Blood RNA panel linked to lung cancer risk up to 10 years before diagnosis

A risk score built from 31 small RNA molecules found in blood was linked to lung cancer up to 10 years before diagnosis in a group of current smokers, according to a study published Aug. 14, 2026, in Advanced Science.

Lung cancer is the leading cause of cancer death in the United States. Five-year survival is about 65% for early-stage disease but about 9% once the cancer has spread, according to American Cancer Society (ACS) figures cited in the study. The U.S. Preventive Services Task Force (USPSTF) recommends annual low-dose computed tomography (LDCT) screening for adults ages 50 to 80 with at least a 20 pack-year smoking history who currently smoke or quit within the past 15 years.

However, uptake remains low. An ACS-led study published in November 2025 in JAMA found that only 18.7% of an estimated 12.8 million eligible U.S. adults were up to date with screening in 2024. These criteria also leave out never-smokers and lighter smokers, the study authors noted.

Lead author Zhuokun Feng, a Ph.D. candidate of the Department of Quantitative Health Sciences at the University of Hawaii at Manoa John A. Burns School of Medicine and the University of Hawaii Cancer Center, and her team focused on exosomes, tiny particles cells release into the blood that carry genetic material for this study. They wrote that most exosome research uses blood drawn at or after diagnosis, and few studies have looked at the years before it. Their goal was a more accurate tool for early risk assessment.

How the study worked

The researchers analyzed stored plasma from 202 current smokers in the Hawaii arm of the Multiethnic Cohort Study. Participants smoked at least 10 cigarettes a day, had no prior cancer and were Japanese American, European American or Native Hawaiian. Blood was drawn between 1994 and 2006.

Out of this group, 68 people later developed lung cancer and 134 did not. Follow-up through tumor registries averaged 13 years. The team sequenced four types of small noncoding RNAs and compared eight machine learning models, testing each on samples held out from training. The risk score was then examined in two ways: a logistic regression adjusted for age, sex, ethnicity and smoking intensity and duration, and a time-to-diagnosis analysis that accounted for deaths from other causes. The team also tested the model on 186 samples from people already diagnosed with early-stage lung cancer and controls.

What the researchers found

The final model used 31 small RNAs across all four types. On a held-out test set of 40 samples, it achieved an area under the curve (AUC) of 0.97, where 1.0 means perfect separation of cases from non-cases, with 93% sensitivity and 100% specificity. In the logistic regression analysis, each one-standard-deviation increase in the risk score was associated with 13.19 times higher odds of lung cancer.

Heavy smoking was the only traditional risk factor tied to lung cancer on its own, and it lost significance after adjustment. In the time-to-diagnosis analysis, a higher score was associated with earlier diagnosis.

Overall discrimination held steady across follow-up windows, with average AUCs of 0.93 to 0.95. But the score's ability to concentrate true cases among top scorers weakened with time. Average precision was 0.73 within five years of diagnosis, 0.67 at five to 10 years and 0.33 at 10 to 15 years, which fell below the baseline for that window.

The authors suggest the panel could eventually become a blood test to help decide who is referred for LDCT, including people outside current eligibility rules, such as never-smokers and lighter smokers. That would require a streamlined assay reading only the 31 markers, validated under federal lab standards, with an estimated turnaround of two to three days. “The test should be inexpensive relative to LDCT and could prove cost-effective by reducing scans and false-positive results, although a formal cost-effectiveness analysis and a possible smaller core panel remain to be addressed,” the authors wrote.

Strengths, limitations and next steps

The study's strengths include blood drawn years before diagnosis, long follow-up and models tested only on held-out samples. Still, with only 68 cases, confidence intervals were wide and analyses by subtype or stage were not possible. There was no outside pre-diagnostic group to confirm the results. In the exploratory check among already-diagnosed patients, AUCs dropped to 0.63 to 0.73. Every participant smoked, which the authors said could overstate the score's added value. About one-third of the cohort developed lung cancer, far above the general population rate, and all participants came from three ethnic groups in Hawaii.

The authors wrote that their top priority is to validate the 31-marker panel in a larger, more ethnically diverse long-term cohort. They recommended that future studies include a substantial number of never-smokers and match cancer cases with controls on age, sex and detailed smoking history to test whether the panel adds value beyond known risk factors. They also called for lab studies on how these RNAs may contribute to early cancer development