AI Models Outperform Standard Biomarkers for Lung Cancer Immunotherapy

Machine learning models can predict which advanced lung cancer patients will respond to immunotherapy more accurately than current clinical biomarkers, according to findings from a large international study published in Nature Medicine.

The I3LUNG project enrolled 2,396 patients with advanced non-small cell lung cancer (NSCLC) across six centers in Italy, Germany, Greece, Israel, Spain, and the United States. Researchers integrated clinical, imaging, pathology, and genomic data to train AI models that predict treatment response and survival outcomes.

Performance exceeds current standards

The study tested two AI model families. The first, using clinical and blood data alone, achieved an Area Under the Curve (AUC) score of 0.77. The second, incorporating clinical, blood, imaging, and digital pathology data, reached an AUC of 0.88—a score considered excellent in machine learning classification. Both consistently outperformed standard clinical biomarkers.

For context, physicians currently rely heavily on PD-L1 expression to guide immunotherapy decisions, despite well-documented limitations. Immunotherapy achieves long-term benefit in only 20% to 30% of NSCLC patients, while most develop treatment resistance. Better prediction tools could spare patients unnecessary toxicity and cost.

Human-AI collaboration shows promise

Twenty physicians—10 lung cancer specialists and 10 from other disciplines—reviewed 100 patient cases twice: first independently, then with AI support. Access to the AI tool improved their ability to identify treatment responders from an AUC of 0.72 to 0.87.

Non-specialist physicians showed the greatest improvement, a finding with direct implications for community oncology settings where thoracic expertise may be limited. Inter-physician agreement also increased from slight to moderate, suggesting the tool promotes more consistent clinical reasoning across experience levels.

"This alignment between machine and clinical logic is essential for building trust in AI-assisted decision-making," said Marina Garassino, thoracic oncologist and professor of medicine at UChicago Medicine, who served as senior author.

Why it matters

Current biomarkers leave physicians unable to reliably predict which lung cancer patients will benefit from immunotherapy at diagnosis. This uncertainty means some patients receive treatments unlikely to help them while missing alternative approaches. AI tools that improve prediction accuracy using routinely available clinical data could enable more personalized treatment decisions without requiring new diagnostic tests or infrastructure investments.

Next phase focuses on prospective validation

The published results represent the retrospective phase of I3LUNG. The project is now prospectively enrolling more than 2,000 additional patients across the same international centers to validate the models in real-world clinical workflows and focus on treatment optimization.

"I3LUNG establishes a new benchmark for AI in thoracic oncology," Garassino said. "For patients, this means fewer missed opportunities for treatment benefit. For community physicians, it means access to expert-level guidance at the point of care."

The European Union's Horizon 2020 research and innovation program funded the study. Additional UChicago Medicine authors included Alexander T. Pearson, Christine Bestvina, Matteo Sacco, Samuel G. Armato III, Alessandra Esposito, Costanza Siniscalchi, and Anna Di Lello.

These findings were first reported by News Medical.