Establishing Pelvimetry-Based Machine Learning Models to Predict Surgical Difficulty in Laparoscopic Intersphincteric Resection in Patients With Low Rectal Cancer

Abstract

Abstract Aim Intersphincteric resection (ISR) is an anus-preserving procedure for the treatment of low rectal cancer. However, some patients have difficult ISR procedures due to pelvic stenosis. We aim to build a machine learning (ML) model to predict the difficulty of ISR.Methods We retrospectively collected information of 163 patients with low rectal cancer who underwent laparoscopic ISR from January 2017 to August 2022. The prediction models of surgical difficulty were constructed by five MLs. External validation of the European MRI and Rectal Cancer Surgery (EuMaRCS) score was also performed.Results Of 163 patients,36 (22.1%) were assessed as difficult, and 127 (77.9%) were assessed as non-difficult. 9 variables were finally included through lasso regression and binary logistic regression. Two main types of models were constructed, with one retaining all variables, with random forest (RF) performing best (accuracy, 0.878; positive predictive value [PPV], 1; negative predictive value [NPV], 0.867; sensitivity, 0.4; specificity, 1; area under the curve [AUC], 0.877; 95% confidence interval [CI], 0.732–1). The other category retained the 9 variables screened, with support vector machine (SVM) performing best(accuracy, 0.857; PPV, 0.636; NPV, 0.921; sensitivity, 0.7; specificity, 0.897; AUC, 0.854; 95% CI, 0.698–1). The EuMaRCS score did not show a better predictive performance in our study.Conclusions The ML models we developed were found to be more accurate in comparison to the EuMaRCS score. The pelvimetry-based ML model can be used as an effective predictive tool for identifying the difficulty of ISR for low rectal cancer.