Predicting the prognosis of patients with sudden sensorineural hearing loss by analyzing the audiometric curve of the unaffected ear.

Objective: This study aims to extract potential information from the audiograms of the unaffected ear in patients with unilateral sudden sensorineural hearing loss (USSNHL). It explores the relationship between the characteristics of the audiograms of the unaffected ear and the treatment effectiveness for USSNHL. Additionally, the research presents the findings in a way that enhances communication and allows for verification. Methods: The study employs piecewise curve fitting to simplify the changing trend of audiograms in the unaffected ear of USSNHL patients into the slopes of three straight lines. Utilizing Python, the research team conducts a cluster analysis on the 229 patients' audiometric characteristics and trains the clustering results into an algorithm model. After clustering, the team applies statistical methods such as regression analysis to explore the correlation between the clustering results and the therapeutic efficacy. Results: The study completes the clustering analysis and encapsulates the trained model into an executable program. The algorithm clusters the patients into Cluster X and Cluster Y based on the audiogram characteristics of the unaffected ear. The clustering results demonstrate a significant correlation with the treatment efficacy. Regression analysis shows that Cluster Y patients achieve an average improvement in hearing threshold post-treatment that is 6.52 dB higher than that of Cluster X. The relative risk of "No improvement" for Cluster Y is half that of Cluster X. Additionally, age and the audiogram type of the affected ear also contribute to the prognosis of USSNHL to varying degrees. Furthermore, the research team submits the trained clustering model and corresponding spreadsheet as attachments, facilitating dissemination and validation. Conclusion: Regression analysis confirms that the clustering results are independent factors indicative of the prognosis in patients with USSNHL. The data exerting the most significant influence on clustering analysis outcomes were derived from the evolving auditory threshold patterns in the posterior segment of audiometric curves obtained from unaffected ears. This observation indicates a strong correlation between mid-to-high frequency threshold progression in the contralateral ear and clinical prognosis among patients with USSNHL. The clustering methodology demonstrated robust classification efficacy for auditory data lacking explicit cutoff values, ultimately enabling refined patient stratification through multidimensional pattern recognition.