Nature Medicine · 1 September 2026 · From issue 36
Clinical usability of an explainable AI decision support tool and evaluation of multimodal models in NSCLC
An AI tool using blood and health data helped doctors predict immunotherapy response in lung cancer better than standard scores.
Arsela Prelaj, Vanja Miskovic et al.
An AI tool built from blood tests and basic health records helped doctors predict how lung cancer patients respond to immune-based treatment. The tool used data from 2,396 patients and worked better than older methods like standard cancer markers or general health scores. Adding scan images and tissue samples made the tool look better in early tests. But this benefit did not clearly hold up when tested on different patient groups.
The AI model's accuracy score in predicting immunotherapy response, in its own test group of patients.
2,396 patients studied, but accuracy dropped in outside patient groups.
Integrity screen: passed (3 checks) Checked 6 October 2026. Retraction record: none. DOI resolves at doi.org. Metadata record found (Nature Medicine). Read the source
This screen checks process signals. It does not tell you the findings are true.