A Machine Learning Approach for Prediction of Surgical Outcomes in Elective Surgery

The aim of this research was to design a Machine Learning (ML) approaches to predict surgical outcome associated with perioperative risks factors among patients undergoing elective surgery. The research employed descriptive cross-sectional survey and a sample size of 292 patients. Only adult patients undergoing elective surgery were considered. Machine Learning (ML) Algorithm such as Logistic regression, Support vector machine, k-nearest neighbors and random forest were used to provide insights into how different factors such as patient related perioperative risk, procedure related perioperative risk and health system related perioperative risk influence the likelihood of successful surgical outcome. The study found that Random Forest model achieved the highest cross validation accuracy of 100%, which means it correctly classified all data points in the test set. It implies that the random Forest model was the most suitable for classifying surgical outcome among elective surgery patient at Chuka County Referral Hospital. It had a Kappa of 1 indicating a perfect agreement between its predictions and the ground truth in comparison with other algorithms. In addition, Random Forest model achieves a perfect score (1.0) for sensitivity, precision, F1-Score, and balanced accuracy. This suggests that the model is doing extremely well at correctly classifying both positive and negative cases. Availability of main surgical supplies (health system related perioperative risk factors) had the highest score indicating that it was more important factor for the models predictions than other perioperative risk factors. In this study, the Machine Learning analysis identified unknown parameters associated with successful surgical outcome. An application of Machine Learning algorithms as a decision support tool could enable the medical health practitioners to predict the surgical outcome of patients undergoing elective surgery and consequently optimize and personalize clinical management of patient.