Characterizing immune profiles in hepatocellular carcinoma patients benefiting from pembrolizumab and lenvatinib using machine learning.
Combination immunotherapies, such as pembrolizumab plus lenvatinib (PL), are commonly used in treatment for unresectable hepatocellular carcinoma (uHCC). However, it remains challenging to predict which patients will benefit from this therapy. This study aimed to address this issue by comparing immune cell profiles (ICPs) between uHCC patients with objective response (responders, R) and those with tumor progression (non-responders, NR) following PL therapy, and to identify the key contributors to ICPs.
We prospectively enrolled 51 uHCC patients between July 2019 and July 2023. Peripheral blood samples were collected prior to initiating PL therapy, and ICPs were analyzed according to tumor response according to RECIST 1.1 criteria. A machine learning (ML) model was developed to differentiate R from NR using baseline ICP data.
16 patients achieved objective tumor responses, while 11 experienced disease progression following PL therapy. Responders exhibited higher levels of total T cells, CD8 T cells, and PD-1+ subpopulations of CD4 T cells, CD8 T cells, and NK cells. In contrast, NR had higher proportions of PD-L1+ monocytes. The trained ICP-based ML model accurately discriminated between the two groups, achieving 100% sensitivity and 66.7% specificity, with CD8 T cells, PD-1+ CD8 NK cells, and PD-L1+ monocytes contributing significantly to the classification.
This study recognized distinct ICPs between uHCC patients with and without tumor response to PL therapy and identified key contributing immune subpopulations. These findings provide a foundation for developing predictive tools for clinical outcomes before initiating combination immunotherapy.
We prospectively enrolled 51 uHCC patients between July 2019 and July 2023. Peripheral blood samples were collected prior to initiating PL therapy, and ICPs were analyzed according to tumor response according to RECIST 1.1 criteria. A machine learning (ML) model was developed to differentiate R from NR using baseline ICP data.
16 patients achieved objective tumor responses, while 11 experienced disease progression following PL therapy. Responders exhibited higher levels of total T cells, CD8 T cells, and PD-1+ subpopulations of CD4 T cells, CD8 T cells, and NK cells. In contrast, NR had higher proportions of PD-L1+ monocytes. The trained ICP-based ML model accurately discriminated between the two groups, achieving 100% sensitivity and 66.7% specificity, with CD8 T cells, PD-1+ CD8 NK cells, and PD-L1+ monocytes contributing significantly to the classification.
This study recognized distinct ICPs between uHCC patients with and without tumor response to PL therapy and identified key contributing immune subpopulations. These findings provide a foundation for developing predictive tools for clinical outcomes before initiating combination immunotherapy.
Authors
Lee Lee, Li Li, Lee Lee, Lin Lin, Wu Wu, Hung Hung, Chen Chen, Chan Chan, Mon Mon, Lee Lee, Chi Chi, Lee Lee, Hou Hou, Chao Chao, Huang Huang, Lee Lee
View on Pubmed