Interobserver Reliability of Baseline Noncontrast CT Alberta Stroke Program Early CT Score for Intra-Arterial Stroke Treatment Selection

The ASPECTS has been shown to predict outcomes of early ischemic patients after intra-arterial therapy by providing semiquantitative data regarding infarction core. In this article the authors assessed the interobserver reliability of this scale in patients with proximal occlusions. CT studies in 155 patients were retrospectively analyzed by 2 independent observers. Among patients with anterior circulation proximal artery occlusions who were eligible for intra-arterial therapy, interrater reliability for ASPECTS grading was substantial across the entire scale. When using the dichotomized ASPECTS (≤ 7 versus >7) for treatment selection, agreement was only moderate, limiting its utility. In the patient cohort, approximately 25% of treatment decisions would have been affected by interrater reliability. BACKGROUND AND PURPOSE: Early ischemic changes on pretreatment NCCT quantified using ASPECTS have been demonstrated to predict outcomes after IAT. We sought to determine the interobserver reliability of ASPECTS for patients with AIS with PAO and to determine whether pretreatment ASPECTS dichotomized at 7 would demonstrate at least substantial κ agreement. MATERIALS AND METHODS: From our prospective IAT data base, we identified consecutive patients with anterior circulation PAO who underwent IAT over a 6-year period. Only those with an evaluable pretreatment NCCT were included. ASPECTS was graded independently by 2 experienced readers. Interrater agreement was assessed for total ASPECTS, dichotomized ASPECTS (≤7 versus >7), and each ASPECTS region. Statistical analysis included determination of Cohen κ coefficients and concordance correlation coefficients. PABAK coefficients were also calculated. RESULTS: One hundred fifty-five patients met our study criteria. Median pretreatment ASPECTS was 8 (interquartile range 7–9). Interrater agreement for total ASPECTS was substantial (concordance correlation coefficient = 0.77). The mean ASPECTS difference between readers was 0.2 (95% confidence interval, −2.8 to 2.4). For dichotomized ASPECTS, there was a 76.8% (119/155) observed rate of agreement, with a moderate κ = 0.53 (PABAK = 0.54). By region, agreement was worst in the internal capsule and the cortical areas, ranging from fair to moderate. After adjusting for prevalence and bias, agreement improved to substantial or near perfect in most regions. CONCLUSIONS: Interobserver reliability is substantial for total ASPECTS but is only moderate for ASPECTS dichotomized at 7. This may limit the utility of dichotomized ASPECTS for IAT selection.