Modeling Tool Guides CRC Treatment Strategies with More Precision

October 27, 2020
Hannah Slater
Hannah Slater

“This algorithm could allow us a better shot at personalized medicine and enhance our ability to tailor the treatments to be as appropriate as possible,” said study author Daniel Chang, MD.

A new modeling tool has high predictive performances at the individual scale for patients with colorectal cancer (CRC) and could be used to discuss treatment strategies, according to a study published in Gut.1

The model harnesses artificial intelligence and evaluates 32 details about an individual patient, such as age, the stage of the cancer, exercise habits, cholesterol levels, and history of chronic disease, and then compares those details to other CRC cases to predict a patient's chance of surviving past 10 years. In addition, the tool also provides context, citing the top reasons behind its calculation.

"Predicting survival of cancer patients as a means to help determine treatments is not new," study leader Jean Emmanuel Bibault, MD, PhD, a radiation oncologist at Stanford Medicine, said in a press release.2 "But current standard techniques are not very accurate, and we're hoping that by using AI we can bring more precise information to doctors as they make crucial decisions about care."

Researchers utilized the prostate, lung, colorectal, and ovarian cancer screening (PLCO) Trial, which randomized 154,900 participants to either screening with flexible sigmoidoscopy, with a repeat screening at 3 or 5 years, or to usual care. Patients who were diagnosed with CRC were then selected during the follow-up to train a gradient-boosted model to predict the risk of death within 10 years after a diagnosis of CRC for any given patient.

Over the course of the follow-up period, a total of 2359 patients were diagnosed with CRC. Median follow-up was 16.8 years (14.4-18.9) for mortality. Moreover, 686 patients (29%) died from CRC during the follow-up.

The dataset was randomly split into a training (n = 1887) and a testing (n = 472) dataset. The area under the receiver operating characteristic was found to be 0.84 (±0.04) and accuracy was 0.83 (±0.04) with a 0.5 classification threshold. Of note, the model is now available online for research use.

"The treatments that we have nowadays are becoming more and more specialized, targeted, in many cases intensified. And the reality is that not everybody is going to benefit from new treatments, therapies or technologies in the same way," study author Daniel Chang, MD, professor of radiation oncology at Stanford Medicine, said in the release. "This algorithm could allow us a better shot at personalized medicine and enhance our ability to tailor the treatments to be as appropriate as possible.”

Though patients could use the tool themselves, Bibault suggested that the ideal application for the model would be for doctors and patients to use the tool together. This way, doctors would be able to contextualize the result and answer any follow-up questions patients may have.

Moving forward, researchers hope to enhance the algorithm's accuracy and find other applications for it.

"We have laid the foundation for this model, and we're hopeful it can apply to other cancer types as well,” said Bibault.

References:

1. Bibault J, Chang DT, Xing L. Development and validation of a model to predict survival in colorectal cancer using a gradient-boosted machine. Gut. doi: 10.1136/gutjnl-2020-321799

2. AI tool created to guide colorectal cancer care with more precision [news release]. Stanford Medicine. Published October 1, 2020. Accessed October 16, 2020. https://scopeblog.stanford.edu/2020/10/01/ai-tool-created-to-guide-colorectal-cancer-care-with-more-precision/