Purpose: Human epidermal growth factor receptor 2 (HER2)-targeted therapies have shown promise in treating HER2-amplified metastatic colorectal cancer (mCRC). Identifying optimal biomarkers for treatment decisions remains challenging. This study explores the potential of artificial intelligence (AI) in predicting treatment responses to trastuzumab plus pertuzumab (TP) in patients with HER2-amplified mCRC from the phase II TRIUMPH trial.
Materials and methods: AI-powered HER2 quantification continuous score (QCS) and tumor microenvironment (TME) analysis were applied to the prescreening cohort (n = 143) and the TRIUMPH cohort (n = 30). AI analyzers determined the proportions of tumor cells (TCs) with HER2 staining intensity and the densities of various cells in TME, examining their associations with clinical outcomes of TP.
Results: The AI-powered HER2 QCS for HER2 immunohistochemistry (IHC) achieved an accuracy of 86.7% against pathologist evaluations, with a 100% accuracy for HER2 IHC 3+ patients. Patients with ≥50% of TCs showing HER2 3+ staining intensity (AI-H3-high) exhibited significantly prolonged progression-free survival (PFS; median PFS, 4.4 v 1.4 months; hazard ratio [HR], 0.12 [95% CI, 0.04 to 0.38]) and overall survival (OS; median OS, 16.5 v 4.1 months; HR, 0.13 [95% CI, 0.05 to 0.38]) compared with the AI-H3-low (<50% group). Stratification among patients with AI-H3-high included TME-high (all lymphocyte, fibroblast, and macrophage densities in the cancer stroma above the median) and TME-low (anything below the median), showing a median PFS of 1.3 and 5.6 months for TME-high and TME-low respectively, with an HR of 0.04 (95% CI, 0.01 to 0.19) for AI-H3-high with TME-low compared with AI-H3-low.
Conclusion: AI-powered HER2 QCS and TME analysis demonstrated potential in enhancing treatment response predictions in patients with HER2-amplified mCRC undergoing TP therapy.