EHR-Based Risk Factors With Prediction Models for Tinnitus Subtypes: A Multicenter Cross-sectional Study

Using a prespecified, clinically informed framework, we will classify tinnitus into relevant subtypes, including somatosensory tinnitus, acute vs. chronic tinnitus, pulsatile tinnitus, and sudden hearing loss-related tinnitus. We will first describe the distribution of these subtypes and characterize their demographic, clinical, and laboratory profiles. We will then evaluate associations between candidate risk factors and subtype membership using multivariable analyses to quantify adjusted effects. Finally, we will develop and validate multivariable prediction models using both conventional statistical approaches and machine learning methods to support tinnitus subtype classification. Model performance will be assessed using discrimination, calibration, and clinical utility metrics. By integrating routine clinical data with biomarker information captured in real-world care, this study aims to provide evidence-based tools to improve tinnitus subtype diagnosis and enable more personalized clinical assessment.