Abstract Background and aims Accurate early differentiation between acute ischemic stroke (AIS) and stroke mimics (SM) remains a key challenge in prehospital and early in-hospital triage. Up to one third of patients with suspected stroke are ultimately diagnosed with non-ischemic conditions such as seizures, migraine, or functional disorders. Identifying reliable predictors available before imaging may improve triage precision, optimise resource use, and enhance patient safety. Methods This study enrolled 1,662 consecutive patients evaluated for suspected acute ischemic stroke over an 18-month period An elastic-net logistic regression model was trained on 1,411 patients using routinely collected pre-imaging variables: demographics, comorbidities, medications, NIHSS, FAST, and FAST-Plus scores. Model hyperparameters were optimised through repeated cross-validation. Numerical missing data were median-imputed with indicators; and categorical missing data were coded as “unknown.” Model performance was evaluated on an independent test set (n=251). Results The model achieved an AUC of 0.73 (95% CI: 0.66–0.81) with good calibration. Predictors of SM included history of epilepsy or headache, headache at presentation, and incomplete documentation (i.e., NIHSS, pre-stroke mRS, or glycemia). Atrial fibrillation, smoking, and antiplatelet therapy predicted AIS/TIA. At the optimal ROC threshold, sensitivity for AIS/TIA was 78.7% and specificity 63.6% (PPV 83.0%, NPV 57.0%). Conclusions A pre-imaging model based on routine clinical data shows clinically meaningful potential to support early diagnostic decisions, especially where imaging is limited or delayed. Sensitivity-oriented thresholds may enhance triage safety and efficiency in acute stroke transport pathways. Supported by the MH CZ (NU23-04-00336), MH CZ - DRO (FNOs/2022) and by MEYS, LRI CZECRIN (LM2023049). Conflict of interest Katerina Dvornikova:nothing to disclose , Veronika Kunesova: nothing to disclose, Svatopluk Ostry: nothing to disclose, Petr Novobilsky: nothing to disclose, Lenka Bártová: nothing to disclose, Linda Machova: nothing to disclose, Beata Stanková:nothing to disclose, István Szegedi:nothing to disclose, Adéla Kondé:nothing to disclose, Jana Licha:nothing to disclose, Kristyna Janotova:nothing to disclose, Martin Reiser:nothing to disclose, Tomas Jonszta:nothing to disclose, Jaroslav Havelka:nothing to disclose, Ondrej Volny: nothing to disclose, Michal Bar1:nothing to disclose
Dvorníková et al. (Fri,) studied this question.