Sudden Hearing Loss Prognosis Predictors Under Integrated Therapy

227 patient records with unilateral sudden sensorineural hearing loss were used to train an explainable machine learning model. The XGBoost model achieved an AUC of 0.718 in validation, identifying key predictors like activated partial thromboplastin time and platelet count. The dataset was created by Xiaoxiao Ye and last updated on June 1, 2026.

Use Cases

• Developing prognostic models for sudden hearing loss based on clinical and laboratory features like APTT and platelet count.
• Investigating non-linear relationships between biomarkers and recovery probability based on the thresholds identified in SHAP analysis.
• Validating machine learning model performance in clinical settings using metrics of discrimination, calibration, and net benefit.
• Exploring personalized care pathways for patients with Qi stagnation and blood stasis syndrome undergoing integrated therapy.

Strengths

• Model development followed rigorous methodology, including outcome-based stratified sampling and strict prevention of information leakage during preprocessing.
• The study provides specific, interpretable thresholds for key predictors, such as a critical intervention window within 10 days of symptom onset.
• Model performance was systematically evaluated across discrimination (AUC 0.718), calibration, and clinical utility.

Limitations

• The dataset is small, containing only 227 patient records, which may limit the model's generalizability.
• The study is retrospective and lacks external validation, as noted by the authors.
• Column-level documentation is absent; field semantics must be inferred from the research paper.

Provenance

Source

figshare

Collection Method

Retrospective study of patients receiving standardized integrated therapy.

Freshness

Last updated 2026-06-01 04:15:11