Researchers have developed a novel artificial intelligence (AI)-based Cancer Survival Calculator for estimating long-term survival in patients with newly diagnosed cancer, according to new findings presented by Janczewski et al at the American College of Surgeons (ACS) Clinical Congress 2023.
Background
Currently, estimating survival rates for patients with cancer may primarily depend on their cancer stage.
“There is a multitude of other factors that may influence a patient’s survival beyond just their staging criteria,” explained lead study author Lauren Janczewski, MD, a clinical scholar at the ACS Cancer Programs and a general surgical resident at the Northwestern University McGaw Medical Center. “We sought to develop this Cancer Survival Calculator to provide a more personalized estimate of what patients can expect regarding their cancer prognosis,” she added.
Study Methods and Results
In the new study, the researchers used machine learning technology to create the Cancer Survival Calculator prototype and tested the tool on 2015 and 2017 data from the National Cancer Database. The initial tests aimed to identify which patient, tumor, and treatment characteristics most influenced 5-year survival in patients with breast cancer (n = 259,485), thyroid cancer (n = 76,624), and pancreatic cancer (n = 84,514).
The researchers noted that about 75% of the data collected in the study were used to train the machine learning algorithms to recognize patterns between characteristics at diagnosis and patients’ survival at 5 years, and then to rank the factors with the greatest impact on survival. With the remaining data, the researchers used statistical methods to test the Cancer Survival Calculator prototype’s accuracy in estimating survival.
The researchers identified multiple characteristics that significantly affected 5-year survival rates. For breast cancer, these included whether the patient had undergone cancer surgery, the patient’s age at diagnosis, their tumor size, and the time from diagnosis to treatment. For thyroid cancer, the factors were the patient’s age at diagnosis, their tumor size, their time to treatment, and lymph node involvement. For pancreatic cancer, whether the patient had undergone cancer surgery, their histology, their tumor size, and their age at diagnosis were found to be the most influential in survival.
Additionally, the researchers found that hormone receptor status and the presence of the Ki-67 biomarker were also significant for survival in patients with breast cancer.
The researchers revealed that the Cancer Survival Calculator differed in several ways from cancer survival estimators already in use. For instance, the novel tool included specific tumor biomarkers and treatment variables that were known to affect patients’ estimated prognoses, which many prior survival calculators often lack. The dataset used to develop the new calculator was also more comprehensive than the datasets used by other calculators. The tool used new data modeling, such as machine learning, to speed up processing. Furthermore, the model’s risk predictions demonstrated improved accuracy compared with predictions generated by older calculators.
Conclusions
Although some of the predictive factors such as tumor size were part of cancer staging, the researchers stressed that their results suggested that many more factors may influence survival beyond the patients’ disease stage.
Further, their validation testing demonstrated that the Cancer Survival Calculator was highly accurate for all three cancer types at estimating cancer survival rates within 9 to 10 months of actual survival.
In the next steps of their development of the novel tool, the researchers plan to finalize a user interface that will allow the use of the Cancer Survival Calculator in clinical practice, followed by pilot testing at selected cancer centers.
The researchers hope to eventually broaden the calculator by incorporating all other cancer types included in the National Cancer Database.
