Predictive modeling of maneuver numbers in BPPV therapy using machine learning.
Mine Baydan-Aran1, Kübra Binay-Bolat1, Emre Söylemez2
1Department of Audiology, Ankara University, Ankara, Türkiye.
Summary
Machine learning models can predict if patients with benign paroxysmal positional vertigo (BPPV) will need more than one maneuver for treatment. This helps in planning care and improving patient outcomes for BPPV.
Area of Science:
- Neurology
- Medical Informatics
Background:
- Benign paroxysmal positional vertigo (BPPV) often requires multiple treatment maneuvers.
- Predicting the number of maneuvers needed is crucial for efficient patient management.
Purpose of the Study:
- To utilize machine learning (ML) to identify predictors for multiple canalith repositioning maneuvers (CRMs) in BPPV patients.
- To enhance the predictability of treatment requirements for BPPV.
Main Methods:
- Retrospective study of 520 BPPV patients (2018-2023).
- Collected data included age, BPPV type, comorbidities, gender, and number of maneuvers.
- Evaluated ML models (GBM, Logistic Regression, XGBoost, SVC) using precision, F1-score, accuracy, recall, and AUC.
Main Results:
- 36% of patients required one maneuver, while 67% needed more than one.
- Gradient Boosting Machine (GBM) showed the best AUC for maneuver number estimation.
- XGBoost achieved the best F1 and recall scores; Support Vector Classifier had the highest accuracy.
Conclusions:
- Machine learning models demonstrate high predictive capability for identifying BPPV patients needing multiple maneuvers.
- ML can facilitate more efficient treatment planning and improve patient outcomes in BPPV management.
