Predicting the optimal therapeutic intervention for tinnitus patients using random forest regression: A preliminary study of UNITI's decision support system model.

Abstract

Tinnitus is the conscious perception of a phantom sound in absence of an external or internal stimulus. More than 1 in 7 adults in the EU experience tinnitus and for a large proportion of them tinnitus is an intrusive, persistent, and disabling condition, which impairs their life quality. Therefore, tinnitus is posed as a major global burden, which requires a precision-medicine approach in terms of treatments that are tailored to individual patients, due to its high heterogeneity. UNITI is a research and innovation project which aims towards this goal, unifying treatments and interventions for tinnitus. In the context UNITI, a randomized controlled trial (RCT) is being conducted and all the participants' data will be utilized for the development of a clinical decision support system (CDSS). This CDSS will predict the optimal therapeutic intervention for a tinnitus patient based on their profile. In this paper, we present a preliminary study of the CDSS model development process. We describe the available input data, the pre-processing steps conducted, the algorithms tested to model the CDSS' prediction, the models' results, and the future work in the context of this project. The R2 score of the selected model is currently 0.65, indicating that its development process is in the right direction but further tuning and hyperparameter optimization is needed. Clinical Relevance- The proposed model will be integrated in a CDSS aiming at indicating the optimal treatment strategy for a tinnitus patient based their personal profile.