Treating cancer requires a deep understanding of a tumor’s biology and genetic features in order to provide personalized therapies. To achieve this, researchers at Charité—Universitätsmedizin Berlin and Humboldt-Universität zu Berlin sought to explore the potential of generative artificial intelligence (AI) tools in assisting with treatment decisions. In their study, recently published in the journal JAMA Network Open, they investigated the capabilities and limitations of large language models like ChatGPT in analyzing scientific literature to identify personalized treatments.
The researchers conducted an experiment involving ten fictitious patient profiles with molecular tumor profiles. Both a human physician specialist and four large language models were assigned the task of recommending a personalized treatment option. These results were then presented to a multidisciplinary team of experts in the “molecular tumor board” for evaluation.
While the AI models were able to identify some promising treatment options in certain cases, they fell short when compared to the expertise of human experts. Dr. Damian Rieke, a doctor at Charité, concluded that the AI models were “not even close” to the abilities of human experts. Additionally, challenges relating to data protection, privacy, and reproducibility were identified as crucial considerations when using AI in real-world patient care.
Despite the limitations, the researchers remained optimistic about the future potential of AI in medicine. They highlighted the continuous improvement of AI models as a promising factor. As AI models advance, they have the potential to provide greater support in complex diagnostic and treatment processes. However, the study emphasized the importance of human oversight and the final decision-making authority in treatment plans.
The integration of AI in precision oncology holds immense possibilities for the future of cancer treatment. While the study revealed the current limitations, ongoing advancements in AI technology may lead to improved models that can aid in the identification of personalized treatment options, ultimately enhancing patient care.
Can AI revolutionize treatment decisions in precision oncology?
While AI shows promise in assisting with treatment decisions, current AI models still fall short compared to human experts. Ongoing advancements in AI technology may lead to improved models that can provide greater support in the future, but human oversight and the final decision-making authority remain crucial.
What were the limitations of AI in the study?
The AI models were found to perform much worse than human experts in identifying personalized treatment options. Challenges such as data protection, privacy, and reproducibility must also be addressed before applying AI to real-world patient care.
What is the “molecular tumor board”?
The “molecular tumor board” is a multidisciplinary team of experts in pathology, molecular pathology, oncology, human genetics, and bioinformatics. They collaborate to analyze patient data and the latest research to determine the most promising treatment options based on the genetic makeup of the tumor tissue.