Automatic Algorithm for the Determination of the Anderson-wilkins Acuteness Score In Patients With St Elevation Myocardial Infarction

Background: The Anderson-Wilkins score (AW-score) is based on quantitative 12-lead electrocardiogram (ECG) criteria to estimate acuteness of ischemia in patients with ST elevation myocardial infarction (STEMI). The score identifies patients with a substantial salvage potential after primary percutaneous coronary intervention regardless of patient reported symptom duration. However, due to the complexity of the score, its manual interpretation is time consuming and therefore has not been applied in clinical practice. Automation of this score could facilitate clinical application. Therefore, we aimed to develop and validate an automatic algorithm for the AW-score. Methods: The AW-score (obtained from presenting ECG), assesses changes in ST-T-segments, T-waves and Q-waves. Each lead is designated an acuteness phase (1A, 1B, 2A or 2B) and the overall score is calculated. AW-score ranges from 1 (late ischemia/least acute) to 4 (early ischemia/most acute) and is calculated from the formula: AW-score = [(4 × (#leads 1A) + 3 × (#leads 1B) + 2 × (#leads 2A) + 1 × (#leads 2B))/(∑#leads with 1A, 1B, 2A or 2B)]. We developed an algorithm to automatically determine AW-score. The algorithm was designed using 50 ECGs. Each ECG lead (except aVR) was manually scored according to AW-score by two independent experts (Exp1 and Exp2) and automatically by our designed algorithm (auto-score). An adjudicated manual score (Adj-score) was determined between Exp1 and Exp2. The inter-rater reliabilities (IRRs) between Exp1 vs Exp2, and Adj-score vs auto-score were assessed by interclass correlation coefficient (ICC). Results: The Adj-score and auto-score had median AW-score 2.7 (1.75–3.52) and 3.0 (2.32–3.80), respectively. The IRR for AW-score between Adj-score and auto-score was ICC = 0.64 (CI 0.36–0.80), p < 0.001). Substantial differences in AW-score between Adj-score and auto-score were due to difference in measures of Q-wave duration. The IRR for AW-score between Exp1 and Exp2 was ICC = 0.89 (confidence interval (CI) 0.79–0.93, p < 0.0001). Conclusion: We have developed an automatic algorithm for measurement of AW-score. The preliminary test result was near acceptable for the inter-rater reliability between manual Adj-score and auto-score. More adjustments are needed to improve the measure of agreements between manual score and automatic algorithm for AW-score.