Development and Evaluation of a Clinical Model to Predict Low-Frequency Hearing Prognosis in Menière’s Disease: A Multicenter Retrospective Cohort Study

Background Patients with Menière's disease (MD) typically present with fluctuating low-frequency sensorineural hearing loss. The progression of hearing loss varies among individuals, making it challenging to predict prognosis and select appropriate treatment strategies. A method to estimate future hearing levels in patients with MD could support personalised treatment decisions; however, such an approach has not yet been established. Accordingly, we aimed to develop and validate a model to predict low-frequency hearing levels 1 year later using clinical data obtained at the initial visit. Methods We retrospectively collected electronic health record data from 124 patients diagnosed with definite MD according to the Barany Society criteria. The development cohort (n = 92) was recruited from two university hospitals, whereas the external validation cohort (n = 32) was obtained from one general hospital. The outcome was the low-frequency hearing level measured 1–2 years after the initial visit. We developed the predictive model using multiple linear regression analysis, selecting predictors based on the smallest root mean square error (RMSE) during K-fold cross-validation. Model performance was evaluated using RMSE and the coefficient of determination (R²). Results The model incorporating age and low- and high-frequency hearing levels at the initial visit showed the highest predictive accuracy. By inputting the values of each predictor into the model, low-frequency hearing levels 1 year later were predicted. Across the overall cohort, 69.4% of predictions fell within the measurement error range of < 10 dB. In internal validation, RMSE ranged from 9.1 to 11.5 dB, with R² values of 0.64–0.78. In external validation, RMSE was 11.2 dB, and R² was 0.60. Conclusions We developed a simple model for MD that can reasonably predict low-frequency hearing levels 1 year later using readily available clinical data. This model may assist clinicians and patients in making informed treatment decisions. Trial registration Not applicable.