Abstract
BACKGROUND: This study sought to identify critical body composition characteristics associated with surgical difficulty in Laparoscopic Total Mesorectal Excision (LaTME) and to develop and validate an interpretable machine learning model using body composition data. METHODS: Patients with pathologically confirmed mid to low rectal cancer treated between January 2017 and December 2022 were enrolled. LASSO regression identified clinical features most predictive of surgical difficulty. Seven machine learning algorithms were developed and validated. Model performance was comprehensively assessed using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, Brier score, calibration curves, and decision curve analysis. SHapley Additive exPlanations (SHAP) values elucidated key feature contributions, and the optimal model was implemented as a dynamic nomogram. RESULTS: A cohort of 387 rectal cancer patients undergoing LaTME was enrolled. LASSO regression identified the following predictors for model development: visceral fat area (VFA), visceral fat ratio (VFR), tumor anal verge distance, subcutaneous fat area (SFA), receipt of neoadjuvant therapy, neutrophil-to-lymphocyte ratio (NLR), and history of abdominal surgery. In the validation cohort, the logistic regression (LR) model demonstrated optimal performance . SHAP analysis revealed that increased VFA, elevated VFR, greater SFA, shorter tumor anal verge distance, administration of neoadjuvant therapy, higher NLR, and prior abdominal surgery were associated with increased surgical difficulty during LaTME. Kaplan-Meier analysis demonstrated significantly reduced 1-year, 3-year, and 5-year overall survival (OS) rates in the difficult surgery cohort compared to the non-difficult cohort (p < 0.05). CONCLUSION: Seven predictive models for LaTME surgical difficulty were constructed and validated. The LR model exhibited the best predictive performance. Survival analysis indicated poorer prognosis in patients experiencing difficult surgery.