There’s now an AI cancer survivor calculator

The multicenter research team developed a prototype tool dubbed the Cancer Survival Calculator using AI and tested it on a national cancer dataset. Initial research predicted a five-year survival rate for pancreatic, thyroid, and breast cancer patients. 

Patients with these three cancers who were diagnosed between 2015 and 2017 provided the researchers with the pertinent data to build their new tool. The National Cancer Database (NCDB), which has data on 72 percent of newly diagnosed cancer cases in the US, provided the patient records.

In order to select the elements that have the biggest effects on survival, three-fourths of the obtained data was used to train the machine learning algorithms to recognize patterns relevant to diagnosis and patient survival at five years. Data from 84,514 pancreatic cancer patients, 76,624 thyroid cancer patients, and 259,485 breast cancer patients was included.

The top four variables in cancer survival

The researchers then tested the prototype's survival estimation accuracy using the remaining data. The following were the top four variables that determined whether patients survived five years after their diagnosis:

• Breast cancer: (1) whether the patient had cancer surgery, (2) the patient’s age at diagnosis, (3) tumor size, and (4) time from diagnosis to treatment. 

• Thyroid cancer: (1) age at diagnosis, (2) tumor size, (3) time to treatment, and (4) lymph node involvement. 

• Pancreatic cancer: (1) cancer surgery; (2) histology, or microscopic analysis of the cancer, (3) tumor size, and (4) age at diagnosis. 

Subsequent validation testing revealed that the calculator's predictions of cancer survival rates were "highly accurate" for all three cancers, falling between nine and ten months of actual survival. 

The new and unique calculator also contains particular tumor biomarkers and therapy factors that are known to influence a patient's anticipated prognosis. Prior tools lacked these details.

The next steps for the device are to complete a user interface that will permit the tool to be used in clinical practice and to conduct a pilot test of the calculator at a few cancer centres. In addition, with the addition of all other cancer sites found in the NCDB, the researchers plan to broaden the calculator's use and reach.