LO011: Identification of mild acute cerebrovascular syndrome (ACVS) in the emergency department: validation of an ACVS clinical classifier to help distinguish mimics

Abstract

Introduction: National guidelines (NICE, AHA) for management of Acute Cerebrovascular Syndrome (ACVS) in the Emergency Department (ED) recommend the use of ABCD2 score to risk stratify patients despite its poor specificity and low diagnostic accuracy. The SpecTRA project previously developed a clinical classifier for ACVS vs. Mimic derived from historical clinical data collected during a 5-year period at an outpatient stroke clinic (Victoria, BC). Here we present a prospective evaluation of the performance of our clinical classifier on prospectively collected ED patient data compared to the industry-standard ABCD2. Methods: The prospective cohort consisted of ED patients (N=555, Male=54%, Mean (SD) Age=68.7(15.5), ACVS=70%) enrolled between Jan 2014 and May 2015 at Victoria General Hospital (BC) and Foothills Medical Centre (Calgary, AB). ABCD2 and clinical classifier scores were calculated from clinical data from the ED. We compared the performance of the two classifiers using DeLong’s test of Dependent Receiver Operating Curves (ROC). In keeping with national guidelines, we used a score of 4 or more to assess sensitivity, specificity and accuracy (sens/spec and acc) of the ABCD2; for our clinical classifier, we used the cut point previously determined to maximize agreement between predictions and true class labels in the historical data. Results: Our new clinical classifier significantly outperformed the ABCD2 (z=2.44, p=0.015) with an AUC of 0.72, (95% CI: 0.68, 0.77) vs. 0.66 (0.61, 0.71). In terms of sens/spec and acc, our classifier achieved 0.78/0.55 with acc 71% compared to 0.75/0.46 with acc 66% for the ABCD2 (using the previously specified cut points). Conclusion: Our ACVS clinical classifier showed better performance than the ABCD2 score on a prospective sample of ED patients. The improved specificity of the clinical classifier relative to existing prognostic tools would reduce the number of non-ACVS patients referred for early treatment as well as conserve medical resources. Our ongoing multi-site study will evaluate the utility of the ACVS classifier embedded in a logic-enabled e-fillable form. This form will also provide risk-based thresholds guiding timely ordering of CTA as well as links to clinical treatment guidelines. Longer-term, the e-form and classifiers will be further enhanced to include plasma-based protein biomarker data.