Clinical Report: A Transparent Machine Learning Approach for Predicting Outcomes
Overview
This study developed an explainable XGBoost model to predict outcomes in patients with sudden sensorineural hearing loss (SSNHL) undergoing integrated therapy. Key predictors identified include activated partial thromboplastin time, disease duration, platelet count, and total protein levels.
Background
Sudden sensorineural hearing loss (SSNHL) is a prevalent otologic emergency with a variable prognosis, affecting over 1.5 billion people globally. Current prognostic models are limited, particularly for patients receiving integrated therapy combining Western and Traditional Chinese Medicine. There is a pressing need for reliable predictive tools to guide clinical decision-making and improve patient outcomes.
Data Highlights
Model
AUC
95% CI
XGBoost
0.718
0.590–0.846
Key Findings
The XGBoost model achieved an AUC of 0.718 in predicting outcomes for SSNHL patients.
Four key prognostic predictors were identified: APTT, disease duration, PLT, and TP.
Optimal recovery was associated with PLT counts of 200–250 × 109/L.
APTT values of 25–30 seconds indicated better recovery chances.
TP levels showed a tri-phasic association with peak prognostic probability at 65–75 g/L.
A critical intervention window was identified within 10 days of symptom onset.
Clinical Implications
The findings suggest that clinicians can utilize the XGBoost model to better predict treatment outcomes in SSNHL patients, allowing for more personalized care. Understanding the key predictors can aid in early intervention and tailored treatment strategies.
Conclusion
This study presents a novel machine learning approach to prognostic prediction in SSNHL, emphasizing the integration of traditional and modern medical practices. Further validation through multicenter studies is necessary to confirm these findings.
