Posterior Canal and Atypical Benign Paroxysmal Positional Vertigo: Development of a Predictive Model for Clinical Decision-Making.

BACKGROUND

Benign paroxysmal positional vertigo (BPPV) is the most common cause of dizziness. Despite its prevalence, clinical diagnosis can be challenging because of difficulty discerning the direction and intensity of eye movements, resulting in diagnostic uncertainty and suboptimal treatment outcomes. The aim of this study was to develop and evaluate a data-driven model designed to differentiate between typical and atypical variants of BPPV.

METHODS

A retrospective cross-sectional design included data from 26 patients diagnosed clinically with either typical BPPV or atypical variants of BPPV and two control subjects with a history of dizziness but no vertigo. Eye and head movement data were extracted using VisualEyes™ (Interacoustics, Middelfart, Denmark) and processed in MATLAB (MathWorks, Natick, Massachusetts, United States). Latency and peak slow-phase velocity (SPV) of torsional nystagmus were used as predictors in a logistic regression model. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, area under the ROC curve (AUC), and repeated five-fold cross-validation.

RESULTS

The final model, based on latency and peak SPV of torsional nystagmus, achieved a mean AUC of 0.936±0.007, indicating excellent discrimination between typical and atypical BPPV. Across cross-validation folds, the model demonstrated a mean accuracy of 88.0%±0.9%, a sensitivity of 78.2%±1.9%, and a specificity of 93.9%±1.1%, reflecting a consistently high rate of correct classification. When evaluated on four previously unseen patient recordings outside the training dataset, the model showed 100% agreement with expert clinical assessments.

CONCLUSION

The developed model demonstrates high diagnostic accuracy in differentiating typical BPPV from atypical variants and shows promise as a clinical decision support tool.