Comprehensive Summary
The study looks at how ChatGPT 4.0 generates molecular tumor board (MTB) recommendations using 20 oncology cases from the past. For each case, ChatGPT was prompted three times to ensure reproducibility, and the data was compared with human MTB based on the factors of content, reproducibility, evidence, and efficiency. The model was found to produce more therapeutic recommendations compared to human MTBs even though the information content was similar between both ChatGPT's MTB and human MTB. One limitation is that ChatGPT used evidence reflecting preclinical evidence, which often fails to accurately portray the correct outcome. However, the AI model was faster at generating the information, doing it in half the time required by the human teams.
Outcomes and Implications
The research findings suggest that ChatGPT 4.0 has the ability to support precision oncology through providing diverse therapeutic options, in a manner that is both timely and information-dense. However, due to its reliance on preclinical data, there is a need for human oversight before clinical application. Through this, ChatGPT may act as a tool in the future for human teams to find therapies quickly. In regards to future work in the field, it should look at the integration of AI into MTB decision-making while addressing the lack of evidence.