ver. 4.4.0). Sample assays with missing values were imputed using the ‘mice’ package with predictive mean matching. Data manipulation and formatting were carried out by the ‘dplyr’ package. Heatmaps were drawn using the ‘pheatmap’ package, Euclidean distance and complete linkage were applied for unsupervised hierarchical clustering. Pairwise protein-protein correlations were calculated using Spearman’s rank correlation coefficient via the ‘stats’ package.

For differential protein expression analyses the ‘OlinkAnalyze’ package’s linear mixed-effects models were utilized. Benjamini-Hochberg false discovery rate (FDR) method was used as the default multiple testing correction method. Volcano plots were drawn using the ‘ggplot2’ package.

To assess potential confounding effects of clinical variables on protein expression, we fitted linear mixed effects models using the ‘lmerTest’ package for each protein. Lung cancer or radiographic findings by NELSON criteria were used as the fixed effect and “PlateID” as the random intercept. We adjusted for age, sex, and smoking as independent covariates. Interaction terms were not included to reduce model complexity given the limited sample size. Covariates included age (binary: over/under 65 years), sex, and smoking burden (binary: over/under 30 pack-years). For each protein, adjusted and unadjusted models were compared by estimating marginal means across NELSON categories.

For the lung cancer prediction analysis, participants diagnosed with non-lung malignancy were excluded (n=5) leaving 188 participants, including six incident lung cancer cases. The binary outcome was incident lung cancer during the study; NELSON category was not used as the outcome of this model. Protein selection for the lung cancer prediction model was performed using penalized regression via the ‘glmnet’ package. Multiple alpha values ranging from 0 (RIDGE) to 1 (LASSO) were compared and least absolute shrinkage and selection operator (LASSO) was chosen as it minimized the deviance. The lambda min value was chosen for extracting non-zero coefficients. A seven-protein panel was selected using the complete study cohort and subsequently treated as a fixed predictor panel. The final lung cancer prediction model was fitted using penalized logistic regression with the ‘glmnet’ package. Internal validation used nested stratified six-fold cross-validation, with one incident lung cancer case assigned to each outer fold. Five folds were used for training, and the remaining one-fold was held out for evaluation. Predictions from all six held-out folds were pooled, providing one out-of-fold probability for each participant. Receiver operating characteristic area under the curve (ROC-AUC) and precision-recall area under the curve (PR-AUC) were calculated from these pooled predictions, and 95% confidence intervals (CIs) were estimated for ROC AUC. Because the seven-protein panel was selected using the complete cohort before cross-validation, the cross-validated performance estimates might be overoptimistic.

Separately, for the forest plot we used a standard unpenalized multivariable logistic regression to report odds ratios (ORs) per 1 standard deviation increase in NPX with Wald 95% CIs. Results were visualized using ‘forestplot’ package.

Clusterprofiler 4.0 package was employed for the over-representation analysis (ORA). Statistically significant proteins in the lung cancer group were ranked according to the difference of log₂ fold change between the two groups. The Hallmark collection from the Molecular Signatures Database (MSigDB) (14), C2 Reactome (15), C5 Gene Ontology (GO) (16) and Kyoto Encyclopedia of Genes and Genomes (KEGG) (17) were used as reference gene set collections. For results, the minimum geneset size was set to 3 and FDR 0.25 were considered significant. ORA results were further visualized as a cnetplot using the ‘enrichplot’ package.

Results

Study population

A total of 200 individuals with a significant smoking history (smoked ≥15 cigarettes/day for ≥25 years or smoked 10 cigarettes/day for ≥30 years) participated in the prospective lung cancer screening study at Oulu University Hospital. The demographic and clinical characteristics are summarized in Table S1. Most of the patients were aged 65 years or younger (60.1%), while 39.9% were older than 65 years. The sex distribution was nearly equal, with 49.2% of participants being male and 50.8% female. Of the total cohort, 44% had a history of smoking less than 30 pack-years, while 56% had smoked more than 30 pack-years. One hundred ninety-three out of the 200 study participants (96.5%) had baseline blood samples available for analysis and this formed the cohort of the current study.

Radiographic findings

At baseline LDCT, 160 study participants (82.9%) had negative results, while the remaining 33 individuals (17.1%) had radiographic findings of NODCAT 1–4 with the most prevalent being NODCAT 3 (7.3%). One-year follow-up LDCT scans were available for 177 participants (91.7% of the cohort) while data were not available for 16 participants due to withdrawal of consent or identification of new lung cancer. Among these individuals, 164 (92.7% with follow-up scans) had negative LDCT results and notable LDCT findings were observed in 19 participants (10.7%). During the entire screening period (minimum follow-up of 16 months), six participants (3.1%) were diagnosed with new lung cancer which were all based on the LDCT examinations of the study. Of the detected lung cancers, five were non-small cell histology and five were detected at pathological stage I (83.3%) (Table S1).

Blood protein expression and its association to LDCT findings and detected cancers

Blood protein expression levels were measured for 92 proteins with the Olink target 96 Immuno-Oncology panel, and the resulting proteomic data was available for all 193 (100%) samples analyzed. We selected the protein panel since we were investigating early detection of cancer, and immunology has been especially linked to lung carcinogenesis (11,18).

Unsupervised hierarchical clustering was performed on the protein expression data. The resulting heatmap illustrates the relationship between protein expression and the following clinical variables: NELSON criteria (negative, indeterminate, positive); new lung cancer. Clustering analysis did not demonstrate a clear separation of protein expression patterns and LDCT findings, suggesting that global protein expression patterns do not strongly reflect radiographic categories in this dataset. Six new lung cancers were identified during follow-up and as expected with the limited number of events, no clear clustering according to lung cancer status was observed.

Proteins were additionally annotated according to their known biological functions: T cell activation; apoptotic process; inflammatory response; natural killer cell mediated cytotoxicity; negative regulation of apoptotic process; peptidase activity; angiogenesis and other. This functional categorization did not reveal major clusters corresponding to specific biological processes (Figure 1).

Figure 1 Plasma proteome of the LDCT cohort. Heatmap depicting the expression levels of 92 proteins with unsupervised hierarchical clustering. Columns represent samples and rows represent the 92 measured proteins. LDCT findings by NELSON and new cancer diagnoses during the follow-up period are annotated on top of the heatmap. Red color indicates high protein expression while blue indicates low protein expression. LDCT, low-dose computed tomography.

Correlation analysis of all the measured proteins

Using a Spearman correlation analysis, all the 92 proteins measured with the Target 96 Immuno-Oncology panel were plotted against each other. A numerical rank correlation coefficient is represented on every cell. Unsupervised hierarchical clustering of the correlation matrix identified five major groups of proteins characterized by shared correlation patterns, broadly corresponding to: Inflammatory and Chemokine signaling; Cytokine Signaling and Immune Modulation; Growth Factor Signaling; Matrix Remodeling and Invasion; Immune Costimulatory Signaling (Figure S1).

Differential protein expression by NELSON criteria and identified new lung cancers

Differential protein expression was analyzed in four distinct contexts: NELSON radiographic categories (negative, indeterminate, and positive) and incident lung cancer diagnosed during follow-up. We excluded all non-lung cancers (n=5) since radiographic assessment was restricted to the thoracic region, and including non-lung cancers could introduce confounding factors and increase noise in the analysis.

When comparing protein expression across the groups defined by NELSON LDCT criteria, we observed that positive cases differed from negative and indeterminate with significant overexpression of multiple proteins, whereas only minor differences were observed between indeterminate and negative categories. Linear mixed models (LMMs) were used to assess differences between the groups. We explored unadjusted models and models adjusted for age (>65 years), sex, and smoking history (≥30 pack-years). When comparing protein expression NELSON positive to negative showed the highest number of overexpressed proteins. In the unadjusted model, nine proteins were overexpressed while in adjusted models, seven out of nine proteins retained their statistical significance. Only minor differences were detected between proteomic profiles of indeterminate vs. negative and positive vs. indeterminate categories with respectively two (CXCL12 and PD-L2) and one protein (IL-6) were significantly upregulated in the comparisons (Figure 2).

Figure 2 Differential plasma protein expression visualized as volcano plots. LDCT findings stratified by NELSON criteria (positive, indeterminate and negative pairwise contrasts). Points represent log₂ fold change vs. −log₁₀(p). Adjusted versions of the analyses include age, sex, and pack-years as covariates; unadjusted plots are shown alongside for comparison. Int, intermediate; LDCT, low-dose computed tomography; Neg, negative; Pos, positive.

When comparing participants who developed lung cancer during follow-up (n=6) with those who did not, 20 proteins were identified as significantly differentially expressed. After adjusting for age, sex and smoking history in the LMM analysis, we found 20 proteins to be significantly overexpressed in lung cancer cases and 16 of these were shared between the unadjusted and adjusted models (Figure 3).

Figure 3 Differential plasma protein expression comparing incident lung cancer cases identified during follow-up vs. participants without cancer. Points represent log₂ fold change vs. −log₁₀(p). Proteins meeting FDR <0.05 are colored red and annotated by name. Adjusted versions of the analyses include age, sex, and pack-years as covariates; unadjusted plots are shown alongside for comparison. FC, fold change; FDR, false discovery rate.

Exploratory prediction of incident lung cancer

A logistic regression classification model was trained and evaluated to explore the ability of circulating blood proteins to predict incident lung cancer (n=6) diagnosed during LDCT follow-up. To limit confounding factors, all participants diagnosed with non-lung cancer during the study follow-up (n=5) were excluded from the analysis. Full cohort was analyzed and seven proteins were selected for the predictive model using LASSO regression with six-fold cross-validation, based on the minimum binomial deviance. Logistic regression model was trained with the following proteins: TNFRSF9, FGF2, CCL20, CD4, MMP12, CXCL11, and CD40.

Prediction performance was evaluated using nested stratified six-fold cross-validation using a fixed full-cohort selected protein panel. The model yielded a pooled out-of-fold ROC and precision-recall (PR) curves. The ROC analysis yielded an AUC of 0.915 (95% CI: 0.829–1.000), while the PR curve showed a modest AUC of 0.249, reflecting the trade-off between precision and sensitivity in the context of heavy class imbalance (Figure S2). As a sensitivity analysis, a prediction model consisting only of the most important clinical lung cancer risk factors (age and pack-years) was evaluated. This model resulted in lung cancer prediction with an AUC of 0.757 (95% CI: 0.618 to 0.897) (not shown). However, given the strong class imbalance, precision-recall analysis was also performed with the model achieving a PR-AUC of 0.062 (not shown).

A separate unpenalized multivariable logistic regression was used to calculate descriptive protein-specific ORs for the individual proteins selected for the full-cohort penalized classification model. The mean ORs with Wald 95% CIs were visualized using a forest plot. All protein-specific CIs crossed the null value of OR =1, reflecting uncertainty associated with small number of incident lung cancer cases. Among the selected proteins, TNFRSF9 exhibited the highest OR (3.43, 95% CI: 0.82 to 14.32) (Figure 4).

Figure 4 Forest plot of adjusted protein-specific odds ratios for theincident lung cancer model. Odds ratios are shown per standard deviation increase in NPX with 95% CIs; vertical line indicates OR =1. Model was adjusted for age, sex, and pack-years. CI, confidence interval; NPX, normalized protein expression; OR, odds ratio; SD, standard deviation.

Assessment of collinearity among selected proteins

Next, we assessed whether the 20 differentially expressed proteins in the lung cancer group and the seven proteins selected for the incident lung cancer predictive machine learning model share correlation. A Spearman rank correlation test was used to assess the pairwise associations and the results were visualized as a correlation heatmap. Overall, most of the protein pairs demonstrated weak to very weak correlation coefficients (CC). Strong correlation was observed only for a few proteins including CXCL1, EGF, PDGF (Figure S3).

Biological pathways enriched in lung cancers

ORA was performed using the MSigDB Hallmark, Reactome, KEGG and Gene Ontology collections. The background universe was restricted to the 92 assayed proteins. Proteins were considered differentially expressed at FDR <0.05. Multiple pathways were significantly enriched after FDR correction (q<0.25). We highlight four representative pathways: TNFα signaling via NFκB, IL-6-JAK-STAT3, cytokine receptor interaction, and chemokine receptor binding (Figure 5).

Figure 5 ORA of plasma proteins associated with incident lung cancer. Cnetplot showing enriched pathways connected to their member proteins. Node color intensity reflects log₂ fold change. FC, fold change; ORA, over-representation analysis.

Discussion

In this prospective study, we investigated the use of proteomic profiles in individuals participating in LDCT lung cancer screening. We hypothesized that blood-based proteomic signatures provide complementary information beyond LDCT findings and may improve the identification of clinically significant radiographic findings and incident lung cancer. Previous studies in the field have identified that lung cancer predisposition is related to many blood-inflammatory proteins (10,19) and systemic inflammation is known to have a central role in lung carcinogenesis (20,21). Based on this evidence, we selected a targeted proteomic panel focusing on inflammatory markers. An important strength of the study was that pre-diagnostic blood samples were collected within a prospective LDCT lung cancer screening trial ensuring a homogeneous well-characterized screening cohort.

Many previous blood biomarker studies in lung cancer screening have primarily investigated prediction of cancer as an endpoint (19), whereas prediction of radiographic findings has received comparatively little attention. In this respect, our study provides an additional perspective on early lung cancer detection by evaluating proteomic associations with LDCT findings. We identified several plasma proteins associated with clinically significant LDCT findings, and these associations remained largely consistent after adjustment for key clinical covariates. This suggests that the detected proteomic signals are not solely driven by age, sex or smoking history.

Among participants classified as positive in the NELSON criteria, nine proteins were differentially expressed in unadjusted analyses, with seven retaining statistical significance after adjustment, whereas only a limited number of differences were observed in comparisons involving the indeterminate category. Differentially expressed proteins in NELSON positive category have previously been observed to be overexpressed in various types of cancer with accumulating evidence about their role in immune evasion and tumor progression (22,23). These findings suggest that negative and positive cases represent biologically distinct profiles, whereas the indeterminate group may reflect a more heterogeneous or transitional state. This observation highlights the practical relevance of the three-class categorization used by the NELSON criteria.

Importantly, the structure of the NELSON screening protocol introduces a degree of temporal validation that strengthens the validity of our observations. Specifically, the NODCAT/GROWCAT-based classification incorporates findings from at least two temporally distinct LDCT scans, enabling exclusion of transient inflammatory conditions, such as respiratory infections, that could confound interpretation based on a single time point. Suspicious findings identified either at baseline or at the 1-year follow-up were reassessed with short-term (~3-month) imaging, and all participants underwent annual follow-up. This longitudinal design reduces the risk of misclassification and enhances the biological relevance of the observed proteomic signatures.

Although the number of incident lung cancer cases in our study was limited, we observed consistent and biologically plausible associations in plasma proteomic profiles. Among participants with newly diagnosed lung cancers 20 proteins were found to be overexpressed with CD40, CXCL9/11, and TNFRSF9 showing largest effect sizes and strongest statistical significance. As expected, the majority of differentially expressed proteins overlapped between NELSON positive cases and individuals diagnosed with lung cancer. Notably, MMP12 and CXCL9 have previously been linked to lung cancer risk in blood proteomics studies (10,19).

Seven proteins (TNFRSF9, FGF2, CCL20, CD4, MMP12, CXCL11, and CD40) were identified to be associated with incident lung cancer and selected for an exploratory machine learning model. The model demonstrated strong discrimination with ROC-AUC =0.915 (95% CI: 0.829–1.000) whereas the lower PR-AUC (0.249) reflects the expected reduction in precision under marked class imbalance and a limited number of cases. In this context, ROC-AUC may overestimate performance, whereas PR-AUC provides a more conservative assessment. For pre-diagnostic cancer screening, high sensitivity is generally prioritized even at the expense of precision. Notably the generated protein model outperformed age- and pack-year-based model (AUC 0.757), suggesting added predictive value beyond conventional risk stratification. Previous studies have reported comparable performance using multi-protein models of varying size (24,25). Beyond proteomics, several other methods have also been explored to predict cancer risk in asymptomatic individuals. Circulating DNA methylation pattern (26,27), genomic profiling (28), lipidomics (29),and integrated multiplatform, such as the Shield platform from Guardant Health (30) have been studied with varying performance in early cancer detection.

Among the enriched biological pathways, the lung cancer-positive group was primarily driven by chemokines CXCL1, CXCL9, CXCL11, CCL20, and the costimulatory receptor TNFRSF9. These proteins are known to participate in pathways such as TNF-α/NF-κB and IL6-JAK-STAT3 signaling which are commonly upregulated in inflamed cancer types (31,32). CXCR3 and the corresponding ligands CXCL9 and CXCL11 are known to promote antitumor response by inducing T and NK cell tumor infiltration (33). However, CXCR3 has also been linked to tumor progression and depending on the tumor type high CXCR3 expression may correlate with more aggressive cancer (34). Given that pathway enrichment was performed on a targeted 92-protein panel with few cases, these findings should be viewed as hypothesis-generating.

The most significant finding of the current study is the characterization of blood proteomic profiles and their potential integration with LDCT lung cancer screening within a prospectively collected clinical trial cohort. To our knowledge, most prior blood-based proteomic studies have been carried out with retrospective material mainly derived from biobank collections. A key strength of this study is the prospective design within an LDCT screening cohort, which more closely resembles a real-life clinical decision-making compared to retrospective biobank-based analyses. This setting enables evaluation of proteomic biomarkers in a pre-diagnostic context, which is critical for early detection strategies. Importantly, this integrative approach which combines imaging and plasma proteomics enables investigation of early disease biology, providing insights into systemic alterations that may precede clinically apparent cancer and reflect early tumor-host interactions.

Although LDCT screening for lung cancer is recommended in several countries, its uptake has been limited due to numerous hurdles. Radiographic imaging can be limited by availability of scanners and trained staff, radiation exposure, and costs. In addition, LDCT screening provides a high rate of false-positive findings, which require additional radiographic follow-up or invasive procedures(2,3). In our study, LDCT positive cases by NELSON criteria differed significantly by blood proteomics from indeterminate or negative cases while there is little difference between indeterminate and negative. Furthermore, some clinical evidence exists that targeting inflammation can reduce the risk for lung cancer (35). Hypothetically, blood proteomic profiling could be used in patient inclusion for studies investigating anti-inflammatory cancer prevention therapies.

There are several limitations in our study that should be acknowledged. The sample size was restricted (n=200) and all participants were recruited in a single center, which may limit generalizability. The follow-up period was relatively short (16–25 months), and only six incident lung cancer cases occurred. Given the small number of events, uncertainty remains in the cancer prediction models and pathway enrichment analyses, and these should be cautiously interpreted. Additionally, the blood samples analyzed were collected only at baseline, reflecting the immunological status at a single time point, transient events, such as infections may have influenced both LDCT findings and the plasma proteome. Although the study protocol helps mitigate these uncertainties by implementing follow-up LDCT scans and study nurse visit before blood sampling, longitudinal sampling would provide more insight into the dynamics of proteomic profiles. Machine learning models bear uncertainty with an unbalanced dataset and internal evaluation is prone to overfitting and should be interpreted as hypothesis-generating. The study results remain hypothesis-generating until they are validated in a separate cohort. The prospective study population will continue to mature, and follow-up will continue for at least three years, enabling updated analyses with a larger number of events and improved statistical power.

Conclusions

Baseline plasma proteomic profiles were associated with cumulative LDCT categories and incident lung cancer in this prospective pilot cohort. While not providing clinical utility validation our study generates a hypothesis that plasma proteomic profiling could identify asymptomatic individuals at higher risk for clinically significant LDCT nodules and lung cancer. If validated in a larger independent cohort, this approach could open possibilities for selecting individuals for lung cancer screening and nodule characterization.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0646/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0646/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0646/prf

Funding: This study was supported by Finnish Medical Foundation, Finnish Cultural Foundation, Emil Aaltonen Foundation, Ida Montin Foundation, Cancer Society of Northern Finland, Cancer Foundation Finland, Sigrid Juselius Foundation, Roche, and AstraZeneca.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2026-0646/coif). S.I. reports personal honoraria for lectures from AstraZeneca, Janssen, and Pierre Fabre; support for congress travel from Janssen, Roche, Novartis, and AstraZeneca; and personal payments for advisory board participation from BMS, Pierre Fabre, and Merck. J.P.K. reports institutional research support for the present study from Roche and AstraZeneca and research funding from the Cancer Foundation Finland and the Sigrid Juselius Foundation. He reports personal consulting fees from Roche, AstraZeneca, Janssen, BMS, Merck, Amgen, Novartis, Merck KGaA, Sanofi, and Pfizer; personal lecture honoraria from Siemens Healthineers and AstraZeneca; part-time employment with Sartar Therapeutics; and former part-time employment with Faron Pharmaceuticals. The other authors have no conflicts of interest to declare.

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. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Northern Ostrobothnia Hospital District, the regional ethics committee responsible for Oulu University Hospital at the time of the study approval (EETTKM 21/2022). All the participants signed an informed consent before any study procedures.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sung H, Filho AM, Laversanne M, et al. Global cancer statistics 2024: GLOBOCAN estimates of incidence and mortality worldwide for 34 cancers in 186 countries. CA Cancer J Clin 2026;76:e70090. [Crossref] [PubMed]
2. Aberle DR, Adams AMNational Lung Screening Trial Research Team, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 2011;365:395-409. [Crossref] [PubMed]
3. de Koning HJ, van der Aalst CM, de Jong PA, et al. Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial. N Engl J Med 2020;382:503-13. [Crossref] [PubMed]
4. Martini K, Chassagnon G, Frauenfelder T, et al. Ongoing challenges in implementation of lung cancer screening. Transl Lung Cancer Res 2021;10:2347-55. [Crossref] [PubMed]
5. Burzic A, O'Dowd EL, Baldwin DR. The Future of Lung Cancer Screening: Current Challenges and Research Priorities. Cancer Manag Res 2022;14:637-45. [Crossref] [PubMed]
6. Tammemägi MC, Katki HA, Hocking WG, et al. Selection criteria for lung-cancer screening. N Engl J Med 2013;368:728-36. [Crossref] [PubMed]
7. Hanash SM, Ostrin EJ, Fahrmann JF. Blood based biomarkers beyond genomics for lung cancer screening. Transl Lung Cancer Res 2018;7:327-35. [Crossref] [PubMed]
8. Guida F, Sun NIntegrative Analysis of Lung Cancer Etiology and Risk (INTEGRAL) Consortium for Early Detection of Lung Cancer, et al. Assessment of Lung Cancer Risk on the Basis of a Biomarker Panel of Circulating Proteins. JAMA Oncol 2018;4:e182078. [Crossref] [PubMed]
9. Davies MPA, Sato T, Ashoor H, et al. Plasma protein biomarkers for early prediction of lung cancer. EBioMedicine 2023;93:104686. [Crossref] [PubMed]
10. Khodayari Moez E, Warkentin MT, Brhane Y, et al. Circulating proteome for pulmonary nodule malignancy. J Natl Cancer Inst 2023;115:1060-70. [Crossref] [PubMed]
11. Fahrmann JF, Marsh T, Irajizad E, et al. Blood-Based Biomarker Panel for Personalized Lung Cancer Risk Assessment. J Clin Oncol 2022;40:876-83. [Crossref] [PubMed]
12. Iivanainen S, Kurtti A, Wichmann V, et al. Smartphone application versus written material for smoking reduction and cessation in individuals undergoing low-dose computed tomography (LDCT) screening for lung cancer: a phase II open-label randomised controlled trial. Lancet Reg Health Eur 2024;42:100946. [Crossref] [PubMed]
13. Yousaf-Khan U, van der Aalst C, de Jong PA, et al. Final screening round of the NELSON lung cancer screening trial: the effect of a 2.5-year screening interval. Thorax 2017;72:48-56. [Crossref] [PubMed]
14. Liberzon A, Subramanian A, Pinchback R, et al. Molecular signatures database (MSigDB) 3.0. Bioinformatics 2011;27:1739-40. [Crossref] [PubMed]
15. Milacic M, Beavers D, Conley P, et al. The Reactome Pathway Knowledgebase 2024. Nucleic Acids Res 2024;52:D672-8. [Crossref] [PubMed]
16. Gene Ontology Consortium. The Gene Ontology resource: enriching a GOld mine. Nucleic Acids Res 2021;49:D325-34. [Crossref] [PubMed]
17. Kanehisa M, Furumichi M, Sato Y, et al. KEGG: biological systems database as a model of the real world. Nucleic Acids Res 2025;53:D672-7. [Crossref] [PubMed]
18. Seijo LM, Peled N, Ajona D, et al. Biomarkers in Lung Cancer Screening: Achievements, Promises, and Challenges. J Thorac Oncol 2019;14:343-57. [Crossref] [PubMed]
19. Lung Cancer Cohort Consortium (LC3). The blood proteome of imminent lung cancer diagnosis. Nat Commun 2023;14:3042. [Crossref] [PubMed]
20. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011;144:646-74. [Crossref] [PubMed]
21. Hill W, Lim EL, Weeden CE, et al. Lung adenocarcinoma promotion by air pollutants. Nature 2023;616:159-67. [Crossref] [PubMed]
22. Tokunaga R, Zhang W, Naseem M, et al. CXCL9, CXCL10, CXCL11/CXCR3 axis for immune activation - A target for novel cancer therapy. Cancer Treat Rev 2018;63:40-7. [Crossref] [PubMed]
23. Yearley JH, Gibson C, Yu N, et al. PD-L2 Expression in Human Tumors: Relevance to Anti-PD-1 Therapy in Cancer. Clin Cancer Res 2017;23:3158-67. [Crossref] [PubMed]
24. Robbins HA, Alcala K, Moez EK, et al. Design and methodological considerations for biomarker discovery and validation in the Integrative Analysis of Lung Cancer Etiology and Risk (INTEGRAL) Program. Ann Epidemiol 2023;77:1-12. [Crossref] [PubMed]
25. Feng X, Wu WY, Onwuka JU, et al. Lung cancer risk discrimination of prediagnostic proteomics measurements compared with existing prediction tools. J Natl Cancer Inst 2023;115:1050-9. [Crossref] [PubMed]
26. Schrag D, Beer TM, McDonnell CH 3rd, et al. Blood-based tests for multicancer early detection (PATHFINDER): a prospective cohort study. Lancet 2023;402:1251-60. [Crossref] [PubMed]
27. Li P, Liu S, Du L, et al. Liquid biopsies based on DNA methylation as biomarkers for the detection and prognosis of lung cancer. Clin Epigenetics 2022;14:118. [Crossref] [PubMed]
28. Malone ER, Oliva M, Sabatini PJB, et al. Molecular profiling for precision cancer therapies. Genome Med 2020;12:8. [Crossref] [PubMed]
29. Wang G, Qiu M, Xing X, et al. Lung cancer scRNA-seq and lipidomics reveal aberrant lipid metabolism for early-stage diagnosis. Sci Transl Med 2022;14:eabk2756. [Crossref] [PubMed]
30. Chung DC, Gray DM 2nd, Singh H, et al. A Cell-free DNA Blood-Based Test for Colorectal Cancer Screening. N Engl J Med 2024;390:973-83. [Crossref] [PubMed]
31. Jorgovanovic D, Song M, Wang L, et al. Roles of IFN-γ in tumor progression and regression: a review. Biomark Res 2020;8:49. [Crossref] [PubMed]
32. Huang B, Lang X, Li X. The role of IL-6/JAK2/STAT3 signaling pathway in cancers. Front Oncol 2022;12:1023177. [Crossref] [PubMed]
33. Russo E, D'Aquino C, Di Censo C, et al. Cxcr3 promotes protection from colorectal cancer liver metastasis by driving NK cell infiltration and plasticity. J Clin Invest 2025;135:e184036. [Crossref] [PubMed]
34. Wang X, Zhang Y, Wang S, et al. The role of CXCR3 and its ligands in cancer. Front Oncol 2022;12:1022688. [Crossref] [PubMed]
35. Ridker PM, MacFadyen JG, Thuren T, et al. Effect of interleukin-1β inhibition with canakinumab on incident lung cancer in patients with atherosclerosis: exploratory results from a randomised, double-blind, placebo-controlled trial. Lancet 2017;390:1833-42. [Crossref] [PubMed]