Large vessel occlusion prediction scales provide high negative but low positive predictive values in prehospital suspected stroke patients

INTRODUCTION: We studied a registry of Emergency Medical Systems (EMS) identified prehospital suspected stroke patients brought to an academic endovascular capable hospital over 1 year to assess the prevalence of disease and externally validate large vessel occlusion (LVO) stroke prediction scales with a focus on predictive values. METHODS: All patients had last known well times within 6 hours and a positive prehospital Cincinnati Prehospital Stroke Scale. LVO prediction scale scores were retrospectively calculated from emergency department arrival National Institutes of Health Stroke Scale scores. Final diagnoses were determined by chart review. Prevalence and diagnostic performance statistics were calculated. We prespecified analyses to identify scale thresholds with positive predictive values (PPVs) ≥80% and negative predictive values (NPVs) ≥95%. A secondary analysis identified thresholds with PPVs ≥50%. RESULTS: Of 220 EMS transported patients, 13.6% had LVO stroke, 15.9% had intracranial haemorrhage, 20.5% had non-LVO stroke and 50% had stroke mimic diagnoses. LVO stroke prevalence was 15.8% among the 184 diagnostic performance study eligible patients. Only Field Assessment Stroke Triage for Emergency Destination (FAST-ED) ≥7 had a PPV ≥80%, but this threshold missed 83% of LVO strokes. FAST-ED ≥6, Prehospital Acute Severity Scale =3 and Rapid Arterial oCclusion Evaluation ≥7 had PPVs ≥50% but sensitivities were <50%. Several standard and lower alternative scale thresholds achieved NPVs ≥95%, but false positives were common. CONCLUSIONS: Diagnostic performance tradeoffs of LVO prediction scales limited their ability to achieve high PPVs without missing most LVO strokes. Multiple scales provided high NPV thresholds, but these were associated with many false positives.