Eleven studies (n = 2,100) were included, of which five were classed as level I and five as level II. Ten studies (n = 1,614) evaluated an algorithm score of 3 or more. Both the SROC curve and diagnostic odds ratio (odds ratio 9.9; Χ2 13.0, p = 0.16) indicated homogeneity among studies and the Spearman correlation was not significant. Summary sensitivity was 20 per cent (95% confidence interval (CI): 18%, 23%), specificity 98 per cent (95% CI: 97%, 99%), positive likelihood ratio 7.9 (95% CI: 4.5, 13.8) and negative likelihood ratio 0.8 (95% CI: 0.8, 0.9). These results were considered to indicate comparable accuracy to other ECG findings used in practice. Subgroup analysis of level I studies gave similar results.
For an algorithm score of 2 or more (seven studies, n = 1,213), the SROC curve and diagnostic odds ratio indicated significant heterogeneity. Sensitivity ranged from 20 per cent (specificity 93 per cent) to 79 per centy (specificity 100 per cent) and specificity ranged from 61 per cent (sensitivity 44 per cent) to 100 per cent (sensitivity 99 per cent). Results for 14 groups of observers (six studies, n = 1,287) revealed substantial inter-observer agreement (kappa = 0.7).