Nature Medicine: I3LUNG AI beats standard markers for NSCLC immunotherapy prediction

The paper is from an international consortium including Fondazione IRCCS Istituto Nazionale dei Tumori di Milano and the University of Chicago thoracic oncology program. Journal metadata lists Arsela Prelaj as corresponding author.

Authors state I3LUNG (NCT05537922) is currently the largest international real-world multimodal AI-based study in this setting, enrolling 2,396 patients. File that as the authors’ claim in Nature Medicine — not this desk’s ranking of every prior cohort.

Modalities in the paper: clinical and blood (CB) data, CT, digital pathology (DP), and genomics, fused with machine-learning early fusion (MLEF) and deep-learning intermediate fusion (DLIF).

CB-only models: the authors report AUC up to 0.77 on an independent TEST set. External validation (EXVAL) AUC is 0.55–0.72; they cite population differences for the drop.

On the independent TEST set, the paper says AI models significantly surpassed PD-L1, ECOG performance status, neutrophil-to-lymphocyte ratio (NLR), lactate dehydrogenase (LDH), and the Lung Immune Prognostic Index (LIPI).

Clinical usability: lung expert and nonexpert physicians improved prediction when given the explainable-AI ML CB-only tool, per the paper.

Multimodal MLEF (CB+CT+DP) was associated with higher performance in some analyses. The authors say the incremental benefit remains uncertain and did not translate in TEST and EXVAL.

A prospective validation of the decision-support system is underway in more than 2,000 patients. This is not approved or standard-of-care. This desk does not claim finished validation or FDA clearance.

CONFIRMED here: the 13 Sep Nature Medicine primary. Distinct from the already-filed deepmind-alphagenome-atlas and openai-codex-chatgpt-antimicrobials — different papers, different tasks.

A same-day Nature Medicine primary puts a named multi-center NSCLC immunotherapy AI study on the record — with stated AUCs, biomarker comparisons, and an honest EXVAL drop — so HEALTH readers can track cancer-AI claims against peer-reviewed evidence.