Lung cancer remains the deadliest cancer in the United States, yet early detection continues to provide one of the strongest opportunities to save lives. Screening with low-dose CT scans can reduce the risk of dying from lung cancer by as much as 20%, per Han. Still, only a fraction of people who qualify for screening actually receive it, and growing evidence suggests that the guidelines determining eligibility may be excluding many of those who need it most.
Stanford Cancer Institute member Summer Han, PhD, associate professor of neurosurgery and of medicine, is helping lead the effort to change lung cancer screening criteria. In a recent study, Han and her team propose a more straightforward approach to identifying high-risk patients, one that could reduce long-standing racial and ethnic disparities in care and make screening more accessible nationwide.
Today, national lung cancer screening eligibility is primarily based on pack-years, a calculation that multiplies the number of cigarette packs smoked per day by the number of years a person has smoked. For example, someone who smoked one pack of cigarettes a day for 20 years, or two packs a day for 10 years, would reach the same total: 20 pack-years. While this metric has been standard for decades, it is difficult to track accurately. Many patients have trouble recalling their smoking intensity over long periods, and clinicians may document smoking history inconsistently across visits. Even electronic health records struggle to maintain reliable smoking data, leaving large information gaps for determining whether a patient qualifies for screening.
These challenges have contributed to profound disparities. The current guidelines, set by the U.S. Preventive Services Task Force (USPSTF), whose recommendations shape national insurance coverage, may unintentionally exclude high-risk groups. Research has shown that Black patients, for example, tend to develop lung cancer at lower levels of smoking exposure compared to white patients. Many never reach the required 20 pack-years threshold, despite having equal or greater risk. Similar trends occur among Latino and Asian populations.
“We know the current criteria miss a lot of people,” Han said. “Pack-years don’t reflect the biological, metabolic, and behavioral differences across groups.”
Han is now exploring an alternative: using the duration of smoking (how many years a person smoked) as the primary measure of exposure. A recent national study suggested that counting years smoked is more intuitive for patients, easier for clinicians to verify, and, critically, more equitable. Intrigued by this proposition, Han and her team set out to test whether a duration-based approach performs better across diverse populations.
Using data from more than 100,000 adults of varied racial and ethnic backgrounds, the team modeled different smoking-duration cutoffs to determine which threshold best predicted lung cancer risk. They found that using 30 years of smoking duration provided the most consistent performance across every racial and ethnic group studied. It significantly reduced disparities in screening eligibility among Black, Latino, Asian, and white patients. It increased the sensitivity of screening, meaning more true cases of lung cancer would be detected.
“This approach isn’t just simpler,” she explained. “It’s fairer and more effective.”
The potential implications are far-reaching. The USPSTF typically updates its lung cancer screening guidelines every five or six years, most recently in 2021. Stanford Cancer Institute researchers contribute to this effort through the National Cancer Institute organization Cancer Intervention and Surveillance Modeling Network (Stanford PI: Sylvia Plevritis, PhD), a modeling consortium whose findings help shape the USPSTF’s decisions. With new evidence pointing toward a more equitable and accessible method of determining eligibility, Han and her team hope these results will play a meaningful role in future national recommendations.
But filling gaps in screening criteria is only part of the story. Smoking data in medical charts is notoriously unreliable. Patients may appear as “current smokers” in one visit and “never smokers” in another. Key details such as intensity, duration, and quit dates are often missing entirely. To correct this, Han trained large language models (LLMs) to extract detailed smoking information from clinical notes, reconstruct complete longitudinal histories, and reconcile contradictory documentation.
The team tested these algorithms across three independent health systems and found that the models achieved 97-98% accuracy in all settings. This cleaned data made screening eligibility easier to calculate and allowed researchers to uncover new insights, such as the risk of developing a second primary lung cancer.
“This work solves real, practical problems,” she said. “If we can reliably extract and clean smoking information, we can finally calculate screening eligibility at scale.”
Shifting national guidelines toward duration-based eligibility may make lung cancer screening more equitable and effective across diverse populations and better aligned with individual risk. Han is optimistic about what this work can mean for patients nationwide.
These studies have the potential to reshape screening in a way that is more accurate, more equitable, and ultimately more lifesaving"
“These studies have the potential to reshape screening in a way that is more accurate, more equitable, and ultimately more lifesaving,” she said.
