P3749Discrimination between normal and abnormal electrocardiograms: agreement of seven built-in automatic diagnostic programs in a cohort of >2000 12-lead traces

Abstract

Abstract Introduction 12-lead ECG is a standard evaluation for any patient admitted to a clinic but also for population screenings program and athlete periodical evaluation. Definition of a normal vs. abnormal ECG is a hard task and carefully trained physicians are needed to avoid inappropriate second level evaluations driven from a claim for ECG abnormality. Almost all current 1ECG recorders provide automatic diagnosis through built-in automatic-diagnostic computer programs (ACP). However, we have limited data comparing different ACP's in discriminating between normal and abnormal ECGs. Aim To assess the agreement of the main world-wide available automatic diagnostic programs implemented in current ECG recorders in discriminating between “normal” vs. “abnormal” ECGs in a large dataset of real-world ECGs. Methods We assessed seven ECG interpretation programs from seven different manufacturers (GE 12SL, Glasgow, MEANS, Midmark, Mortara VERITAS, Philips DXL and Schiller). We created a large set of representative ECGs converted from previously recorded digital ECGs acquired with equipment that complied with the requirements of International Electrotechnical Commission standard IEC 60601–2-51:2003 and were representative of those in hospital settings. We decided to exclude ECGs from pacemaker carriers. We used a specific device for playing back ECGs to 12-lead ECG recorders in appropriately setting to avoid interferences. Each statement from automatic diagnosis provided by each device was recorded and combined appropriately for the purpose of this analysis, identifying three group of ECGs: abnormal/substantially abnormal (ABN), normal/substantially normal (NRM) and borderline. Results 2155 ECGs of 10s duration were analyzed by the 7 different ACPs: 513 from a pediatric population and 1642 from patients >16 years old consecutively collected mainly in hospital settings. Figure 1 evidences the prevalence of normal to abnormal grading according to each ACP in both groups of ECGs. Focusing in adult group we found that a NRM diagnosis was reported in a range of 129 (7.9%) to 478 (29.1%) among 1642 adult ECGs. On the contrary, ABN statement was reported in a range of 774 (47.1%) to 1271 (77.4%). Notably, agreement between the 7 ACPs was present in 36 ECGs (2.2%) for NRM diagnosis, while the agreement for ABN diagnosis was present in 661 (40.3%) of the ECGs. We performed a sensitivity analysis by repeating the same calculation after taking out one of the device at turn reaching a maximum of 6.5% for NRM and 41.2%% for ABN diagnosis with 6/6 agreement. Figure 1 Conclusions In our large cohort of almost unselected hospital ECGs the agreement on “normal” and “abnormal” among programs of different manufacturers is rather low. This should be carefully considered when using automatic ACP diagnosis as a screening or priority tool for ECG interpretation. Tailor-made review by physicians is still necessary for both clinical and research purposes.