Abstract 36: Acute Stroke Evolution is in the Eye of the Beholder

    loading  Checking for direct PDF access through Ovid

Abstract

Background: MR WITNESS was a safety trial giving tPA to acute ischemic stroke (AIS) patients with unwitnessed onset with MRI findings of early stroke. A DWI-positive, FLAIR negative pattern has been shown to identify strokes <4.5 hr duration, but visual inspection alone may be unreliable and insensitive. Signal intensity ratios (SIR) of manual outlines (lesion/contralateral side) increase sensitivity but can lead to variability in patient selection. We investigated the influence of interrater variability on clinical and safety outcomes.

Methods: Core readers blinded to enrollment status and clinical presentation reviewed MRI of screened patients. The MR WITNESS algorithm enrolled subjects with no visible FLAIR or a pre-specified SIR <1.15. Good outcome was defined as 90 Day modified Rankin Scale (mRS) <2. SIR consistency was measured with intraclass coefficient (ICC) and reader agreement assessed with Fleiss’ Kappa. Statistical analysis included 2-sided Fisher’s Exact test or Wilcoxon Exact test as appropriate.

Results: 201 subjects were screened. 153 baseline MRIs with DWI lesions were reviewed. ICC was 71% (P<.001) and kappa was 67% (P<.0001). Among the 80 subjects enrolled using SIR <1.15 by site reading, 15 (18.8%) subjects would have been excluded based on Core readings. These 15 subjects were younger with worse NIHSS and 90-day mRS (Table). Subset analyses of subjects with pre-stroke mRS<2 showed no statistical difference (P=0.19) between subjects deemed eligible and not eligible by core readers. No difference in hemorrhagic transformation (HT) or good outcome rates was found. Using SIR <1.25 would have only excluded 3 subjects by Core readings, all of whom had HT, and poor 90 day outcomes.

Discussion: Although SIR reads between Core and sites showed good agreement, there was still variability in the absolute values. This demonstrates the potential limitation of subjective lesion volume delineation. Automated approaches should be investigated.

Related Topics

    loading  Loading Related Articles