Advancing Education, Research, and Quality of Care for the Head and Neck oncology patient.
Introduction: Deciding upon the appropriate discharge location following head and neck cancer surgery requires an in-depth evaluation of patient-specific factors. While some of these factors can be somewhat conspicuous such as the patient’s age, pre-operative functional status, comorbidities, or postoperative complications, other more subtle factors can oftentimes be overlooked with ill-defined predictive factors as to which patients should be discharged directly to home versus which patients necessitate a higher level of post-operative care with discharge to a post-acute care facility. Specifically, the importance of discharge location is multifaceted with important implications in terms of not only clinical outcomes but also healthcare resource utilization. This study aims to describe a novel machine learning model that can be easily integrated into clinical workflows, allowing clinicians to input data and receive predictions on discharge location to help facilitate early discharge planning and patient counseling, ensuring that appropriate resources and support are arranged in advance.
Methods: A retrospective chart review was conducted of patients surgically treated by head and neck otolaryngologists at Shand’s Hospital of the University of Florida between January 1st, 2014, and January 1st, 2023 with only patients who underwent inpatient head and neck surgery being included in this study. Charts were queried for both clinical and social pre-operative factors such as age, comorbidities, nutrition status, smoking status, functional status, social support as determined by who the patient lives with, occupational status, and distance lived from the hospital. Furthermore, surgery specific and post-operative factors were also collected including the surgery severity, post-operative length of stay, discharge location, type of surgery and whether it involved a laryngectomy, free flap, tracheotomy, or PEG/G-Tube.
Results: Of the 1,895 patients included within this study, 1,554 (82.0%) patients were discharged to home while 341 (17.99%) patients were discharged to an acute-care facility. Of the clinical and social factors that were considered, a multivariate analysis by backward selection determined age at encounter, preoperative functional status, social support, surgical severity, surgeries involving PEG/G-tube or tracheotomy, comorbidities, length of hospital stay, distance lived from the hospital, and occupational status to be the statistically significant predisposing factors that prompted a higher level of postoperative care at a facility rather than home. A functional decision tree model was designed with the future plan to integrate these components into a neural network model in which patient preoperative and perioperative factors can be passed through the network from the input to output layers in a nuanced and non-linear manner, allowing machine-learning of complex patterns and ultimately the ability to generate predictions. Neural network models' ability to learn intricate patterns from large datasets has made it the cornerstone of modern artificial intelligence (AI) applications and is specifically applicable in the context of deciding discharge planning to enable personalized care plans based on predicted risks and enhancing efficiency while also optimizing patient outcomes. We hope to train this dataset through multiple iterations to enable continued learning and improvement of the predictive value.