Development and external validation of nomogram associated with gastroparesis syndrome after subtotal gastrectomy...

Postsurgical gastroparesis syndrome (PGS) significantly diminishes the quality of life for patients following surgery. With the evolution of robotic surgery, there is a debate on whether it can offer a novel treatment modality for gastric cancer and reduce the incidence of gastric paralysis syndrome. This study utilizes machine learning techniques and traditional logistic regression to construct and validate predictive models, with the aim of providing guidance for clinical practitioners. This study included two cohorts from one medical centers based on the surgical timing for division (Cohort 1: n = 619 for model building and internal validation; Cohort 2: n = 312 for external validation). In Cohort 1, a 3:1 ratio was used for training and validation in random forest and a 7:3 ratio for logistic regression. After analyzing the Receiver Operating Characteristic curves (ROC), we chose classical logistic regression to build the prediction model followed by evaluation with calibration and decision curve analysis (DCA). Finally, we performed external validation on Cohort 2. The model incorporated 7 factors including: Pre-operative TBIL (OR = 2.99), Pre-operative DBIL (OR = 2.35), Pre-operative potassium (OR = 6.8), Surgical type (OR = 3.76), Gastric tube removal time (OR = 3.48), Reconstruction mode (OR = 4.63) and Operative time (OR = 2.21). The model performed well in ROC, with AUC values of 0.892 in the training set, 0.858 in the inner validation set (Cohort 1), and 0.849 in the exterior validation set (Cohort 2). All three datasets' calibration curves revealed a high level of agreement between projected and actual probability. DCA suggested that the model had great clinical usefulness. We have established a more reliable predictive model for PGS which can provide guidance for clinical practitioners. Robotic surgery is also considered to be one of the factors that can reduce the occurrence of PGS.