Outcome prediction in endoscopic surgery for chronic rhinosinusitis - a multidimensional model

Abstract

Purpose: Chronic rhinosinusitis (CRS) is one of the most common diseases in the modern society. In recent years endoscopic sinus surgery (ESS) has become the treatment of choice for patients with CRS refractory to medical therapy. ESS proved to be successful in most, but not all patients with CRS. Currently there is no direct method available to distinguish between patients who are likely to benefit from ESS and those who are not. The aim of this study was to build multidimensional models (artificial neural networks) to predict early outcomes of ESS in individual patients.

Material/methods: The study group comprised of 115 patients operated for CRS in the Department of Otolaryngology, Jagiellonian University Collegium Medicum, Cracow. The neural models were created using the Statistica Neural Network computer software package. The models required only information easily achievable for every patient before surgery. Consequently, the models could be readily applicable in everyday clinical practice. To define the results of surgery three different mathematical descriptions were compared. The models' predictions were compared with the actual results of surgery 3-6 months postoperatively.

Results: The models were able to predict the early outcome of surgery in 90% of the patients but their quality depended on mathematical representation of the surgery result. The best models were characterized by 93% sensitivity and 86% specificity.

Conclusions: The results of ESS depend on many factors, so reliable outcome prognoses can be produced only by multidimensional models. Artificial neural networks are a promising multidimensional tool facilitating clinical decision making in patients with CRS.

Keywords: Artificial neural networks; Chronic rhinosinusitis; Endoscopic sinus surgery; Modelling; Outcome prediction.