Application of Artificial Intelligence in Surgical Decision-Making: Developing a Machine Learning Model for Predicting Anastomotic Leak in Nigerian Patients

Anastomotic leak following colorectal surgery is a life-threatening complication for which clinicians currently lack validated risk prediction tools tailored to Nigerian patient populations. Artificial intelligence and machine learning offer promising approaches to developing context-specific predictive models. This dissertation develops and validates a machine learning model for predicting anastomotic leak in Nigerian colorectal surgery patients. A training dataset of 900 colorectal surgery cases from six teaching hospitals across three geopolitical zones was assembled (2016 to 2022). Twenty-four clinical, laboratory, operative, and imaging variables were examined as candidate predictors. Four machine learning algorithms (random forest, gradient boosting, support vector machine, and logistic regression) were compared using 10-fold cross-validation. External validation was performed on 200 prospective cases from three independent hospitals in 2022 to 2023. The gradient boosting model demonstrated the best predictive performance (AUC 0.84, sensitivity 79%, specificity 83%). The five most informative features were distal anastomosis location, preoperative haemoglobin, intraoperative blood loss, absence of defunctioning stoma, and serum albumin. The model was packaged as a free mobile application for clinical use and validated on smartphone platforms. The dissertation represents the first AI surgical prediction tool developed for and validated in a Nigerian patient population, and recommends integration into perioperative surgical decision support systems. Keywords: artificial intelligence, machine learning, anastomotic leak, colorectal surgery, prediction model