Systematic review and pooled analysis of published and unpublished validations of the ABCD and ABCD2 transient ischemic attack risk scores.

Abstract

The ABCD system was derived to predict early risk of stroke after transient ischemic attack. Independent validations have reported conflicting results. We therefore systematically reviewed published and unpublished data to determine predictive value and generalizability to different clinical settings and users. Validations of the ABCD and ABCD2 scores were identified by searching electronic databases, reference lists, relevant journals, and conference abstracts. Unpublished tabulated data were obtained where available. Predictive value, expressed as pooled areas under the receiver operator characteristic curves (AUC), was calculated using random-effects meta-analysis, and analyses for heterogeneity were performed by categorization according to study setting and method. Twenty cohorts were identified reporting the performance of the ABCD system in 9808 subjects with 456 strokes at 7 days. Among the 16 studies of both the ABCD and ABCD2 scores, pooled AUC for the prediction of stroke at 7 days were 0.72 (0.66 to 0.78) and 0.72 (0.63 to 0.82), respectively (P diff=0.97). The pooled AUC for the ABCD and ABCD2 scores in all cohorts reporting relevant data were 0.72 (0.67 to 0.77) and 0.72 (0.63 to 0.80), respectively (both P<0.001). Predictive value varied significantly between studies (P<0.001), but 75% of the variance was accounted for by study method and setting, with the highest pooled AUC for face-to-face clinical evaluation and the lowest for retrospective extraction of data from emergency department records. Independent validations of the ABCD system showed good predictive value, with the exception of studies based on retrospective extraction of nonsystematically collected data from emergency department records.