Abstract: Objective This study aimed to explore the predictive efficacy of body adipose composition analysis using CT imaging for adverse outcomes in patients with cirrhosis and to enhance predictive performance by integrating a radiomics model based on liver and spleen regions. Method 243 patients with liver cirrhosis were retrospectively enrolled from two centers between September 2016 and December 2020, including 153 cases of hepatitis B or C virus-related cirrhosis. Patients were classified into non-adverse outcome group (n=119) and adverse outcome group (n=124) based on the occurrence of severe complications such as spontaneous bacterial peritonitis during the follow-up period (over 3 months). Quantitative measurements of body adipose composition were performed at the third lumbar vertebra level, and radiomic features of the liver and spleen were extracted. Multiple machine learning models were developed to evaluate predictive performance. Result The area under the receiver operating characteristic curve (AUC) for the adipose tissue model at the two centers was 0.804 and 0.749, respectively, outperforming the radiomics model (AUC=0.767 and 0.712). Shapley additive explanations (SHAP) analysis identified intermuscular fat ratio and subcutaneous fat volume as key predictors of adverse outcomes. The predictive efficacy of the combined model was significantly improved (AUC=0.856 and 0.780; integrated discrimination improvement index=0.125, 0.163, P<0.05). Conclusion body adipose composition can effectively predict the risk of adverse outcomes in patients with viral hepatitis-related cirrhosis. The integration of liver and spleen radiomics features further enhances predictive accuracy, offering a promising approach for individualized risk management and early intervention in viral liver diseases.

Key words: Liver cirrhosis, Prognostic prediction, Computed tomography, Adipose Tissue, Machine learning