A validation study of a machine learning model for predicting clinical benefit in patients with hepatocellular carcinoma treated with atezolizumab and bevacizumab.

Hepatocellular carcinoma (HCC) is a type of cancer that originates in the main liver cells, called hepatocytes. It is the third leading cause of cancer-related death globally, and is the most common primary liver cancer, constituting approximately 80% of cases. Many patients are diagnosed at advanced stages, and systemic therapy (a treatment that uses drugs to target the entire body, reaching cells through the bloodstream) is the mainstay of advanced HCC treatment.

The IMBrave150 clinical trial used a combination of systemic drugs – atezolizumab and bevacizumab. Atezolizumab works by binding to a protein called programmed death-ligand 1 (PD-L1) on the surface of cancer cells. This allows the immune system to recognize and attack cancer cells, which can help slow tumor growth. Bevacizumab works by blocking a protein called vascular endothelial growth factor (VEGF), which prevents the formation of new blood vessels, which reduces the blood supply to tumors and limits their growth.

The IMBrave150 trial demonstrated significantly improved overall survival (OS) (the length of time from either the date of diagnosis or the start of treatment that patients are still alive) and progression-free survival (PFS) (the length of time during and after treatment that a patient lives with the disease but it does not get worse) compared with another drug called sorafenib which works by blocking cancer cell growth. Accordingly, the combination therapy of atezolizumab and bevacizumab (ate/bev) has been approved as the new standard of care (recommended treatment) for patients with unresectable (cannot be removed through surgery) HCC who have not received previous systemic therapy.

However, identification of patients who could clinically benefit from ate/bev therapy is less clear. To predict clinical response of this treatment, real-world studies have used various scoring systems and blood tests. In this study, to predict clinical response of ate/bev therapy better than previous methods, we have developed a model based on 313 patients with HCC at CHA Bundang Medical Center and this model outperformed previous clinical factors or methods. The model developed and proposed for validation was based on machine-learning techniques. Machine learning is the process of using computers to detect patterns in large datasets and then make predictions based on what the computer learns from those patterns.

We plan to use the patient cohort from the IMbrave150 study as an external validation set for the model developed in our study.