Multimodal AI Improves Treatment Response Prediction in Lung Cancer

Clinicians still struggle to predict who will respond, and treatment decisions often rely on PD-L1 despite well-recognized limitations. More precise, data-driven tools could help tailor therapy and reduce unnecessary toxicity and cost. New findings demonstrate that multimodal artificial intelligence (AI) can improve prediction of treatment response and survival in this setting.

The I3LUNG multimodal AI tool is an international effort to support decision-making for immunotherapy in metastatic NSCLC. Built to outperform standard clinical biomarkers, the technology uses routinely collected patient information to classify likely responders and estimate survival outcomes. The study describing this approach was published in Nature Medicine on September 13, 2026.

At the University of Chicago Medicine, investigators and international partners integrated clinical, imaging, pathology, and genomic data from patients with advanced NSCLC into a shared database and trained two families of predictive models. Area under the curve (AUC), a standard machine-learning metric of discriminative performance, was used to evaluate accuracy. One model using clinical and blood data reached an AUC of 0.77, while a model that added imaging and digital pathology achieved an AUC of 0.88.

The retrospective analysis encompassed 2,396 patients treated with immunotherapy across six centers in Italy, Germany, Greece, Israel, Spain, and the United States. The AI models consistently outperformed standard biomarkers used in current practice. In a reader study, 20 physicians (10 lung cancer experts and 10 non-experts) reviewed 100 real cases first without and then with AI support; performance improved from an AUC of 0.72 to 0.87, and inter-physician agreement rose from slight to moderate.

The project has now moved into a prospective phase, enrolling more than 2,000 patients across the same international centers. This next step focuses on treatment optimization and on evaluating not only model performance but also clinical usability, aligning the tool with real-world decision-making needs.

“I3LUNG establishes a new benchmark for AI in thoracic oncology. Decision support tools built even from routinely available clinical data can outperform the biomarkers we rely on today. 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. For the field, it provides a rigorous, fair, and explainable framework — validated across diverse healthcare systems and populations — that can serve as a global platform for the next generation of precision immunotherapy,” said Marina Garassino, MD, Professor of Medicine at UChicago Medicine and senior author of the study.

Related Links
University of Chicago Medicine