Researchers in Canada say they've developed an AI model that can predict the survival chance of cancer patients more accurately than previous methods.
The model relies on natural language processing to read a cancer
doctor's notes on a patient that contain "details relevant to survival."
- The model was trained on data from 47,625 patients from sites run by BC Cancer, a government agency that's part of the British Columbia Provincial Health Services Authority.
- The
model "essentially reads the consultation document similar to how a
human would read it," according to Dr. John-Jose Nunez, the lead author
of the study.
- The documents contain data such as the patient's age, underlying health issues, family history, and substance abuse.
- In
follow-up research, the model was able to predict patients' six-month,
36-month, and 60-month survival changes at over 80% accuracy.
- According
to Nunez, using such a tool could help health providers refer patients
to support services earlier or pursue a "more aggressive treatment
option up front."
