Risk stratification for stroke in acute persistent vertigo: development and internal validation of a multivariable prediction model.

Background: Despite the fact that acute persistent vertigo can be a presenting sign of stroke, early bedside risk assessment continues to be particularly difficult. In order to predict clinically detected strokes among patients who were experiencing acute persistent vertigo, our objective was to create and internally validate a multivariable model. Additionally, we wanted to evaluate the model's performance in comparison to scores that are routinely utilized. Methods: We retrospectively analyzed 689 consecutive patients with acute persistent vertigo, operationally defined as vertigo lasting longer than 24 h, including 165 clinically diagnosed strokes (23.9%). Candidate predictors were prespecified and selected using LASSO-regularized logistic regression, followed by multivariable logistic modeling. Internal validation was performed using 5-fold cross-validation, with 95% confidence intervals obtained via bootstrap resampling ( Results: The whole cohort had a mean age of 60.3 ± 12.9 years and included 376 males (54.6%) and 313 females (45.4%); the stroke subcohort had a mean age of 65.0 ± 14.6 years and included 89 males (53.9%) and 76 females (46.1%). In the multivariable model, older age (analyzed as a continuous variable; OR 1.050 per 1-year increase), smoking (OR 2.147), hypertension (OR 5.452), hyperlipidemia (OR 2.621), diabetes mellitus (OR 3.918), coronary heart disease (OR 4.361), history of atrial fibrillation (OR 10.376), higher CNS score (OR 1.679 per point), and nausea/vomiting (OR 2.020) were associated with increased stroke odds, whereas tinnitus was inversely associated (OR 0.382). The model showed excellent discrimination (AUC 0.902; 95% CI 0.874-0.927), with sensitivity 0.873 and specificity 0.792 at the Youden threshold, good calibration (intercept -0.036; slope 0.973), and a low Brier score (0.096). Performance exceeded ABCD2 (AUC 0.642) and Triage-Plus (AUC 0.514), and was higher than CNS alone (AUC 0.846). Risk tertiles (cut-points 0.039 and 0.242) yielded stroke event rates of 2.6, 10.5, and 58.7% in low-, intermediate-, and high-risk groups, respectively. Conclusion: Our findings suggest that a multivariable prediction model may assist in risk stratification and diagnostic decision-making, as well as providing tailored stroke risk assessment for acute persistent vertigo. It would be beneficial to perform impact studies and external validation.