Comprehensive Summary
In this study, a machine-learning model was developed to analyze preoperative factors to create a predictive model for resource-intensive revision joint surgery. The model made predictions on the duration of surgery (DOS), length of stay (LOS), and unplanned hospital readmissions within 30 days. Artificial neural networks (ANNs) were developed to interpret data from national and institutional databases of patient outcomes, and performance was compared against traditional mathematical models. Mean-squared error analysis revealed that variation of performance existed between data sets, with the institutional model outperforming the national model in the prediction of DOS, and the national model outperforming the institutional model in LOS. However, the models also showed similar accuracy to regression modeling, meaning potential technical improvements for optimization are still necessary. Overall, this paper holds tremendous value in hospital administration and cost-saving analysis. The only limitations are inherent to the large datasets used to train the ANNs, such as low predictive capability due to being unable to interpret comprehensive patient disease severity.
Outcomes and Implications
This research represents further advancements in the possibility of AI to streamline the financial management of surgical and hospitalization costs. With high predictive capability and the ability to interpret large datasets on patient data, rather than relying on mathematically calculated traditional statistical prediction models, this study presents a clear ability for further AI research in statistical analysis within healthcare settings. Revision joint surgery is one of the highest cost orthopedic procedures, meaning less complex orthopedic surgeries have an even higher potential for efficient predictive models to be developed for clinical use. Given time to improve technical capabilities and incorporate wider sets of data, machine learning models have the potential to dramatically improve surgical cost predictions.