Artificial Intelligence in Cancer Care: Opportunities, Challenges, and Governance

TL;DR

A panel of experts at the NCCN 2025 Annual Conference surveyed the current and projected use of AI in cancer care, emphasizing that documentation and administrative tools are already delivering value while governance, evidence generation, and trustworthiness remain unresolved. The panelists argue that robust multidisciplinary governance, transparency, and demonstrated value are prerequisites for responsible adoption.

Key findings

• Current AI applications already in use include LLM-based record summarization, described by Dr. Haddad as "life-changing" and acting like a "digital fellow," and ambient listening "AI scribes," which Mr. Chornenky said save physicians time — "even a few minutes per encounter adds up significantly" — and improve their experience.
• Oncology's data intensity, multimodal data (genomics, phenomics, imaging, pathology, clinical notes), rapid drug discovery, and high patient engagement distinguish AI use in cancer care; Dr. Haddad cited nearly half of new FDA drug approvals and expanded indications occurring annually in oncology and benign hematology, with the rate accelerating from "<1 per month a decade ago to approximately 1 per week now."
• Nearer-term (2–5 year) predictions include precise decision support for targeted treatments, treatment de-escalation, remote patient monitoring, and cancer care navigation; Mr. Chornenky anticipates models becoming "orders of magnitude more capable in reasoning and coordination" within 5 years, and predicts AI will drive a significant portion of major research discoveries within 10 years.
• Risks identified include hallucinations (which Dr. Stetson attributed to models extrapolating beyond training data, with retrieval-augmented generation as a possible but not foolproof mitigation), model drift, regulatory lag, patient privacy concerns, and unresolved liability questions.

Methods

• Narrative summary of an expert panel convened at the NCCN 2025 Annual Conference, moderated by Clifford S. Goodman, PhD, with panelists Dennis Chornenky, Tufia C. Haddad, and Peter D. Stetson.
• No study design, dataset, or formal analysis is reported; the content is the panelists' expert opinion and predictions, organized into sections on what makes AI in cancer care unique, the current state, future directions, and governance.

Limitations

• Not stated in the paper: the text does not include a limitations section.
• Evident from the text: the content is expert panel commentary rather than empirical research, so claims about current impact (eg, scribe time savings) are attributed to panelists and cited only as "data from health systems" without study details; forward-looking statements are predictions, and Mr. Chornenky himself cautioned that "human minds struggle to predict exponential change."
• Evident from the text: disclosed conflicts include Dr. Chornenky being employed by Domelabs AI and Dr. Goodman receiving consulting fees from Bayer HealthCare.