Development and validation of a prediction model for metastasis in colorectal cancer based on LncRNA CRNDE and radiomics

Abstract

Accurate prediction of metastasis is an important determinant for selecting appropriate treatment for advanced colorectal cancer (CRC). In this study, 1250 patients in two hospitals from 2014 to 2019 histologically diagnosed with CRC were enrolled. We performed the transcriptome analysis on 141 CRC patients. RNA-seq analysis revealed that long noncoding RNA (LncRNA) colorectal neoplasia differentially expressed (CRNDE) played an important role in CRC metastasis. The least absolute shrinkage and selection operator regression was used to select features and develop radiomics model. Multivariate logistic regression analysis was used to develop combined model. The radiomics model with 13 filtered radiomics features had good discrimination in predicting expression level of LncRNA CRNDE in training set (receiver operating characteristic [AUC] = 0.809) and testing set (AUC = 0.755). Furthermore, the radiomics model could predict the metastasis of CRC in internal validation set (AUC, 0.665) and in external validation set (AUC = 0.690). The combined model developed with radiomics score and carcinoembryonic antigen had better performance, and the AUC was 0.708, 0.700 in internal validation set and in external validation set, respectively. In conclusion, we proposed a radiomics model and combined model, which could predict the expression level of LncRNA CRNDE and further predict CRC metastasis, thereby helping clinician make treatment decisions.