A Machine Learning Model Based on Counterfactual Theory for Treatment Decision of Hepatocellular Carcinoma Patients

Abstract

Xiaoqin Wei,1,* Fang Wang,2,* Ying Liu,3 Zeyong Li,4 Zhong Xue,2 Mingyue Tang,5 Xiaowen Chen1 1School of Medical Imaging, North Sichuan Medical College, Nanchong City, Sichuan Province, People’s Republic of China; 2Department of Research and Development, Shanghai United Imaging Intelligence Co., Ltd, Shanghai, People’s Republic of China; 3Department of Radiology, The First Affiliated Hospital of Chengdu Medical College, Chengdu City, Sichuan Province, People’s Republic of China; 4Department of Radiology, Bishan Hospital of Chongqing Medical University, ChongQing, People’s Republic of China; 5Department of Physics, School of Basic Medicine, North Sichuan Medical College, Nanchong, People’s Republic of China*These authors contributed equally to this workCorrespondence: Xiaowen Chen, School of Medical Imaging, North Sichuan Medical College, 234 Fujiang Road, Nanchong City, Sichuan Province, People’s Republic of China, 637001, Email [email protected]: To predict the efficacy of patients treated with hepatectomy and transarterial chemoembolization (TACE) based on machine learning models using clinical and radiomics features.Patients and Methods: Patients with HCC whose first treatment was hepatectomy or TACE from June 2016 to July 2021 were collected in the retrospective cohort study. To ensure a causal effect of treatment effect and treatment modality, perfectly matched patients were obtained according to the principle of propensity score matching and used as an independent test cohort. Inverse probability of treatment weighting was used to control bias for unmatched patients, and the weighted results were used as the training cohort. Clinical characteristics were selected by univariate and multivariate analysis of cox proportional hazards regression, and radiomics features were selected using correlation analysis and random survival forest. The machine learning models (Deathhepatectomy and DeathTACE) were constructed to predict the probability of patient death after treatment (hepatectomy and TACE) by combining clinical and radiomics features, and an optimal treatment regimen was recommended. In addition, a prognostic model was constructed to predict the survival time of all patients.Results: A total of 418 patients with HCC who received either hepatectomy (n=267, mean age, 58 years ± 11 [standard deviation]; 228 men) or TACE (n=151, mean age, 59 years ± 13 [standard deviation]; 127 men) were recruited. After constructing the machine learning models Deathhepatectomy and DeathTACE, patients were divided into the hepatectomy-preferred and TACE-preferred groups. In the hepatectomy-preferred group, hepatectomy had a significantly prolonged survival time than TACE (training cohort: P < 0.001; testing cohort: P < 0.001), and vise versa for the TACE-preferred group. In addition, the prognostic model yielded high predictive capability for overall survival.Conclusion: The machine learning models could predict the outcomes difference between hepatectomy and TACE, and prognostic models could predict the overall survival for HCC patients.Keywords: radiomics, hepatocellular carcinoma, prognosis, hepatectomy

Keywords

• radiomics
• hepatocellular carcinoma
• prognosis
• hepatectomy