Automatic Holter electrocardiogram analysis in ischaemic stroke patients to detect paroxysmal atrial fibrillation: ready to replace physicians?

Abstract

Background and purpose: The detection of paroxysmal atrial fibrillation (pAF) in patients presenting with ischaemic stroke shifts secondary stroke prevention to oral anticoagulation. In order to deal with the time- and resource-consuming manual analysis of prolonged electrocardiogram (ECG)-monitoring data, we investigated the effectiveness of pAF detection with an automated algorithm (AA) in comparison to a manual analysis with software support within the IDEAS study [study analysis (SA)]. Methods: We used the dataset of the prospective IDEAS cohort of patients with acute ischaemic stroke/transient ischaemic attack presenting in sinus rhythm undergoing prolonged 72-h Holter ECG with central adjudication of atrial fibrillation (AF). This adjudicated diagnosis of AF was compared with a commercially available AA. Discordant results with respect to the diagnosis of pAF were resolved by an additional cardiological reference confirmation. Results: Paroxysmal AF was finally diagnosed in 62 patients (5.9%) in the cohort (n = 1043). AA more often diagnosed pAF (n = 60, 5.8%) as compared with SA (n = 47, 4.5%). Due to a high sensitivity (96.8%) and negative predictive value (99.8%), AA was able to identify patients without pAF, whereas abnormal findings in AA required manual review (specificity 96%; positive predictive value 60.6%). SA exhibited a lower sensitivity (75.8%) and negative predictive value (98.5%), and showed a specificity and positive predictive value of 100%. Agreement between the two methods classified by kappa coefficient was moderate (0.591). Conclusion: Automated determination of ‘absence of pAF’ could be used to reduce the manual review workload associated with review of prolonged Holter ECG recordings.