A new study that reports on the clinical performance of the Cardiolund ECG Parser has been published: Åke Olsson, Symptom-Ruled Real World Arrhythmic Recordings With an Internet Based System, Circulation 2018;138:A12893.
The study involved manual interpretation by a trained cardiologist of 1000 consecutive anonymous printouts of chest- and thumb ECG waveforms recorded using a Coala Heart Monitor. The printouts contained three 10 second strips of ECG at 25 mm/s and included mean heart rate, RR median and any user annotation, but with personal identification and algorithm analysis results removed (blind). The recordings came from actual Coala users in Sweden with no training, control or influence. The manual interpretation was compared with the analysis performed by the Cardiolund ECG Parser algorithm, which is integrated in the Coala system.
The study results:
- Prevalence of atrial fibrillation in the recordings: 14.4 % (143 of 990 recordings)
- Sensitivity for detecting atrial fibrillation: 97,2% (95% CI = 0.930 - 0.992)
- Specificity for detecting atrial fibrillation 94,6% (95% CI = 0.928 - 0.960)
- Negative predictive value for detecting atrial fibrillation 99,5% (95% CI = 0.987 - 0.999)
- Positive predictive value for detecting atrial fibrillation 75,1% (95% CI = 0.683 - 0.812)
- Kappa coefficient 0.818 (95% CI = 0.769 - 0.866)
These results agrees well with ECG Parser performance reported in other studies, such as the STROKESTOP study, see Safe automatic screening for AF using Cardiolund ECG Parser.