Abstract
OBJECTIVES: The composition of abdominal adipose tissue and muscle mass has been strongly correlated with the prognosis of rectal cancer. This study aimed to develop and validate a machine learning (ML) predictive model for 3-year all-cause mortality after laparoscopic total mesorectal excision (LaTME). METHODS: Patients who underwent LaTME surgery between January 2018 and December 2020 were included and randomly divided into training and validation cohorts. Preoperative computed tomography (CT) image parameters and clinical characteristics were collected to establish seven ML models for predicting 3-year survival post-LaTME. The optimal model was determined based on the area under the receiver operating characteristic curve (AUROC). The SHAPley Additive exPlanations (SHAP) values were utilized to interpret the optimal model. RESULTS: A total of 186 patients were recruited and divided into a training cohort (70%, n = 131) and a validation cohort (30%, n = 55). In the training cohort, the AUROCs of the seven ML models ranged from 0.894 to 0.949. In the validation cohort, the AUROCs ranged from 0.727 to 0.911, with the XGBoost model demonstrating the best predictive performance: AUROC = 0.911. SHAP values revealed that subcutaneous adipose tissue index (SAI), visceral adipose tissue index (VAI), skeletal muscle density (SMD), visceral-to-subcutaneous adipose tissue ratio (VSR), and subcutaneous adipose tissue density (SAD) were the five most important variables influencing all-cause mortality post-LaTME. CONCLUSION: By integrating body composition, multiple ML predictive models were developed and validated for predicting all-cause mortality after rectal cancer surgery, with the XGBoost model exhibiting the best performance.