Development and validation of a machine learning-based risk model for prolonged length of stay after laparoscopic gastrointestinal surgery in women: A secondary analysis of the FDP-PONV trail

Background Prolonged length of stay (PLOS) after surgery is associated with several clinical risks and increased medical costs. We aimed to develop a risk model for PLOS based on clinical features throughout pre-, intra-, and post-operative periods in women undergoing laparoscopic gastrointestinal surgery. Methods Women who underwent laparoscopic gastrointestinal surgery in the FDP-PONV randomized controlled trial were eligible for this secondary analysis. PLOS was defined as a duration longer than the median postoperative length of stay. All 96 clinical features prospectively collected in the FDP-PONV trial were used to generate the models. Six machine learning algorithms were employed: logistic regression, K-nearest neighbor, gradient boosting machine, random forest, support vector machine, and extreme gradient boosting. The model performance was assessed using numerous metrics and evaluated using bootstrapping with 1000 replicates. Results In total, 770 and 331 patients were assigned to the training and validation cohorts, respectively. The logistic regression model performed best in the validation cohort [area under the receiver operating characteristic curve (AUC) 0.751, 95% confidence interval (CI), 0.699–0.804] and the bootstrapping (AUC 0.760, 95% CI 0.722–0.796). A nomogram based on ten factors, including education level, preoperative hypoalbuminemia, preoperative insufficient sleep, duration of surgery, blood loss, postoperative hypotension, postoperative albumin infusion, postoperative serum phosphorus level, highest pain score during 73–120 h after surgery, and postoperative infection, was established. Conclusions A risk model was established using machine learning to characterize the key perioperative drivers and identify women at high risk of PLOS after laparoscopic gastrointestinal surgery. Trial registration: The FDP-PONV trial was registered at clinicaltrials.gov (NCT04853147) on 2021-04-27.