A data-driven decision tree for diagnosing somatosensory tinnitus

Emilie Cardon^1,2, Sarah Michiels^2,3, Annick Gilles^1,2,4, Hazel Goedhart⁵, Markku Vesela⁵, Winfried Schlee⁶

¹ Department of Translational Neuroscience, Faculty of Medicine and Health Science, University of Antwerp, Antwerp, Belgium; ² Department of Otorhinolaryngology, Antwerp University Hospital, Edegem, Belgium; ³ REVAL Rehabilitation Research Center, Faculty of Rehabilitation Sciences, Hasselt University, Diepenbeek, Belgium; ⁴ Department of Education, Health and Social Work, University College Ghent, Ghent, Belgium; ⁵ Tinnitus Hub Ltd, London, United Kingdom; ⁶ Department of Psychiatry and Psychotherapy of the University Regensburg at Bezirksklinikum Regensburg, Regensburg, Germany

Background: The influence of the somatosensory system on the perception of tinnitus is well-documented. International experts have defined somatosensory tinnitus (ST) as a tinnitus influenced by the cervical or temporomandibular somatosensory system, and have recently agreed on a set of diagnostic criteria. However, a straightforward diagnosis of ST has proven elusive. Here, we present an uncomplicated model for diagnosing ST, based on the results of a large-scale online survey.

Methods: A survey was launched on the online forum Tinnitus Talk, managed by Tinnitus Hub, in September 2019. The survey included 42 questions, both on the presence of diagnostic criteria for somatosensory tinnitus and on other potentially influencing factors. Individuals were identified as having ST if they reported a known diagnosis of ST and/or experienced considerable influence from cervical spine or temporomandibular problems on their tinnitus. A decision tree was constructed to classify participants with and without ST using the rpart package in R. Tree depth was optimized during a five-fold cross-validation. Finally, model performance was evaluated on a subset containing 20% of the original dataset.

Results: In total, 7981 participants filled out the questionnaire completely, of whom 681 individuals (8.5%) were identified as having ST. The final model classified participants with an accuracy of 82.24% (82.54% sensitivity, 79.02% specificity). This decision tree was based on only four parameters: simultaneous improving or worsening of tinnitus and neck or jaw complaints, tension in the suboccipital muscles, the ability to modulate the tinnitus sound with certain movements, and the presence or absence of bruxism.

Conclusions: Based on a large dataset, we developed a straightforward tool to diagnose ST with over 80% accuracy. The final decision tree can be readily used in the clinic, and may improve diagnostic flow and expedite the start of appropriate treatment for somatosensory tinnitus.