AI lung cancer risk model validated in predominantly Black population at hospital

A new study presented at the International Association for the Study of Lung Cancer 2025 World Conference on Lung Cancer (WCLC) validates the use of Sybil, a deep learning artificial intelligence model, for predicting future lung cancer risk in a predominantly Black population.

The study, conducted by the University of Illinois Hospital & Clinics, (UI Health), the academic health enterprise of the University of Illinois Chicago (UIC), highlights Sybil's strong performance in a real-world clinical setting with racially and socioeconomically diverse patients.

The Sybil Implementation Consortium comprises UIC, Mass General Brigham, Baptist Memorial Health Care, Massachusetts Institute of Technology, and WellStar Health System.

While prior United States Sybil validations were conducted in cohorts that were more than 90% white, this new analysis focused on a population where 62% of participants identified as Non-Hispanic Black, 13% Hispanic, and 4% Asian. The model demonstrated high predictive accuracy for lung cancer risk up to six years after a single low-dose CT (LDCT) scan.

"This study confirms that Sybil performs well in a racially and socioeconomically diverse setting, supporting its broader utility for lung cancer screening," said Mary Pasquinelli, lead author, nurse practitioner and the Director of the Lung Screening Program at UI Health and a member of the University of Illinois Cancer Center.

"It shows promise as a tool for improving early detection and addressing disparities in lung cancer outcomes."

Pasquinelli and her colleagues evaluated 2,092 baseline LDCTs from UI Health's lung screening program between 2014 and 2024. Of these, 68 patients were diagnosed with lung cancer, with follow-up times ranging from 0 to 10.2 years, she reported.

The study found that Sybil's Area Under the Curve (AUC) performance for years one through six were:

• 0.94 (1-year)
• 0.90 (2-year)
• 0.86 (3-year)
• 0.85 (4-year)
• 0.80 (5-year)
• 0.79 (6-year)

If a lung cancer screening model has an AUC of 0.94, that means there's a 94% chance the model will correctly rank a randomly chosen patient who develops cancer in the future as higher risk than a randomly chosen patient who does not develop cancer in the near future.

She reported that the results remained strong when restricted to Black participants and after excluding cancers diagnosed within three months of screening.

According to Pasquinelli, the study affirms Sybil's clinical generalizability and suggests that the model may be unbiased with respect to factors like race and ethnicity, demonstrating strong performance in underrepresented communities.

The Sybil Implementation Consortium will now proceed with prospective clinical trials to integrate Sybil into real-world clinical workflows, she said.

Provided by International Association for the Study of Lung Cancer