Of note, AI is great but that and radiomics are very hard to standardize so as to be able to apply to diverse populations across the world.

This review highlights the limitations of relying on PD-L1 and TMB alone to predict response to immunotherapy in NSCLC. Combining ctDNA, immune profiling, and other multi-omic data may provide a more complete picture of each patient’s tumor and immune response. AI-based approaches could further support personalized treatment decisions, although more research is needed before these tools become part of routine clinical practice.

I believe we get better outcomes using targeted agents rather than chemotherapy. PD-L1 status plays a vital role in my management.

It would be interesting to see what the multi-omic ICI biomarker data DID predict. The linked article didn't mention the response rate. We do know that PD-L1 has some predictive value. For example PD-L1 over 50% predicts a OS advantage. I think we would need to drill down on the benefit of a new marker over that. Surely, there are better biomarkers, we just need the data

https://www.sciencedirect.com/science/article/abs/pii/S1525730423000578