Sudden sensorineural hearing loss (SSHL) is a critical otologic emergency with rising bilateral ‑ onset rates. Using 14 ‑ year data from 939 SSHL patients, we built a machine ‑ learning model to predict healthy ‑ ear incidence. Key predictors included age, hearing loss severity, gender, cardiovascular and inflammatory indices. The optimal model yielded AUC 0.77 and robust metrics, filling a research gap. Future multi ‑ center validation and multi ‑ omics integration will enhance its clinical utility.
In the field of otology research, sudden sensorineural hearing loss (SSHL) is a highly concerned clinical emergency. It refers to a sudden-onset sensorineural hearing loss of unknown cause, with a decrease of ≥20 dB in two consecutive frequencies. Bilateral sudden sensorineural hearing loss (BSSHL) is more complex. According to the onset interval between the two ears, it can be further divided into simultaneous bilateral sudden sensorineural hearing loss (Si-BSSHL, with an interval ≤ 3 days) and sequential bilateral sudden sensorineural hearing loss (Se-BSSHL, with an interval > 3 days).
In recent years, the incidence of SSHL has shown a significant upward trend, and the same is true for BSSHL. As early as 1996, it was reported that the incidence of BSSHL accounted for 1.7% of the total number of SSHL patients. By 2024, a meta-analysis indicated that this proportion had reached 11.9%. Due to the unclear etiology, unknown mechanism, lack of standardization, and poor prognosis of this disease, it has received increasing attention. Hearing loss not only causes great physical and psychological trauma to patients but also has a serious negative impact on their social activities, significantly reducing their quality of life.
Against this backdrop, the team led by Director Wang Qiuju from the Department of Otology, Chinese PLA General Hospital, has presented exciting research results. They have successfully established the first model for predicting the risk of healthy-ear incidence after SSHL through machine-learning algorithms. The research team conducted an in-depth analysis of the clinical data of 939 inpatients with SSHL accumulated over 14 years. These patients were divided into two groups: unilateral sudden sensorineural hearing loss (USSHL) and sequential bilateral sudden sensorineural hearing loss (Se-BSSHL). Stratified sampling was carried out according to the ratio of the number of patients in the unilateral and bilateral groups, and the data were divided into an internal dataset (accounting for 70%) and an external dataset (accounting for 30%). During the research process, the team used a decision-tree-based recursive feature elimination method to screen variables. Five-fold cross-validation was performed on the internal dataset, and at the same time, SMOTE data augmentation was carried out on each fold of data, increasing the ratio of unilateral to bilateral from 1:7 to 1:2. Four machine-learning algorithms, namely random forest (RF), extreme gradient boosting (XGB), logistic regression (LR), and support vector machine (SVM), were employed to establish the prediction model, and then, the model was verified on the external dataset. The performance of the model was comprehensively evaluated through a series of professional indicators, such as the area under the curve (AUC), calibration curves, decision curve analysis (DCA) curves, as well as accuracy, precision, recall, and F-value. Finally, the model results were visually presented using a Shapley diagram.
The research results show that the incidence of completely normal healthy ears in USSHL patients is 12.57%. A total of 19 variables were included in the prediction model for the incidence of healthy ears in SSHL. Among them, age, degree of hearing loss, and gender are the three most important characteristics affecting the recurrence of healthy ears in USSHL. An increase in age and a more severe degree of hearing loss are both risk factors for incidence, and men have a higher risk of incidence. In addition, the high-risk test indicators for incidence can be summarized into two categories of risk factors: cardiovascular (body mass index, total cholesterol, high-density lipoprotein cholesterol, apolipoprotein A1, fibrinogen) and inflammatory (eosinophils, lymphocyte count, neutrophil count). The model performed excellently in external testing. The optimal AUC reached 0.77, the accuracy rate was 88%, the precision rate was 84%, the recall rate was 88%, and the F1 value was 0.85, fully demonstrating the superior performance of this model.
This research is of great significance. It uses machine-learning algorithms that are more capable of handling complex data, integrates multiple factors such as patients' basic information, medical history, symptoms, and test indicators for comprehensive analysis, achieves more comprehensive disease prediction, and successfully fills the research gap in this field. Looking ahead, in the research of this field, the development direction of the prediction model is promising. On the one hand, efforts can be focused on establishing a multi-center database, conducting more extensive external verification and adjustment of the model, and then promoting and applying the model through web-based forms. On the other hand, efforts can focus on further deepening the integration of multimodal data by incorporating genomic, radiomic, and other types of data into the model to continuously improve its ability to diagnose and predict diseases. With continued research in this area, these advances may provide greater benefits to patients with SSHL.
***
Reference
DOI: 10.1097/CM9.0000000000004230
About Prof. Qiuju Wang from PLA General Hospital
Prof. Qiuju Wang is a renowned otology expert and a pioneer in establishing the otology subspecialty in China and the People’s Liberation Army (PLA). Her research focuses on the diagnosis and treatment of deafness, tertiary prevention, and the genetic mechanisms underlying hearing loss. She has led more than 30 national and military key research projects. Her major contributions include establishing domestic deafness-screening models, identifying the world’s first reported case of Y-linked hereditary deafness, and leading multiple national guideline development and genome research initiatives.
Funding information
This study received support from the National Natural Science Foundation of China (General Program) (No. 82171130), and the National Key Research and Development Program of China (Nos. 2023YFC2508400, 2023YFF1203504, and 2023YFC2509800).
Chinese Medical Journal
Observational study
People
Development of machine learning-based predictive models for evaluating risk in the contralateral ear among patients with unilateral sudden sensorineural hearing loss
20-Sep-2026
None