Following thorough screening based on inclusion and exclusion criteria, patients from the two sizable medical centers were split up into two cohorts for this study. Cohort 1 served primarily as the training and internal validation set, while Cohort 2 was used for external validation of the predictive model constructed from Cohort 1. We used six distinct machine learning methodss, including DT, RF, XGBOOST, SVM, lightGBM, and SHLNN, in addition to conventional logistic regression to create the predictive model. We chose the approach with the best sensitivity and specificity by comparing the concordance index(C-index) akin to the area under the ROC curve (AUC) of these seven distinct model-building methods. The predictive model for Cohort 1 was then built using this method, and internal validation was finished. Lastly, Cohort 2 underwent external validation of the predictive model

Inclusion Criteria:(1) rectal adenocarcinoma (2) minimally invasive sphincter-preserving surgery (taTME/ISR/LAR) (3) intact baseline anal function (4) no emergent presentations or metastases.

-

Exclusion Criteria:emergent presentations or metastases

-

This retrospective analysis included 3,937 radical rectal cancer cases from two Chinese university hospitals (Northern Jiangsu People's Hospital 2015-2023, n=2612; Jilin University's China-Japan Union Hospital 2021-2023, n=1325), with rigorous selection criteria ensuring cohort homogeneity