Abstract 2650: Reliability and Validity of Retrospective Extraction of the ABCD
2
Score From Medical Records
Background and Purpose: The ABCD 2 score is increasingly used for risk stratification of TIA patients. Many of the studies assessing the utility of the ABCD 2 score have retrospectively extracted the score from medical records, but the reliability of estimation based on chart review has not been reported. We sought to determine whether the ABCD 2 score can be estimated retrospectively from medical records. Methods: We compared ABCD 2 scores that were prospectively determined by a board certified vascular neurologist to scores determined retrospectively from medical record review. All ED records and initial neurology consult resident notes were abstracted with explicit ABCD 2 scoring redacted. ABCD 2 scores were then estimated by two independent blinded trained raters using these records. Estimated component, total scores, and ABCD 2 risk category (0-3, 4-5, 6-7), were compared both between raters and with the prospectively obtained scores. Reliability was assessed using unweighted kappa statistics. Results: Records were analyzed for 55 consecutive TIA patients. For the two raters of retrospective ABCD 2 scores, interrater reliability was good with 75% exact agreement in total score between raters (κ=0.68) and 84% agreement for ABCD 2 category (κ=0.74). Interrater agreement on individual items was greatest for age and diabetes, and poorest for clinical features and duration. Retrospectively estimated total ABCD 2 scores were identical to prospective scores in 52% of patients. Agreement between raters and the prospectively obtained score was >90% for age, blood pressure, and diabetes, but was only ∼70% for clinical features and duration. Retrospectively estimated ABCD 2 category was identical to the prospective category in 64% of patients ( Table ). Dichotomizing at an ABCD 2 score ≥4, retrospective assessment had a sensitivity of 83% (CI 70-95%) and specificity of 67% (CI 40-95%) compared to the prospective score. Conclusions: The ABCD 2 score can be abstracted from medical records with good interrater reliability but limited validity, which may lead to misclassification of risk category in over one-third of patients.