Atrial fibrillation (AF) is a common irregular heart rhythm associated with a fivefold increase in stroke risk. It is often not recognised as it can occur intermittently and without symptoms. A promising approach to detect AF is to use a handheld electrocardiogram sensor (ECG, measuring heart activity) multiple times a day for 1-4 weeks. However, the ECG recordings must be manually reviewed, which is time-consuming and costly. Our aims were to: (i) evaluate the manual review workload; and (ii) evaluate strategies to reduce the workload.

2,141 older adults were asked to record their ECG four times per day for 1-3 weeks in the SAFER Feasibility Study, producing 162,515 recordings. Patients with AF were identified by: (i) an algorithm identifying recordings exhibiting anything other than a high quality signal and a regular rhythm; (ii) a nurse reviewing recordings to correct algorithm misclassifications; and (iii) two cardiologists independently reviewing recordings from patients with any evidence of irregular heart beats.

A total of 30,165 reviews were anticipated to be required (20,155 by the nurse, and 5,005 by each cardiologist). After cardiologists reviewed a subset of recordings to identify AF patients, 813 recordings were found with AF. The number of reviews would have been reduced to: 25,160 by using only one cardiologist reviewer; 18,144 by using a stricter algorithm to only identify recordings with an irregular rhythm throughout; and 14,946 by using one cardiologist and the stricter algorithm. These changes would reduce the number of reviews per AF recording from 35 to 29, 24, and 19 respectively. The number of AF patients identified would not have fallen considerably: from 54 to 54, 53 and 53.

In conclusion, it may be possible to reduce the manual workload by almost half whilst still identifying a very similar number of patients with undiagnosed, clinically relevant AF.