Development and Validation of a Modified Sudbury Vertigo Risk Score for Predicting Central Causes of Dizziness in the Emergency Department.

Abstract

Background: Distinguishing central causes of dizziness from peripheral causes in emergency department (ED) patients is a challenging clinical problem. The Sudbury Vertigo Risk Score is a clinical prediction model that supports decision-making by predicting central causes of dizziness. However, its predictor "benign paroxysmal positional vertigo (BPPV) diagnosis" may introduce uncertainty and limit clinical usability. Therefore, we developed a modified model in which this predictor was replaced with variables that can be assessed through patient history.

Methods: We retrospectively included consecutive patients aged ≥ 15 years who presented with dizziness to an ED between April 2013 and March 2023. The outcome was dizziness due to central lesions, defined as abnormalities identified on neuroimaging and judged by a relevant specialist to be the cause of dizziness. We developed a modified model using multivariable logistic regression in which "BPPV diagnosis" was replaced with two predictors-"trigger (provoked by changes in head position)" and "history of dizziness"-while retaining the other original predictors. We evaluated discrimination using the area under the receiver operating characteristic curve (AUROC) and calibration using a calibration plot. We compared efficiency and safety between the modified model and the Sudbury model.

Results: Among 3606 patients, a total of 2958 were eligible. Dizziness due to central lesions was identified in 155 patients (5.2%). The AUROC was 0.85 (95% CI, 0.82-0.88) for the Sudbury model and 0.81 (95% CI, 0.77-0.85) for the modified model. The modified model showed calibration comparable to that of the original model. Safety and efficiency were 0.0% (95% CI, 0.0%-0.5%) and 26.8% (95% CI, 25.2%-28.4%) for the Sudbury model, and 0.9% (95% CI, 0.4%-2.1%) and 18.7% (95% CI, 17.3%-20.1%) for the modified model.

Conclusion: The modified model showed comparable performance after replacing BPPV diagnosis with predictors that can be more easily assessed through patient history.