A four‐gene‐based model predicts TACE efficacy for hepatocellular carcinoma patients

Abstract Background Primary liver cancer, of which hepatocellular carcinoma (HCC) accounts for over 90%, is the sixth most common cancer and the fifth leading cause of cancer mortality worldwide. Transcatheter arterial chemoembolization (TACE) is widely performed globally as an effective treatment for HCC patients at intermediate and advanced stages, but the efficiency varies amongst patients. Objective This study aimed to develop models that predict the efficacy of TACE for HCC patients. Methods GSE104580 from Gene Expression Omnibus was used as training and validation sets. Tumor samples were randomly allocated to training and validation sets in a ratio of 7:3. Differentially expressed genes were screened in the training sets, and efficacy-related genes were identified. Their common genes were used in LASSO L1 regression and support vector machine (SVM)-based recursive feature elimination learner to identify the best predictive four genes. After gene screening, we established the predictive model for two sets using SVMs, random forests models, and logistic regression analysis. Result IFIT1, LIN28B, S100A9, and SPARCL1 were significantly related to the efficacy of TACE for HCC patients. The accuracy of the training sets was 90.3%, 100%, and 91.3% using SVMs, random forests models, and logistic regression analysis, respectively. The four-gene predictive signature showed larger area under the curve values (> 80%) after receiver operating characteristic curve analysis, indicating a high predictive capacity. Conclusion The gene-based model constructed in this study acts as a reliable efficacy assessment tool for clinicians and will aid treatment decision-making for HCC patients.