Document resource
Background/Objectives Vestibular Neuritis (VN) and Posterior Circulation Stroke (PCS) are the two common causes of Acute Vestibular Syndrome. Expert clinicians distinguish between them using history, examination, and vestibular tests. Machine learning models capable of expert-level classification could expand access to diagnostic expertise.Methods We recruited Emergency Department (ED) patients with acute vestibular syndrome who received a final diagnosis of VN or PCS. Data from history, bedside examination and four vestibular tests (videonystagmography, video head impulse test (VHIT), vestibular-evoked myogenic potentials (VEMP) and subjective visual horizontal) were used for model development.To ensure clinical applicability across EDs with varying neuro-otology resources, we tailored our models for three scenarios, simulated by restricting available data to predefined tiers. Tier 1 represented an ED with neuro-otology support (history, neuro-otological examination, videonystagmography, VHIT, ocular VEMP), Tier 2 an ED with VHIT (history, bedside examination, VHIT) and Tier 3 an ED using only history and bedside examination. Model performance was also compared against HINTS (head impulse, nystagmus, test-of-skew) by experts.Results Our dataset included 163 VN and 131 PCS patients. The best-performing model in each tier used either the CatBoost or XGBoost algorithms and identified PCS with accuracies of 96.6% (95% CI: 93.3–99.9%), 94.6% (95% CI: 90.5–98.6%) and 88.8% (95% CI: 86.0–91.6%) in Tiers 1, 2 and 3. HINTS achieved 94.6% accuracy.Conclusion Machine learning models can differentiate PCS from VN with high accuracy and have potential to improve the diagnosis of acute vestibular syndrome for clinicians in EDs with varying neuro-otology expertise and resources.