Decision support system to improve postoperative discharge: A novel multi-class classification approach

Postoperative discharge decision-making is a critical process that determines not only postoperative patient outcomes and, in some cases, their survival, but also the management of the hospital resources, both financial and human ones. Existing decision-making support systems for aiding postoperative discharge heavily rely on statistical-based methods that lack objectivity in predicting optimal recovery area on a subject-specific basis. Machine Learning (ML)-based methods can enable these predictions, but current modelling implementations are inaccurate to be applied clinically or too sophisticated for the relatively low gain in classification performance. As an accurate and reliable method to predict where patients in a postoperative recovery area should be sent to next, the clinical potential of a novel hybrid multi-class classification algorithm was assessed. Data on 90 patients regarding their body temperature, oxygen saturation, blood pressure and perceived comfort upon discharge were obtained from the University of California-Irvine (UCI) Machine Learning repository. A multi-class classification was performed on such data using a 'controlled' All-vs-All approach by optimising kernel and hyperparameters via Genetic Algorithms. The novel hybrid algorithm was found to yield the highest classification accuracy, improving the highest accuracy from the literature by almost 12%. Achieving maximum accuracy and reliability, whilst retaining the lowest computational cost amongst the classifiers tested, the hybrid model is deemed an accurate, reliable and clinically viable solution to assist clinicians and nurses in improving postoperative discharge decision making.