Biomarker selection of liver metastatic colorectal patients for anti-EGFR monoclonal antibodies: A machine learning analysis

Abstract

Background: New predictive biomarkers for cetuximab-resistance for patients with RAS wild-type colorectal liver metastases (CRLM). Methods: 216 patients with initially unresectable liver-limited RAS wild-type CRLM were identified from previous clinical studies. Among these patients, 103 patients received chemotherapy (mFOLFOX6 or FOLFIRI)plus cetuximab, and 113 received chemotherapy alone as first-line treatment. Next-generation sequencing of primary tumors was done for single nucleotide polymorphism according to custom panel. The patients receiving cetuximab-based chemotherapy were divided into two groups: one group was cetuximab-resistant group and the other group was cetuximab-sensitive group. Potential predictive biomarkers were determined by "random forest" machine learning. Results: Ten potential predictive genes, namely RET, PTPN11, FLT3, AKT1, ACTN4, ERBB4, FGFR3, MDC1, CUL9, and ZNF462, were identified.Inthe cohort ofcetuxi-mab-based chemotherapy, patients with all-wild-type genes had markedly improved median progression-free survival (12.0 vs. 4.0 months, P < 0.0001) and overall survival (37.0 VS. 24.0 months, P< 0.0001) compared with those with gene mutation; In the cohort of at-least-one gene mutation, patients receiving chemotherapy alone had comparable median PFS (4.0 VS. 4.0 months, P = 0.9498). Moreover, mutated are more common in patients with right-sided colon cancer. Conclusions: Ten new predictive mutations help to refine the selection of RAS and BRAF wild-type metastatic colorectal cancer patients candidates for anti-EGFRs.