UBC, BC Cancer researchers develop AI model to predict cancer survival rates

The AI model can predict survival outcomes within six months, 36 months, 60 months.

In recent years, the healthcare sector has been widening its adoption of artificial intelligence (AI) applications in cancer treatment and diagnosis. Among others, AI has been used to forecast cancer risk, tumour regrowth, and exploring treatment options for breast cancer.

”Our hope is that a tool like this could be used to personalize and optimize the care a patient receives right away.”
– Dr. John-Jose Nunez, UBC/BC Cancer

In British Columbia (BC), researchers from the University of British Columbia (UBC) and BC Cancer have developed an AI model that makes predictions on cancer patient survival rates using oncologists’ notes.

The project was funded in part by the UBC Institute of Mental Health Marshall Fellowship, as well as by research funding from the BC Cancer Foundation, with some funds originating from the Pfizer Canada Innovation Fund.

Using natural language processing , the researchers conducted a prognostic study of over 47,600 patients with cancer and predicted their survival rates over the course of six, 36 and 60 months.

Lead author Dr. John-Jose Nunez explained that the research defines “survival” in cancer patients based upon whether or not the patient passes away before these time periods, starting from the patient’s initial appointment with the oncologist. Nunez added that the model doesn’t factor in the different definitions of remission, which refers to a decrease or disappearance of cancer symptoms.

Cancer survival rates are typically calculated retrospectively based on information gathered from hundreds or thousands of people with a specific type of cancer, and can be categorized by only a few generic factors such as cancer site and tissue type. This could pose a challenge for oncologists to accurately predict an individual patient’s survival, according to Nunez, as the data is not specific to the patient.

UBC and BC Cancer’s AI model uses a statistical language model referred to as ‘bag-of-words’ to analyze the text in an oncologist’s notes. It counts the word occurrences in a document, determining which of 5,000 different medical or health terms are used within the patient’s initial consultation document to make a prediction. As well, it takes into account how words are presented together, like phrases or sentences.

The AI model also factors in a variety of data, such as the type of cancer and its staging, medical comorbidities, a patient’s current functioning capabilities, their age, and their current symptoms, among others. With a large number of variables considered in its approach, the researchers claim this AI model can produce a more nuanced assessment than traditional methods.

RELATED: Oncology innovators Canexia Health, Imagia merge, raise $20 million

“The AI essentially reads the consultation document similar to how a human would read it,” said Nunez. “These documents have many details like the age of the patient, the type of cancer, underlying health conditions, past substance use, and family histories. The AI brings all of this together to paint a more complete picture of patient outcomes.”

Before this project, Nunez was researching AI use in predicting cancer patients’ need for psychosocial services such as psychiatry and counselling. Seeing the opportunity to also predict survival rates, Nunez worked more closely with BC Cancer and other collaborators to make this initiative possible.

The team working on this AI model include multiple professionals in the field of cancer research such as oncology nurse practitioner Bonnie Leung, as well as Dr. Cheryl Ho, medical oncologist at BC Cancer and clinical associate professor in the UBC division of medical oncology. It also includes Dr. Alan Bates, practice lead for psychiatry and acting head of supportive care at BC Cancer.

The data used to test and train the model comes from BC Cancer sites located across the province. In the future, the researchers said this technology can be applied in cancer clinics across Canada and beyond.

“Our hope is that a tool like this could be used to personalize and optimize the care a patient receives right away, giving them the best outcome possible,” said Nunez.

Featured image courtesy Pexels. Photo by Cottonbro Studio.

Charlize Alcaraz

Charlize Alcaraz

Charlize Alcaraz is a staff writer for BetaKit.

0 replies on “UBC, BC Cancer researchers develop AI model to predict cancer survival rates”