Two Blood Proteins Linked to Lung Cancer Risk Years Before Diagnosis in People Who Never Smoked

Key Takeaways

• Researchers examined 2,821 blood proteins in 360 people who had never smoked, including 180 who developed lung cancer within five years of giving a blood sample.
• Two proteins, CEACAM5 and CEACAM6, remained associated with subsequent lung cancer after researchers accounted for the large number of proteins tested.
• For each standard-deviation increase in protein levels, the odds of lung cancer were 65% higher for CEACAM5 and 72% higher for CEACAM6 in the matched case-control analysis.
• The associations were also examined in a larger UK Biobank population. However, only 16 lung cancer cases occurred among never-smokers in that dataset, limiting independent confirmation.
• The study did not establish a blood test capable of diagnosing lung cancer, predicting an individual’s absolute risk, or determining who should undergo screening.

Introduction

Researchers identified two blood proteins, CEACAM5 and CEACAM6, associated with future lung cancer diagnosis among people who have never smoked. Further studies are needed before these markers can be used for screening.

Lung cancer is strongly associated with smoking, but it also affects people who have never used tobacco. Approximately 20% to 25% of lung cancers worldwide occur in people who have never smoked, creating a challenge for screening programs that primarily focus on people with substantial smoking histories.

Most people who have never smoked face a relatively low risk of lung cancer. Screening everyone in this population with repeated CT scans could expose many people to unnecessary testing, false alarms, and treatment of cancers that might never have caused harm.

Researchers are therefore investigating whether blood-based biomarkers could help distinguish people who might benefit from screening from those unlikely to need it.

A study published November 2, 2026, in JAMA Network Open identified two circulating proteins associated with lung cancer diagnosed as long as five years after blood collection.

The research, led by Hana Zahed and colleagues, offers a possible starting point for more individualized risk assessment. However, the evidence remains preliminary, particularly when it comes to independently confirming the findings among people who have never smoked (Zahed et al., 2026).

What the Study Examined

Blood samples collected before lung cancer diagnosis

The researchers used a nested case-control design, examining blood samples already collected through four long-running population studies in the United States and China.

The discovery analysis included 180 people who developed lung cancer within five years of blood collection and 180 matched participants who served as controls.

Participants were matched by age, sex, study cohort, and blood collection date to reduce differences that could influence the results. Researchers then used conditional logistic regression, a statistical method designed for matched case-control studies, to examine the relationship between protein levels and subsequent lung cancer diagnosis.

The discovery analysis did not include additional adjustments for other potential risk factors, such as environmental exposures or family history. Therefore, the findings establish statistical associations rather than proving that the proteins independently predict lung cancer risk beyond these factors.

All 360 participants were classified as people who had never smoked, although the precise definition differed slightly between cohorts. Three studies used a lifetime threshold of fewer than 100 cigarettes, while one allowed fewer than 400 cigarettes.

The participants had a median age of approximately 66 years. Women accounted for 256 participants, or 71.1%, while 82.2% were Asian and 17.8% were non-Hispanic White.

The blood samples were collected between 1996 and 2006, depending on the cohort. Among participants who developed lung cancer, the median interval between blood collection and diagnosis was 2.7 years.

Researchers measured circulating proteins using the Olink Explore 3072 platform. After excluding proteins that could not be adequately measured and resolving duplicate measurements, they analyzed 2,821 unique proteins.

The measurements represented relative protein concentrations, rather than the absolute concentrations that might eventually be used in a clinical test.

A separate population for replication

The researchers then examined UK Biobank data to determine whether the proteins identified in the discovery analysis were associated with lung cancer in another population.

This dataset contained 49,291 participants, including 157 who developed lung cancer during five years of follow-up.

The replication population differed substantially from the original group. Most participants were White, and most of the lung cancer cases occurred in people who currently or previously smoked.

Of the 157 lung cancer cases, only 16 occurred among never-smokers.

That difference matters because the study’s central question concerned people without a smoking history. Although the larger dataset could test whether the proteins were associated with lung cancer more generally, it offered much less certainty about whether the results would hold specifically among never-smokers.

What Researchers Found

Only two of 2,821 proteins survived statistical correction

The initial analysis identified 115 proteins associated with subsequent lung cancer at a conventional statistical significance threshold of P below .05.

However, testing thousands of proteins creates many opportunities for apparently significant results to arise by chance.

After the researchers applied a false discovery rate correction, only two proteins remained statistically significant: carcinoembryonic antigen-related cell adhesion molecule 5 (CEACAM5) and carcinoembryonic antigen-related cell adhesion molecule 6 (CEACAM6).

Both had a false discovery rate-adjusted P value of .04.

These results correspond to 65% higher odds of lung cancer for CEACAM5 and 72% higher odds for CEACAM6 per standard-deviation increase in the measured protein level.

They do not mean that a person with an elevated protein level has a 65% or 72% chance of developing lung cancer.

The study compared people who developed cancer with matched controls, so the odds ratios cannot directly establish an individual’s absolute probability of disease.

The associations were also examined in UK Biobank

In the larger replication dataset, the researchers observed similar associations after adjusting for age, sex, smoking status, smoking intensity, and smoking duration.

For each standard-deviation increase in protein level:

• CEACAM5: Hazard ratio of 1.92 (95% CI, 1.66–2.22).
• CEACAM6: Hazard ratio of 1.86 (95% CI, 1.59–2.16).

These results indicate higher relative rates of lung cancer diagnosis during follow-up among participants with higher protein levels.

But the findings require an important qualification.

Among UK Biobank participants who had never smoked, the hazard ratio was 1.84 (95% CI, 1.16–2.91) for CEACAM5 and 1.49 (95% CI, 0.88–2.52) for CEACAM6.

The never-smoking subgroup included only 16 people who developed lung cancer, making these estimates considerably less precise than those from the full UK Biobank population.

CEACAM5 remained statistically associated with lung cancer in this subgroup, but its confidence interval was relatively wide. CEACAM6 showed an association in the same direction, although its confidence interval included 1, meaning the result was not statistically conclusive.

The findings therefore provide some supporting evidence for CEACAM5 among never-smokers, while the evidence for CEACAM6 remains less certain. Neither protein has yet been adequately validated for clinical risk prediction specifically in people who have never smoked.

The proteins were associated with cancer diagnosed several years later

The researchers also examined whether the associations weakened when blood samples had been collected further from the eventual cancer diagnosis.

Among the original never-smoking participants, the associations remained evident even when blood had been collected more than 2.5 years before diagnosis.

The associations were numerically weaker at longer intervals, but the study did not establish a statistically significant trend in their strength over time.

This is important because a marker associated with cancer years before diagnosis could potentially contribute to future risk assessment.

However, it does not establish when a tumor first developed or whether the proteins would detect cancer at a stage when treatment is more effective.

The two proteins may provide different information

CEACAM5 and CEACAM6 belong to the same family of proteins involved in cell adhesion and other biological processes relevant to cancer.

Despite their similarities, their measured levels were only weakly correlated in the discovery population, with a Pearson correlation coefficient of 0.25.

When the researchers included both proteins in the same statistical model, their associations with lung cancer remained, although their regression coefficients decreased by approximately 30%.

This raises the possibility that the two markers provide partly complementary information rather than simply measuring the same biological signal.

Whether combining them meaningfully improves an individual’s risk prediction remains unanswered.

How the Findings Compare With Earlier Research

Many biomarkers associated with smoking-related lung cancer did not perform similarly

One of the study’s most interesting findings came from a comparison with earlier research on people with a smoking history.

A previous analysis by the Lung Cancer Cohort Consortium identified 36 blood proteins associated with future lung cancer among people who had smoked (Lung Cancer Cohort Consortium, 2023).

In that earlier population, the median odds ratio across the 36 proteins was 1.47.

When the current researchers examined those same markers among people who had never smoked, the median odds ratio was only 1.01, indicating little overall association.

Just four of the 36 proteins showed statistically significant associations in the same direction among never-smokers: CEACAM5, ALPP, SPINT1, and WFDC2.

CEACAM6 was not measured in the earlier analysis.

The comparison suggests that biomarkers identified in people with smoking-related lung cancer may not automatically be useful for people who have never smoked.

However, the studies used different protein measurement platforms and different prediction periods: three years in the earlier smoking-related analysis versus five years in the current never-smoking analysis. Those differences limit direct comparisons.

Why existing lung cancer screening cannot simply be expanded to everyone

Large randomized trials have demonstrated that low-dose CT screening can reduce lung cancer deaths in appropriately selected people with substantial smoking histories.

The National Lung Screening Trial reported a reduction in lung cancer mortality with low-dose CT compared with chest radiography (National Lung Screening Trial Research Team, 2011).

The European NELSON trial also demonstrated a lung cancer mortality benefit in a high-risk screening population (de Koning et al., 2020).

Those findings do not establish that routine screening provides the same balance of benefits and harms for people who have never smoked.

A major concern is overdiagnosis, in which screening detects cancers that would not have caused symptoms or death during a person’s lifetime.

A study examining CT screening patterns among Asian women reported findings consistent with substantial overdiagnosis following expanded screening (Gao et al., 2022).

The challenge is therefore not simply to detect more abnormalities. It is to identify people whose risk is high enough that screening is likely to provide more benefit than harm.

What the Results May Mean

A possible route toward more personalized risk assessment

CEACAM5 and CEACAM6 could eventually contribute to a model that combines blood protein measurements with other information, such as age, family history, environmental exposures, and other established risk factors.

The goal would be to estimate an individual’s probability of developing lung cancer over a defined period, rather than relying on a single laboratory measurement.

But the current study did not demonstrate that these proteins can achieve that goal.

It did not report a validated clinical cutoff, establish a reliable absolute-risk prediction model for never-smokers, or show that using these markers to select people for CT screening improves outcomes.

The distinction between relative risk and absolute risk is particularly important.

A marker associated with a substantial relative increase in risk may still have limited clinical usefulness when the underlying probability of disease is low.

Researchers would need to demonstrate that the biomarkers improve prediction beyond conventional risk factors, accurately distinguish higher-risk from lower-risk individuals, and produce decisions that benefit patients.

The proteins are biologically plausible, but not specific to lung cancer

CEACAM5 and CEACAM6 are involved in cell adhesion and processes related to tumor growth and progression.

CEACAM5 is already used in clinical medicine as a marker in certain cancer-monitoring settings, particularly colorectal cancer.

CEACAM6 has also been investigated in several malignancies, including lung, pancreatic, and gastric cancers.

Their involvement in cancer biology makes the observed associations plausible, but it also creates a potential challenge: an elevated protein level may not uniquely indicate lung cancer.

Further studies would need to determine whether these markers can distinguish lung cancer risk from other conditions and whether they add clinically useful information when measured together.

Should People Who Never Smoked Get Tested for These Proteins?

Not based on this study alone.

Although CEACAM5 and CEACAM6 were associated with lung cancer diagnosed within five years of blood collection, researchers have not established a reliable blood test that uses these proteins to determine who should undergo lung cancer screening.

The study did not identify a clinical cutoff for either protein or demonstrate that measuring them improves survival or reduces lung cancer deaths.

For people who have never smoked, decisions about lung cancer risk should consider individual circumstances, including family history, exposure to radon or secondhand smoke, and relevant occupational or environmental exposures.

People concerned about their lung cancer risk can discuss these factors with a healthcare professional. Anyone experiencing persistent symptoms, such as an unexplained cough, coughing up blood, or unexplained weight loss, should seek medical evaluation rather than relying on an experimental biomarker test.

For now, CEACAM5 and CEACAM6 remain research markers, not established tools for routine lung cancer screening.

Limitations

Several limitations prevent these findings from being translated directly into screening recommendations.

First, the discovery sample was small relative to the number of proteins tested. The researchers examined 2,821 proteins using only 180 lung cancer cases and 180 matched controls. The statistical correction reduced the likelihood of false-positive findings, but the study may have missed proteins with weaker associations.

Second, independent validation among never-smokers was limited. Only 16 never-smokers developed lung cancer in the UK Biobank replication dataset. Most of the replication evidence therefore came from people with a smoking history.

Third, the study could not establish whether the proteins reliably identify early-stage lung cancer. In the discovery population, 55 cases (30.6%) were diagnosed at stage I or II, 58 (32.2%) at stage III or IV, and 67 (37.2%) had unknown or missing stage information. Cancer stage was unavailable for all 157 cases in the UK Biobank replication population.

Because of these gaps, the researchers could not determine whether CEACAM5 and CEACAM6 are particularly effective at identifying cancers at an early, potentially curable stage. Detecting a statistical association years before diagnosis does not necessarily mean that a biomarker can detect cancer early enough to improve treatment outcomes.

Fourth, the discovery population was not representative of every population of never-smokers. More than four-fifths of participants were Asian, and women made up over 70%. The UK Biobank population was predominantly White. Further research is needed across populations with different backgrounds and exposures.

Fifth, the study did not establish clinical prediction performance. It did not demonstrate a screening threshold, the number of false positives that might occur in practice, or the number of cancers that a biomarker-guided screening strategy could detect.

Finally, the observational design cannot determine whether elevated proteins contribute to cancer development, reflect biological changes associated with an existing undiagnosed tumor, or are related to other processes.

Frequently Asked Questions

Can people who have never smoked develop lung cancer?

Yes. Approximately 20% to 25% of lung cancers occur in people who have never smoked. Possible risk factors include radon, secondhand smoke, air pollution, and certain occupational exposures.

What are CEACAM5 and CEACAM6?

They are proteins involved in cell adhesion and other biological processes. Researchers found that higher levels in blood were associated with a greater likelihood of a subsequent lung cancer diagnosis.

Does having high levels of these proteins mean I have lung cancer?

No. Elevated levels do not confirm cancer. These proteins can be associated with other conditions, and the study did not establish a diagnostic cutoff.

What does a 65% or 72% increase in odds mean?

It means higher relative odds of lung cancer for each standard-deviation increase in the respective protein level. It does not mean a person has a 65% or 72% chance of developing cancer.

What is a standard deviation?

It is a statistical measure of how much values vary within a group. Researchers used it to compare differences in protein levels across participants.

What is the difference between an odds ratio and a hazard ratio?

An odds ratio compares the odds of an outcome between groups. A hazard ratio compares the relative rate at which an outcome occurs during follow-up. Neither directly tells someone their personal probability of developing cancer.

Can these proteins detect lung cancer five years early?

Not necessarily. The proteins were associated with cancers diagnosed up to five years after blood collection, but researchers did not establish whether they can reliably detect early-stage tumors.

How reliable were the findings?

Both proteins showed statistically significant associations in the original study. However, independent confirmation specifically among never-smokers was limited because only 16 developed lung cancer in the replication dataset.

Are blood tests for CEACAM5 and CEACAM6 available?

Some clinical tests measure CEACAM5, commonly called CEA, for other medical purposes. However, no validated screening test using these two proteins has been established for predicting lung cancer in people who have never smoked.

Should people who have never smoked get a CT scan to check for lung cancer?

Not routinely based on these findings. Established screening recommendations primarily target people with substantial smoking histories. Individual concerns and risk factors should be discussed with a healthcare professional.

What symptoms of lung cancer should people watch for?

A persistent or worsening cough, coughing up blood, unexplained weight loss, chest pain, or unexplained shortness of breath can warrant medical evaluation. These symptoms have many possible causes and do not necessarily indicate cancer.

What needs to happen before these biomarkers could be used in screening?

Larger studies must confirm their accuracy in never-smokers, establish meaningful risk thresholds, and determine whether using the proteins to guide screening actually improves patient outcomes.

Final Thoughts

The identification of CEACAM5 and CEACAM6 offers a potentially useful direction for research into lung cancer among people who have never smoked.

The most notable result is that both proteins were associated with lung cancer diagnosed years after blood collection, and they appeared to provide partly independent information.

Equally important, most protein markers previously linked to smoking-related lung cancer showed little association in the never-smoking discovery population. This supports investigating risk assessment approaches specifically designed for people without a smoking history.

The findings offer a promising direction for future lung cancer research, particularly for people who have never smoked and are not typically included in established screening programs. But important questions remain about accuracy, individual risk prediction, and whether biomarker-guided screening would improve outcomes. Until larger studies answer those questions, CEACAM5 and CEACAM6 should be considered research biomarkers rather than blood tests ready for routine clinical use.

Funding and Disclosures

The study was supported by grants from the US National Cancer Institute, including INTEGRAL-AT U19CA203654 and LEAP R01CA262164. Hana Zahed also received partial support from the International Association for the Study of Lung Cancer/International Lung Cancer Foundation.

The authors reported that the funders had no role in study design, data collection, analysis, interpretation, manuscript preparation, or the decision to submit the findings for publication.

Three authors, Hana Zahed, Mattias Johansson, and Hilary A. Robbins, were named inventors on a planned provisional patent application related to the INTEGRAL-Risk model, a protein-based lung cancer risk assessment tool. They reported that they would not receive royalties or other compensation related to the potential patent.

Loïc Le Marchand reported grant support from the National Cancer Institute through the University of Hawaiʻi, and Xiao-Ou Shu reported National Institutes of Health grant support. No other disclosures were reported.

References

Zahed, H., Bulos-Salim, C., Alcala, K., AlWaheidi, S., Feng, X., Wang, Y., Patel, A. V., Le Marchand, L., Zheng, W., Shu, X.-O., Johansson, M., & Robbins, H. A. (2026). Protein biomarkers of lung cancer risk among individuals who never smoked. JAMA Network Open, 9(11), e2641494. https://doi.org/10.1001/jamanetworkopen.2026.41494

Lung Cancer Cohort Consortium. (2023). The blood proteome of imminent lung cancer diagnosis. Nature Communications, 14, Article 3042. https://doi.org/10.1038/s41467-023-37979-8

National Lung Screening Trial Research Team. (2011). Reduced lung-cancer mortality with low-dose computed tomographic screening. The New England Journal of Medicine, 365(5), 395–409. https://doi.org/10.1056/NEJMoa1102873

de Koning, H. J., et al. (2020). Reduced lung-cancer mortality with volume CT screening in a randomized trial. The New England Journal of Medicine, 382(6), 503–513. https://doi.org/10.1056/NEJMoa1911793

Gao, W., Wen, C. P., Wu, A., & Welch, H. G. (2022). Association of computed tomographic screening promotion with lung cancer overdiagnosis among Asian women. JAMA Internal Medicine, 182(3), 283–290. https://doi.org/10.1001/jamainternmed.2021.7769