A Canadian study suggests that it is possible to predict cancer patient survival by reading a document and without training separate models for each cancer type.

Researchers at the University of British Columbia (Canada) have developed an artificial intelligence (AI) technique that predicts the survival of cancer patients simply by reading the oncologist’s notes! The models, applicable to all cancer types, achieved an accuracy of 85% for predicting 6-month survival and 84% for 36-month and 60-month survival, according to study results published in JAMA network open.

The developed model uses natural language processing – a branch of AI that can decipher complex human language – to analyze the patient’s first consultation document given after the diagnosis of cancer. Information from more than 47,000 patients (53.4% ​​women) who received cancer treatment between 2011 and the end of 2016 was used in this retrospective study.

AI, a virtual assistant to support oncologists

Without AI, oncologists rely on cancer survival rates calculated and ranked based on a few factors such as cancer location and tissue type. However, due to the many complex factors that influence it, it can be difficult for these specialists to accurately predict a patient’s individual survival.

” I see AI almost like a virtual assistant for doctors », called dr Nunez, psychiatrist and clinical researcher. ” As medicine becomes more advanced, AI helping to sort and understand all the data will help make decisions for doctors. Ultimately, this will help improve quality of life and patient outcomes. »