AI-based “oncologist’s assistant” to help match patients with targeted therapies

Researchers at the Dana-Farber Cancer Institute, led by former Damon Runyon Clinical Investigator Eliezer M. Van Allen, MD, PhD, have built an experimental artificial-intelligence tool designed to help oncologists keep up with the rapidly evolving field of precision cancer therapy. With over 100 precision therapies (and counting) approved for the treatment of cancer, each of which targets a specific cancer-driving mutation, clinicians can find it difficult to stay abreast of new FDA approvals, even when one might provide a life-saving treatment option.

This new tool, dubbed an “oncologist’s assistant,” is designed to help clinicians identify targeted therapies that may benefit a given patient based on their cancer’s genetic profile. To create it, the team trained a large language model (the one behind ChatGPT) on a highly curated database of information about targeted medicines, known as the Molecular Oncology Almanac. In early testing with about 100 realistic scenarios from physicians, it provided accurate responses 93% of the time. Further tuning ensued—including, crucially, the instruction to report when no approved options were found, an answer people-pleasing AI tools are often loath to give.

The goal, Dr. Van Allen clarified, is “to be helpful to an oncologist without taking away the autonomy, decision-making, and relationship between the patient and provider.” The tool is currently intended for research purposes only, as more refinement is needed before any model is used for clinical care.

This research was published in Cancer Cell.