Machine learning based prediction of Survival and Progression to atezolizumab-bevacizumab in Hepatocellular Carcinoma (HCC) patients

Hepatocellular carcinoma (HCC) is a type of liver cancer and is the third leading cause of cancer deaths around the world. Sadly, many people with HCC are diagnosed too late, when it’s harder to treat and cure. For patients with more advanced stages of HCC, one of the main treatments is a combination of two drugs: atezolizumab and bevacizumab. These medicines work by helping the body’s immune system fight the cancer and by reducing the blood flow to the tumor, which the cancer needs to grow. However, only about 30% of patients respond well to this treatment, and right now, doctors have no easy way to predict who will benefit from it.

The goal of this research is to find a way to predict which patients are more likely to respond to atezolizumab and bevacizumab. To do this, we gathered data from 1,399 patients who were treated with these two drugs at 12 different hospitals and research centers worldwide, including both Western and Eastern countries. We then divided the data into two groups: one to develop our prediction models and the other to test how well the models work.

In the first part of the study, we looked at the patients’ characteristics when they started treatment and grouped them into three categories based on their outcomes. We called this the "A-B-C" classification. Next, we used advanced computer techniques, called machine learning, to develop a scoring system that predicts whether a patient’s cancer will progress quickly (get worse) after their first follow-up scan. Machine learning is a type of technology that allows computers to learn from data and make predictions without being directly programmed to do so.

To make sure our findings are reliable, we are also looking to test our methods on data from a well-known clinical trial called IMbrave150. This will help us confirm how well our models work and compare outcomes based on the treatment patients received. Our ultimate aim is to create tools that physicians can use in everyday practice to choose the best treatments for their patients.