Bundang CHA Hospital Develops AI to Predict Effectiveness of Liver Cancer Immunotherapy

Predicts response to atezolizumab plus bevacizumab combination therapy using six blood test indicators
Validated with data from a global Phase 3 trial and real-world clinical practice

A research team at Bundang CHA Hospital has developed an artificial intelligence (AI) model that predicts the effectiveness of immunotherapy for liver cancer based on blood test results.

Bundang CHA Hospital announced on October 19 that a research team comprising Professor Jeon Hongjae of Medical Oncology and Professor Hwang Sohyun of Pathology at the hospital, and doctoral student Jo Gyehun of CHA University, had developed an AI model called CAPTAIN to predict the effectiveness of atezolizumab plus bevacizumab combination therapy. The findings were published in the international journal Cancer Communications.

Jeon Hongjae, Professor of Medical Oncology, and Hwang Sohyun, Professor of Pathology, at Bundang CHA Hospital.

CAPTAIN uses machine learning to analyze six blood test indicators—C-reactive protein (CRP), alpha-fetoprotein (AFP), platelet count, total bilirubin, lymphocyte count, and neutrophil count—to predict treatment response. All six indicators are routinely measured in clinical practice, so the model can help predict treatment effectiveness for individual patients while reducing the burden of additional testing.

The research team developed the model using data from 354 patients at Bundang CHA Hospital and the Medical University of Vienna in Austria. They then conducted external validation using data from 99 patients who participated in a global Phase 3 clinical trial and 184 patients treated in routine clinical practice at Severance Hospital. Data from 100 patients treated with sorafenib or lenvatinib were also used as a comparison group to evaluate the model's predictive performance for a specific combination therapy. After reviewing data from 949 patients in total, the team analyzed 737 patients.

In external validation, CAPTAIN achieved an AUC of 0.70 based on the IMbrave150 data. AUC is a measure of how well a predictive model distinguishes patients who experience a particular outcome from those who do not. The research team said CAPTAIN outperformed CRAFITY and ALBI, established prognostic assessment models. It also performed better than the existing models on real-world clinical data from Severance Hospital. Significant differences were also found in overall survival (OS) and progression-free survival (PFS) between patient groups classified according to CAPTAIN's predictions.

For advanced hepatocellular carcinoma, atezolizumab plus bevacizumab combination therapy and other treatments are used as first-line systemic therapy. However, patients can have different responses and prognoses even when receiving the same treatment, creating a need for tools that can predict treatment effectiveness before therapy begins. Existing models such as CRAFITY and ALBI are primarily used to assess prognosis and have limitations in predicting the effectiveness of specific treatments.

Professor Jeon Hongjae said, "This study is significant because it presents a model that can predict the effectiveness of atezolizumab plus bevacizumab combination therapy using only blood tests that are readily available in routine clinical practice. We will continue to develop it into a predictive tool that can help guide treatment selection for individual patients and establish personalized treatment strategies through further clinical validation."

This content was produced with the assistance of AI translation services.