An explainable machine learning model for prognosis prediction in sudden sensorineural hearing loss under integrated...

Area of Science:

• Otolaryngology
• Medical Informatics
• Traditional Chinese Medicine

Background:

• Sudden sensorineural hearing loss (SSNHL) is a common otologic emergency with unpredictable outcomes.
• Current prognostic tools for SSNHL, especially those integrated with therapies like Traditional Chinese Medicine (TCM), are limited.

Purpose of the Study:

• To develop and validate a machine learning model for predicting prognosis in SSNHL patients receiving integrated therapy.
• To identify key clinical and laboratory predictors influencing SSNHL recovery.

Main Methods:

• Retrospective analysis of 227 unilateral SSNHL patients treated with integrated therapy.
• Development and optimization of eight machine learning models using a training set (70%) and validation set (30%).
• Utilized SHapley Additive exPlanations (SHAP) for model interpretability and predictor identification.

Main Results:

• The XGBoost model achieved the best performance (AUC: 0.718) in the validation set, demonstrating good calibration and clinical utility.
• SHAP analysis identified activated partial thromboplastin time (APTT), disease duration, platelet count (PLT), and total protein (TP) as crucial prognostic factors.
• Identified optimal thresholds for intervention timing (within 10 days), PLT (200-250 × 10^9/L), APTT (25-30 s), and TP (65-75 g/L) for improved SSNHL recovery.

Conclusions:

• An explainable XGBoost model provides a framework for SSNHL prognosis, integrating machine learning with TCM-informed treatments.
• The model highlights the potential for personalized care by identifying specific prognostic thresholds.
• Further multicenter prospective studies are needed to validate the model's generalizability and the identified biological predictors.