Clin Res Cardiol (2021)
DOI DOI https://doi.org/10.1007/s00392-021-01843-w
Patient characteristics influence performance of automated single-lead ECG atrial fibrillation detection
M. Zink1, B. Freedman2, K. Mischke3, A. P. Keszei4, C. Rummey5, N. Marx1
1Med. Klinik I - Kardiologie, Angiologie und Internistische Intensivmedizin, Uniklinik RWTH Aachen, Aachen; 2Heart Research Institute Charles Perkins Centre, University of Sydney, Sydney, AU; 3Medizinische Klinik I, Leopoldina-Krankenhaus Schweinfurt, Schweinfurt; 4Uniklinik RWTH Aachen, Aachen; 5Clinical Data Science GmbH, Basel, CH;

Introduction: Automated AF detection by a single-lead electrocardiogram stick can be used for AF screening.

Aim: The aim of this analysis was to determine the accuracy of automated AF detection of a single-lead ECG device and to identify factors associated with diagnostic performance.

Methods: A single-time point screening for AF using a handheld single-lead ECG stick was performed in 6842 customers in community-pharmacies. ECGs were analyzed by a cardiologist blinded to the results, graded according to signal quality, and compared to the automated analysis.

Results: Automated ECG stick analysis showed good prediction of AF with an area under the receiver operating curve of 0.89; sensitivity 80%; specificity 98%; positive predictive value 71%; negative predictive value 99%. ECG signal quality was highly associated with correct measurement, while low signal quality leads to incorrect measurements in many cases. In a multivariate model (Table) we determined factors associated with excellent signal quality and incorrect automatic AF identification. The Odds’ ratio (OR) for correct analysis was strongly associated with female sex (OR 1.92; 95%CI 1.52-2.41; P<0.001), lower age [years] (step per year OR 0.97; 95%CI 0.96-0.98; P<0.001), lower height [cm] (OR 0.98; 95%CI 0.97-0.99; P=0.003), and higher body weight index [kg/cm2] (OR 1.04; 95%CI 1.03-1.06; P<0.001). 

Conclusion: Based on our data, there is a significant effect of sex, age, height and body mass index on the accuracy of automated AF-detection by a single-lead ECG device. The potential target population – patient groups with a high prevalence of AF – have a higher probability of incorrect automatic measurement. Thus, an expert over-read of the recorded ECGs is recommended for all algorithm-positive tests. To further improve sensitivity and acceptance of results an over-read for all recorded ECGs would be ideal.

 

 

 

 

 

 

Table Uni- and Multivariate statistics for excellent signal quality and incorrect measurement of automated AF screening. OR – odd’s ratio, CI – confidence interval

 

 

 

Excellent signal quality

Incorrect measurement

Univariate analysis

Multivariate analysis

Univariate analysis

Multivariate analysis

Parameter

OR

95% CI

P

OR

95% CI

P

OR

95% CI

P

OR

95% CI

P

Sex [Female]

2.33

1.97-2.75

<0.001

1.92

1.53-2.41

<0.001

0.64

0.49-0.84

0.001

0.57

0.43-0.76

<0.001

Age [years]

0.97

0.96-0.98

<0.001

0.97

0.96-0.98

<0.001

1.07

1.04-1.09

<0.001

1.06

1.04-1.09

<0.001

Height [cm]

0.96

0.95-0.97

<0.001

0.98

0.97-0.99

0.003

1.01

0.99-1.03

0.068

Weight [kg]

0.99

0.99-1.00

0.418

0.99

0.98-0.99

0.003

BMI [kg/cm2]

1.04

1.03-1.06

<0.001

1.04

1.03-1.06

<0.001

0.91

0.88-0.95

<0.001

0.91

0.87-0.94

<0.001

CHADSVASC 

1

0.95-1.06

0.912

1.06

0.97-1.17

0.205

Heart failure

0.62

0.41-0.93

0.022

1.86

1.13-3.05

0.015

Hypertension

0.96

0.83-1.11

0.58

1.06

0.80-1.39

0.689

Diabetes mellitus

0.85

0.68-1.07

0.159

0.82

0.54-1.25

0.359

Stroke / TIA

0.82

0.66-1.01

0.066

1.19

0.83-1.69

0.341

Vascular disease

0.89

0.75-1.07

0.213

1.31

0.98-1.77

0.70
https://dgk.org/kongress_programme/jt2021/aP600.html