Stroke Risk Assessment in the Emergency Department: Prognostic Models, Test Modalities and Variation of Practice

Abstract

Introduction: Stroke is a common and serious disorder and a leading cause of death and disability. Most strokes are ischemic with a significant portion preceded by transient ischemic attack (TIA). Early TIA diagnosis and stroke prevention are critical to prevent subsequent poor outcomes. There may be deficiencies and opportunities for improvements in current TIA patient management. Objectives: The main objectives of this thesis were threefold: to 1) examine reporting quality of derivation and validation studies for prognosis of stroke in patients with TIA, 2) examine practice variation in Canadian stroke prevention clinics, and 3) derive and validate a clinical prediction score for clinically significant symptomatic carotid artery disease (i.e., stenosis ≥ 50%) in patients with TIA. Methods: To achieve these objectives we conducted a systematic review of prediction model studies for TIA patients, conducted an electronic survey of stroke prevention clinic leads, and derived and internally validated a clinical prediction tool. Results: For the systematic review, 7,026 articles were screened; 60 were retained consisting of 100 derivation and validation studies. Among the 100 derivation and validation studies, few reported whether assessment of outcome (24%) and predictors (12%) was blinded; sample size justifications (49%), description of method for handling missing data (16%), and calibration method (5%) were seldom reported. Thirty-eight clinicians at 76 eligible clinics responded to the survey (response rate 50%). The majority of clinics (66%) were open at least five days a week. Most high-risk patients were not seen by the clinics within 48 hours. COVID-19 had a negative impact on routine patient care including longer wait times. The clinical prediction model for clinically significant symptomatic carotid artery disease included 13 predictors with an optimism-corrected C-statistic of 0.781 (95% CI: 0.769-0.798) and good calibration. We simplified the model into a score (Symcard Score), with suggested cut-points for high (15+ points), medium (12-14), and low-risk (≤11) stratification. A substantial portion (38%) of TIA patients were classified as low-risk [sensitivity 92.9% (95% CI: 91.0%-94.5%), specificity 41.1% (95% CI: 40.1%-42.1%), and diagnostic yield 1.7% (95% CI: 1.3%-2.1%)]. Conclusions: Many studies deriving or validating clinical prediction models for the prognosis of stroke in patients with TIA do not report essential items adequately, precluding effective critical appraisal of such tools for potential use in clinical settings. There is wide variation in management of TIA patients among stroke prevention clinics and potentially suboptimal patient care. Only a portion of TIA patients require vascular imaging - yet, no tool was previously available to identify such patients, leading to unnecessary imaging and delayed care. We derived and internally validated a prediction score for the diagnosis of clinically significant carotid artery disease that has good discrimination and calibration. After external validation, the Symcard Score can potentially support clinicians and avoid or delay a significant percentage of vascular imaging for patients with TIA, especially in crowded emergency departments.