Exploring the possibility of combining the Swedish Perioperative Register with machine learning to predict surgery durations

Sammanfattning: The operating room is one of the most expensive resources at a hospital and it is therefore important that the scheduling of surgeries is done in an optimal way. Underand overestimations can lead to both risks and costs associated with inefficient use of the operating room. Having accurate estimations for the surgery durations is therefore one critical element in the planning process. In this study, machine learning techniques are applied to data from the Swedish Perioperative Register (SPOR) to investigate the possibility of predicting durations with less error than current methods used in hospitals. Using linear regression and neural networks, the result shows that feature engineering is an important factor and that the method demonstrates a potential improvement of almost 50% in terms of the mean squared error. This shows that by using surgery-related data from SPOR to train predictive models, it is possible to improve upon existing estimations. These improvements could then lead to better scheduling processes and amore optimal use of surgery resources.