Developing AI device to predict cancer patient survival

Dr. Lauren Janczewski, a surgeon at Northwestern University McGaw Medical Center in Chicago (USA), and colleagues have developed an AI-based device to estimate the chances of survival of new patients after being diagnosed with cancer.

The Cancer Survival Calculator uses machine learning to provide highly accurate, patient-specific prognoses, refining assessments that are valid within 9-10 months of detection for a variety of cancers, a key factor in improving patient survival.

Unlike current conventional assessment methods that rely primarily on cancer stage, this tool incorporates a multitude of influencing factors, from patient age and tumor size, to specific treatment variables, resulting in more comprehensive and personalized prognoses.

The study focused primarily on patients diagnosed with breast, thyroid, and pancreatic cancers, because these cancers have large patient populations and rich clinical profiles.

Using an extensive dataset from the US National Cancer Database, the computer was optimized using patient records diagnosed in 2015 and 2017. The dataset included 259,485 breast cancer patients, 76,624 thyroid cancer patients, and 84,514 pancreatic cancer patients, allowing machine learning algorithms to identify and rank different factors that influence patient survival.

“There are countless objective factors that can influence a patient’s survival beyond their own condition,” said Dr. Lauren Janczewski. The device comprehensively assesses tumor-specific biomarkers and treatment variables, improving its accuracy and predictive power compared to traditional prognostic tools.

The current development process involves using three-quarters of the collected data to train machine learning algorithms. The remaining data is used for validation, ensuring the accuracy and reliability of the estimated patient survival.

Next, the team plans to improve the computer's user interface, facilitate implementation in clinical practice, and conduct pilot trials at selected cancer centers.

The ultimate goal of the research is to expand diagnostic and prognostic capabilities to other types of cancer, helping doctors improve cancer prognosis and patient care.