Abstract
Surgical site infection (SSI) is one of the common complications after surgeries. It results in various consequences, from additional treatment to morbidity and mortality. Therefore, it increases the probability and duration of hospitalization. In Norway, SSI is counted for one-quarter of all healthcare-associated infections. Aside from physical impacts, it threatens patients' emotional well-being. It also has negative effects on hospitals, for instance, more work pressure on the hospital personnel. Furthermore, SSI puts a substantial financial burden on the patient and hospital. It is estimated that the cost of treating SSI can
vary from 400$ to 30,000$ per patient. Therefore, SSI is a potential threat to patient safety.
This thesis aims to develop a data-driven model based on different factors for predicting SSI, considering infection as an undesired result of surgery that can negatively affect a patient's safety. This study suggests how different models can be selected in different predictions. Many factors lead to infection. Patient-related reasons, like the well-being of the patient and factors related to each patient. Environmental factors include factors that are presented in the operating room or from the people presented in the surgery. Surgical-related factors which happen during the surgery or because of the preparation for the surgery. Lastly, heating and ventilation factors that influence the air quality. In this study, the focus is on environmental factors. The study begins with discussions about risk analysis and SSI and how SSI can be investigated in risk analysis studies. For performing any data-driven study, data is required. As a result, an experiment to collect indoor data was performed. The experiment was done on a pig in an acute operating room in St.Olav's hospital. Due to the novelty of the work, dataset faces some problems, like not enough observations; therefore a part of this thesis is dedicated to solve these issues before beginning the risk analysis part. In the end, different measures were used to evaluate the efficiency of different models in different settings. Finally, when coming up with data-driven models and calculating their efficiency in prediction, suggested models can differ depending on what healthcare personnel require from prediction.
