Abstract:Statistically-based smoothing techniques are described which have been applied to the existing framework of the Glasgow ECG Analysis program. These methods have been designed with the aim of improving repeatability in the computer interpretation of ECGs which have been recorded either several minutes or 24 hours apart from patients in a clinically stable condition. With respect to the ECG diagnosis of Left Ventricular Hypertrophy (LVH), these flexible methods have the effect of reducing the number of inconsistent day-to-day interpretations by 36% from 33 to 21 in330 pairs of ECGs recorded one day apart. Similarly, when comparing agreement in the diagnosis of LVH in 249 pairs of ECGs which were recorded several minutes apart, the number of discordant computer interpretations was 6 using the new methodology, compared with 13 using conventional criteria, i. e. there was a 54% reduction in disagreements.