The multi-omics signature to predict sensitive to anti-epidermal growth factor receptor therapy in RAS wild-type colorectal cancer liver metastases.

Abstract

e15578 Background: There are nearly half patients with Ras wild-type metastatic colorectal cancer (mCRC) do not response to Anti-EGFR therapy. Identification of patients who are sensitive to anti-EGFR therapy may increase the response rate and reduce the adverse effect. In this study, we aimed to develop and validate a multi-omics deep learning model to predict cetuximab efficacy in RAS wild type mCRC patients. Methods: In this study, we retrospectively analyzed 213 Ras wild type mCRC patients. The training and testing cohorts comprised patients in Arm A (FOLFOX + cetuximab) of the CHINESE study (NCT01564810) and patients in the CHINESE follow-up study (PMID: 30305811). The external and negative validation cohorts were derived from an independent cohort and Arm B (FOLFOX) of the CHINESE study, respectively. Based on the deep learning framework Pytorch, we first built the radiomic and genetic signature. Next, we passed the CT images and gene data into the trained ResNet18 and Random Forest and then we sum the output probabilities of two models with a weight of 3:7 to obtain the classification probability of the fusion model. Results: The signature successfully predict sensitivity to anti-EGFR therapy (The area under the curve (AUC) of the radiomic signature: 0.75; the genetic signature: 0.81; the fusion signature: 0.86) but failed with chemotherapy (The fusion signature: 0.54). In cetuximab-containing sets, the fusion signature outperformed existing biomarkers for detection of treatment sensitivity (OR = 17.9, 95% CI 3.22–154.31, P = 0.003) and was strongly associated with progression free survival (mPFS: 12.0 vs. 15.0 months, HR: 0.44, 95% CI 0.20–0.99, P = 0.047). Conclusions: The multi-omics signature can serve as an intermediate surrogate marker of anti-EGFR treatment sensitivity and survival. The signature outperformed known biomarkers in providing an early prediction of treatment sensitivity and could be used to RAS wild type mCRC treatment decisions.